diff --git "a/Dadar.json" "b/Dadar.json" new file mode 100644--- /dev/null +++ "b/Dadar.json" @@ -0,0 +1,149208 @@ +{ + "measurement": [ + { + "key": "model.layers.0.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.021142661571502686, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.020939838141202927, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.011009960435330868, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.010954434052109718, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.010889262892305851, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.007545556873083115, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03587133437395096, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.02072034403681755, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.010905690491199493, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.01086618285626173, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.01103656180202961, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.012844415381550789, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.010848687961697578, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.008263043127954006, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0075215501710772514, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0082784965634346, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007515629753470421, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007302894722670317, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007514655590057373, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.007301408797502518, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0074328589253127575, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007514357101172209, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0071812286041677, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00730051239952445, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.011009960435330868, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.011009960435330868, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.02278142049908638, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.022498682141304016, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.01045851781964302, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.010385480709373951, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.010305531322956085, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.005549488589167595, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03020278923213482, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.022238967940211296, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.010338634252548218, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.010264958254992962, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.010355242528021336, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.011595640331506729, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.010243366472423077, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.00666029192507267, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.005507700610905886, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.006500467658042908, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.005493086762726307, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.005037783179432154, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.00549065787345171, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.005034215282648802, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005113386549055576, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005489745177328587, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0045533436350524426, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005032145418226719, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.01045851781964302, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.01045851781964302, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.059329308569431305, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.04993274807929993, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.039446137845516205, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.033451490104198456, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.026580795645713806, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.01884789764881134, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05039023235440254, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.038410067558288574, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.028147036209702492, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.02214955724775791, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.023377206176519394, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02732221595942974, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.018885686993598938, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.013176311738789082, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.011411351151764393, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013838585466146469, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007352108135819435, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0065230438485741615, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.006572898011654615, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0055047618225216866, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007055348716676235, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006035428959876299, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0039216019213199615, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004274086561053991, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.013176311738789082, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.013176311738789082, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.0234044399112463, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.014719506725668907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.008571576327085495, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.009822294116020203, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.009358822368085384, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.004201150964945555, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.015570491552352905, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.013835351914167404, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.010891251266002655, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.00706202257424593, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0074327075853943825, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.007872577756643295, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.006875296588987112, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.005041754338890314, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.004502446856349707, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.004157891497015953, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0034351281356066465, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0032828859984874725, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0031540975905954838, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0029055410996079445, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0026170569472014904, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0031346732284873724, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0021506352350115776, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0027795457281172276, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.014719506725668907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.014719506725668907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.054336003959178925, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.05128583312034607, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.05029866471886635, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.04614023491740227, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.024213315919041634, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.023304058238863945, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.026925748214125633, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.02492045983672142, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.024497732520103455, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.022351989522576332, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.02152225747704506, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.013653308153152466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.012007252313196659, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.011716498993337154, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.011646251194179058, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0068907649256289005, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0064295632764697075, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.006406490225344896, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.006101508159190416, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.006058343220502138, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.0038880696520209312, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.004431148059666157, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.003790528979152441, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.003550601191818714, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.013653308153152466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.013653308153152466, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.07981983572244644, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.07543476670980453, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.07408834248781204, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.06802623718976974, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.03559218719601631, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.034265898168087006, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.039480071514844894, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.03649414703249931, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.035996634513139725, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.03284517303109169, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.031591422855854034, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.019895363599061966, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.017315562814474106, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0169284138828516, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.016830487176775932, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.009932068176567554, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.008820096962153912, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.00878819264471531, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.008287526667118073, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.008235279470682144, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.005299142561852932, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.005490544252097607, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.005158089101314545, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.0038627199828624725, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.009932068176567554, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.009932068176567554, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.0.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.059150565415620804, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.051768913865089417, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.03075847402215004, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.02872263640165329, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.025077061727643013, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.01575944945216179, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.03679404780268669, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.04704505205154419, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.02635985240340233, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.02275823801755905, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.02259666845202446, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.01647006906569004, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.021438099443912506, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.01646599918603897, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.012687984853982925, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.010903947055339813, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.011160343885421753, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.009305139072239399, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.010907551273703575, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.008943879045546055, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.006447474937886, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01077430322766304, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.004275696352124214, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.00863436609506607, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.012687984853982925, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.012687984853982925, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.013749438337981701, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.011027819477021694, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.008474353700876236, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.007440617308020592, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0058911764062941074, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.003966988530009985, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.010745959356427193, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.008591282181441784, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.006390497088432312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.004902511835098267, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.005219644866883755, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.00566359655931592, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.00425231596454978, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.002959925215691328, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.002534608356654644, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0029170254711061716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0016896759625524282, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.001494365744292736, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0015255478210747242, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.001269925618544221, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0015746194403618574, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.001425333903171122, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0009947959333658218, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0010336959967389703, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.013749438337981701, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.013749438337981701, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.012928780168294907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.010153806768357754, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.007883635349571705, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.006891157943755388, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.005389999132603407, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0037327734753489494, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.008812309242784977, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.00775895407423377, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.005921275354921818, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.004369613714516163, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.004423936828970909, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.004473527893424034, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.003717705374583602, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.002667284570634365, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.002358691766858101, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0022548765409737825, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0015124917263165116, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0013800860615447164, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.001338761649094522, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0011486185248941183, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0012275846675038338, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0012362347915768623, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0008585245232097805, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0009091442334465683, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.012928780168294907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.012928780168294907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.07471964508295059, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.06347496807575226, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.05861077085137367, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05053030699491501, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03332144394516945, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.02860242687165737, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0412747897207737, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03788434714078903, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.034915562719106674, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.027154332026839256, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.02540450356900692, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02107536606490612, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.01817837730050087, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.016087014228105545, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.015562934800982475, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01057194173336029, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.008510490879416466, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.008313678205013275, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007328389212489128, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.006978808436542749, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005587832536548376, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0057409075088799, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004850481636822224, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004023311659693718, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.015562934800982475, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.015562934800982475, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.14653237164020538, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08701067417860031, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.05459456890821457, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06397979706525803, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06291370838880539, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03385240212082863, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08892786502838135, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07943753153085709, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06962589174509048, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04002094641327858, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0426422618329525, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04588712006807327, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.039139524102211, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03183668479323387, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029894452542066574, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.023469066247344017, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018901024013757706, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01831195130944252, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015388867817819118, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014157118275761604, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013605027459561825, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015275383368134499, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.011392737738788128, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012627528980374336, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015388867817819118, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015275383368134499, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.10376954823732376, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.0990917757153511, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.09773150831460953, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.09020891785621643, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.045774854719638824, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.04453372582793236, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.050392232835292816, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.04654310643672943, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0461466945707798, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.04282640665769577, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.04157702252268791, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.025477634742856026, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.022184189409017563, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.021861307322978973, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.02178928256034851, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.012791640125215054, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.011707495898008347, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.011687085963785648, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.011173180304467678, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.011138941161334515, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.007105211727321148, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0076963454484939575, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.00700906477868557, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.005912180058658123, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.012791640125215054, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.012791640125215054, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1280508041381836, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.12263572216033936, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.12108975648880005, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.11217786371707916, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.056674204766750336, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.055194027721881866, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.062250785529613495, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.05753578245639801, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.05709199607372284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.05313339829444885, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.05153237283229828, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.031206397339701653, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.02709297090768814, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.02672751620411873, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.026642700657248497, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.015521103516221046, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.013592596165835857, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.013566441833972931, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.012907606549561024, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01285548321902752, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.008081435225903988, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.007960855960845947, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.007951149716973305, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00507302675396204, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.015521103516221046, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.015521103516221046, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.1.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.1085464134812355, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.09727673977613449, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.08524850010871887, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.07879536598920822, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.053090404719114304, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.037440232932567596, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.0732206180691719, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.0664939284324646, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.054546959698200226, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.04879599064588547, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.046564456075429916, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.04038897156715393, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.03928178548812866, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.024011977016925812, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.021035490557551384, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.019052736461162567, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.013681295327842236, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.012181870639324188, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.01294846460223198, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.01127406395971775, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.011677141301333904, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.011292295530438423, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.00914674811065197, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.008336943574249744, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.013681295327842236, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.013681295327842236, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.030219439417123795, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.027836473658680916, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.02673930674791336, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.02399485744535923, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.013620596379041672, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.012649565003812313, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.016490954905748367, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.014729301445186138, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.01388130895793438, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.012124264612793922, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.011592209339141846, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.008418428711593151, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.007063637021929026, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.00653997203335166, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.006407102104276419, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.004255062434822321, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.003434294369071722, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0033747253473848104, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.00314412172883749, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.003062761388719082, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0022831005044281483, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.002237070119008422, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0020401952788233757, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0015670977300032973, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.014729301445186138, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.014729301445186138, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.02411750890314579, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.02216612920165062, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.02128482796251774, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.01906261406838894, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.010826833546161652, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.010058638639748096, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.012864258140325546, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.011701677925884724, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.011044775135815144, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.009606740437448025, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.009130085818469524, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0064873844385147095, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0055716391652822495, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.005184563342481852, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.005091956816613674, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0032568525057286024, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.002708474639803171, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.00266812345944345, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.002470653969794512, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0024123117327690125, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00173541612457484, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0017404068494215608, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0015924146864563227, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0012031348887830973, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.012864258140325546, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.012864258140325546, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11153673380613327, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10420429706573486, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10186365991830826, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09241075068712234, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.051056962460279465, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.048535965383052826, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.057128049433231354, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05276504158973694, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.051811691373586655, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04600463807582855, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043769802898168564, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02891877479851246, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025092503055930138, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024315033107995987, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02413042075932026, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014424688182771206, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012387150898575783, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012327728793025017, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011360462754964828, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011242873966693878, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007439783774316311, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007323249243199825, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007145291194319725, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0045889294706285, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014424688182771206, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014424688182771206, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1469869166612625, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11692947149276733, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10557082295417786, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08503539115190506, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06534410268068314, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05399661138653755, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08193392306566238, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07426045835018158, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06950372457504272, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0488712415099144, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04478650167584419, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.042453523725271225, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.036460138857364655, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03254874795675278, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03155902028083801, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02184949815273285, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018673639744520187, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018393702805042267, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01590123400092125, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015294034965336323, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012843498960137367, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01410398818552494, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0117145711556077, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.011614379473030567, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015294034965336323, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015294034965336323, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.13921448588371277, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.13293041288852692, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1311313360929489, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.12145477533340454, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.06316579878330231, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.06132059544324875, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.06959617137908936, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.06416822224855423, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.06370473653078079, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.05886060371994972, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.05717746540904045, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.035297147929668427, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.030548488721251488, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.030108574777841568, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.030009159818291664, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.017684422433376312, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.015704788267612457, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.015677986666560173, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.014887051656842232, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.014824679121375084, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.009600327350199223, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.0097741037607193, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.009468022733926773, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.006958102807402611, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.014887051656842232, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.014887051656842232, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17181338369846344, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1643010824918747, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16215869784355164, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15039078891277313, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0781591460108757, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07593553513288498, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08610156178474426, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07937924563884735, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07881759852170944, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07285385578870773, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07068659365177155, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.043394654989242554, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03761911764740944, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.037091709673404694, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.036968886852264404, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021667156368494034, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.018901098519563675, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.018862897530198097, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0178450345993042, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017766930162906647, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011386290192604065, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011092258617281914, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01121925562620163, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.007090485189110041, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011386290192604065, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011386290192604065, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.2.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.16010473668575287, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.1438024938106537, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.13723637163639069, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.12450767308473587, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.07177986949682236, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.06556273996829987, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.08500610291957855, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.07801337540149689, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.07408987730741501, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.06313980370759964, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.060795821249485016, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.04317328333854675, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.03726717829704285, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.034477971494197845, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.0337960422039032, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.02170597016811371, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.01817869208753109, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.01793881505727768, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.016521917656064034, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.016091343015432358, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.011852017603814602, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.011926124803721905, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.010962003841996193, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.008499711751937866, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.011852017603814602, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.011852017603814602, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.04287330061197281, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.03981925919651985, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0383320115506649, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.03460371866822243, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.01958618499338627, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.01827283389866352, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.02363884449005127, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.02108159475028515, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.019915200769901276, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.017552629113197327, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.016858121380209923, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.012127168476581573, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.010134052485227585, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.009408203884959221, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.009223727509379387, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.006133665330708027, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.00492473877966404, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.004839757457375526, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.004529518075287342, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.004415153991430998, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0032868245616555214, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.003183376509696245, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0029463833197951317, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.002207877580076456, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.012127168476581573, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.012127168476581573, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.03450768068432808, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.032028231769800186, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.030800793319940567, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.02779122069478035, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.015734216198325157, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.01470672432333231, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.01854804717004299, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.01695043221116066, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.016005443409085274, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.01408619899302721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.013434773311018944, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.009381916373968124, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.008094241842627525, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.007559302728623152, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.00743148336187005, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.004703462589532137, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.00397183233872056, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.003915846347808838, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0036527730990201235, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0035728365182876587, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.002514715539291501, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.002573698293417692, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.002330936724320054, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0018197811441496015, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.01408619899302721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.01408619899302721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1552315354347229, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14559844136238098, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1424374133348465, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12947170436382294, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07165532559156418, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0682746022939682, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08066421002149582, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07392265647649765, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07265772670507431, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06468116492033005, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06173944100737572, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04085516184568405, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03519866615533829, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03418239951133728, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03392660617828369, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.020399754866957664, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.017411815002560616, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01733240857720375, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01598953828215599, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015837831422686577, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010579499416053295, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01028620358556509, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010181792080402374, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006473226938396692, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010579499416053295, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010579499416053295, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13675054907798767, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11429456621408463, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10379485040903091, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07903659343719482, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06227891147136688, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.052587416023015976, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08352449536323547, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0717543438076973, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0646350160241127, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.046751074492931366, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.042826104909181595, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04287036135792732, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03598091006278992, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.031806837767362595, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0307020116597414, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.022891046479344368, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.019303711131215096, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018889302387833595, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016890916973352432, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016262220218777657, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014369234442710876, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015470285899937153, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012938336469233036, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013345412909984589, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014369234442710876, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014369234442710876, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16956490278244019, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1613760143518448, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1588488221168518, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14631812274456024, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07872775197029114, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07607310265302658, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08681810647249222, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08020635694265366, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07951660454273224, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0726415291428566, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0700804591178894, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.044426094740629196, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.038682129234075546, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0380147248506546, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.037871286273002625, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02228422276675701, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02038579434156418, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.020344745367765427, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019243501126766205, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019145455211400986, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012344086542725563, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013437374494969845, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012143445201218128, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010313124395906925, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012344086542725563, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012344086542725563, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.202090784907341, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19235728681087494, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18950560688972473, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17459286749362946, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09355492144823074, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09045106172561646, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10316252708435059, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09531811624765396, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09449402242898941, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08630596846342087, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08319167047739029, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.052422840148210526, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04544258862733841, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04470690339803696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04451723396778107, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02620484121143818, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023102842271327972, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023061349987983704, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021668659523129463, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021559491753578186, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013948247767984867, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014074762351810932, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013707175850868225, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00963863916695118, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013948247767984867, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013948247767984867, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.3.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.03318595141172409, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.03023177944123745, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.018315667286515236, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.016054820269346237, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.010726246982812881, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.009506909176707268, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.028117161244153976, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.024798493832349777, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.01121361181139946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.008851379156112671, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.008388693444430828, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.006937493570148945, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.005990931764245033, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0053130220621824265, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.005531100556254387, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.0035846130922436714, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0035770810209214687, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.0029205328319221735, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.0032925563864409924, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.002542572794482112, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0027891912031918764, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0029304588679224253, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.001847843755967915, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.0017468706937506795, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.01121361181139946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.01121361181139946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.05790776014328003, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.053388725966215134, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.051542192697525024, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.04635641351342201, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.026406392455101013, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.024685829877853394, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.030496761202812195, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.02809026464819908, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.026924187317490578, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.02346361055970192, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.022306423634290695, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.015432258136570454, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.013377521187067032, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.012635268270969391, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.012453760020434856, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.007710774429142475, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.006534556392580271, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.006464008707553148, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.00595423998311162, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.005839291960000992, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.004040078725665808, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.004076373763382435, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0037904763594269753, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0027333814650774, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.015432258136570454, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.015432258136570454, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.050263457000255585, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.04631085693836212, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.044635478407144547, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.04015309736132622, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.02290390431880951, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.021370185539126396, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.026700017973780632, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.02446962706744671, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.023360298946499825, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.02034299075603485, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.019406389445066452, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.013506499119102955, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.011648801155388355, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.010960262268781662, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.010790841653943062, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.006751623935997486, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.005662614945322275, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.005592792760580778, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.00515505438670516, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.005048341117799282, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.003533918410539627, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0035306415520608425, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.003302720608189702, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0023482334800064564, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.013506499119102955, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.013506499119102955, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.16403502225875854, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15330363810062408, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14975735545158386, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.13582953810691833, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0759282335639, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07212687283754349, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.08507917076349258, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.07849186658859253, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.07709431648254395, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06821555644273758, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0648195818066597, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04308934509754181, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.037389133125543594, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03621766343712807, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03593987971544266, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.021505847573280334, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.018444392830133438, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.018348317593336105, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01686505600810051, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016687888652086258, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011092493310570717, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01088648196309805, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010661118663847446, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.006799729540944099, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011092493310570717, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011092493310570717, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1771346926689148, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15617875754833221, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1475946456193924, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11875736713409424, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08258672803640366, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07420355081558228, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09900571405887604, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08988474309444427, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08510972559452057, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06614314764738083, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05821774899959564, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05109618604183197, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.043954845517873764, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04063599556684494, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03983663022518158, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02598416991531849, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022447651252150536, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022160159423947334, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01943969912827015, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01892738603055477, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014857269823551178, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015901004895567894, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013859588652849197, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012560012750327587, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014857269823551178, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014857269823551178, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17145487666130066, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1625717282295227, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1599205732345581, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14676178991794586, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0796261727809906, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07667405903339386, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08776376396417618, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08122159540653229, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08051083981990814, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07294349372386932, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06989219784736633, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04453711584210396, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.038728952407836914, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.038008272647857666, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03783830627799034, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.022225894033908844, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.01954836957156658, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01950404979288578, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.018210984766483307, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.018102670088410378, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011670373380184174, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011761248111724854, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011430987156927586, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.007843973115086555, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011670373380184174, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011670373380184174, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21010041236877441, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19934719800949097, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.19611437618732452, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18014535307884216, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0975685641169548, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0939563512802124, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.107455313205719, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0994720384478569, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09862612932920456, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08940544724464417, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08564507961273193, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05442718416452408, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0473305843770504, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04646702855825424, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.046256035566329956, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.027119485661387444, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023609401658177376, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023555558174848557, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021937919780611992, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.021806642413139343, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014005081728100777, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013744523748755455, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013714604079723358, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00855645164847374, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014005081728100777, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014005081728100777, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.4.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2028440237045288, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.18134987354278564, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1732012778520584, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1537920981645584, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09229366481304169, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08420698344707489, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1077827662229538, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.09918859601020813, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.09509645402431488, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.07938877493143082, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07494426518678665, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.05483534187078476, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.047414328902959824, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.044313739985227585, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04356233775615692, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.027480533346533775, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.023189179599285126, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02292986772954464, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.0206769909709692, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.020175041630864143, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.014813981018960476, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.014898160472512245, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.013839036226272583, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.010371360927820206, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.014813981018960476, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.014813981018960476, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.06620289385318756, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.061059948056936264, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.05883551016449928, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05284591019153595, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.030277805402874947, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.028226830065250397, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0350399911403656, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.032366227358579636, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.030871042981743813, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.02685805968940258, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.02545277215540409, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.017753111198544502, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.015451068989932537, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.01451953686773777, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.014294641092419624, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.00887822825461626, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007566077634692192, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007474391255527735, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.006892933510243893, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.006751342210918665, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.004685243591666222, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0048079900443553925, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0043762014247477055, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003311406821012497, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.015451068989932537, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.015451068989932537, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.05651655048131943, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.05206291750073433, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.05003992095589638, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.04498325660824776, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.02584335021674633, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.024051405489444733, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03034755028784275, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.027814239263534546, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.026363808661699295, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.022935330867767334, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.021883685141801834, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.015362922102212906, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.013260964304208755, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.012393392622470856, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.012183674611151218, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.007684895768761635, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.006435086019337177, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.006343618966639042, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0058612702414393425, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.005725291091948748, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0040414175018668175, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0040724799036979675, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.003753840457648039, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0027557567227631807, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.015362922102212906, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.015362922102212906, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.17554409801959991, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16390550136566162, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1600421965122223, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1448434740304947, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08141086995601654, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07726603746414185, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09154224395751953, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.08434540033340454, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08268812298774719, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07299734652042389, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06922921538352966, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.04641064256429672, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04019628092646599, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.038885194808244705, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03856542333960533, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02314716950058937, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.019796811044216156, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01969541236758232, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01806977577507496, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.017866840586066246, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0119528379291296, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.011710569262504578, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.011461074464023113, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007333521265536547, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0119528379291296, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0119528379291296, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18164081871509552, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15734359622001648, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14892999827861786, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12213047593832016, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08413688093423843, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07504945993423462, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09964723885059357, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09043912589550018, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08698073029518127, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06738607585430145, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05960662662982941, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05135239288210869, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04415710270404816, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04124924913048744, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04058201238512993, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026153836399316788, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022805742919445038, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02260526828467846, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019788939505815506, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.019359499216079712, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014974523335695267, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016107887029647827, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014089441858232021, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012821095064282417, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014974523335695267, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014974523335695267, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16524630784988403, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15630164742469788, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15361729264259338, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14071837067604065, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07700292766094208, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07398233562707901, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08514704555273056, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07875529676675797, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07793337106704712, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07027451694011688, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06728172302246094, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04330980405211449, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.037651803344488144, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.036874257028102875, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.036694858223199844, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021645989269018173, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.01915283128619194, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.019106131047010422, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.017814934253692627, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017708783969283104, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011518552899360657, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011806806549429893, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011277337558567524, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008202334865927696, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011518552899360657, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011518552899360657, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21740595996379852, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20572452247142792, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20231053233146667, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18524932861328125, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10125711560249329, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0973210409283638, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11186102777719498, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10342365503311157, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10244480520486832, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09241407364606857, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0883590430021286, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.056711968034505844, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04928293824195862, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0482921302318573, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04805964231491089, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028302941471338272, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024546485394239426, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024486016482114792, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0227296594530344, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022584784775972366, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014682176522910595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014330334030091763, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014347490854561329, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008936863392591476, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014682176522910595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014682176522910595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.5.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.20839691162109375, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.18605414032936096, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.17742136120796204, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.15729376673698425, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09493361413478851, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08639904856681824, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1114407554268837, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.10248725861310959, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.09789835661649704, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08157806843519211, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07691333442926407, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.056831590831279755, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0490790531039238, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04571069777011871, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.0448884591460228, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.028554623946547508, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.024085093289613724, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02379979006946087, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.021510589867830276, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.020962461829185486, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01559504121541977, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.015724387019872665, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014539326541125774, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011182633228600025, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01559504121541977, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01559504121541977, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08217410743236542, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07655660063028336, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0743413046002388, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06724544614553452, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03796424716711044, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.035806991159915924, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04322861507534981, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03993489220738411, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.038583576679229736, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03403520584106445, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.032312020659446716, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.021926341578364372, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.019091229885816574, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.018208859488368034, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.01800130121409893, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010969571769237518, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0094757666811347, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009393097832798958, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008695902302861214, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008565579541027546, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005786660127341747, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005945772398263216, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005493988282978535, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004097263794392347, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010969571769237518, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010969571769237518, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.06709007918834686, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.06256610155105591, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0606621615588665, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05489445850253105, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.031013889238238335, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.02919299528002739, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.03577041253447533, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03281622752547264, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.031507182866334915, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.02783474512398243, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.026579443365335464, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.01815369911491871, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.015666618943214417, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014862860552966595, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.014669420197606087, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009080236777663231, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007686643395572901, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007607220206409693, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007050391752272844, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.006927970796823502, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.004758815746754408, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.004774894565343857, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004488012287765741, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0031977221369743347, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014862860552966595, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014862860552966595, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19136665761470795, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1793876588344574, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1755158007144928, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1594138741493225, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08909548819065094, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08479941636323929, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0996512845158577, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09192963689565659, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09032373130321503, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08023399859666824, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07620598375797272, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05052808299660683, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.043790336698293686, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04251207411289215, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04220359027385712, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.025184668600559235, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02160150557756424, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021503105759620667, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01978042908012867, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.019579516723752022, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012950983829796314, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01267844159156084, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012471376918256283, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00784915778785944, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012950983829796314, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012950983829796314, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1787324994802475, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1549779623746872, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14552347362041473, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12151403725147247, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08293335884809494, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07313454151153564, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10008793324232101, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09103737771511078, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08552979677915573, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06490203738212585, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06108904257416725, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05159416049718857, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04431791976094246, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04064910486340523, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.039775025099515915, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026007220149040222, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022343771532177925, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022006111219525337, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019049042835831642, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.018465086817741394, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014626748859882355, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015782130882143974, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013498677872121334, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012330802157521248, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014626748859882355, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014626748859882355, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1620061993598938, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15301769971847534, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15027841925621033, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13739782571792603, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07564721256494522, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07256622612476349, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08360707014799118, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07743897289037704, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07659178227186203, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06883829087018967, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0657208189368248, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04247787967324257, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03697681054472923, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03616180270910263, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03598254919052124, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021203387528657913, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.018588818609714508, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.018534570932388306, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01720951497554779, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017092274501919746, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01109648123383522, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011181860230863094, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01083169411867857, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.0074093337170779705, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01109648123383522, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01109648123383522, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22024545073509216, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20812250673770905, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20447036623954773, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1870582103729248, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10285068303346634, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.098679319024086, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.113711416721344, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10519963502883911, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10412998497486115, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09359598159790039, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0894194170832634, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0577080212533474, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05014042928814888, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04906545951962471, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04882029443979263, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02878820151090622, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024962596595287323, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02489352598786354, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023060429841279984, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022899432107806206, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01494569331407547, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01460457406938076, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014584317803382874, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00912573840469122, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01494569331407547, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01494569331407547, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.6.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.21313993632793427, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19079597294330597, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1820962131023407, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16168059408664703, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09728305786848068, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08872295171022415, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.11424291878938675, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.10499918460845947, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10028673708438873, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08379308879375458, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07913151383399963, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.05817977711558342, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05028887093067169, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04682933911681175, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.045985147356987, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.029205363243818283, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.024667706340551376, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.024371985346078873, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02206612005829811, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.021508604288101196, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01586003415286541, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0160884577780962, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014763521030545235, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011435589753091335, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014763521030545235, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014763521030545235, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08073227852582932, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07530633360147476, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07312541455030441, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06611112505197525, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.037293750792741776, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03522225841879845, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04254970699548721, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03925569728016853, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.03789189085364342, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.033439721912145615, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03176737204194069, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.021568449214100838, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.018737776204943657, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.017871834337711334, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.017658378928899765, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010783442296087742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009228279814124107, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009143069386482239, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008451094850897789, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008318371139466763, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005652627907693386, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005698298569768667, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005355119239538908, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003805025015026331, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010783442296087742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.010783442296087742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.06719447672367096, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.0627356767654419, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.06076928973197937, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.05497640371322632, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03108184039592743, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.029260648414492607, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.035777464509010315, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.032977137714624405, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.031582269817590714, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.027882881462574005, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.026566307991743088, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.018134618178009987, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.015731537714600563, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014882729388773441, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.014681820757687092, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009062389843165874, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.007681982591748238, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.007596186362206936, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.007040384225547314, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.006914486642926931, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.004741458222270012, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00475799897685647, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.004459603223949671, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003155155573040247, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014882729388773441, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.014882729388773441, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18882369995117188, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.17701928317546844, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17314141988754272, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15702660381793976, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08790645003318787, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08368292450904846, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09865841269493103, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09080586582422256, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.089169442653656, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07911421358585358, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07518874108791351, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05008986219763756, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0432765819132328, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.041975706815719604, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04166487604379654, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.024972105398774147, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02133096382021904, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02123222127556801, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019518213346600533, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.019322926178574562, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012856317684054375, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.012531374581158161, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012340538203716278, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.007752280682325363, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012856317684054375, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.012856317684054375, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2011307328939438, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16807715594768524, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.15678279101848602, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1289709508419037, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09170123189687729, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07933132350444794, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.109882652759552, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10005206614732742, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09577181190252304, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0709734559059143, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06501491367816925, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0566389337182045, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0489388108253479, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.045155975967645645, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.044240809977054596, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02878667786717415, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025231454521417618, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024986444041132927, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02158963680267334, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021013658493757248, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016540413722395897, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018197976052761078, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01540962141007185, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014655051752924919, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01540962141007185, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01540962141007185, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.15140500664710999, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.14296339452266693, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.14034968614578247, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.12828919291496277, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07084289193153381, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.06788413971662521, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.07855360209941864, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0726199746131897, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07171504944562912, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06439696997404099, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.061527349054813385, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03994014114141464, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03473003953695297, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03393281623721123, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.033741295337677, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.019981026649475098, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.017545649781823158, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01749151013791561, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.016253458335995674, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.016135483980178833, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010596212930977345, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01072295568883419, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.010331596247851849, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.0073058088310062885, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010596212930977345, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010596212930977345, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21790827810764313, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20586590468883514, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2021743506193161, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18493995070457458, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10207448899745941, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09792476892471313, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11306525021791458, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10449965298175812, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10334590077400208, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09284669160842896, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0887785330414772, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.057515859603881836, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.049958180636167526, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04885834828019142, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0486048199236393, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028731394559144974, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02524021826684475, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02516983263194561, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023375390097498894, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02322101593017578, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015157720074057579, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015345992520451546, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01480168104171753, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010388539172708988, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015157720074057579, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015157720074057579, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.7.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.20222841203212738, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.180251806974411, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1717403531074524, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.15243856608867645, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09212897717952728, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08375764638185501, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1092754676938057, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.0996684655547142, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.09505395591259003, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.07923810929059982, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07480589300394058, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.05547351390123367, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.04792598634958267, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04458877444267273, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04376932233572006, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.028067972511053085, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02391824685037136, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.023624958470463753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02149026095867157, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.020952550694346428, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01570120081305504, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.016124224290251732, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01465616375207901, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01201304979622364, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01465616375207901, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01465616375207901, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.091714046895504, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08571990579366684, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08328723162412643, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.075371652841568, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.042405202984809875, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04009192809462547, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04828733578324318, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04455902427434921, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.043064530938863754, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03809633105993271, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.036258820444345474, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.024468999356031418, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.021247267723083496, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.020301563665270805, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.020071182399988174, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012234543450176716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010480868630111217, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010390055365860462, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009612633846700191, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009471739642322063, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006413259077817202, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0064596147276461124, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006093267817050219, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004312924575060606, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012234543450176716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012234543450176716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.07418617606163025, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.0693117156624794, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0672277882695198, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.060879409313201904, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03433344513177872, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03236956521868706, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.039398614317178726, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03632459044456482, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0348564088344574, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03085022233426571, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.02940441109240055, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.01998019963502884, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.017335228621959686, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.016438277438282967, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.016230348497629166, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009985437616705894, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.00848501268774271, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.008394780568778515, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.00778614217415452, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.007652296219021082, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005234950687736273, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005249131470918655, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00494465883821249, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0034842819441109896, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009985437616705894, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.009985437616705894, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.20047980546951294, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18804821372032166, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18396221101284027, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16700807213783264, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09336916357278824, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08894231915473938, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10496827214956284, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09638077020645142, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09469199925661087, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08412241190671921, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08011581748723984, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05330429971218109, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.045939087867736816, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04456577077507973, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.044247888028621674, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02658041939139366, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022656382992863655, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022549986839294434, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020753299817442894, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.020547328516840935, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013699725270271301, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01330533716827631, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01313673984259367, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00825249683111906, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013699725270271301, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013699725270271301, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18996559083461761, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16774404048919678, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.160415917634964, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.13651616871356964, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08829710632562637, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07994595915079117, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10187266767024994, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09354391694068909, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09080905467271805, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0729069709777832, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0656907707452774, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.052741888910532, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04564258083701134, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04327944666147232, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04270029067993164, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026807475835084915, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02372516132891178, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02356577478349209, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020892828702926636, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02054653689265251, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015338867902755737, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01648728735744953, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01464625634253025, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013029171153903008, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015338867902755737, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015338867902755737, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.15662670135498047, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.14780019223690033, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.14506831765174866, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13255061209201813, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0734463781118393, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07036874443292618, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08145078271627426, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07534117251634598, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07436175644397736, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06675571203231812, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06374640017747879, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04157351702451706, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03611774370074272, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.035273030400276184, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03507960960268974, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.020783929154276848, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.01840142533183098, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01834462210536003, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01707989163696766, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0169551819562912, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.0111033134162426, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011481043882668018, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.010826393030583858, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008101249113678932, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.0111033134162426, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.0111033134162426, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21923862397670746, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20709167420864105, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2033410668373108, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1858355551958084, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1025485098361969, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09831200540065765, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11388988792896271, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10503605008125305, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10381564497947693, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09323704242706299, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0890943855047226, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05779585987329483, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05012999475002289, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04900159686803818, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.048735663294792175, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02890254557132721, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02501809597015381, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024941593408584595, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02310306765139103, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02293536812067032, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015156754292547703, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014781307429075241, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014782429672777653, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009426407516002655, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015156754292547703, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015156754292547703, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.8.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.21480470895767212, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19161206483840942, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18251588940620422, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1618397831916809, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09822896122932434, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.0891827642917633, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.11598771810531616, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.10626215487718582, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10126250237226486, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08422048389911652, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07938553392887115, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.059080664068460464, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05111163854598999, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.047518134117126465, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04662951081991196, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.029780607670545578, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02538882941007614, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02507811412215233, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.022731037810444832, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022156503051519394, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.016422703862190247, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017014101147651672, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015288923867046833, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012576676905155182, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015288923867046833, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015288923867046833, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.097457155585289, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.0910518541932106, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08853727579116821, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08019024133682251, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04522405564785004, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.042754095047712326, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05145503953099251, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.047453925013542175, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.045899685472249985, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.040662284940481186, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03871383145451546, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02618204429745674, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022784996777772903, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.021772466599941254, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021527128294110298, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013123322278261185, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011442155577242374, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011343722231686115, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010546725243330002, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010400386527180672, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007008129730820656, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00733191380277276, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00666667427867651, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005237936973571777, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013123322278261185, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013123322278261185, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08232472091913223, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07699882984161377, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07472875714302063, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06773263961076736, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0381934717297554, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.036064088344573975, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04371536523103714, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.040318362414836884, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0387716218829155, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03436722233891487, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03279104828834534, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.022197505459189415, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.019248904660344124, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.018303953111171722, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.018077146261930466, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011099018156528473, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009467190131545067, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009373882785439491, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008700194768607616, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008557597175240517, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0058318548835814, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005873026791960001, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005518003832548857, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003944140858948231, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011099018156528473, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011099018156528473, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19996285438537598, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18770088255405426, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18356560170650482, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16673117876052856, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09324940294027328, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08880820125341415, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10489240288734436, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09630094468593597, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.094558484852314, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08406826108694077, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08004121482372284, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.053274840116500854, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.045926354825496674, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04454878717660904, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04420491307973862, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026572715491056442, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022692011669278145, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0225819144397974, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020808083936572075, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02059645764529705, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013742639683187008, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.013403050601482391, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013189814984798431, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.008420297876000404, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013742639683187008, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013742639683187008, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2016715407371521, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1732465624809265, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1620025783777237, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.13028287887573242, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09266113489866257, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0815221518278122, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11176305264234543, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10238257795572281, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09608825296163559, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07325532287359238, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0643545538187027, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05788201466202736, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05015166848897934, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0458366759121418, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04477350413799286, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02945820428431034, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025801243260502815, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02544596418738365, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02240097150206566, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021751251071691513, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017062779515981674, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018874000757932663, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01577812246978283, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015327316708862782, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015327316708862782, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015327316708862782, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1634591668844223, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15428030490875244, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15136834979057312, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13832131028175354, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0766574814915657, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07340628653764725, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08499667048454285, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07865890115499496, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07760971784591675, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06964904069900513, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06645341217517853, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04332125931978226, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03768070787191391, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03677830472588539, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.036570217460393906, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02164665050804615, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.019126368686556816, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01905778795480728, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.017736516892910004, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01759985089302063, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01148331817239523, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011843577958643436, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011187233962118626, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008247969672083855, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01148331817239523, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01148331817239523, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22267186641693115, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21028006076812744, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20643514394760132, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1886356621980667, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10424645990133286, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09991898387670517, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11555458605289459, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1068103238940239, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10553941130638123, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09472360461950302, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09034696966409683, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.058633603155612946, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05095555633306503, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04979483410716057, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04953061044216156, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02926700934767723, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0253891721367836, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025308648124337196, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023437172174453735, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02325885184109211, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015208457596600056, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014957782812416553, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014811670407652855, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009465225972235203, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015208457596600056, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014957782812416553, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.9.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22569230198860168, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20178551971912384, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19241617619991302, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17034222185611725, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10356053709983826, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.0941825658082962, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12174684554338455, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11183721572160721, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10675264149904251, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0887855738401413, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08375426381826401, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06239055469632149, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05395042523741722, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0502278134226799, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04932064563035965, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03147714212536812, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.027049317955970764, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026722414419054985, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.0242727380245924, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023682793602347374, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017506582662463188, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.018359532579779625, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016319790855050087, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013837689533829689, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013837689533829689, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013837689533829689, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10456246137619019, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09777526557445526, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0951140969991684, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08610859513282776, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04849035292863846, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04587553068995476, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05499295890331268, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05077260732650757, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04920423775911331, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04354242980480194, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0413818396627903, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027923576533794403, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024276841431856155, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023249097168445587, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02300448901951313, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013965451158583164, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01207034569233656, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01197777409106493, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0110931983217597, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010940546169877052, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007359105162322521, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007525892928242683, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00701010599732399, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005156621336936951, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013965451158583164, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013965451158583164, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08441966772079468, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.0789380595088005, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07659907639026642, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06933913379907608, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03909188136458397, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.036911141127347946, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04474934563040733, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04130029305815697, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.03969525173306465, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03514521196484566, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03348029777407646, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02269558236002922, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.019717417657375336, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.01873750425875187, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.018498899415135384, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01134201418608427, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009684219025075436, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009583861567080021, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008890202268958092, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.00874413549900055, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005954031366854906, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006005138158798218, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005636605434119701, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004013675265014172, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01134201418608427, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01134201418608427, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.20021645724773407, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18767903745174408, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18335981667041779, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1663431078195572, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09325656294822693, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08866076916456223, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10488861054182053, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0966673493385315, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09457036852836609, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08396996557712555, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07972899824380875, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05318000540137291, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.046094901859760284, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04453667253255844, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04416414350271225, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026542728766798973, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022727951407432556, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022595679387450218, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020829251036047935, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02058897726237774, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013735915534198284, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.013515020720660686, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013170458376407623, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00853533111512661, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013735915534198284, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013735915534198284, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2130201756954193, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1927948147058487, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1845562756061554, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1588420271873474, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09982931613922119, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0917690247297287, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11718827486038208, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1073451042175293, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10204091668128967, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08509305864572525, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07715042680501938, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06040044501423836, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05236869677901268, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04891752451658249, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04808653146028519, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03057227097451687, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02685742825269699, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026525719091296196, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024138882756233215, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023624958470463753, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017211169004440308, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018790792673826218, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016173357143998146, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01477669645100832, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01477669645100832, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01477669645100832, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.15612506866455078, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.147150456905365, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.14427974820137024, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13169646263122559, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07327581197023392, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07011618465185165, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08133818954229355, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0752510353922844, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07421476393938065, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06645115464925766, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0634230226278305, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.041561152786016464, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.036130521446466446, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.035241901874542236, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03503415361046791, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.020798400044441223, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.018467983230948448, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0184085201472044, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.017130419611930847, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017006203532218933, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011179227381944656, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011647327803075314, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.010893377475440502, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008354898542165756, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011179227381944656, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011179227381944656, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21917614340782166, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2067241221666336, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20289312303066254, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18520943820476532, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10254974663257599, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09818735718727112, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11386127024888992, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10510538518428802, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10381831228733063, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09301745891571045, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08869629353284836, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05766249820590019, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05015106871724129, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04898601397871971, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04871656373143196, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02883944660425186, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024991454556584358, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02491188608109951, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023035481572151184, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02285849116742611, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015056464821100235, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014739925041794777, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0146636376157403, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00935562513768673, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015056464821100235, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015056464821100235, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.10.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22719088196754456, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2040921300649643, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1952897608280182, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17343071103096008, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10424100607633591, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.0954694151878357, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12128248810768127, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11166062951087952, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10723093897104263, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08984574675559998, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08459922671318054, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06198975071310997, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05365573987364769, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.050315406173467636, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04951252043247223, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031244339421391487, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02672392502427101, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026447037234902382, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023972705006599426, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023441361263394356, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0173043180257082, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0176501777023077, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016301218420267105, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012902220711112022, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012902220711112022, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012902220711112022, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10317052155733109, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09647532552480698, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09385068714618683, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08489885926246643, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04786273464560509, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04528749734163284, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05440358445048332, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05020473152399063, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04858320951461792, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04299869015812874, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04087061434984207, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027592703700065613, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02397807501256466, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.022922566160559654, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02267318218946457, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01379353553056717, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011846345849335194, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011742942035198212, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010866900905966759, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010708536952733994, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007235962897539139, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007311348803341389, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00687790010124445, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004898599348962307, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01379353553056717, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01379353553056717, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08486348390579224, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07943452149629593, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07702988386154175, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06976167112588882, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03942243009805679, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03718354552984238, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04509251192212105, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0416564904153347, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.039990879595279694, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.035425279289484024, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03371651470661163, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.022861801087856293, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.019877519458532333, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.01886560581624508, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.018620280548930168, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011419462971389294, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009722895920276642, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009617677889764309, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008918370120227337, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008764058351516724, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00597565621137619, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005985206458717585, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00565147353336215, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003942972514778376, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011419462971389294, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.011419462971389294, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2027350813150406, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19005250930786133, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18557608127593994, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16817961633205414, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0944676324725151, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08977337926626205, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.106478750705719, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0980154424905777, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09582360833883286, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.084998220205307, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08063694089651108, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05399081110954285, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04674318805336952, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.045158445835113525, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04477272182703018, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026942040771245956, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02304038405418396, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0229030754417181, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02109646238386631, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02085319720208645, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013944404199719429, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.013704154640436172, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013355541974306107, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.008658566512167454, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013944404199719429, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013944404199719429, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21146996319293976, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18817578256130219, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17777882516384125, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15170468389987946, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0975104570388794, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08821087330579758, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11825411021709442, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10860395431518555, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10142068564891815, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08245459198951721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0762142613530159, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06109611690044403, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05277968570590019, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.047736916691064835, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.046486347913742065, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030721111223101616, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026132728904485703, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025689981877803802, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023352721706032753, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02258995547890663, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016973575577139854, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018456000834703445, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015401508659124374, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014224136248230934, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015401508659124374, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015401508659124374, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.15879836678504944, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.14963199198246002, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.14665189385414124, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1336161345243454, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07456298172473907, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07129788398742676, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08294078707695007, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07667744159698486, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0755319818854332, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06753024458885193, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0643680989742279, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04233188554644585, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03684118017554283, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.035893648862838745, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.035669904202222824, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021189860999584198, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.018853820860385895, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.018785784021019936, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.017477218061685562, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017345335334539413, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01139116007834673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011960258707404137, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011082664132118225, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008645087480545044, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01139116007834673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01139116007834673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2181585133075714, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20558732748031616, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20167388021945953, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1837933510541916, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10209137946367264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0976664200425148, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11337652802467346, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10476523637771606, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10339032113552094, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09244144707918167, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08796358108520508, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05749540403485298, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04996280372142792, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04876408353447914, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04847585782408714, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028686007484793663, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024865666404366493, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024779440835118294, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0228851567953825, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022707190364599228, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014864407479763031, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014652595855295658, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014448934234678745, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009262170642614365, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014864407479763031, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014864407479763031, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.11.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2324669361114502, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20954862236976624, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.20085959136486053, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17764629423618317, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10691794008016586, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09810172766447067, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12448678910732269, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11445441097021103, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1099424660205841, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09228745102882385, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08670790493488312, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06365051865577698, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05502577871084213, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05157751590013504, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05074825882911682, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03210793808102608, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.0273366030305624, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.027047783136367798, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02452278882265091, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023986293002963066, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017754442989826202, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017969956621527672, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016695253551006317, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013063684105873108, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013063684105873108, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013063684105873108, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10988983511924744, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10274941474199295, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09995619207620621, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09034251421689987, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05103498324751854, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0482792966067791, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05782800912857056, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05342589318752289, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05179275572299957, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.045793429017066956, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04346473887562752, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02934250235557556, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025493722409009933, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024421216920018196, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024164145812392235, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014657648280262947, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012558842077851295, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01245777029544115, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01149783842265606, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011336615309119225, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007655104156583548, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007658498827368021, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0072897057980299, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0050259800627827644, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014657648280262947, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014657648280262947, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08959811925888062, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08381426334381104, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08133213222026825, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07360076904296875, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04160456359386444, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03928445652127266, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.047534361481666565, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04392601177096367, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04222919046878815, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03739919140934944, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03556361421942711, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.024062521755695343, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.020965255796909332, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.01991993561387062, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.01966690830886364, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012022741138935089, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01024826243519783, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01014114823192358, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009399134665727615, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.00924014300107956, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006280533969402313, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006289571989327669, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005947064608335495, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0041192579083144665, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012022741138935089, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012022741138935089, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.20804600417613983, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19486567378044128, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1901821494102478, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17218516767024994, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09700745344161987, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09207575023174286, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10941784083843231, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10072116553783417, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09838749468326569, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08714845776557922, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08270911127328873, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05550448223948479, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.048116397112607956, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04638633877038956, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04596993699669838, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.027733957394957542, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02373935654759407, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.023581501096487045, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.021733703091740608, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021473616361618042, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014446237124502659, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014242156408727169, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013828086666762829, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009148433804512024, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014446237124502659, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014446237124502659, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21827208995819092, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1968671679496765, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1889481246471405, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1644047349691391, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10187831521034241, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0936807245016098, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11808528006076813, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10823516547679901, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1044069156050682, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08672747015953064, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07970711588859558, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06069467216730118, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05250738933682442, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04964025318622589, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04896295443177223, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030524926260113716, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026894675567746162, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026666829362511635, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024063263088464737, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023638857528567314, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01688924990594387, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018317563459277153, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016011955216526985, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014122401364147663, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014122401364147663, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014122401364147663, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16102415323257446, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15169018507003784, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1486644595861435, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13539601862430573, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07566724717617035, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07234277576208115, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08414128422737122, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0778115764260292, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07666425406932831, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0685066431760788, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06535353511571884, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04298108071088791, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03740821033716202, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03645041957497597, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03622030094265938, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021524932235479355, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.019185462966561317, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.019116120412945747, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.017782429233193398, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017648721113801003, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011600875295698643, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012219484895467758, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011291594244539738, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008885226212441921, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011600875295698643, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011600875295698643, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21957607567310333, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20684489607810974, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20283764600753784, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18476074934005737, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10281416028738022, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0983111709356308, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11419866234064102, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10553118586540222, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1041545569896698, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09303887188434601, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0885753408074379, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05799293518066406, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05034077540040016, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04910966381430626, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04882263392210007, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028948256745934486, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025015288963913918, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02492741495370865, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023004457354545593, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022821100428700447, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015024464577436447, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014708382077515125, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01459454745054245, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00924922525882721, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015024464577436447, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015024464577436447, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.12.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2337893545627594, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2107868492603302, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.20214027166366577, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17935842275619507, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10778568685054779, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09893730282783508, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12528246641159058, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11501210182905197, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.11070045083761215, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0930987149477005, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08753260970115662, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06416881084442139, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.055389758199453354, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05210355669260025, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.051301855593919754, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03246599808335304, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02775191143155098, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.027481788769364357, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.024948135018348694, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.024435538798570633, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.018131688237190247, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01838480681180954, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01712695136666298, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013577812351286411, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013577812351286411, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013577812351286411, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11732746660709381, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10968134552240372, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1067919060587883, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09664180874824524, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.054509032517671585, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.051646724343299866, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06176506727933884, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05698643624782562, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.055336449295282364, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04896508529782295, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04652116447687149, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03135935962200165, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02727196179330349, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02614922821521759, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02588701993227005, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01567789912223816, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013565224595367908, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013458590023219585, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012453758157789707, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012286990880966187, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00825477298349142, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008426121436059475, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007877957075834274, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005748092662543058, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013565224595367908, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013565224595367908, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09238270670175552, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08637935668230057, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08369576930999756, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07574307173490524, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04287010058760643, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04040028154850006, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.049166202545166016, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04542592540383339, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.043492041528224945, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03851982578635216, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03665067255496979, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.024927325546741486, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.021685540676116943, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.020532632246613503, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02025495283305645, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012452910654246807, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010587324388325214, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010464196093380451, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009713509120047092, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009537800215184689, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006525248754769564, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00654389476403594, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006163462996482849, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004323048982769251, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012452910654246807, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012452910654246807, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21713247895240784, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20342424511909485, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19867894053459167, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1800878942012787, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10134866088628769, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09626772999763489, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1141698807477951, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10504263639450073, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10278903692960739, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09111838042736053, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0864708423614502, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.057955071330070496, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05017274245619774, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.048462867736816406, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04805499315261841, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.028962310403585434, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.024792812764644623, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024644536897540092, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.022708119824528694, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02244287170469761, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015065006911754608, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014853623695671558, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01443386822938919, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009517856873571873, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015065006911754608, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015065006911754608, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21786442399024963, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.192824125289917, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1847137212753296, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15779276192188263, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10152340680360794, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0921952947974205, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11769231408834457, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10749158263206482, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10460374504327774, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08464328199625015, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07718328386545181, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06081802770495415, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0524309016764164, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.049734245985746384, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04909549653530121, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03071933053433895, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027149638161063194, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026966385543346405, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024013735353946686, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02360803447663784, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017226161435246468, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018707897514104843, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01641552522778511, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014662347733974457, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014662347733974457, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014662347733974457, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16005507111549377, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15070560574531555, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.14767789840698242, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1345244199037552, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07526747137308121, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07194554805755615, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.0836828351020813, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07739096134901047, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07624755054712296, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06812415271997452, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06484334170818329, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.042728353291749954, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03720936179161072, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03625916317105293, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03603590279817581, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021393023431301117, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0190554391592741, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.018992986530065536, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01766159012913704, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017524689435958862, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011512446217238903, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012115320190787315, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01120498776435852, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008764440193772316, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011512446217238903, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011512446217238903, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21983495354652405, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20707957446575165, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2031526267528534, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18503133952617645, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10294507443904877, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09845445305109024, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11437168717384338, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10562584549188614, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10428935289382935, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09315858781337738, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0887407660484314, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05806954205036163, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05039861798286438, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0491897277534008, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0488969162106514, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029012365266680717, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025064636021852493, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024983234703540802, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02304520271718502, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022864744067192078, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015087698586285114, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014741746708750725, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014677049592137337, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009274113923311234, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015087698586285114, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015087698586285114, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.13.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2328951358795166, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20977844297885895, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.20091979205608368, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17836375534534454, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10702081024646759, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09811806678771973, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12521584331989288, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11471126973628998, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.11007998883724213, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09245080500841141, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08711569756269455, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06377121806144714, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0550612136721611, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05162854120135307, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05078880861401558, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03202136605978012, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.027314860373735428, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.027018949389457703, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.024496544152498245, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023944782093167305, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01743587665259838, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01789381355047226, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01633925922214985, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012926709838211536, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012926709838211536, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012926709838211536, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11956329643726349, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11191877722740173, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10906265676021576, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09878288954496384, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055612072348594666, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05271562933921814, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06283247470855713, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05804668366909027, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.056396618485450745, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05000974237918854, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.047502465546131134, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.031893808394670486, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027713559567928314, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026612216606736183, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026352960616350174, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01592310704290867, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013699600473046303, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013596845790743828, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012569119222462177, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012403803877532482, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008313233032822609, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008358842693269253, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007934220135211945, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00551474466919899, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013699600473046303, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013699600473046303, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09309856593608856, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08717434108257294, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0846034586429596, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07667101919651031, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04327094554901123, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04086245596408844, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.04945650324225426, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.045733559876680374, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04390089586377144, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.038933515548706055, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03704453632235527, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.025034358724951744, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02180701494216919, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.020710021257400513, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0204347912222147, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012505811639130116, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01063506118953228, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010519234463572502, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009757641702890396, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009590418078005314, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006512661464512348, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00649936031550169, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006160159595310688, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004219836555421352, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012505811639130116, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012505811639130116, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2197917103767395, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2062726765871048, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20130757987499237, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1826842874288559, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10278373211622238, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09765598922967911, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11610852926969528, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1068153828382492, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10421384125947952, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09255972504615784, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08799208700656891, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05892232060432434, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05104988440871239, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04915963113307953, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.048702653497457504, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02943081595003605, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025115886703133583, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024940915405750275, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023014796897768974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02272709831595421, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015313960611820221, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015004013665020466, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014643726870417595, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009515909478068352, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015313960611820221, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015004013665020466, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22023378312587738, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20000043511390686, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.19362592697143555, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.16818328201770782, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10249608010053635, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09518302977085114, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11689970642328262, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10702075809240341, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10468123108148575, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08780093491077423, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08103397488594055, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06041662022471428, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05235864967107773, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.050376445055007935, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04988838732242584, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03086557425558567, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027751198038458824, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027623482048511505, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025043027475476265, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024745821952819824, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01790439523756504, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019302161410450935, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017325423657894135, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015462849289178848, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015462849289178848, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015462849289178848, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16490136086940765, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15529687702655792, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15222947299480438, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13866066932678223, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07752715051174164, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0740756243467331, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08611328154802322, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07970274984836578, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07852388173341751, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07016313076019287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06680941581726074, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04397917911410332, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.038261447101831436, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.037284206598997116, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03705749660730362, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021982869133353233, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.019507670775055885, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.019433677196502686, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0180522333830595, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0179084874689579, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011746808886528015, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012260045856237411, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011423935182392597, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008722114376723766, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011746808886528015, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011746808886528015, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22018608450889587, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20745576918125153, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2035139501094818, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18545807898044586, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10318895429372787, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09869951754808426, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1144658699631691, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10587696731090546, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10453344136476517, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0934406965970993, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08888907730579376, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05811358243227005, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05051054060459137, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.049298934638500214, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.049004927277565, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028992297127842903, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025098133832216263, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025018062442541122, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02308046817779541, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02290288731455803, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015007861889898777, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014741125516593456, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014595448970794678, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00925415474921465, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015007861889898777, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015007861889898777, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.14.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2331739217042923, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20862062275409698, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19923825562000275, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1763341873884201, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10688390582799911, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09742972999811172, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1251458376646042, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11511756479740143, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.11018414795398712, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09185176342725754, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08630117774009705, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06390448659658432, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.055230334401130676, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051518071442842484, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.050597891211509705, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03219187259674072, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.027158742770552635, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02684972994029522, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.024246487766504288, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023644903674721718, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017598172649741173, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01772647723555565, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016439350321888924, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012640233151614666, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012640233151614666, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012640233151614666, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11526329815387726, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.107967309653759, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10521118342876434, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09547494351863861, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05364985018968582, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05087431147694588, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06064083054661751, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05600494518876076, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0544128380715847, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04827507212758064, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.045965634286403656, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030792327597737312, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026744652539491653, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.025676913559436798, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025423932820558548, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015391232445836067, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013196209445595741, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013094599358737469, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012111585587263107, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011952158063650131, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008037170395255089, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008027929812669754, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007681000046432018, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0052626547403633595, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015391232445836067, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015391232445836067, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09271707385778427, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08687126636505127, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08435720205307007, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07648380845785141, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04311025142669678, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04073205590248108, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.049147509038448334, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04551149904727936, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04372774064540863, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03880561515688896, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03693993389606476, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.024899205192923546, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02171177603304386, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.020643077790737152, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.020383670926094055, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012436691671609879, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010614241473376751, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01050649769604206, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009746711701154709, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009583170525729656, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006498660426586866, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00651022931560874, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006160431541502476, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0042541129514575005, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012436691671609879, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012436691671609879, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22484490275382996, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21105685830116272, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20637977123260498, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18732789158821106, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10515642911195755, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10007105022668839, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11840745806694031, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1089724525809288, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10665254294872284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09484308212995529, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09016985446214676, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.060107987374067307, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.052057985216379166, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.050298914313316345, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04988177493214607, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03002685308456421, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02567972242832184, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025527574121952057, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023553814738988876, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02328106202185154, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01558724045753479, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015299682505428791, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014940041117370129, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009684796445071697, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01558724045753479, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015299682505428791, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21009773015975952, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1893259584903717, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18032610416412354, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15859133005142212, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09810363501310349, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0892157107591629, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11707118153572083, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10685831308364868, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1006987914443016, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08388473093509674, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07826372981071472, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.060154665261507034, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05158510059118271, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04759917035698891, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04662175104022026, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03014077991247177, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02533423714339733, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024953197687864304, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02257908321917057, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02198183164000511, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016214735805988312, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016924381256103516, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014909092336893082, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012358146719634533, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014909092336893082, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014909092336893082, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16390615701675415, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.154362291097641, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15129482746124268, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13783089816570282, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07710260152816772, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07371299713850021, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.0856536403298378, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07920189946889877, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07811220735311508, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06977846473455429, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06645028293132782, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.043734047561883926, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03807413578033447, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03710928186774254, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.036885153502225876, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0218916367739439, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.01947370357811451, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01940802112221718, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.0180371031165123, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017899157479405403, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011756232008337975, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012317119166254997, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011447152122855186, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008858060464262962, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011756232008337975, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011756232008337975, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22072160243988037, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20780396461486816, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20381556451320648, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18574929237365723, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10341985523700714, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09890161454677582, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11475417762994766, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10606563091278076, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1047423779964447, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09359163045883179, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08902592211961746, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05823222175240517, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.050610847771167755, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.049408022314310074, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0491250604391098, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02909461036324501, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02519032545387745, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025112349539995193, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02317005954682827, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022986551746726036, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015140308067202568, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014839914627373219, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014733792282640934, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009379872120916843, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015140308067202568, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015140308067202568, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.15.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2325097620487213, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20836219191551208, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19907712936401367, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1764025092124939, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1067085936665535, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09735620766878128, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1250370442867279, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11488907784223557, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10998683422803879, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09180055558681488, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08650021255016327, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06389191001653671, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05510886013507843, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051375553011894226, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.05047930032014847, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03217354789376259, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02703440934419632, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026714850217103958, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02414114400744438, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02353435754776001, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017628075554966927, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017572959885001183, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016459207981824875, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012423846870660782, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012423846870660782, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012423846870660782, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1218518614768982, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11424717307090759, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1114024743437767, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10110047459602356, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05682007223367691, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05397096648812294, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06411692500114441, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05916299670934677, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.057622358202934265, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.051227349787950516, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.048773545771837234, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03261547163128853, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028294211253523827, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027239566668868065, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026985034346580505, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016315773129463196, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014085567556321621, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01398973073810339, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012961422093212605, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012802965939044952, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008575042709708214, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008678635582327843, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00821690447628498, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005859851837158203, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014085567556321621, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014085567556321621, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09674753993749619, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09068109095096588, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08813092112541199, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07995234429836273, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.045046012848615646, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.042610809206962585, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05120226368308067, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.047384537756443024, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.045684173703193665, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04058392345905304, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03862893581390381, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.025963926687836647, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022623896598815918, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02155657857656479, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021301494911313057, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01297134067863226, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01108875684440136, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010977967642247677, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010187292471528053, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010026331059634686, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006781911943107843, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006781033705919981, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006446167826652527, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0044375150464475155, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01297134067863226, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01297134067863226, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23536475002765656, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.221400648355484, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2165663242340088, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1969345510005951, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11037105321884155, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10516510158777237, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12376037985086441, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11407263576984406, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11184178292751312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09962984919548035, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09478434920310974, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06289169937372208, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.054488398134708405, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.052758507430553436, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05235015228390694, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03138053044676781, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02689700946211815, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026748647913336754, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024695493280887604, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02442692220211029, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016242152079939842, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015929967164993286, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015608692541718483, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010019022971391678, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015608692541718483, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015608692541718483, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2146003097295761, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19038225710391998, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18249791860580444, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15350309014320374, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09894023835659027, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08995507657527924, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1152300164103508, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10493699461221695, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10204386711120605, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08080001920461655, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07335027307271957, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05932265520095825, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05113137140870094, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0484406016767025, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04778224229812622, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030232228338718414, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02658938243985176, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02642061933875084, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023253027349710464, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022857720032334328, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01728052832186222, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.018480712547898293, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016490021720528603, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014627157710492611, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014627157710492611, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.014627157710492611, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1614486128091812, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15208281576633453, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.14911577105522156, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13598467409610748, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07585880160331726, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07253158837556839, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08429743349552155, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07794266194105148, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07686646282672882, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06871313601732254, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06551343947649002, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04297863692045212, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03741279989480972, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03646747022867203, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.036246463656425476, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021502016112208366, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.01908293552696705, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01902022212743759, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01767074130475521, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017534440383315086, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011501346714794636, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011998089030385017, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011195859871804714, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008550404570996761, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011501346714794636, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011501346714794636, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21791291236877441, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2054225504398346, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2015647143125534, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18374380469322205, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10209748148918152, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09771154075860977, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11327753216028214, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10470039397478104, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10342603921890259, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09247623383998871, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08812712877988815, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05756595730781555, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04996493458747864, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04878552258014679, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04850493371486664, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02875630557537079, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02487112767994404, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024795683100819588, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022891735658049583, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02271563932299614, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014992862939834595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014654891565442085, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014596696943044662, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009267251007258892, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014992862939834595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014992862939834595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.16.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22575663030147552, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20136506855487823, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19157862663269043, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16941803693771362, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10331787168979645, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09365814924240112, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12226693332195282, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11219093948602676, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10670483857393265, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08844597637653351, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08336219936609268, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06246786192059517, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.053782690316438675, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.049747541546821594, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04877358675003052, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031441401690244675, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026191474869847298, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02584085613489151, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02329917810857296, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022656334564089775, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01717558689415455, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017100822180509567, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01588607206940651, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012085446156561375, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012085446156561375, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012085446156561375, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11863381415605545, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11144153773784637, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10869687050580978, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0987510085105896, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055271122604608536, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05252788960933685, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06245896592736244, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.057606495916843414, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.056020911782979965, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04994640871882439, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04765330255031586, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03174055740237236, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027531765401363373, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026463789865374565, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02621360495686531, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01585904136300087, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01360171940177679, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013501115143299103, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012524861842393875, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012367073446512222, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008281851187348366, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008266258984804153, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007928559556603432, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005423617083579302, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01360171940177679, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01360171940177679, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.0948738306760788, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08908629417419434, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08650548756122589, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07857156544923782, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.044140394777059555, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04177345708012581, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.050306107848882675, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.046552520245313644, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04475481063127518, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03985022008419037, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.037987612187862396, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.025499749928712845, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022210322320461273, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02112901397049427, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.020870449021458626, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012745918706059456, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010880131274461746, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010764900594949722, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010012405924499035, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009849694557487965, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006669242400676012, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006673474796116352, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006331778597086668, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004380489699542522, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012745918706059456, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012745918706059456, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22943666577339172, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21609169244766235, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21157203614711761, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19258123636245728, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10749461501836777, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10255926102399826, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1203833669424057, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11094798892736435, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1089048758149147, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09722073376178741, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09262143075466156, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06115426868200302, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05296453461050987, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05136212706565857, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05099218338727951, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03051793947815895, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02616170048713684, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0260335560888052, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024056900292634964, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023814130574464798, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01577828638255596, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015447953715920448, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015183521434664726, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009677964262664318, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015447953715920448, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015183521434664726, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21430464088916779, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1909434050321579, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18341532349586487, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15089735388755798, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10033378005027771, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09207728505134583, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11496128141880035, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10545824468135834, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10274717956781387, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08176206052303314, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07134168595075607, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05921497941017151, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.051244717091321945, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04895223304629326, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.048398613929748535, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029835132881999016, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02645067125558853, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02629290707409382, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02295857109129429, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022609084844589233, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01656571961939335, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017877886071801186, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01588614284992218, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013758767396211624, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013758767396211624, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013758767396211624, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16817474365234375, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1585509479045868, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1555594503879547, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14187780022621155, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07906081527471542, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07565069943666458, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08775165677070618, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.081142358481884, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08007246255874634, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07165145128965378, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06835117191076279, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.044707659631967545, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03891313821077347, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03797835484147072, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03776269033551216, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.022401364520192146, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.019800158217549324, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01973685994744301, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.018329061567783356, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01818971149623394, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011980942450463772, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012334081344306469, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011682012118399143, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008668314665555954, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011980942450463772, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011980942450463772, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22538641095161438, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21261276304721832, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20870742201805115, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19044163823127747, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10564397275447845, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10117338597774506, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11725731939077377, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10828520357608795, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10700739175081253, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09577878564596176, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09135383367538452, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05950024351477623, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.051686763763427734, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05048713833093643, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05021422728896141, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029791558161377907, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02577093429863453, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025692351162433624, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023743774741888046, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023564646020531654, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01564842090010643, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01520655956119299, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015257911756634712, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009675450623035431, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01564842090010643, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01520655956119299, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.17.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.229674831032753, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2040916085243225, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19381290674209595, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17120911180973053, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10516653209924698, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09495057910680771, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12410863488912582, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11439771205186844, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10865157842636108, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08966439217329025, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08442167192697525, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06332170218229294, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05505693703889847, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.050788819789886475, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.049735717475414276, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031853437423706055, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02695629373192787, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026567768305540085, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023974180221557617, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023284895345568657, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01714414730668068, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01789231039583683, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015751194208860397, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012918990105390549, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012918990105390549, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012918990105390549, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11929770559072495, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11219025403261185, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10944517701864243, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09959162026643753, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055673133581876755, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.052934788167476654, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06285827606916428, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05797616392374039, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05641641467809677, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05033959448337555, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.048052772879600525, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.031934142112731934, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027705341577529907, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026655135676264763, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026402993127703667, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015961719676852226, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013725418597459793, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01362534612417221, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012647482566535473, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0124895591288805, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008364703506231308, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00836791843175888, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008019722066819668, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005527167581021786, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013725418597459793, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013725418597459793, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09827547520399094, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09239369630813599, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08990512043237686, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08176378160715103, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.045822873711586, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.043452441692352295, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05202934890985489, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04808623343706131, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.046426594257354736, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04140152409672737, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03950624167919159, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02637816034257412, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02294905111193657, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.021919013932347298, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02166924998164177, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013178892433643341, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011256352998316288, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011154220439493656, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010365763679146767, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010211859829723835, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006878571584820747, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006854130886495113, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006553004961460829, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004458482377231121, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013178892433643341, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013178892433643341, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2327301800251007, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21924027800559998, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21481317281723022, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19565704464912415, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1090347021818161, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1041683778166771, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12181441485881805, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11229456961154938, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11045902967453003, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09868761152029037, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09407089650630951, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.061833687126636505, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05357838422060013, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05207677558064461, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05171623453497887, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030840769410133362, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026455461978912354, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026338253170251846, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024326415732502937, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024094335734844208, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015882249921560287, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01550137996673584, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015318810939788818, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009578724391758442, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01550137996673584, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015318810939788818, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2241036742925644, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.19306664168834686, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.18299412727355957, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.15270037949085236, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10393228381872177, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09199348092079163, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12053027004003525, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11056652665138245, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10737114399671555, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.08309124410152435, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07362592965364456, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.062196243554353714, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05396429821848869, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05096805840730667, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05026174709200859, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031602248549461365, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028073187917470932, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02787749283015728, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024287909269332886, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023844385519623756, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01810535416007042, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.019637465476989746, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.017225775867700577, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015606832690536976, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015606832690536976, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.015606832690536976, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17452964186668396, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.16459199786186218, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16146817803382874, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14733463525772095, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08203282207250595, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07850024849176407, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09085635840892792, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08413945138454437, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0830819383263588, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07437913119792938, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07089733332395554, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.046312373131513596, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04032173752784729, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03936489671468735, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.039142414927482605, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.023125797510147095, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.020436592400074005, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.020372534170746803, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.018906572833657265, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.018765242770314217, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012216451577842236, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012610022909939289, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011908037588000298, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008729871362447739, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012216451577842236, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012216451577842236, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22913062572479248, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2161797434091568, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21222369372844696, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1936916559934616, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10741913318634033, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10293152183294296, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11883817613124847, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11006175726652145, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10880892723798752, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09746832400560379, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09280562400817871, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06032632663846016, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0525064691901207, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.051315855234861374, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.051029305905103683, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.030087530612945557, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.026124509051442146, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02604502998292446, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02406701073050499, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02388683892786503, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015543855726718903, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015321687795221806, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015142902731895447, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009607849642634392, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015543855726718903, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015142902731895447, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.18.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23282597959041595, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2063758224248886, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19559097290039062, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17254436016082764, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10641995072364807, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09582389891147614, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12647928297519684, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11624164134263992, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1100715920329094, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09047282487154007, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0851692408323288, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06432832032442093, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05566512420773506, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051263824105262756, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.050167061388492584, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.032247524708509445, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026940232142806053, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026540957391262054, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02383190020918846, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02310209907591343, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017309201881289482, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017585787922143936, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01588541828095913, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012305778451263905, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012305778451263905, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012305778451263905, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11826781928539276, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11114151775836945, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10837125033140182, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09855827689170837, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05512615665793419, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.052388932555913925, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.062345024198293686, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05751236528158188, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05586191266775131, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.049843549728393555, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.047605834901332855, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03169474005699158, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027496472001075745, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026414738968014717, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026158396154642105, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01586555503308773, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013634268194437027, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01353734266012907, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01257123239338398, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012413831427693367, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008324390277266502, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008370818570256233, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007971400395035744, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005599253810942173, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013634268194437027, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013634268194437027, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.0985196977853775, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09252485632896423, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08996763080358505, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08179312199354172, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04589581862092018, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.043481603264808655, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05231131240725517, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04836048558354378, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04652981087565422, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.041469771414995193, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.039568133652210236, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.026518741622567177, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.023076355457305908, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.021966250613331795, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021702274680137634, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013251131400465965, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011290659196674824, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011173807084560394, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010395858436822891, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010226059705018997, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006923270877450705, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006897428072988987, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006574657279998064, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004493627697229385, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013251131400465965, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013251131400465965, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23297908902168274, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2194424569606781, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.214863121509552, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19591739773750305, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10914289206266403, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10422355681657791, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12205169349908829, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11250956356525421, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11056068539619446, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09878247231245041, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09416079521179199, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06199568882584572, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05367047339677811, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.052124299108982086, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.051770444959402084, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03091302514076233, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026521911844611168, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026403583586215973, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02438502572476864, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024153517559170723, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015938421711325645, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015589901246130466, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015354393050074577, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00969819724559784, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015589901246130466, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015354393050074577, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.20783357322216034, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.18215635418891907, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.17375129461288452, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.14392511546611786, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0963950976729393, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08674530684947968, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11163836717605591, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10201772302389145, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09947226196527481, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07800869643688202, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06866470724344254, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05744494870305061, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04945499822497368, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04692619666457176, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04631795361638069, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02901652082800865, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02523096837103367, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025071974843740463, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.021739087998867035, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021350353956222534, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01622387394309044, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01695188321173191, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01545137632638216, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012867736630141735, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01545137632638216, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01545137632638216, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17822028696537018, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.168178990483284, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16507655382156372, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15071600675582886, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0837734192609787, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0802110955119133, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09299536049365997, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08591847121715546, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08485238254070282, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07600525766611099, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07261441648006439, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04738051816821098, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04120752960443497, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04024173691868782, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04001413285732269, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02374184876680374, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02097717486321926, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.020910007879137993, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019434962421655655, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019291870296001434, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012733843177556992, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013067751191556454, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012420067563652992, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009185193106532097, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012733843177556992, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012733843177556992, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22826401889324188, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21542681753635406, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21147359907627106, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19304701685905457, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10704470425844193, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10253428667783737, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11863832920789719, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10962136089801788, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10837838053703308, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.097093865275383, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09260574728250504, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.060262974351644516, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05233599618077278, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05115655064582825, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05087482929229736, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03014986589550972, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02610253170132637, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02602263353765011, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02405780367553234, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023882072418928146, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015810677781701088, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015387555584311485, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015428343787789345, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009776477701961994, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015387555584311485, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015387555584311485, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.19.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23025958240032196, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20303986966609955, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19158422946929932, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16890014708042145, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10507895797491074, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09402281790971756, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1256009042263031, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11566789448261261, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10889340937137604, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08893170952796936, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0837533175945282, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06397056579589844, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.055363476276397705, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05059757083654404, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04941236227750778, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03203196078538895, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026551879942417145, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026117615401744843, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023395182564854622, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02261168882250786, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017167866230010986, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01734716258943081, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015653446316719055, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012042693793773651, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015653446316719055, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015653446316719055, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1140889897942543, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10724538564682007, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10456347465515137, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0951174646615982, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05313635990023613, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.050496794283390045, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06014399230480194, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.055514857172966, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.053867727518081665, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04806572198867798, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04591457173228264, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030531957745552063, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02649461105465889, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.025418410077691078, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025163188576698303, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01525547169148922, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013048039749264717, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012940289452672005, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012022921815514565, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011858450248837471, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007976582273840904, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007914985530078411, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007624846883118153, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005145109258592129, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01525547169148922, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01525547169148922, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09565360099077225, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08986705541610718, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08734433352947235, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07945568859577179, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04450799152255058, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04214787483215332, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05077829957008362, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04693063348531723, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.045133642852306366, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04024757444858551, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03842679411172867, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02575792372226715, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022403977811336517, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02130916900932789, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021037040278315544, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012868466787040234, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010949249379336834, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010835201479494572, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010085477493703365, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009919623844325542, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006723690312355757, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006695483811199665, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006373236887156963, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004359712358564138, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012868466787040234, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012868466787040234, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22553522884845734, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21233826875686646, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20785099267959595, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18929672241210938, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10537274181842804, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10054770857095718, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11808023601770401, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10890369862318039, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10675108432769775, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09539908915758133, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09097975492477417, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05992536246776581, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05191212520003319, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05032947659492493, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.049944210797548294, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029885681346058846, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02561230957508087, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025478314608335495, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023548245429992676, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023305751383304596, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015437851659953594, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015086371451616287, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014856265857815742, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009402950294315815, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015437851659953594, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015086371451616287, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21236105263233185, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1816667914390564, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.171450674533844, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.14341731369495392, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09804601222276688, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.08649525046348572, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11503831297159195, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10525330901145935, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10182351619005203, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07723347097635269, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.07051139324903488, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05926041677594185, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.050996169447898865, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.047785770148038864, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.046990178525447845, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029985470697283745, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025795582681894302, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025592369958758354, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.021894091740250587, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.021384909749031067, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016878431662917137, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01754879392683506, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015918653458356857, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013361524790525436, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013361524790525436, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013361524790525436, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17942465841770172, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.16937652230262756, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1662169247865677, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1518816351890564, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08424624055624008, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08073072880506516, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09349705278873444, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08639731258153915, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08532790094614029, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07647513598203659, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07301405072212219, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04764527827501297, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.041412971913814545, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.040456466376781464, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04022597894072533, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.023867670446634293, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02103506028652191, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02096867933869362, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019483113661408424, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019342023879289627, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012754262425005436, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013012772426009178, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012443738989531994, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009052099660038948, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012754262425005436, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012754262425005436, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2255493551492691, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2128925323486328, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20910406112670898, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1909853219985962, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10567443072795868, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10128092765808105, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11709191650152206, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10825192183256149, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1070096418261528, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09593993425369263, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09156649559736252, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05943937227129936, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.051677193492650986, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05051806569099426, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0502474345266819, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029700255021452904, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02581534907221794, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025730576366186142, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023813173174858093, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023638131096959114, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015523111447691917, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015280152671039104, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015136713162064552, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009776479564607143, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015523111447691917, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015136713162064552, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.20.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23012210428714752, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2021659016609192, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19038976728916168, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16757218539714813, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10491860657930374, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09364768862724304, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12683536112308502, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11583441495895386, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10887707769870758, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08862534165382385, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08357112854719162, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06448858976364136, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0556950569152832, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05076742544770241, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04953894764184952, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.032432883977890015, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.027066607028245926, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02661939524114132, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023964257910847664, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023162156343460083, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01769528165459633, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01823021098971367, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016121797263622284, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013256838545203209, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013256838545203209, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.013256838545203209, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1167672872543335, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10975220799446106, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10695912688970566, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09737245738506317, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05440245568752289, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05170023813843727, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0615299791097641, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.056774869561195374, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05513281002640724, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04919830337166786, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.047027476131916046, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03126739710569382, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027126701548695564, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02603444829583168, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025772014632821083, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015632281079888344, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013402231968939304, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013302858918905258, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012359032407402992, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012202308513224125, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008200590498745441, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008185043931007385, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007847657427191734, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005409185774624348, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015632281079888344, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015632281079888344, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09649484604597092, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09068259596824646, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0880727469921112, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08014623820781708, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04488298296928406, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04246840253472328, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05141633003950119, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04750138521194458, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04550687596201897, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04060405120253563, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03878308832645416, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.026054752990603447, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022671494632959366, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.021495111286640167, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0212103184312582, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01303551159799099, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011072052642703056, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010945864953100681, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010211716406047344, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010028974153101444, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006840532645583153, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006817750632762909, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006472757551819086, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004479952156543732, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01303551159799099, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01303551159799099, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23505344986915588, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22144868969917297, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21700212359428406, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19776643812656403, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11001387238502502, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10510358959436417, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12295330315828323, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11330755054950714, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11145646870136261, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0996459349989891, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09497039020061493, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06242421269416809, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.054033197462558746, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05251085013151169, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05216594412922859, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031108923256397247, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026687659323215485, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02656319923698902, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024539316073060036, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02430862933397293, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01603776589035988, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015629569068551064, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015467995777726173, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009657450020313263, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015629569068551064, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015467995777726173, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1921083927154541, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16458040475845337, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1557546854019165, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12872812151908875, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08818475902080536, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07840065658092499, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10294634103775024, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09430938959121704, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09155179560184479, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.07030350714921951, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06152147799730301, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05289188399910927, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04614689573645592, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04338369518518448, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04270389676094055, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026889150962233543, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.024065785109996796, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.023912716656923294, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020855367183685303, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.020443866029381752, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015437697060406208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017083274200558662, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014616092666983604, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013728156685829163, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015437697060406208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015437697060406208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17254282534122467, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1629609614610672, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15995225310325623, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14614319801330566, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08105569332838058, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07766713201999664, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08980792760848999, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08307033777236938, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08207815885543823, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07362806051969528, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07023156434297562, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04572661966085434, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.039763715118169785, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03884860500693321, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03863370418548584, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.022846495732665062, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0201072059571743, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.020044049248099327, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.018612058833241463, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.018477164208889008, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012063887901604176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012302517890930176, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011764167807996273, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008401891216635704, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012063887901604176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012063887901604176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.21871379017829895, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2066463977098465, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20290346443653107, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18544863164424896, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10255548357963562, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09827066957950592, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11344324052333832, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10498805344104767, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1038142740726471, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09316244721412659, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08884027600288391, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.057593606412410736, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05007869005203247, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04896854981780052, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04869658499956131, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02872176095843315, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024911053478717804, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.024838820099830627, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022975919768214226, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02280794270336628, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014854569919407368, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014590286649763584, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.014481987804174423, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009126259945333004, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014854569919407368, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014854569919407368, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.21.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22059814631938934, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19280856847763062, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18095842003822327, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.15948189795017242, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10020095109939575, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08889483660459518, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12136876583099365, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11127787828445435, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10417386144399643, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0843401849269867, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0798516720533371, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06188656762242317, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.053370069712400436, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04837343841791153, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04713451862335205, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031122924759984016, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02561257965862751, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025156734511256218, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02256220206618309, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.021738100796937943, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.016961244866251945, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01709326170384884, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015354481525719166, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012160031124949455, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015354481525719166, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015354481525719166, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12035864591598511, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11353747546672821, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.110953189432621, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1012335866689682, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05620183050632477, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05356840416789055, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06311281025409698, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.058395761996507645, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.056930333375930786, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05099955201148987, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04868712276220322, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0320090651512146, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027848731726408005, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02686181664466858, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02662861905992031, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015992147848010063, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013723660260438919, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01363377459347248, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012660132721066475, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012509594671428204, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008322209119796753, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008203382603824139, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00799679197371006, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0052336775697767735, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013723660260438919, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013723660260438919, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10411060601472855, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09813208878040314, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09575202316045761, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08737374097108841, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04861678555607796, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.046282801777124405, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05500536039471626, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05074317380785942, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04926247149705887, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.044114649295806885, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.042199328541755676, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027901824563741684, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024202842265367508, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023234520107507706, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02300763688981533, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01392379030585289, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011877389624714851, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011783725582063198, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010958234779536724, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010810588486492634, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007233100943267345, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007125925738364458, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006917855702340603, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004536100197583437, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01392379030585289, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01392379030585289, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23395349085330963, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2209339141845703, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21671679615974426, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1979547142982483, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10959061235189438, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10491713881492615, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12192854285240173, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11259607970714569, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11100628972053528, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09953315556049347, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09496860206127167, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06186913698911667, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05367538705468178, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05230552703142166, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05197872221469879, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030857430770993233, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02655627578496933, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02645743265748024, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02448202483355999, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.0242724921554327, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015884997323155403, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015513104386627674, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015379313379526138, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00956171564757824, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015513104386627674, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015379313379526138, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2067519575357437, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1685662716627121, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.15535949170589447, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12121881544589996, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09409978985786438, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07954682409763336, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1123976781964302, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1031169444322586, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09899814426898956, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06759953498840332, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.061471231281757355, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05777427554130554, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.049818772822618484, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04572780057787895, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04472823068499565, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029073631390929222, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.024508167058229446, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024243107065558434, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019527804106473923, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.018816908821463585, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016042185947299004, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016585366800427437, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014779685065150261, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01227651722729206, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014779685065150261, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014779685065150261, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.180776447057724, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1708047240972519, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16770173609256744, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1533554494380951, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08492494374513626, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08140409737825394, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09405873715877533, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08699561655521393, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08597135543823242, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07720528542995453, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07369109243154526, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0478944331407547, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0416361503303051, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04069036617875099, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04046986997127533, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.023934975266456604, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02102932706475258, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0209672674536705, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019473135471343994, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01933140493929386, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012629021890461445, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012824445962905884, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012322412803769112, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008704818785190582, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012629021890461445, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012629021890461445, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2251274585723877, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2127707302570343, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20903605222702026, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19118750095367432, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10550235211849213, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10113883763551712, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11700330674648285, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10800231248140335, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10681276768445969, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09595086425542831, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09171248972415924, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05943547934293747, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05154036730527878, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05040278285741806, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05014064162969589, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029717931523919106, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025719907134771347, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025641389191150665, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023758968338370323, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023588942363858223, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01558676641434431, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015166480094194412, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015209239907562733, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00964205153286457, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01558676641434431, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015166480094194412, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.22.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22270214557647705, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19505900144577026, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18247371912002563, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1610974371433258, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1013394445180893, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08962170779705048, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12367186695337296, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11351695656776428, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10539914667606354, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0855879932641983, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08096331357955933, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06289941072463989, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.054481539875268936, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04897378757596016, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04759305715560913, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03155461698770523, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.025966014713048935, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025438200682401657, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.022929832339286804, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022027557715773582, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01703675091266632, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01743421144783497, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01530099380761385, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012391201220452785, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01530099380761385, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01530099380761385, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11825110763311386, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11135740578174591, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10872351378202438, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09912790358066559, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05520705133676529, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0525301918387413, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06214766204357147, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05747060850262642, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05593074485659599, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.050002261996269226, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04773101583123207, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03152981400489807, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.027430718764662743, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026383785530924797, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026136361062526703, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015751365572214127, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013513332232832909, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01341375894844532, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01245624665170908, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012298809364438057, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00821093749254942, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008129989728331566, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007864488288760185, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005229176953434944, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013513332232832909, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013513332232832909, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09912967681884766, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.093365877866745, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09094903618097305, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08294123411178589, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.046273138374090195, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04392436891794205, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05267058685421944, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04858918488025665, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04688301682472229, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.041939277201890945, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04014863073825836, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.026704132556915283, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.023178672417998314, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.022135451436042786, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021883424371480942, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013346475549042225, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01133626140654087, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011231024749577045, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010458837263286114, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010304354131221771, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00695749232545495, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006857678294181824, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006618022918701172, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004403062164783478, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013346475549042225, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013346475549042225, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23267331719398499, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21937373280525208, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21510963141918182, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1964021623134613, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10893981903791428, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1041727066040039, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1213655173778534, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11203712970018387, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11030109226703644, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09879463911056519, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09429065138101578, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06156586483120918, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05341225489974022, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05199846625328064, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05166265368461609, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0307100061327219, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026417819783091545, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0263034887611866, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024326568469405174, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024111421778798103, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01580093987286091, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015450266189873219, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015274327248334885, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00953450333327055, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015450266189873219, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015274327248334885, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18947511911392212, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15043067932128906, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13572180271148682, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11009286344051361, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08607134222984314, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07006927579641342, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10460060089826584, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09610606729984283, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0904378592967987, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06101410835981369, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0573405995965004, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05393369495868683, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04654344916343689, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04202381893992424, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04090655967593193, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02719985321164131, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022814422845840454, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022477589547634125, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01825406774878502, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01747242361307144, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015126476064324379, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015932591632008553, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013766231946647167, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012067340314388275, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015126476064324379, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015126476064324379, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18459069728851318, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1744854748249054, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17132355272769928, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1567678600549698, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08672915399074554, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08314106613397598, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09597855806350708, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08879867196083069, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08778280764818192, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07884585857391357, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07528755068778992, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.048849351704120636, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.042468272149562836, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04153069853782654, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04130478948354721, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02440032735466957, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02143864706158638, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.021377595141530037, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01985197514295578, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019715970382094383, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012857591733336449, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013039613142609596, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012553201988339424, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008814584463834763, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012857591733336449, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012857591733336449, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2267567217350006, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2143452763557434, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2105807363986969, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19267725944519043, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10627245903015137, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10193148255348206, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11750705540180206, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10877732932567596, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10756547003984451, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0966711938381195, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09227393567562103, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05966634675860405, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05186764895915985, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.050734762102365494, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05046854913234711, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029754813760519028, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025798341259360313, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025723233819007874, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02381664514541626, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.0236468818038702, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015400753356516361, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015076644718647003, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015028377994894981, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009403948672115803, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015400753356516361, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015028377994894981, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.23.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2243480086326599, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.1970827728509903, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18446317315101624, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1633087396621704, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1021050363779068, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09054777771234512, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12497635185718536, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11452878266572952, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.106076680123806, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0866403579711914, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08212821185588837, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06357378512620926, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05493459478020668, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04938877001404762, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.047995224595069885, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03229053318500519, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026256268844008446, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025713128969073296, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023282933980226517, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022381041198968887, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.018088318407535553, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017711330205202103, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016436003148555756, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012682637199759483, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012682637199759483, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012682637199759483, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11941072344779968, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11251987516880035, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10987227410078049, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10025784373283386, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05576492100954056, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05309117212891579, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06312566250562668, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0581248477101326, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0564955472946167, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05060107260942459, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0484105683863163, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.031990714371204376, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02775450237095356, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026686575263738632, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02642526850104332, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016019480302929878, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013681338168680668, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013579721562564373, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012628260999917984, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012467843480408192, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008387383073568344, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008254887536168098, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008029954507946968, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005337625276297331, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013681338168680668, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013681338168680668, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10198905318975449, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09608238190412521, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0936894565820694, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08545544743537903, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.047634582966566086, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04530461132526398, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05399448424577713, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04984761402010918, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04826275631785393, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04320890083909035, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04130803421139717, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027375683188438416, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.023781834170222282, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.022782325744628906, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02254890277981758, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013673052191734314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011643961071968079, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011546263471245766, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01073853112757206, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010589521378278732, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0070919468998909, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006991587579250336, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006767278537154198, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00443815253674984, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013673052191734314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013673052191734314, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2286880910396576, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2157660871744156, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2115895301103592, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19314457476139069, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1070861965417862, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10242398828268051, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11929365247488022, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11010856926441193, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10842600464820862, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0971827432513237, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09274128824472427, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.060521673411130905, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05251378193497658, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0511217936873436, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.050800107419490814, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030184365808963776, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025952985510230064, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025853503495454788, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023910557851195335, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023704199120402336, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015524528920650482, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01517638098448515, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015010674484074116, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009358295239508152, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015524528920650482, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015010674484074116, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19536259770393372, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15586614608764648, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1430925875902176, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.12471206486225128, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08708679676055908, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0719870775938034, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1052478775382042, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09479910135269165, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09198594093322754, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0665653795003891, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.06003033369779587, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.053801726549863815, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04636954516172409, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.042936403304338455, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04209145903587341, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02796521969139576, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02390674129128456, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02373833768069744, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.020271899178624153, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.019735250622034073, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016768453642725945, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017123844474554062, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015837235376238823, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013721234165132046, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013721234165132046, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.013721234165132046, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1917884200811386, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18136504292488098, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1781114637851715, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16296032071113586, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09012613445520401, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.0864165872335434, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09981239587068558, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09229744970798492, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09123089164495468, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08193575590848923, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07828444987535477, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05083385482430458, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04413895308971405, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.043163444846868515, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04292905330657959, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025394847616553307, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022240031510591507, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022173887118697166, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02058909460902214, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020437443628907204, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013387427665293217, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01345944032073021, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013067272491753101, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009016760624945164, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013387427665293217, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013387427665293217, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2320726215839386, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.21945294737815857, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21558137238025665, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19725735485553741, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10888072848320007, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10442972183227539, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12057961523532867, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11142566055059433, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11020221561193466, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09904244542121887, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09467953443527222, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06134159490466118, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.053204018622636795, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05204089730978012, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.051774706691503525, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.030660128220915794, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.026585550978779793, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.026503169909119606, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02456168457865715, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02439408376812935, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01607581414282322, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015721525996923447, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01569497585296631, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010062916204333305, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010062916204333305, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010062916204333305, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.24.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2293412834405899, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20255516469478607, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19031964242458344, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16872398555278778, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10445168614387512, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09319508075714111, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12641336023807526, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11645438522100449, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10834752768278122, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08904217183589935, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08436398953199387, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06431566923856735, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05566048249602318, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05025281012058258, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04891365393996239, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.032229259610176086, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026235636323690414, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025714989751577377, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02319810725748539, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02230161614716053, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01722993515431881, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017082374542951584, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015516247600317001, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011553678661584854, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015516247600317001, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015516247600317001, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12164775282144547, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11458469182252884, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11192958056926727, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10217031836509705, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.056830111891031265, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05410263314843178, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06400223076343536, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05914854630827904, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0575840063393116, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0515090748667717, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04920772835612297, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03247366473078728, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028226714581251144, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027172023430466652, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026923349127173424, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016216708347201347, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013916628435254097, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013818192295730114, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012836558744311333, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012673930265009403, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00845189020037651, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008367589674890041, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008109216578304768, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005387276876717806, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013916628435254097, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013916628435254097, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10297940671443939, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09700930118560791, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09455686807632446, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08631760627031326, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04810350015759468, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04571658745408058, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05462462082505226, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05035531148314476, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.048749808222055435, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04363319277763367, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.041747309267520905, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02770073711872101, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024027056992053986, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023016387596726418, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.022766267880797386, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013836762867867947, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011762474663555622, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011662713252007961, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010847017168998718, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010691260918974876, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007185596041381359, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00707236398011446, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006852967664599419, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004494198597967625, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013836762867867947, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013836762867867947, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2342536300420761, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2208966612815857, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21675996482372284, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19797620177268982, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10969207435846329, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10494816303253174, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12205708026885986, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11265797913074493, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11108220368623734, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09949414432048798, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0949486568570137, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06192069873213768, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05371854454278946, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05234774574637413, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05202107131481171, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030885916203260422, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026573041453957558, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02647203952074051, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024464910849928856, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024255875498056412, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01586192287504673, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015493418090045452, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015349914319813251, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009517126716673374, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015493418090045452, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015349914319813251, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.20193921029567719, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15581060945987701, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13877424597740173, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1167660653591156, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08964253216981888, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06838749349117279, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11093482375144958, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10213428735733032, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09649115800857544, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06630268692970276, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05978729575872421, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05698051303625107, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04914366826415062, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04349830746650696, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04208136349916458, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.028675967827439308, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.023142121732234955, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022778118029236794, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.018698859959840775, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01768806017935276, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015631547197699547, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01576581969857216, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013906188309192657, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01123626809567213, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015631547197699547, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015631547197699547, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19226914644241333, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18180809915065765, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17855660617351532, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16339313983917236, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0903877317905426, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08666490018367767, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10016190260648727, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09256777912378311, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09150061011314392, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08220412582159042, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0786694586277008, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05101316049695015, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04430675506591797, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.043321643024683, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04309278354048729, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02552201971411705, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02239302545785904, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022328969091176987, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020739354193210602, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020595161244273186, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013505691662430763, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013654503040015697, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01319210510700941, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009275335818529129, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013505691662430763, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013505691662430763, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2341417521238327, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22135445475578308, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21742375195026398, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19901201128959656, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1099221408367157, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10544315725564957, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12172488123178482, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11250654608011246, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.1112811267375946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09998855739831924, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09564251452684402, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06184717267751694, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.053742069751024246, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05257324129343033, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05229931324720383, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0308931153267622, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.026906747370958328, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.026828650385141373, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02486969716846943, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024696743115782738, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016184670850634575, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015987129881978035, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015798646956682205, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01032953429967165, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01032953429967165, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.01032953429967165, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.25.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22719433903694153, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20131295919418335, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18874098360538483, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16774782538414001, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10337347537279129, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.0922073945403099, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1263229101896286, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11631602793931961, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10729774087667465, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08869867771863937, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08425965905189514, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06410897523164749, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05555756390094757, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04974917322397232, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.048307906836271286, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.0320562943816185, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026001939550042152, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02541317418217659, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023111484944820404, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022149134427309036, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017021428793668747, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017036907374858856, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015150784514844418, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011495277285575867, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015150784514844418, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015150784514844418, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11973077058792114, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11282515525817871, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11009538918733597, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10049764811992645, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055954642593860626, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05326475575566292, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06331001222133636, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05832911282777786, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05671272799372673, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05075039342045784, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0485917367041111, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032126668840646744, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0278506837785244, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02677352912724018, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0265125073492527, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016053644940257072, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013693561777472496, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013586794026196003, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.012624918483197689, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01245800033211708, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00835342239588499, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008210232481360435, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00800199992954731, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005233060568571091, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013693561777472496, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013693561777472496, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10321708023548126, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09731976687908173, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09492046386003494, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0865989699959755, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.048240967094898224, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.045860495418310165, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.055040497332811356, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.050513606518507004, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04888248071074486, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04374375566840172, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04203085973858833, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02789793536067009, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02411067858338356, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02309349924325943, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02285235933959484, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013946225866675377, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011824041604995728, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011723719537258148, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01090103480964899, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01074935495853424, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0072768814861774445, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007110798731446266, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006943459622561932, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0045443144626915455, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013946225866675377, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013946225866675377, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23464836180210114, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22157084941864014, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21723692119121552, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19846321642398834, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11005671322345734, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10527929663658142, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12258961796760559, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11309392750263214, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11146144568920135, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09988705068826675, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09534583985805511, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06226983293890953, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05394429713487625, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05256029963493347, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05223315954208374, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031062958762049675, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02677636221051216, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0266782958060503, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024672605097293854, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024468213319778442, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016050398349761963, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015757840126752853, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015528429299592972, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009891870431602001, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015528429299592972, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015528429299592972, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18578562140464783, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.15427839756011963, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.14201097190380096, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11324711889028549, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08464951813220978, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07118292152881622, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10327041149139404, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09433282911777496, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08892054855823517, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06432301551103592, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05692795664072037, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.053275249898433685, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04590030014514923, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.041574444621801376, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.040499571710824966, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.026891468092799187, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02286890335381031, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02254665456712246, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01919090561568737, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01850571669638157, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015166232362389565, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016276784241199493, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013825410045683384, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01269126869738102, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015166232362389565, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015166232362389565, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19318467378616333, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18259942531585693, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17933763563632965, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.164057657122612, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09085331112146378, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08708642423152924, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10056483000516891, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09306363761425018, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09199082106351852, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08258432149887085, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07889019697904587, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.051223210990428925, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.044507771730422974, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04350971430540085, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04327687621116638, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025585701689124107, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022426821291446686, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02235831692814827, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020754586905241013, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02060719206929207, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013467497192323208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013576536439359188, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013148459605872631, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009096885100007057, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013467497192323208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013467497192323208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23446913063526154, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22166644036769867, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.217748761177063, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19924430549144745, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1099826768040657, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10546373575925827, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1217622309923172, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1126137301325798, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11136814951896667, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.1000451073050499, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09558878093957901, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06181574612855911, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.053730111569166183, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05255019664764404, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.0522671714425087, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.030866187065839767, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02675410732626915, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02667991816997528, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024701561778783798, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024526380002498627, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01604786328971386, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015698816627264023, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015660976991057396, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009859762154519558, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009859762154519558, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009859762154519558, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.26.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22662562131881714, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.2009958177804947, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18873745203018188, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.168113112449646, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1033577099442482, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.0923825278878212, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1258518546819687, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11593905836343765, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1072470024228096, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08894555270671844, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08453544974327087, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06427305936813354, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.055618952959775925, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04988447204232216, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.048461973667144775, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03239501640200615, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026317385956645012, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025756951421499252, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023512795567512512, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022595994174480438, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01754896529018879, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017521966248750687, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015719451010227203, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01225636713206768, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01225636713206768, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01225636713206768, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12925873696804047, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12182336300611496, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11909577995538712, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10869532078504562, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06046822667121887, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05763576924800873, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06784837692975998, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06279627978801727, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06127694621682167, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05482606217265129, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05231916159391403, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034384679049253464, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02994585782289505, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028903363272547722, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028645917773246765, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017158428207039833, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014736163429915905, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014642342925071716, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013568335212767124, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01341375894844532, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008880717679858208, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008753065019845963, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008531732484698296, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005512421950697899, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014736163429915905, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014736163429915905, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1097167432308197, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10339714586734772, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10100352764129639, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09219054132699966, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05134494602680206, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04889284074306488, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05803639069199562, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05350654572248459, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05203389748930931, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.046573590487241745, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.044569533318281174, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029449008405208588, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02552657388150692, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024561507627367973, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024323174729943275, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014705886133015156, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01253464911133051, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012444364838302135, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011556306853890419, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011410810053348541, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007624843157827854, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007477511651813984, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007303727325052023, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004718224983662367, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014705886133015156, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014705886133015156, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23875725269317627, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2253197431564331, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22107014060020447, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20185524225234985, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11189030110836029, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10708945989608765, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12439281493425369, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11488878726959229, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11335739493370056, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10149366408586502, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09691418707370758, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0631529688835144, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.054790470749139786, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05342789366841316, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053106412291526794, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03149190917611122, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027133328840136528, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027033833786845207, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02497508004307747, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02476656436920166, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016213510185480118, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01583394967019558, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015719115734100342, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009769441559910774, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009769441559910774, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009769441559910774, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.17630678415298462, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13591337203979492, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1190049946308136, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09658197313547134, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.07750015705823898, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0618070587515831, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09937594830989838, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09153516590595245, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08413855731487274, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05574783310294151, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.053022369742393494, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05107784643769264, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.043962955474853516, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03759090602397919, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03592517971992493, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02563338354229927, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.019890960305929184, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.019384261220693588, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.015730610117316246, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014534354209899902, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013871630653738976, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.013683722354471684, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01186087355017662, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009425354190170765, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014534354209899902, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.014534354209899902, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19335821270942688, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1827067732810974, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1794157177209854, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16408446431159973, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09096994251012802, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08714224398136139, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10077463835477829, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0931776687502861, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0921066626906395, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08263599872589111, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07886786758899689, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.051315341144800186, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04456127807497978, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.043560516089200974, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.043321263045072556, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02564515918493271, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022402405738830566, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022339623421430588, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020714767277240753, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020566556602716446, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013509055599570274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013508787378668785, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013181773014366627, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008982345461845398, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013509055599570274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013509055599570274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23564352095127106, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22264066338539124, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21871797740459442, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2000848352909088, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11061049997806549, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10599828511476517, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12234195321798325, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11324501037597656, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11197346448898315, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10048612207174301, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09594186395406723, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06216531619429588, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.054025281220674515, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05283027142286301, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.052541207522153854, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.030998073518276215, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.026870176196098328, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.026787638664245605, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024786947295069695, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024599717929959297, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016030337661504745, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015715941786766052, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015636075288057327, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009794099256396294, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015636075288057327, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015636075288057327, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.27.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2258296012878418, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20109206438064575, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1890491247177124, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16858072578907013, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10305501520633698, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09231284260749817, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12534099817276, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.1155516505241394, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10678531229496002, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08887965232133865, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08447764068841934, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06363298743963242, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.055178795009851456, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0495469830930233, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.048136644065380096, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031844764947891235, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.025773227214813232, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02519679069519043, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.022955019026994705, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02202865108847618, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.016861500218510628, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.016708675771951675, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015037667006254196, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011098488233983517, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015037667006254196, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015037667006254196, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12642526626586914, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11908510327339172, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1162932962179184, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10604224354028702, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05919995903968811, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05632304027676582, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06672109663486481, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06163909658789635, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05998027324676514, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.053621627390384674, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05116968974471092, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03385477513074875, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02944258786737919, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028325727209448814, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028057366609573364, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016917314380407333, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014505526050925255, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014398004859685898, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013360155746340752, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01319375354796648, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008820394985377789, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008723986335098743, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008454019203782082, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005606541875749826, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014505526050925255, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014505526050925255, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10716887563467026, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10087031871080399, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09842032939195633, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08981204777956009, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0501590333878994, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04766390845179558, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05680765211582184, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.052382517606019974, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05081220343708992, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.045435186475515366, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04344587028026581, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02881760336458683, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024991489946842194, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02398681454360485, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02373402938246727, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014396358281373978, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01225191354751587, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012153802439570427, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011280319653451443, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011130369268357754, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007472326513379812, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007335019297897816, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007131348364055157, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004643112421035767, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014396358281373978, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014396358281373978, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23384687304496765, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22032169997692108, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21603211760520935, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19727589190006256, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10953739285469055, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10472945123910904, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12195351719856262, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11259293556213379, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11097497493028641, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09930089861154556, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0947045311331749, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06191710755228996, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05367505922913551, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05228760093450546, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.051948364824056625, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.030848905444145203, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026523299515247345, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026425829157233238, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0243963822722435, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.024182718247175217, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015858376398682594, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0154632069170475, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015338366851210594, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009484795853495598, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0154632069170475, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015338366851210594, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1908651441335678, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.13758602738380432, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1158333420753479, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09174464643001556, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08012054860591888, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05679891258478165, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10745342075824738, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09750332683324814, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08997350186109543, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0558989942073822, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05420171096920967, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05450029298663139, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.047328732907772064, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03959563747048378, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03756376728415489, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.027761785313487053, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022056400775909424, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0215877965092659, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01781480759382248, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01644136570394039, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01565464772284031, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016494615003466606, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013353744521737099, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012687545269727707, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01565464772284031, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01565464772284031, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19498980045318604, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18411797285079956, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18077898025512695, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16529572010040283, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09182059019804001, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08794264495372772, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10181804746389389, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09405338764190674, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09295958280563354, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08336102217435837, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0796140655875206, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05184946209192276, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.045044101774692535, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04402221366763115, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04377736523747444, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025991247966885567, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02274230495095253, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.022671958431601524, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021031111478805542, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020878929644823074, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013848412781953812, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013843154534697533, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013511264696717262, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009365294128656387, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013848412781953812, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013848412781953812, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23796366155147552, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22480209171772003, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22077101469039917, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20193256437778473, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11178091168403625, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10714159905910492, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12395095825195312, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11449597030878067, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11321800202131271, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10153433680534363, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0969938263297081, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06296521425247192, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05470640957355499, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.053477976471185684, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05319000408053398, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031517744064331055, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027330175042152405, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.027246424928307533, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02522238716483116, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.025033853948116302, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01656813733279705, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01618698053061962, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016171040013432503, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010371634736657143, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010371634736657143, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010371634736657143, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.28.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23052917420864105, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20645250380039215, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19433681666851044, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17354890704154968, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10562673956155777, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09491699188947678, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12823635339736938, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11824630945920944, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10918724536895752, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0916290283203125, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08706055581569672, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06523310393095016, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05681280046701431, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051049716770648956, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04961446300148964, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.032780200242996216, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02703574113547802, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026452641934156418, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.024322574958205223, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023412657901644707, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017483869567513466, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.018107635900378227, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015608038753271103, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012838700786232948, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015608038753271103, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015608038753271103, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12971478700637817, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12222728878259659, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11951971799135208, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1090998500585556, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.060794733464717865, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05794605240225792, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06815622746944427, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06305409967899323, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06158679723739624, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05506725609302521, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05254749581217766, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03454721346497536, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030078720301389694, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029047654941678047, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02881273254752159, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017238792032003403, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014801019802689552, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014707960188388824, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013616259209811687, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013462129049003124, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00890316441655159, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008765854872763157, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008557828143239021, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005484649445861578, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014801019802689552, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014801019802689552, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1111396923661232, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10471481829881668, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10229069739580154, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09330575913190842, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05206859111785889, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04960139840841293, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.058708544820547104, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.054141685366630554, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05275944992899895, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04717462137341499, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.045095294713974, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029801707714796066, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025836316868662834, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024897610768675804, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02467270940542221, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014878802932798862, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012690994888544083, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012602584436535835, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011683803051710129, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01154533214867115, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0077051459811627865, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007539329119026661, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007389433681964874, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004731907043606043, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014878802932798862, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014878802932798862, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23883609473705292, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2253577709197998, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22104091942310333, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20177516341209412, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1120312288403511, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10715458542108536, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1245182454586029, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11502440273761749, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1134767234325409, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10153622925281525, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09686223417520523, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06317874789237976, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05484973266720772, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.053481973707675934, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053163010627031326, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03151126578450203, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027123427018523216, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02702067419886589, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024932969361543655, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02472977340221405, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01618252694606781, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015759408473968506, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015681294724345207, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009624870494008064, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009624870494008064, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009624870494008064, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2040657103061676, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14721529185771942, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12108694761991501, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0990721732378006, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0859975591301918, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06493185460567474, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11826105415821075, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10875631123781204, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09794741868972778, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.061765361577272415, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.060137614607810974, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.061029884964227676, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05258423462510109, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04224381223320961, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03944741562008858, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0307607501745224, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.023157676681876183, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02240944467484951, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.018788959830999374, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.016884395852684975, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01699678972363472, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01729504019021988, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013880090788006783, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012664670124650002, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013880090788006783, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.013880090788006783, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.19532057642936707, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.18438604474067688, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1810099333524704, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1654449999332428, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.09194069355726242, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08805437386035919, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.10198985040187836, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.0942290872335434, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09312064945697784, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08340922743082047, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07966681569814682, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05200672522187233, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04508465155959129, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04405958205461502, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04380899667739868, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02600834146142006, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02272523008286953, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02265815995633602, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020999446511268616, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020843198522925377, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013782015070319176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013780714012682438, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013454067520797253, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009262030944228172, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013782015070319176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013782015070319176, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2383173704147339, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2250421643257141, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.2209397703409195, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.2019495815038681, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11192820966243744, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10721282660961151, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12391632050275803, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11466703563928604, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11335347592830658, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10159517824649811, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09689400345087051, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06295929104089737, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05472955107688904, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.053478240966796875, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.053190406411886215, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03142053633928299, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02722085826098919, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.0271425973623991, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.025096558034420013, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024904053658246994, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016312329098582268, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015967726707458496, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015903543680906296, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010004359297454357, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010004359297454357, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010004359297454357, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.29.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22661487758159637, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20327551662921906, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1909986138343811, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17050175368785858, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10378743708133698, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09321609139442444, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.128197580575943, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11708611994981766, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10725262016057968, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09020637720823288, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08608072251081467, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06528066843748093, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05609378218650818, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.050075799226760864, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04857994616031647, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.033135831356048584, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026357358321547508, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025709107518196106, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023706790059804916, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022723734378814697, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.018370943143963814, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017523599788546562, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016461536288261414, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012099622748792171, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012099622748792171, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012099622748792171, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.13467630743980408, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12698043882846832, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1241367980837822, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11325322836637497, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06316322088241577, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0601801760494709, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.07093625515699387, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0655471682548523, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06398177891969681, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.057177912443876266, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.054571494460105896, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035988546907901764, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03130142763257027, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03020109049975872, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.029937226325273514, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017980417236685753, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015423227101564407, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015318498946726322, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01419681403785944, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01403226237744093, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.009334664791822433, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.009185084141790867, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008965446613729, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005816007498651743, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015423227101564407, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.015318498946726322, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11521710455417633, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10858093202114105, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10603804886341095, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09674137085676193, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05403821915388107, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05145246163010597, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06099905073642731, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.056208763271570206, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.054756224155426025, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04895181953907013, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.046837761998176575, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030979324132204056, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02682381309568882, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.025855956599116325, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.025615110993385315, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01547978725284338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01320427656173706, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013113181106746197, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0121581656858325, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012009896337985992, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008045338094234467, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007872683927416801, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0077196448110044, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004982744809240103, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01547978725284338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01547978725284338, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24172767996788025, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22794415056705475, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22358033061027527, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20405344665050507, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11337227374315262, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.1085289716720581, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12602603435516357, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1163729876279831, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11486649513244629, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.1027376726269722, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09801220893859863, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06397823244333267, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.055492084473371506, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05412682145833969, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05380556359887123, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03188805654644966, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027473464608192444, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027379341423511505, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025259187445044518, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025058524683117867, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016403092071413994, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016007445752620697, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015900198370218277, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009841794148087502, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009841794148087502, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009841794148087502, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21824140846729279, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.16320094466209412, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13943207263946533, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11257971078157425, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.09681306034326553, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07380323857069016, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12487667798995972, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11455632746219635, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10470530390739441, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0638500526547432, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0643637552857399, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06418248265981674, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05516679957509041, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04714705049991608, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04509282484650612, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03218857944011688, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02532792277634144, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0246772151440382, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.019277891144156456, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01773093268275261, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.017437171190977097, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.017854031175374985, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014935212209820747, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.012850964441895485, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014935212209820747, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014935212209820747, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1904001086950302, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1797698736190796, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17645947635173798, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.16126340627670288, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08966562896966934, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08583974838256836, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09933575987815857, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09190331399440765, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09081120043992996, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0813496857881546, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07761455327272415, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05061757564544678, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.043972380459308624, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04296181723475456, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04272785782814026, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02530544623732567, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.022129129618406296, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02206127718091011, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020444363355636597, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.020293401554226875, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013334888964891434, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013389193452894688, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012999149039387703, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008949115872383118, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013334888964891434, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013334888964891434, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23807403445243835, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22479559481143951, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22066368162631989, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20165209472179413, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11190308630466461, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10717322677373886, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12400007992982864, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11462166160345078, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11331641674041748, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.101530060172081, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09687349945306778, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06309732049703598, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.054746564477682114, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.053501665592193604, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05321669951081276, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.031491219997406006, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027320152148604393, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.027231009677052498, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.025182895362377167, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02499312162399292, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016434483230113983, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.016137463971972466, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.016026383265852928, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010277168825268745, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010277168825268745, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010277168825268745, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.30.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22841796278953552, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20504027605056763, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19312016665935516, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1727021038532257, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10462910681962967, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09429504722356796, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1274598091840744, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11742302775382996, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10814650356769562, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09113040566444397, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08662254363298416, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06468741595745087, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.056223753839731216, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.050461072474718094, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04901585355401039, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03248601406812668, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026507003232836723, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025903142988681793, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02385672740638256, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02292393520474434, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.017410140484571457, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017509447410702705, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015587319619953632, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012062238529324532, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015587319619953632, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015587319619953632, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12975329160690308, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12212473899126053, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11929887533187866, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1087859570980072, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06080949679017067, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.057896096259355545, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06827069818973541, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06316947937011719, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06162474676966667, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05500989407300949, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05239670351147652, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0345991887152195, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030131394043564796, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029070930555462837, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028817525133490562, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017283014953136444, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014829862862825394, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014734199270606041, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01363044697791338, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013470558449625969, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008956436067819595, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00880874041467905, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.00860157422721386, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005542714148759842, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014829862862825394, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014829862862825394, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11010733246803284, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10360824316740036, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10116507112979889, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09221935272216797, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.051583025604486465, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04907795786857605, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05824318900704384, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.053711023181676865, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05228489264845848, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.046707089990377426, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.044584039598703384, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029559487476944923, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02564118802547455, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02468087710440159, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024454014375805855, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014763187617063522, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012609091587364674, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012517370283603668, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01159781962633133, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011452395468950272, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007673502899706364, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007529526948928833, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007357013877481222, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004767830949276686, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014763187617063522, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014763187617063522, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24248072504997253, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22847038507461548, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22414296865463257, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20448307693004608, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11377733945846558, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10876195877790451, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12649983167648315, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11680522561073303, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11527887731790543, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10298031568527222, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09813813865184784, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06416000425815582, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.055711157619953156, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05432179570198059, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05401581898331642, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.032001033425331116, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027548501268029213, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02745049260556698, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02530461549758911, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025098586454987526, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016429530456662178, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016018889844417572, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01591794192790985, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009779904969036579, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009779904969036579, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009779904969036579, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18596072494983673, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1392035335302353, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11751357465982437, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10245901346206665, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08428731560707092, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0642489567399025, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10852072387933731, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09941872209310532, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.0893055871129036, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.059143152087926865, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0567881278693676, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.056085918098688126, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04791863262653351, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.040924109518527985, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03914478421211243, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0280148908495903, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.021776139736175537, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021074926480650902, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.017025813460350037, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015718936920166016, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015013016760349274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015106437727808952, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012715299613773823, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010561391711235046, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015013016760349274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015013016760349274, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18818479776382446, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1775306761264801, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17415931820869446, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1590150147676468, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08860904723405838, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08480265736579895, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09817271679639816, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09086115658283234, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08973456919193268, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08029232919216156, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07652221620082855, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04999380186200142, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04349232465028763, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04248230531811714, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04223727807402611, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025001175701618195, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02190544083714485, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.021839898079633713, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.02022228203713894, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02007436752319336, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013165400363504887, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013294818811118603, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012835601344704628, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008933689445257187, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013165400363504887, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013165400363504887, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23755048215389252, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22415199875831604, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.22003693878650665, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.20087632536888123, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11160802841186523, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10684502869844437, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12354584783315659, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.11434867978096008, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.113021619617939, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10116815567016602, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.0964319258928299, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06276330351829529, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05458267405629158, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.05332612991333008, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05303628742694855, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03129737079143524, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.027140159159898758, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02706167846918106, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024987421929836273, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02479623630642891, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016177961602807045, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015907159075140953, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01576194539666176, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009951220825314522, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009951220825314522, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009951220825314522, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.31.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23063001036643982, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20682916045188904, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1943751722574234, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.17377790808677673, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10564731806516647, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09496551752090454, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1310010403394699, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11917916685342789, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10932641476392746, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.09200862795114517, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08769772201776505, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06624966859817505, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05719226971268654, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.051082611083984375, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.049554817378520966, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.033495236188173294, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.027071848511695862, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.026440324261784554, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.024431137368083, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.023458188399672508, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01837834157049656, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.018206613138318062, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01644800789654255, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012877393513917923, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012877393513917923, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012877393513917923, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1316300332546234, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.12378174811601639, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12094909697771072, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.11019133776426315, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06167389452457428, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.058675944805145264, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06931968033313751, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0641196072101593, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.062496837228536606, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05572608485817909, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05308234319090843, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.035144176334142685, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030591078102588654, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.029488440603017807, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02921939454972744, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017546702176332474, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015049443580210209, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014943892136216164, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01382722333073616, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013660960830748081, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.0090813422575593, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008967297151684761, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008709742687642574, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005660414230078459, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015049443580210209, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.015049443580210209, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11066450923681259, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1041194498538971, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10158850997686386, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09259964525699615, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05189003422856331, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.0493224561214447, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05873847007751465, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.054123684763908386, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05260675773024559, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04691850021481514, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04484381526708603, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0298234224319458, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025834962725639343, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02482304908335209, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02457551658153534, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0148951206356287, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012681642547249794, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012583581730723381, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011662893928587437, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01150604709982872, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007735108025372028, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.0075845373794436455, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007399122696369886, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0048021916300058365, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0148951206356287, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0148951206356287, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24776603281497955, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.23338668048381805, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22884568572044373, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20870277285575867, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11628358066082001, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.11114861816167831, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1293257474899292, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.1193845346570015, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11787078529596329, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10519715398550034, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.10022842884063721, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06564664095640182, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.056962285190820694, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.055532798171043396, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05518767237663269, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03273238241672516, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.028196129947900772, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02808992564678192, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025883972644805908, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025666221976280212, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01683577336370945, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016437944024801254, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01631075330078602, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010103171691298485, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010103171691298485, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010103171691298485, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.20204085111618042, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14640450477600098, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11976540088653564, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09767348319292068, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08496534824371338, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06263522058725357, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1188955008983612, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10891303420066833, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09673264622688293, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.059611979871988297, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.060488492250442505, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06133345514535904, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05239180475473404, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04138646647334099, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03837209939956665, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03067430667579174, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022121572867035866, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02123202197253704, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01736689917743206, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015159486792981625, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01643621362745762, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016133205965161324, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012912877835333347, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010950866155326366, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015159486792981625, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015159486792981625, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18456794321537018, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17404915392398834, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17079244554042816, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15595583617687225, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08691726624965668, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08316131681203842, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09637906402349472, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08914586156606674, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08807051926851273, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0787118449807167, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07510094344615936, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04911739379167557, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04267497360706329, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04167149215936661, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.041433535516262054, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.024556798860430717, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.021498462185263634, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.021432926878333092, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019839584827423096, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019693754613399506, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012965373694896698, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013058232143521309, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012633384205400944, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008791353553533554, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012965373694896698, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012965373694896698, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.23512619733810425, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.221748948097229, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21767660975456238, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1987030804157257, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.11056028306484222, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10580457746982574, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12242061644792557, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1132887527346611, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11198274046182632, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.10015254467725754, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09549969434738159, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.06231869012117386, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05410835146903992, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.052863262593746185, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.05255637317895889, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.03107699751853943, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.0269489549100399, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02686789259314537, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024804623797535896, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.024617541581392288, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.016156762838363647, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.01587529480457306, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015742307528853416, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010043679736554623, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010043679736554623, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010043679736554623, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.32.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22299160063266754, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19996874034404755, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18765977025032043, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1679213047027588, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10205098241567612, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09159542620182037, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.1257721483707428, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11567043513059616, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10566052794456482, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08893644064664841, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08475270867347717, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.0640631690621376, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05535602569580078, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04922698438167572, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04768265783786774, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03228766843676567, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.025853855535387993, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.025203322991728783, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023281605914235115, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02228703536093235, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01752563752233982, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.0171586275100708, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015581903979182243, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01175005454570055, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015581903979182243, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015581903979182243, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1266811192035675, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11916613578796387, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11630252003669739, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10594823211431503, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.059396933764219284, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05647329241037369, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06696394830942154, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.061875421553850174, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06021504104137421, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05368978902697563, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05115243047475815, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03399011865258217, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0295438040047884, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.028410639613866806, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028143959119915962, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016974780708551407, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014511432498693466, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.014400612562894821, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01332961954176426, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013157655484974384, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008802777156233788, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008667592890560627, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008432368747889996, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005482440814375877, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014511432498693466, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014511432498693466, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10817748308181763, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1017284169793129, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09923205524682999, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09038098156452179, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05067359283566475, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04818045720458031, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0572592094540596, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05285857990384102, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05139071121811867, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04581984132528305, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043704804033041, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029072463512420654, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025231359526515007, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024247953668236732, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02401875890791416, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014516601338982582, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012388762086629868, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01229759119451046, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011389215476810932, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011238264851272106, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007536695338785648, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007413665298372507, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007210783660411835, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004702826030552387, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014516601338982582, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014516601338982582, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24008280038833618, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2261432558298111, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22172600030899048, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.202132910490036, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11264369636774063, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10767549276351929, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12526555359363556, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11575093865394592, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.114165760576725, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10190115123987198, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0970216691493988, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06356392800807953, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0552172027528286, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.053813133388757706, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05348135530948639, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03170348331332207, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027285562828183174, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02718500979244709, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025049297139048576, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02483314648270607, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016274936497211456, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015861257910728455, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015759766101837158, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009677359834313393, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009677359834313393, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009677359834313393, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1883629560470581, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14599600434303284, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.12689094245433807, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1032489538192749, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08351753652095795, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06408779323101044, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11025868356227875, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09950514137744904, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08985689282417297, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.060447461903095245, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.057967204600572586, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05618520826101303, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.0483572855591774, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.041147563606500626, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.0392889641225338, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.028581222519278526, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.022937776520848274, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.022325776517391205, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.018902061507105827, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.017698079347610474, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016413047909736633, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01703829877078533, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014344624243676662, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01322196051478386, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014344624243676662, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014344624243676662, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18191063404083252, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17151682078838348, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16824640333652496, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1535244584083557, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08572150766849518, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08198793232440948, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09506171196699142, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08792954683303833, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08684773743152618, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07761374115943909, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07397200912237167, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04847020283341408, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04210282862186432, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04110017418861389, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04086485877633095, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.024222150444984436, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.021217823028564453, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02115347608923912, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.019582444801926613, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019431067630648613, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012777147814631462, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012909946963191032, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012448291294276714, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.00870900135487318, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012777147814631462, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012777147814631462, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2333211898803711, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.22001537680625916, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.21592889726161957, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.19705639779567719, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1097780093550682, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10500648617744446, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.12188281118869781, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.1125466525554657, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.11119064688682556, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.0994241014122963, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09476318210363388, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.062016457319259644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.0537882074713707, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.0525287389755249, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.052228767424821854, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.030993647873401642, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.026863452047109604, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.026786278933286667, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.024739086627960205, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02454971894621849, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01627974770963192, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015951041132211685, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.015864873304963112, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010273153893649578, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010273153893649578, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010273153893649578, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.33.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.23125720024108887, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.20731501281261444, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.19504894316196442, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.1742541491985321, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10594042390584946, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.0952315628528595, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.13020122051239014, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11921466886997223, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1096280962228775, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.0921502560377121, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.0876619964838028, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06607712805271149, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05710276588797569, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.05108489841222763, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04955944046378136, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.03353404626250267, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.026782911270856857, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02615581639111042, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.024072134867310524, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02309250459074974, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.018502231687307358, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.017672287300229073, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.016661154106259346, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01207818929105997, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01207818929105997, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01207818929105997, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12861411273479462, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1208692193031311, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11794501543045044, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10732971876859665, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06028839945793152, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.057250477373600006, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06787911057472229, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06281646341085434, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06110946834087372, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.0543658621609211, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05174098536372185, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.034408003091812134, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.029977256432175636, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02884933352470398, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028568081557750702, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.017182733863592148, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014740203507244587, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0146316047757864, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013525951653718948, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013356311246752739, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008917239494621754, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008817272260785103, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008544581942260265, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005603593308478594, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014740203507244587, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014740203507244587, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.107152558863163, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10073255747556686, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09818178415298462, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08934157341718674, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05027061328291893, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04769875109195709, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.056965552270412445, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05258599668741226, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05097727105021477, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.045367296785116196, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.043279532343149185, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.028947971761226654, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025119595229625702, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02406596951186657, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.023822421208024025, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014465157873928547, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012321248650550842, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012218627147376537, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011318044736981392, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011159481480717659, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007535313721746206, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00742869870737195, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007191765587776899, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004752773325890303, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014465157873928547, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014465157873928547, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.24441245198249817, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.22998149693012238, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.22546188533306122, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20544975996017456, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11476331204175949, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10960014909505844, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12765656411647797, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11790849268436432, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11629350483417511, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10368193686008453, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09864966571331024, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06478599458932877, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05624300613999367, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.054805051535367966, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05446527153253555, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03231244906783104, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02780347689986229, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.027696993201971054, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02549799345433712, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025275949388742447, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016594350337982178, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016186635941267014, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.016053931787610054, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009895019233226776, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009895019233226776, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009895019233226776, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.19261258840560913, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14959749579429626, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13247516751289368, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0993957668542862, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08633997291326523, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06850409507751465, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.10724664479494095, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09886887669563293, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.09205595403909683, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05914273113012314, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05521628260612488, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.05488117039203644, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04739642143249512, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.041704680770635605, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04027068242430687, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02744300290942192, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02191007323563099, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.021448230370879173, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016698064282536507, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015594291500747204, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.014564081095159054, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014653951860964298, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012741957791149616, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009986422024667263, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015594291500747204, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.015594291500747204, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17965291440486908, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.16927355527877808, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16604651510715485, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.151448592543602, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08466571569442749, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08093395829200745, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09384626150131226, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08685589581727982, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08579044044017792, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07660511136054993, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07296894490718842, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04784102737903595, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.041592516005039215, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04059438407421112, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04036158323287964, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02389862760901451, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.02096560038626194, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02089584991335869, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01933569461107254, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.01918831840157509, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01258432399481535, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012768338434398174, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012257445603609085, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008629201911389828, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01258432399481535, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.01258432399481535, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.22504116594791412, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.2121695727109909, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20816916227340698, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.18986114859580994, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.10588856786489487, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.10128609836101532, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11743797361850739, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10858256369829178, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10729759186506271, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09584953635931015, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.09124752879142761, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05976405739784241, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05189249664545059, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.050676602870225906, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.050385426729917526, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.029845135286450386, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.025902284309267998, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.025824081152677536, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.023840855807065964, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.023660793900489807, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015589388087391853, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.015367492102086544, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01518333237618208, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009867722168564796, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.015589388087391853, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.01518333237618208, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.34.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22245842218399048, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19820022583007812, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18597197532653809, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16605640947818756, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.1015043556690216, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09092501550912857, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12466425448656082, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11473019421100616, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10551486164331436, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08799733966588974, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08374647796154022, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06340024620294571, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05488625541329384, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04894557595252991, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04745227098464966, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031802333891391754, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.02567588910460472, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02508031204342842, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.023019392043352127, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.022058973088860512, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.01704840362071991, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.016971949487924576, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01515984907746315, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01159740425646305, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01515984907746315, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.01515984907746315, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12444379180669785, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11702074110507965, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.11418507248163223, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1038743406534195, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05833101272583008, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05542483925819397, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.06567898392677307, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06080024316906929, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05914395675063133, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.052629806101322174, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05006062984466553, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.033317212015390396, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02902555838227272, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027906615287065506, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.027640998363494873, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016637273132801056, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014256083406507969, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.01415332779288292, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.013089025393128395, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01291677076369524, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008635947480797768, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008528555743396282, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.008266880176961422, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0054085999727249146, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014256083406507969, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.014256083406507969, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1033477708697319, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09713122248649597, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09472320228815079, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08615525811910629, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0484316423535347, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.045987166464328766, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05485132709145546, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.050627369433641434, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04910547286272049, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.043728720396757126, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04167107492685318, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.027834484353661537, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02416837587952614, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023174172267317772, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.022947445511817932, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013906428590416908, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011865639127790928, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011773282662034035, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010900704190135002, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010751535184681416, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007242195308208466, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007137789390981197, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.0069173406809568405, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004569135140627623, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013906428590416908, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013906428590416908, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.2415202260017395, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2275058627128601, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2229648232460022, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.20308950543403625, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11347310990095139, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10842534154653549, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12615209817886353, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11655110120773315, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.1149708554148674, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10253893584012985, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09754012525081635, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06403467059135437, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.055611446499824524, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.0541924349963665, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.053871553391218185, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.031950242817401886, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.027494024485349655, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.02739502303302288, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.025226809084415436, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.025011779740452766, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016417499631643295, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.016006752848625183, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.015901843085885048, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009803567081689835, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009803567081689835, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009803567081689835, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1820167601108551, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.14886324107646942, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.13760331273078918, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10651319473981857, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.08237391710281372, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.07010714709758759, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09864809364080429, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09057596325874329, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08711037784814835, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.06104833632707596, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05239402875304222, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.050653260201215744, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04354051500558853, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03984157368540764, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.038925185799598694, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.025362662971019745, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02099655382335186, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.020771944895386696, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0168879684060812, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01623958721756935, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013693522661924362, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.013834496960043907, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.012532858178019524, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009756042622029781, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013693522661924362, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013693522661924362, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.17726996541023254, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.167044997215271, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.16382034122943878, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14941418170928955, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08354980498552322, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07986921817064285, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.09262232482433319, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.085713230073452, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08466678857803345, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07557953894138336, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07194764912128448, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04716115817427635, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04103608429431915, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.040048275142908096, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03981107100844383, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.023568177595734596, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.020642630755901337, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02057594805955887, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01902666501700878, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.018878325819969177, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012379257939755917, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012510815635323524, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012054945342242718, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008382570929825306, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012379257939755917, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.012379257939755917, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.2175205945968628, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.20498423278331757, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.20112833380699158, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1834028661251068, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.1023772731423378, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09787485003471375, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.11345186829566956, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.10498305410146713, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.10371337831020355, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.09259193390607834, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08820655196905136, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05770297348499298, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.05013246834278107, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.04894731566309929, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04866034537553787, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.028788354247808456, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.024953164160251617, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.02487281709909439, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.022934982553124428, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.022757479920983315, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014942185953259468, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.014694263227283955, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.0145461056381464, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009277084842324257, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014942185953259468, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014942185953259468, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.35.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.22296026349067688, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.19696615636348724, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.18476274609565735, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.16503478586673737, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.10130342096090317, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.09038122743368149, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12403116375207901, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11422891914844513, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.10576644539833069, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08744281530380249, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.08314908295869827, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06301487237215042, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.05460971221327782, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.0487835556268692, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.047322824597358704, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.031570471823215485, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.025454597547650337, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02489623613655567, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.022708024829626083, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.02174323797225952, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.016754524782299995, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.016646765172481537, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014874324202537537, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011151948012411594, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014874324202537537, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014874324202537537, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11973598599433899, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11237038671970367, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10944391041994095, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0995141938328743, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.056053102016448975, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05312791466712952, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0633932426571846, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.058675363659858704, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05685151740908623, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.050516773015260696, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04805779084563255, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.032159510999917984, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.028014741837978363, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.026818279176950455, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026534682139754295, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016075706109404564, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013723784126341343, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013611926697194576, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01258906815201044, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.012410617433488369, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.008361796848475933, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.008255845867097378, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007979683578014374, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005260406993329525, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013723784126341343, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.013723784126341343, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09730343520641327, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09128578007221222, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.0888211578130722, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08072982728481293, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04553623124957085, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04310624673962593, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05175723880529404, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.047870032489299774, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.046195466071367264, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.041036996990442276, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03905694559216499, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.026258544996380806, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.022847658023238182, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.021803582087159157, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.021552035585045815, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013120625168085098, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011170810088515282, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0110636530444026, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010250107385218143, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01009137462824583, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006841689348220825, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00676138186827302, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006504086311906576, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004333841614425182, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013120625168085098, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.013120625168085098, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.23664738237857819, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2226274609565735, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.2181582748889923, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.1987123191356659, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.11112318187952042, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10609669983386993, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.12381504476070404, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11425674706697464, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.11264973878860474, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.10034079849720001, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09548412263393402, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0629030391573906, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.054524101316928864, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.053099844604730606, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05276798456907272, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03135986253619194, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.026997612789273262, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.026891300454735756, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.024754298850893974, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.02453894168138504, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.016140855848789215, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.01580231636762619, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01561540924012661, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009787339717149734, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01561540924012661, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01561540924012661, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.18023985624313354, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.1425650715827942, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1291586309671402, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.10032641142606735, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0812513455748558, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.06695907562971115, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.09900949895381927, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.09059752523899078, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.08678919076919556, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.05712403729557991, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.05176466330885887, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0510379858314991, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04369253292679787, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.03943532705307007, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.03837277740240097, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.02566256746649742, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02098318748176098, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.020738013088703156, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.016420086845755577, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01564781554043293, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.013965531252324581, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014161239378154278, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01259977463632822, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.010200290009379387, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01564781554043293, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01564781554043293, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16959819197654724, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15973249077796936, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.15662819147109985, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.14278759062290192, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07989166676998138, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07635606825351715, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08862211555242538, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.08202056586742401, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.08097606897354126, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.07223491370677948, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06876573711633682, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04517607390880585, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.039261769503355026, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.03829888254404068, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.03807634860277176, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02257171832025051, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.019770227372646332, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.01970386505126953, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.018214881420135498, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.018072864040732384, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011882662773132324, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.012024796567857265, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011565454304218292, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008091188967227936, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011882662773132324, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011882662773132324, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.20364969968795776, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.19184616208076477, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.18821054697036743, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.17156293988227844, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0958058312535286, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.09160676598548889, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.1062573492527008, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09829919785261154, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09708794951438904, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08665726333856583, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.08248837292194366, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.05407198145985603, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.046952154487371445, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.045821502804756165, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.04555731639266014, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.026984386146068573, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.023377904668450356, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.023304101079702377, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.021487150341272354, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.02131613716483116, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014033898711204529, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013788416050374508, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.013656596653163433, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008734533563256264, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014033898711204529, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.014033898711204529, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.36.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.21618297696113586, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.18933194875717163, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.17692962288856506, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.15782369673252106, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.0981074869632721, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.08685066550970078, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.12123117595911026, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.11087516695261002, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.1026579737663269, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08396100997924805, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07991816103458405, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.06162773445248604, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0531773716211319, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.047462183982133865, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04602177068591118, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.030966244637966156, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.025147799402475357, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.02462022937834263, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.02241307869553566, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.021478647366166115, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.016717689111828804, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.016923408955335617, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014853237196803093, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.01194140687584877, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014853237196803093, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.014853237196803093, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.11178873479366302, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10482224822044373, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10200511664152145, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09260449558496475, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.0522625669836998, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04947050288319588, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05937054753303528, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05481429770588875, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.053003113716840744, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04704668000340462, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04478373005986214, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.030149586498737335, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.026196395978331566, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02502608485519886, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.024753110483288765, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015080902725458145, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.012873723171651363, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012760961428284645, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01181759499013424, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011643391102552414, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007912716828286648, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00786127895116806, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007540141697973013, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.005158350802958012, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015080902725458145, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.015080902725458145, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09400661289691925, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08817408978939056, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08563174307346344, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07776447385549545, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04394623264670372, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04153391346335411, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05022728815674782, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.04637929052114487, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.044563520699739456, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03959416598081589, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.0377143956720829, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.025497961789369583, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02215430699288845, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02105727232992649, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.020795119926333427, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012758747674524784, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.010830093175172806, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010718763805925846, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009945710189640522, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.009778762236237526, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006696614436805248, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.00663002859801054, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006342983804643154, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004324819892644882, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012758747674524784, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.012758747674524784, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22255530953407288, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.2091771811246872, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.20492124557495117, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.18627609312534332, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10442258417606354, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09960779547691345, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.1164863258600235, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10750127583742142, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10584389418363571, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09414902329444885, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08952245116233826, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.059200625866651535, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.051282986998558044, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04989194497466087, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04956304281949997, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029524533078074455, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.025383982807397842, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025287160649895668, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.0232620257884264, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023050688207149506, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01524311862885952, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014914136379957199, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014741115272045135, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009283594787120819, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01524311862885952, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.01524311862885952, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.12002889811992645, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10326747596263885, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09567616879940033, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07769699394702911, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.055475085973739624, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.04820000007748604, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0685974732041359, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.062426161020994186, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05696577578783035, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04326571151614189, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03932991996407509, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03522350639104843, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.030792294070124626, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.027732238173484802, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.026992343366146088, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01796039566397667, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.016074571758508682, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.0157603919506073, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01402800902724266, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.01358105894178152, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010473336093127728, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.012235256843268871, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009578174911439419, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.01022335235029459, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01402800902724266, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.01402800902724266, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.16073788702487946, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.15128538012504578, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.1483633816242218, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.13519755005836487, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07569096237421036, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.07228736579418182, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.08409243822097778, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07773469388484955, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.0767260417342186, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06844320893287659, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.06517179310321808, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04291520640254021, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03726569190621376, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.036332231014966965, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.036107949912548065, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.021465446799993515, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.018828894942998886, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.018764223903417587, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01736985146999359, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.017234010621905327, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011394664645195007, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011574680916965008, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011084931902587414, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.007937584072351456, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011394664645195007, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011394664645195007, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.18905222415924072, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.17800651490688324, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.17458221316337585, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.15916481614112854, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.08894917368888855, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.08498556166887283, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.0987902358174324, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.09133515506982803, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.09015920013189316, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.08042580634355545, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.07653776556253433, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.0503447987139225, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.04365934804081917, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.042586132884025574, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.042329978197813034, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.025138404220342636, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.021818194538354874, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.021744895726442337, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.020070604979991913, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.019905654713511467, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013186950236558914, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013017648831009865, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.012830655090510845, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.008440003730356693, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013186950236558914, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.013186950236558914, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.37.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.2090531885623932, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.1816970556974411, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.17003843188285828, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.15183110535144806, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.09493529796600342, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.083793044090271, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.11648617684841156, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.10624366998672485, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.0994565337896347, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.08063696324825287, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.07673592865467072, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.05951269343495369, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.0511590838432312, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.04606444388628006, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.04479798674583435, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.030100340023636818, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.024673672392964363, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.024240590631961823, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.021955139935016632, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.021141525357961655, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0166921503841877, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.016834866255521774, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015116887167096138, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.012311779893934727, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015116887167096138, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.015116887167096138, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10933876037597656, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.10270211845636368, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09998389333486557, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.09084180742502213, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.05118207260966301, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.048490896821022034, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.05839700996875763, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05373752862215042, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.05189114809036255, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.046164605766534805, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.044092122465372086, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.029657676815986633, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.025686297565698624, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.024532675743103027, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.02426185831427574, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014859016053378582, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01261498685926199, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.012498477473855019, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011593800969421864, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011417718604207039, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.007815098389983177, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007692386396229267, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007451863959431648, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.0050191027112305164, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014859016053378582, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014859016053378582, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.09027672559022903, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.08477318286895752, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.08235728740692139, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.07483018934726715, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04224760830402374, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03993561491370201, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.048342738300561905, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.0446554534137249, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04284553602337837, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03808680176734924, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03627679497003555, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.02453717030584812, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.021327901631593704, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.020250229164958, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.019987337291240692, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01228710450232029, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01041458360850811, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.010299037210643291, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.009565811604261398, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.00940305832773447, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.006457902491092682, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.006379372905939817, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.006108459085226059, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.004156715702265501, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01228710450232029, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01228710450232029, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.22854529321193695, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.21498499810695648, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.21064579486846924, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.19171521067619324, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.1073080375790596, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.10246016830205917, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11946430057287216, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.11025959253311157, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10876588523387909, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09683068096637726, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.09209276735782623, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.06066859886050224, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.05261655151844025, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.05126679316163063, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.05094362050294876, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.03026648238301277, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02608657069504261, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.025993360206484795, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.023914538323879242, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.023713894188404083, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015589645132422447, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.015308867208659649, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01508971955627203, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009528438560664654, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015589645132422447, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.01508971955627203, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.1345375031232834, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.11169203370809555, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.10216592997312546, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.08184927701950073, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.06035898998379707, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.05224823206663132, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.0760149136185646, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.06958377361297607, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.06440974771976471, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04648662731051445, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04192350059747696, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.03919261321425438, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.03392001986503601, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.02985658496618271, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.028825057670474052, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.01991783082485199, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.01675155758857727, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.016444966197013855, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014321665279567242, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.013679280877113342, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.011336779221892357, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.012393549084663391, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.010141255334019661, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009860522113740444, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014321665279567242, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.014321665279567242, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1336803287267685, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1258753538131714, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.12335581332445145, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.1124463602900505, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.06326310336589813, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.060438670217990875, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.07023631036281586, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.06497331708669662, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.06412457674741745, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.05724042281508446, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.05459938570857048, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03623806685209274, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03172891214489937, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.030965711921453476, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.030778346583247185, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.018272824585437775, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.017056066542863846, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.017007526010274887, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.015949873253703117, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.015848107635974884, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010362613946199417, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.011808331124484539, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.010131003335118294, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.009518058970570564, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010362613946199417, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.010362613946199417, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.1515822410583496, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.14272701740264893, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.13994258642196655, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.12782356142997742, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.07173623889684677, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.06855063140392303, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.07986018061637878, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.07364307343959808, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.07271125167608261, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.06495294719934464, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.062043581157922745, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.04120704531669617, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.03596898540854454, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.035116370767354965, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.034917451441287994, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.02085869386792183, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.019370976835489273, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.019313843920826912, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.01812463067471981, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.018005801364779472, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011955120600759983, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.013410939835011959, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.011697732843458652, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.010829640552401543, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011955120600759983, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.011955120600759983, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.38.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.15652161836624146, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.13439396023750305, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.1254788190126419, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.11151937395334244, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.0709870457649231, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.06219358369708061, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.08797427266836166, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.07945400476455688, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.07442279160022736, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.059812020510435104, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.057103563100099564, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.045107923448085785, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.03883045166730881, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.03496745973825455, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.03404736891388893, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.02315242402255535, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.01957758516073227, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.019234711304306984, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.01757388934493065, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.01696249470114708, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.013637298718094826, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.01424270961433649, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.012357688508927822, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.011310583911836147, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.013637298718094826, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.013637298718094826, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.q_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10310821235179901, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.09676172584295273, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.09402769058942795, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.0855177640914917, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.04813633859157562, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.045520998537540436, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.055354245007038116, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.050794605165719986, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.04886089265346527, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.04344790801405907, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.04159194231033325, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.028105536475777626, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.024267200380563736, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023058274760842323, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.022773023694753647, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014064598828554153, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.011863067746162415, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.011740465648472309, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.010916761122643948, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.010732308961451054, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.00740261934697628, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.007268137764185667, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.007013336289674044, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.00475698197260499, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014064598828554153, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.014064598828554153, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.k_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.08053074032068253, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07559973746538162, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.07334744185209274, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.06663903594017029, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.03759574145078659, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.03546753525733948, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.043269477784633636, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.03992148116230965, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.038137856870889664, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03391224518418312, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03234225511550903, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.021970976144075394, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.019081657752394676, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.018025444820523262, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.017775405198335648, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0110047347843647, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.009304617531597614, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.009191269986331463, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.008559172041714191, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.008399336598813534, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.005818168632686138, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.005762959364801645, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.005479338113218546, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.003807955887168646, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0110047347843647, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.0110047347843647, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.v_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.21350771188735962, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.20066337287425995, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.1965872347354889, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.17903339862823486, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.10025198757648468, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.09560483694076538, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.11327316612005234, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.10323713719844818, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.10162556916475296, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.09065651893615723, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.08671537786722183, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.057681161910295486, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.04954686015844345, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.04809310659766197, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.04771940037608147, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.029184598475694656, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.02476716972887516, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.024643205106258392, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.02280929684638977, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.022574927657842636, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015326760709285736, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.014923862181603909, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.014752613380551338, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009729327633976936, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015326760709285736, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.015326760709285736, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.self_attn.o_proj", + "numel": 26214400, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.17529296875, + "total_bits": 57024000.0, + "err": 0.10456958413124084, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 62266880.00000001, + "err": 0.07672525942325592, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 68820480.0, + "err": 0.06604012846946716, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.72529296875, + "total_bits": 71441920.0, + "err": 0.059429753571748734, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.22529296875, + "total_bits": 84549120.0, + "err": 0.046381138265132904, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.72529296875, + "total_bits": 97656320.0, + "err": 0.034836187958717346, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 79464320.0, + "err": 0.060174789279699326, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 81927680.0, + "err": 0.05261130630970001, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.17529296875, + "total_bits": 83238400.0, + "err": 0.049900900572538376, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.52529296875, + "total_bits": 92413440.0, + "err": 0.03271136060357094, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.662646484375, + "total_bits": 96014080.0, + "err": 0.03255438059568405, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 105678720.0, + "err": 0.0313204750418663, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 108142080.0, + "err": 0.02624346874654293, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.22529296875, + "total_bits": 110763520.0, + "err": 0.023504041135311127, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.32529296875, + "total_bits": 113384960.00000001, + "err": 0.022818898782134056, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 131893120.0, + "err": 0.016507990658283234, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0139192845672369, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.32529296875, + "total_bits": 139599360.0, + "err": 0.013772455044090748, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.52529296875, + "total_bits": 144842240.0, + "err": 0.011726281605660915, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.72529296875, + "total_bits": 150085120.0, + "err": 0.011311345733702183, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 158107520.0, + "err": 0.010193143971264362, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 160570880.0, + "err": 0.010830895975232124, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.2313232421875, + "total_bits": 163350400.0, + "err": 0.009413917548954487, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 212999679.99999997, + "err": 0.009213367477059364, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0139192845672369, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.22529296875, + "total_bits": 136977920.0, + "err": 0.0139192845672369, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.mlp.gate_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.11875869333744049, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1117323487997055, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.10956680774688721, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.09988520294427872, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.05574474856257439, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.05324440076947212, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.06202537938952446, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.057272110134363174, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.056503523141145706, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.050403788685798645, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.048075269907712936, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03159602731466293, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.027414929121732712, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.026734312996268272, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.026565948501229286, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.01580216735601425, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.013840186409652233, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.013793252408504486, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.012763350270688534, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.012659767642617226, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.008363259956240654, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.008485114201903343, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.00813320092856884, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.005790180526673794, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.013840186409652233, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.013840186409652233, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.mlp.up_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1751085069444445, + "total_bits": 153951744.0, + "err": 0.11922353506088257, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.3751085069444446, + "total_bits": 168107520.0, + "err": 0.1122065857052803, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.6251085069444446, + "total_bits": 185802240.0, + "err": 0.11002721637487411, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7251085069444443, + "total_bits": 192880128.0, + "err": 0.10034994781017303, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.2251085069444443, + "total_bits": 228269568.0, + "err": 0.0563276968896389, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7251085069444443, + "total_bits": 263659008.0, + "err": 0.05382771044969559, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.031277126736111, + "total_bits": 214550400.0, + "err": 0.06274999678134918, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.1251085069444446, + "total_bits": 221191680.0, + "err": 0.057850368320941925, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1751085069444445, + "total_bits": 224730624.0, + "err": 0.05708220601081848, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.5251085069444446, + "total_bits": 249503232.0, + "err": 0.05095374956727028, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6625542534722224, + "total_bits": 259231488.0, + "err": 0.048629265278577805, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.031277126736111, + "total_bits": 285329280.0, + "err": 0.03230659291148186, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.125108506944445, + "total_bits": 291970560.0, + "err": 0.02807946689426899, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.225108506944444, + "total_bits": 299048448.0, + "err": 0.027395455166697502, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.325108506944445, + "total_bits": 306126336.0, + "err": 0.027234699577093124, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.031277126736111, + "total_bits": 356108160.0, + "err": 0.0162977185100317, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.014815588481724262, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.325108506944445, + "total_bits": 376905216.0, + "err": 0.014772791415452957, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525108506944444, + "total_bits": 391060992.0, + "err": 0.013802962377667427, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.725108506944444, + "total_bits": 405216768.0, + "err": 0.013705243356525898, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.031277126736111, + "total_bits": 426887040.0, + "err": 0.009151780977845192, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.125108506944445, + "total_bits": 433528320.0, + "err": 0.009934045374393463, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.231277126736111, + "total_bits": 441042816.0, + "err": 0.008934489451348782, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.125108506944445, + "total_bits": 575086080.0, + "err": 0.007727102376520634, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.014815588481724262, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.225108506944444, + "total_bits": 369827328.0, + "err": 0.014815588481724262, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + }, + { + "key": "model.layers.39.mlp.down_proj", + "numel": 70778880, + "options": [ + { + "desc": "0.05:3b_32g/0.95:2b_32g s4", + "bpw": 2.1762188946759258, + "total_bits": 154030336.0, + "err": 0.06615719199180603, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:3b_32g/0.75:2b_32g s4", + "bpw": 2.37529296875, + "total_bits": 168120576.0, + "err": 0.0563923642039299, + "qparams": { + "group_size": { + "3": 32, + "2": 32 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.25:4b_32g/0.75:2b_32g s4", + "bpw": 2.62529296875, + "total_bits": 185815296.0, + "err": 0.05165039747953415, + "qparams": { + "group_size": { + "4": 32, + "2": 32 + }, + "bits": [ + 4, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.4:3b_32g/0.5:2b_32g s4", + "bpw": 2.7294596354166667, + "total_bits": 193188096.0, + "err": 0.045769624412059784, + "qparams": { + "group_size": { + "4": 32, + "3": 32, + "2": 32 + }, + "bits": [ + 4, + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.4, + 0.5 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:4b_32g/0.9:3b_32g s4", + "bpw": 3.227144820601852, + "total_bits": 228413696.0, + "err": 0.030140548944473267, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.2:6b_32g/0.8:3b_32g s4", + "bpw": 3.7294596354166667, + "total_bits": 263966976.0, + "err": 0.025833070278167725, + "qparams": { + "group_size": { + "6": 32, + "3": 32 + }, + "bits": [ + 6, + 3 + ], + "bits_prop": [ + 0.2, + 0.8 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_128g s4", + "bpw": 3.0313232421875, + "total_bits": 214553664.0, + "err": 0.039676129817962646, + "qparams": { + "group_size": { + "3": 128 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:3b_32g s4", + "bpw": 3.12529296875, + "total_bits": 221204736.0, + "err": 0.03499327972531319, + "qparams": { + "group_size": { + "3": 32 + }, + "bits": [ + 3 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:4b_32g/0.95:3b_32g s4", + "bpw": 3.1762188946759258, + "total_bits": 224809216.0, + "err": 0.03169520944356918, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:4b_32g/0.6:3b_32g s4", + "bpw": 3.525755931712963, + "total_bits": 249549056.0, + "err": 0.02533142827451229, + "qparams": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.6:4b_64g/0.4:3b_64g s4", + "bpw": 3.6644983362268517, + "total_bits": 259369088.0, + "err": 0.024366345256567, + "qparams": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.6, + 0.4 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_128g s4", + "bpw": 4.0313232421875, + "total_bits": 285332544.0, + "err": 0.020537858828902245, + "qparams": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:4b_32g s4", + "bpw": 4.12529296875, + "total_bits": 291983616.0, + "err": 0.017411718145012856, + "qparams": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.015265805646777153, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:4b_32g s4", + "bpw": 4.3289966724537035, + "total_bits": 306401536.0, + "err": 0.014711566269397736, + "qparams": { + "group_size": { + "6": 32, + "4": 32 + }, + "bits": [ + 6, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:5b_128g s4", + "bpw": 5.0313232421875, + "total_bits": 356111424.0, + "err": 0.010659663937985897, + "qparams": { + "group_size": { + "5": 128 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:6b_32g/0.9:5b_32g s4", + "bpw": 5.227144820601852, + "total_bits": 369971456.0, + "err": 0.009097772650420666, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.05:8b_32g/0.05:6b_32g/0.9:5b_32g s4", + "bpw": 5.3289966724537035, + "total_bits": 377180416.0, + "err": 0.00890104565769434, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.05, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.4:6b_32g/0.6:5b_32g s4", + "bpw": 5.525755931712963, + "total_bits": 391106816.0, + "err": 0.008310851640999317, + "qparams": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.4, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", + "bpw": 5.731774450231481, + "total_bits": 405688576.0, + "err": 0.007998298853635788, + "qparams": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.3, + 0.6 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_128g s4", + "bpw": 6.0313232421875, + "total_bits": 426890304.0, + "err": 0.0064046201296150684, + "qparams": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:6b_32g s4", + "bpw": 6.12529296875, + "total_bits": 433541376.0, + "err": 0.007193233352154493, + "qparams": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + }, + { + "desc": "0.1:8b_128g/0.9:6b_128g s4", + "bpw": 6.235026945891204, + "total_bits": 441308224.0, + "err": 0.005715166684240103, + "qparams": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "desc": "1.0:8b_32g s4", + "bpw": 8.12529296875, + "total_bits": 575099136.0, + "err": 0.006113196723163128, + "qparams": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1.0 + ], + "scale_bits": 4 + } + } + ], + "best_option_max": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.015265805646777153, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + "best_option": { + "desc": "0.1:5b_32g/0.9:4b_32g s4", + "bpw": 4.227144820601852, + "total_bits": 299192576.0, + "err": 0.015265805646777153, + "qparams": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + } + } + ], + "last_module_idx": 82, + "base_perplexity": 8.391836633786518 +} \ No newline at end of file