{ "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": 0.021202892065048218, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.020985381677746773, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.010785783641040325, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.010726727545261383, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.010659269988536835, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.007223472464829683, "qparams": { "group_size": { "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": 0.034815117716789246, "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": 0.020750347524881363, "qparams": { "group_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": 0.010676338337361813, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.010632657445967197, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.010781798511743546, "qparams": { "group_size": { "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": 0.012539306655526161, "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": 0.010615689679980278, "qparams": { "group_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": 0.00798160582780838, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.007198468782007694, "qparams": { "group_size": { "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": 0.007988816127181053, "qparams": { "group_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": 0.007191397715359926, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006937352940440178, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.007190441247075796, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.00693620927631855, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007094105705618858, "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": 0.00719006173312664, "qparams": { "group_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": 0.006803487427532673, "qparams": { "group_size": { "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": 0.006934960372745991, "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": 0.010785783641040325, "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": 0.010785783641040325, "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": 0.021860815584659576, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.02156759798526764, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.010023864917457104, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.009948465041816235, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.009871145710349083, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.005188138224184513, "qparams": { "group_size": { "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": 0.028260698541998863, "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": 0.021303392946720123, "qparams": { "group_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": 0.009903477504849434, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.00982646644115448, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.009988830424845219, "qparams": { "group_size": { "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": 0.011272343806922436, "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": 0.009805425070226192, "qparams": { "group_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": 0.006261791102588177, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.005143146961927414, "qparams": { "group_size": { "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": 0.006174010690301657, "qparams": { "group_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": 0.0051265438087284565, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.004726627841591835, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.005123877432197332, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.004722295794636011, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.004789573140442371, "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": 0.005122820846736431, "qparams": { "group_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": 0.00427989149466157, "qparams": { "group_size": { "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": 0.00471986411139369, "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": 0.010023864917457104, "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": 0.010023864917457104, "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": 0.059263017028570175, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.04975202679634094, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.039356719702482224, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.03325041010975838, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.026450132951140404, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.018893174827098846, "qparams": { "group_size": { "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": 0.04986594244837761, "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": 0.03815804049372673, "qparams": { "group_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": 0.028060566633939743, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.021919293329119682, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.023016538470983505, "qparams": { "group_size": { "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": 0.026856372132897377, "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": 0.018667228519916534, "qparams": { "group_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": 0.013112807646393776, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.011433795094490051, "qparams": { "group_size": { "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": 0.013412212021648884, "qparams": { "group_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": 0.007366604637354612, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006562788505107164, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.006569164339452982, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.005534512922167778, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007008373271673918, "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": 0.0060387784615159035, "qparams": { "group_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": 0.003945592790842056, "qparams": { "group_size": { "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": 0.004316315054893494, "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": 0.013112807646393776, "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": 0.013112807646393776, "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": 0.02393254265189171, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.015167731791734695, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.009232545271515846, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.01017993874847889, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.00973146315664053, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.004663009196519852, "qparams": { "group_size": { "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": 0.01585477404296398, "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": 0.014012311585247517, "qparams": { "group_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": 0.011100118048489094, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.007205547299236059, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.007594630122184753, "qparams": { "group_size": { "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": 0.008033831603825092, "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": 0.007013114169239998, "qparams": { "group_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": 0.005231048911809921, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.004738816525787115, "qparams": { "group_size": { "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": 0.004255794920027256, "qparams": { "group_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": 0.003581995377317071, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.003421901259571314, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.003273251000791788, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0030259648337960243, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.002694491995498538, "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": 0.0032537884544581175, "qparams": { "group_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": 0.0022420091554522514, "qparams": { "group_size": { "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": 0.0028903037309646606, "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": 0.015167731791734695, "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": 0.015167731791734695, "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": 0.05548559129238129, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.052377525717020035, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05138841271400452, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.04712851718068123, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.024718934670090675, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.02378925494849682, "qparams": { "group_size": { "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": 0.02745315432548523, "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": 0.02541438862681389, "qparams": { "group_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": 0.02498909831047058, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.02281545288860798, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.021958788856863976, "qparams": { "group_size": { "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": 0.013927304185926914, "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": 0.0122431181371212, "qparams": { "group_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": 0.011943605728447437, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.011873168870806694, "qparams": { "group_size": { "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": 0.007025633938610554, "qparams": { "group_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": 0.0065508694387972355, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006528760772198439, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.006220705807209015, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006177263334393501, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00396348861977458, "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": 0.004514445550739765, "qparams": { "group_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": 0.003863505320623517, "qparams": { "group_size": { "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": 0.0036135385744273663, "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": 0.013927304185926914, "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": 0.013927304185926914, "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": 0.08095692098140717, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0765504539012909, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07515482604503632, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06898830831050873, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03607723116874695, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03474889323115349, "qparams": { "group_size": { "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": 0.03999360650777817, "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": 0.0369882732629776, "qparams": { "group_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": 0.03646296262741089, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.033291641622781754, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.032032277435064316, "qparams": { "group_size": { "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": 0.020158981904387474, "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": 0.01754208281636238, "qparams": { "group_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": 0.01714251935482025, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.01705298200249672, "qparams": { "group_size": { "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": 0.0100611113011837, "qparams": { "group_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": 0.008928373456001282, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008898665197193623, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008403067477047443, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008344130590558052, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005357518792152405, "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": 0.005563011392951012, "qparams": { "group_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": 0.005218195263296366, "qparams": { "group_size": { "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": 0.003916956018656492, "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": 0.0100611113011837, "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": 0.0100611113011837, "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": 0.06026403605937958, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05268482118844986, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.030340122058987617, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0282885804772377, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.02458367682993412, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.014980440959334373, "qparams": { "group_size": { "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": 0.05315728858113289, "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": 0.047969356179237366, "qparams": { "group_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": 0.025914983823895454, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.022179342806339264, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.022158659994602203, "qparams": { "group_size": { "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": 0.022717226296663284, "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": 0.020823698490858078, "qparams": { "group_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": 0.01569865457713604, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.011718223802745342, "qparams": { "group_size": { "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": 0.01413507480174303, "qparams": { "group_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": 0.010049003176391125, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008041114546358585, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009768679738044739, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0076109981164336205, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.009573650546371937, "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": 0.009614992886781693, "qparams": { "group_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": 0.007161712273955345, "qparams": { "group_size": { "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": 0.00724476994946599, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } } ], "best_option_max": { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7294596354166667, "total_bits": 263966976.0, "err": 0.014980440959334373, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "scale_bits": 4 } }, "best_option": { "desc": "0.2:6b_32g/0.8:3b_32g s4", "bpw": 3.7294596354166667, "total_bits": 263966976.0, "err": 0.014980440959334373, "qparams": { "group_size": { "6": 32, "3": 32 }, "bits": [ 6, 3 ], "bits_prop": [ 0.2, 0.8 ], "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": 0.013800338841974735, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.011059217154979706, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.008490897715091705, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.007456210907548666, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.005903761833906174, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.003981192130595446, "qparams": { "group_size": { "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": 0.01080199982970953, "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": 0.008618746884167194, "qparams": { "group_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": 0.006413100752979517, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.004913201555609703, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.005253023002296686, "qparams": { "group_size": { "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": 0.0056937094777822495, "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": 0.004263679962605238, "qparams": { "group_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": 0.0029698472935706377, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.002543544862419367, "qparams": { "group_size": { "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": 0.0029323240742087364, "qparams": { "group_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": 0.0016959398053586483, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0015037613920867443, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0015313747571781278, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0012776657240465283, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0015843010041862726, "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": 0.0014306114753708243, "qparams": { "group_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": 0.0009989275131374598, "qparams": { "group_size": { "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": 0.0010411156108602881, "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": 0.013800338841974735, "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": 0.013800338841974735, "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": 0.012921983376145363, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.010141377337276936, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.00787154957652092, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.00689074257388711, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.005388949997723103, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0037297627422958612, "qparams": { "group_size": { "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": 0.00882683228701353, "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": 0.007767348550260067, "qparams": { "group_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": 0.005930052604526281, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.004374371841549873, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.004425331484526396, "qparams": { "group_size": { "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": 0.004474958870559931, "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": 0.0037234127521514893, "qparams": { "group_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": 0.002670199843123555, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.002360186306759715, "qparams": { "group_size": { "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": 0.002255213214084506, "qparams": { "group_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": 0.0015133855631574988, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.001382838119752705, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0013403905322775245, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.001152324490249157, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0012290537124499679, "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": 0.0012378694955259562, "qparams": { "group_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": 0.0008615900296717882, "qparams": { "group_size": { "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": 0.0009126540389843285, "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": 0.012921983376145363, "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": 0.012921983376145363, "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": 0.07498732209205627, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06368767470121384, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05882773920893669, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.050763972103595734, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03344590589404106, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.028737341985106468, "qparams": { "group_size": { "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": 0.04150170087814331, "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": 0.03804910182952881, "qparams": { "group_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": 0.03507094457745552, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.027272744104266167, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0255386121571064, "qparams": { "group_size": { "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": 0.021200431510806084, "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": 0.018263760954141617, "qparams": { "group_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": 0.016160231083631516, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0156300887465477, "qparams": { "group_size": { "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": 0.010636264458298683, "qparams": { "group_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": 0.008553151041269302, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008352949284017086, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.007367831654846668, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0070158494636416435, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005626037251204252, "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": 0.005775377154350281, "qparams": { "group_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": 0.0048872982151806355, "qparams": { "group_size": { "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": 0.004051154479384422, "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": 0.0156300887465477, "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": 0.0156300887465477, "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": 0.14847548305988312, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08803825080394745, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.054806217551231384, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06469431519508362, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.06351549178361893, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03313765674829483, "qparams": { "group_size": { "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": 0.09087412059307098, "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": 0.08077963441610336, "qparams": { "group_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": 0.0708092749118805, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04072410240769386, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04339105635881424, "qparams": { "group_size": { "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": 0.046769239008426666, "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": 0.03978902846574783, "qparams": { "group_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": 0.032094672322273254, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.030011026188731194, "qparams": { "group_size": { "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": 0.023899046704173088, "qparams": { "group_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": 0.0189785398542881, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018363622948527336, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.015435508452355862, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014152288436889648, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01380126178264618, "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": 0.015311653725802898, "qparams": { "group_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": 0.011483555659651756, "qparams": { "group_size": { "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": 0.012596729211509228, "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": 0.015435508452355862, "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": 0.015311653725802898, "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": 0.1048794686794281, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10011803358793259, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0987236276268959, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09121353179216385, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.046276941895484924, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04501384124159813, "qparams": { "group_size": { "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": 0.05100836977362633, "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": 0.0470396913588047, "qparams": { "group_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": 0.046636663377285004, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0433017760515213, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.042045798152685165, "qparams": { "group_size": { "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": 0.025795526802539825, "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": 0.022427957504987717, "qparams": { "group_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": 0.02210012450814247, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022024741396307945, "qparams": { "group_size": { "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": 0.012935280799865723, "qparams": { "group_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": 0.011836270801723003, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011810198426246643, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011302469298243523, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011253380216658115, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0071821920573711395, "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": 0.00777489272877574, "qparams": { "group_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": 0.0070852236822247505, "qparams": { "group_size": { "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": 0.005971741396933794, "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": 0.012935280799865723, "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": 0.012935280799865723, "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": 0.12809467315673828, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12266608327627182, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11226952075958252, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0567081943154335, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05521427467465401, "qparams": { "group_size": { "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": 0.062333445996046066, "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": 0.057565782219171524, "qparams": { "group_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": 0.05712226778268814, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0531698577105999, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05155787244439125, "qparams": { "group_size": { "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": 0.031236223876476288, "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": 0.027104021981358528, "qparams": { "group_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": 0.026745252311229706, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02665279433131218, "qparams": { "group_size": { "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": 0.01553166750818491, "qparams": { "group_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": 0.013603577390313148, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013574600219726562, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012917081825435162, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012864500284194946, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008074061013758183, "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": 0.007969081401824951, "qparams": { "group_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": 0.007943965494632721, "qparams": { "group_size": { "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": 0.005072641186416149, "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": 0.01553166750818491, "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": 0.01553166750818491, "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": 0.10944453626871109, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09805439412593842, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0857418105006218, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07925485074520111, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05364730954170227, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03784159570932388, "qparams": { "group_size": { "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": 0.0742437094449997, "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": 0.06745034456253052, "qparams": { "group_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": 0.055086974054574966, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.049424052238464355, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04335743188858032, "qparams": { "group_size": { "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": 0.03695453330874443, "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": 0.03995112329721451, "qparams": { "group_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": 0.024291539564728737, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02144867368042469, "qparams": { "group_size": { "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": 0.019069254398345947, "qparams": { "group_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": 0.014221075922250748, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012659364379942417, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013521423563361168, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011783426627516747, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01232221256941557, "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": 0.011923865415155888, "qparams": { "group_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": 0.010013914667069912, "qparams": { "group_size": { "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": 0.008972232230007648, "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": 0.014221075922250748, "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": 0.014221075922250748, "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": 0.03025543875992298, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.027885394170880318, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.026768561452627182, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0240163654088974, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.013644217513501644, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.012663494795560837, "qparams": { "group_size": { "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": 0.01769404485821724, "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": 0.014773204922676086, "qparams": { "group_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": 0.013907965272665024, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.012150258757174015, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.011632815934717655, "qparams": { "group_size": { "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": 0.008977544493973255, "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": 0.007089379709213972, "qparams": { "group_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": 0.00655581196770072, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.006423586048185825, "qparams": { "group_size": { "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": 0.004513085819780827, "qparams": { "group_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": 0.0034528146497905254, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.003390770172700286, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0031646625138819218, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.003080795519053936, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0024103443138301373, "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": 0.0022627294529229403, "qparams": { "group_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": 0.0020656310953199863, "qparams": { "group_size": { "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": 0.0015977482544258237, "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": 0.014773204922676086, "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": 0.014773204922676086, "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": 0.024146614596247673, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.022182609885931015, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.02130400389432907, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.019084647297859192, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.010836167261004448, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.010068726725876331, "qparams": { "group_size": { "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": 0.01286209188401699, "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": 0.011711402796208858, "qparams": { "group_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": 0.011061047203838825, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.009619361720979214, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.009143197908997536, "qparams": { "group_size": { "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": 0.0064904470928013325, "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": 0.005576282739639282, "qparams": { "group_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": 0.005191403906792402, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.005097467452287674, "qparams": { "group_size": { "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": 0.0032528769224882126, "qparams": { "group_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": 0.0027129976078867912, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0026725817006081343, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0024745522532612085, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.002416333183646202, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.001729413983412087, "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": 0.0017439806833863258, "qparams": { "group_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": 0.001586024067364633, "qparams": { "group_size": { "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": 0.0012082575121894479, "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": 0.01286209188401699, "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": 0.01286209188401699, "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": 0.11150060594081879, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10416185110807419, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1018199548125267, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09236536175012589, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05101640895009041, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04854818060994148, "qparams": { "group_size": { "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": 0.05714975669980049, "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": 0.05276293307542801, "qparams": { "group_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": 0.05180966109037399, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04598690941929817, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.043762024492025375, "qparams": { "group_size": { "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": 0.028912387788295746, "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": 0.025093194097280502, "qparams": { "group_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": 0.024314573034644127, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.024133658036589622, "qparams": { "group_size": { "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": 0.014420507475733757, "qparams": { "group_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": 0.012380385771393776, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01232088916003704, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011354221031069756, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011236789636313915, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007440446875989437, "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": 0.0073187341913580894, "qparams": { "group_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": 0.007147862110286951, "qparams": { "group_size": { "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": 0.0045944806188344955, "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": 0.014420507475733757, "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": 0.014420507475733757, "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": 0.15129774808883667, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1201387345790863, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10795219242572784, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08656097948551178, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.06748484820127487, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05542352795600891, "qparams": { "group_size": { "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": 0.08497297763824463, "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": 0.07672315090894699, "qparams": { "group_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": 0.07159145176410675, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.049838561564683914, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04572904109954834, "qparams": { "group_size": { "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": 0.04387909546494484, "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": 0.037649769335985184, "qparams": { "group_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": 0.03358932584524155, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03259576857089996, "qparams": { "group_size": { "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": 0.022594206035137177, "qparams": { "group_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": 0.019298918545246124, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018996836617588997, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01632770337164402, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.015701865777373314, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013343525119125843, "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": 0.014570394530892372, "qparams": { "group_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": 0.012162444181740284, "qparams": { "group_size": { "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": 0.012017550878226757, "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": 0.013343525119125843, "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": 0.013343525119125843, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "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": 0.14215046167373657, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.13578851521015167, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.13390672206878662, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12397031486034393, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.06450054049491882, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06264141201972961, "qparams": { "group_size": { "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": 0.07101243734359741, "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": 0.06553234159946442, "qparams": { "group_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": 0.06505747139453888, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06010470911860466, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.058396223932504654, "qparams": { "group_size": { "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": 0.0360577292740345, "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": 0.031220119446516037, "qparams": { "group_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": 0.030771298334002495, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.030675353482365608, "qparams": { "group_size": { "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": 0.01805778034031391, "qparams": { "group_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": 0.016096843406558037, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01606886461377144, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.015255806967616081, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.015195848420262337, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.009819190949201584, "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": 0.010083694942295551, "qparams": { "group_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": 0.009672419168055058, "qparams": { "group_size": { "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": 0.007233769632875919, "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": 0.015255806967616081, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 0.6 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:8b_32g/0.3:6b_32g/0.6:5b_32g s4", "bpw": 5.725108506944444, "total_bits": 405216768.0, "err": 0.015195848420262337, "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.up_proj", "numel": 70778880, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.1751085069444445, "total_bits": 153951744.0, "err": 0.172312930226326, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.164768785238266, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16259810328483582, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1507827788591385, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07838116586208344, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07615084201097488, "qparams": { "group_size": { "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": 0.08634962886571884, "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": 0.07961016893386841, "qparams": { "group_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": 0.0790674164891243, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07307237386703491, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0709301233291626, "qparams": { "group_size": { "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": 0.0435960553586483, "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": 0.037757664918899536, "qparams": { "group_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": 0.037218280136585236, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.037094246596097946, "qparams": { "group_size": { "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": 0.021756451576948166, "qparams": { "group_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": 0.01899932324886322, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01896934024989605, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017949087545275688, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017869142815470695, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011474319733679295, "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": 0.011210056953132153, "qparams": { "group_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": 0.011298885568976402, "qparams": { "group_size": { "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": 0.007245246320962906, "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": 0.011474319733679295, "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": 0.011474319733679295, "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": 0.16043372452259064, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1440732777118683, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.13742995262145996, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12462581694126129, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07191069424152374, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06572327017784119, "qparams": { "group_size": { "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": 0.08506167680025101, "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": 0.07816919684410095, "qparams": { "group_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": 0.07423346489667892, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06326790899038315, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0608760230243206, "qparams": { "group_size": { "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": 0.04319823905825615, "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": 0.037339214235544205, "qparams": { "group_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": 0.034562237560749054, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03387371078133583, "qparams": { "group_size": { "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": 0.021708209067583084, "qparams": { "group_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": 0.018251460045576096, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01801292784512043, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01660245843231678, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016164733096957207, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011815056204795837, "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": 0.012011352926492691, "qparams": { "group_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": 0.01094065885990858, "qparams": { "group_size": { "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": 0.008590441197156906, "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": 0.011815056204795837, "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": 0.011815056204795837, "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": 0.04330871254205704, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.04021662101149559, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.03873389959335327, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.03493190556764603, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.019773676991462708, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.01845172792673111, "qparams": { "group_size": { "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": 0.023867899551987648, "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": 0.021275624632835388, "qparams": { "group_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": 0.020109251141548157, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.017727218568325043, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.017005031928420067, "qparams": { "group_size": { "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": 0.012243693694472313, "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": 0.01021482516080141, "qparams": { "group_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": 0.009496429935097694, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.009314551018178463, "qparams": { "group_size": { "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": 0.006190046668052673, "qparams": { "group_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": 0.004969624802470207, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.004886413458734751, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.00456984993070364, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.004457038827240467, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.003314275061711669, "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": 0.0032085415441542864, "qparams": { "group_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": 0.002971077337861061, "qparams": { "group_size": { "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": 0.0022309443447738886, "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": 0.012243693694472313, "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": 0.012243693694472313, "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": 0.03489084914326668, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.03237300366163254, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.03114987164735794, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.028066501021385193, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.015890736132860184, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.014845590107142925, "qparams": { "group_size": { "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": 0.018775714561343193, "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": 0.017115497961640358, "qparams": { "group_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": 0.01616685278713703, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.014208918437361717, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.013607771135866642, "qparams": { "group_size": { "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": 0.009493101388216019, "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": 0.008168748579919338, "qparams": { "group_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": 0.00763067277148366, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.007500023115426302, "qparams": { "group_size": { "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": 0.0047574033960700035, "qparams": { "group_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": 0.004007335286587477, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.003950802609324455, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.003683066926896572, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0036039811093360186, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0025427264627069235, "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": 0.0025973718147724867, "qparams": { "group_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": 0.0023584922309964895, "qparams": { "group_size": { "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": 0.001836654031649232, "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": 0.014208918437361717, "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": 0.014208918437361717, "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": 0.1540720909833908, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14457949995994568, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14140436053276062, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12850596010684967, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07111044228076935, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06777268648147583, "qparams": { "group_size": { "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": 0.08006619662046432, "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": 0.07334581017494202, "qparams": { "group_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": 0.07215122878551483, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06422196328639984, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06131464242935181, "qparams": { "group_size": { "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": 0.040559072047472, "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": 0.034929726272821426, "qparams": { "group_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": 0.033916931599378586, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.033663928508758545, "qparams": { "group_size": { "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": 0.020253479480743408, "qparams": { "group_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": 0.01726711541414261, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017191162332892418, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.015860578045248985, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01570536196231842, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.010500354692339897, "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": 0.010191721841692924, "qparams": { "group_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": 0.010104267857968807, "qparams": { "group_size": { "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": 0.006389035377651453, "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": 0.010500354692339897, "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": 0.010500354692339897, "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": 0.1380569487810135, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11522123962640762, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10492502897977829, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08025722205638885, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0627397820353508, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05321507528424263, "qparams": { "group_size": { "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": 0.08295504003763199, "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": 0.07204614579677582, "qparams": { "group_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": 0.06509257853031158, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04733336344361305, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04288752004504204, "qparams": { "group_size": { "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": 0.042290691286325455, "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": 0.03616490215063095, "qparams": { "group_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": 0.03204493969678879, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.030993293970823288, "qparams": { "group_size": { "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": 0.0228141937404871, "qparams": { "group_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": 0.01947779767215252, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019072275608778, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017094086855649948, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01647215150296688, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014302390627563, "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": 0.015612808987498283, "qparams": { "group_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": 0.012950746342539787, "qparams": { "group_size": { "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": 0.013509218581020832, "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": 0.014302390627563, "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": 0.014302390627563, "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": 0.16584928333759308, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15777990221977234, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15534767508506775, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1430993676185608, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07701686769723892, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07438729703426361, "qparams": { "group_size": { "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": 0.0848969966173172, "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": 0.0784548670053482, "qparams": { "group_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": 0.07774702459573746, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07100661844015121, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06851396709680557, "qparams": { "group_size": { "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": 0.043486323207616806, "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": 0.03783460333943367, "qparams": { "group_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": 0.03719086945056915, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03703558072447777, "qparams": { "group_size": { "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": 0.021780656650662422, "qparams": { "group_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": 0.019945837557315826, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019900426268577576, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01882397197186947, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018741898238658905, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012052122503519058, "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": 0.01314886100590229, "qparams": { "group_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": 0.011855969205498695, "qparams": { "group_size": { "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": 0.010104409419000149, "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": 0.012052122503519058, "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": 0.012052122503519058, "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": 0.20030736923217773, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19057154655456543, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18767352402210236, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17290817201137543, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0927043929696083, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08959336578845978, "qparams": { "group_size": { "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": 0.10222076624631882, "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": 0.09447227418422699, "qparams": { "group_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": 0.09365791082382202, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0855054184794426, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08240392059087753, "qparams": { "group_size": { "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": 0.05197454243898392, "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": 0.04505231976509094, "qparams": { "group_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": 0.044276148080825806, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0441075935959816, "qparams": { "group_size": { "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": 0.025949807837605476, "qparams": { "group_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": 0.022901363670825958, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022845925763249397, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021470770239830017, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02135465294122696, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013829305768013, "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": 0.013958575204014778, "qparams": { "group_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": 0.013568496331572533, "qparams": { "group_size": { "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": 0.00954378955066204, "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": 0.013829305768013, "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": 0.013829305768013, "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": 0.03455114737153053, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.03140078857541084, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.018515393137931824, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.01620076224207878, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.010812642052769661, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.009579608216881752, "qparams": { "group_size": { "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": 0.02987922541797161, "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": 0.02600944973528385, "qparams": { "group_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": 0.011305619962513447, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.008927970193326473, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.008459007367491722, "qparams": { "group_size": { "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": 0.0069830333814024925, "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": 0.0060219899751245975, "qparams": { "group_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": 0.005348717328161001, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0055630262941122055, "qparams": { "group_size": { "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": 0.003586934646591544, "qparams": { "group_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": 0.0035955754574388266, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.002934409538283944, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.003291728440672159, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0025527747347950935, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.002783553209155798, "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": 0.002921327482908964, "qparams": { "group_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": 0.0018523100297898054, "qparams": { "group_size": { "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": 0.0017423790413886309, "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": 0.011305619962513447, "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": 0.011305619962513447, "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": 0.057486023753881454, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05301937088370323, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05119858682155609, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.04602786898612976, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.026223646476864815, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.024517731741070747, "qparams": { "group_size": { "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": 0.030285555869340897, "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": 0.027890395373106003, "qparams": { "group_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": 0.026734890416264534, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.023309459909796715, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.022149013355374336, "qparams": { "group_size": { "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": 0.01533686462789774, "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": 0.013285534456372261, "qparams": { "group_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": 0.012548279017210007, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.012372109107673168, "qparams": { "group_size": { "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": 0.0076637426391243935, "qparams": { "group_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": 0.00649535097181797, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006424936931580305, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.005915922112762928, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.005803593900054693, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.004013463389128447, "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": 0.004054196644574404, "qparams": { "group_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": 0.003765638917684555, "qparams": { "group_size": { "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": 0.002723206765949726, "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": 0.01533686462789774, "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": 0.01533686462789774, "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": 0.04990556463599205, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.04598783329129219, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.04431527480483055, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.03984898701310158, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.022755129262804985, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.02122313342988491, "qparams": { "group_size": { "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": 0.026520175859332085, "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": 0.024321366101503372, "qparams": { "group_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": 0.023194899782538414, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.020219966769218445, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.019285332411527634, "qparams": { "group_size": { "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": 0.01342853344976902, "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": 0.011573673225939274, "qparams": { "group_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": 0.01088440977036953, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.010723481886088848, "qparams": { "group_size": { "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": 0.006708718370646238, "qparams": { "group_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": 0.005627221893519163, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0055593084543943405, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.005124042741954327, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.005018381867557764, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.003513956908136606, "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": 0.0035091436002403498, "qparams": { "group_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": 0.0032840941566973925, "qparams": { "group_size": { "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": 0.0023361644707620144, "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": 0.01342853344976902, "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": 0.01342853344976902, "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": 0.16420191526412964, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1533917337656021, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14990776777267456, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1358426958322525, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07600144296884537, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07220117747783661, "qparams": { "group_size": { "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": 0.08518511801958084, "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": 0.07858771085739136, "qparams": { "group_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": 0.07717865705490112, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06833525747060776, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06490814685821533, "qparams": { "group_size": { "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": 0.04316513612866402, "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": 0.03742890805006027, "qparams": { "group_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": 0.036268990486860275, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.035982776433229446, "qparams": { "group_size": { "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": 0.02153824269771576, "qparams": { "group_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": 0.01845891959965229, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01836972124874592, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01688573509454727, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016702894121408463, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011101849377155304, "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": 0.010897837579250336, "qparams": { "group_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": 0.010674430988729, "qparams": { "group_size": { "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": 0.0068075088784098625, "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": 0.011101849377155304, "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": 0.011101849377155304, "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": 0.17949844896793365, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15744026005268097, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14851504564285278, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1196131631731987, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08371719717979431, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07494652271270752, "qparams": { "group_size": { "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": 0.10025729238986969, "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": 0.09110808372497559, "qparams": { "group_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": 0.08626905083656311, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06646627187728882, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05858651548624039, "qparams": { "group_size": { "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": 0.051926884800195694, "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": 0.04453092813491821, "qparams": { "group_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": 0.04115152731537819, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04031107947230339, "qparams": { "group_size": { "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": 0.026338547468185425, "qparams": { "group_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": 0.0226645115762949, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022363267838954926, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01951441913843155, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018997114151716232, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015019647777080536, "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": 0.016002431511878967, "qparams": { "group_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": 0.013993985950946808, "qparams": { "group_size": { "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": 0.01261040661484003, "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": 0.015019647777080536, "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": 0.015019647777080536, "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": 0.17408989369869232, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1650402843952179, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16231517493724823, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.14893795549869537, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0808740109205246, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07783946394920349, "qparams": { "group_size": { "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": 0.08916707336902618, "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": 0.0825156420469284, "qparams": { "group_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": 0.08175243437290192, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07403997331857681, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07094595581293106, "qparams": { "group_size": { "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": 0.045220714062452316, "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": 0.039323970675468445, "qparams": { "group_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": 0.038601018488407135, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03842291235923767, "qparams": { "group_size": { "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": 0.022557126358151436, "qparams": { "group_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": 0.01983720436692238, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019794747233390808, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01847126893699169, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018362510949373245, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011833014898002148, "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": 0.011929784901440144, "qparams": { "group_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": 0.01158815436065197, "qparams": { "group_size": { "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": 0.00793811958283186, "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": 0.011833014898002148, "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": 0.011833014898002148, "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": 0.21067360043525696, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19981220364570618, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1965722143650055, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18039681017398834, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09781256318092346, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09420429915189743, "qparams": { "group_size": { "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": 0.10778894275426865, "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": 0.09977622330188751, "qparams": { "group_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": 0.09889471530914307, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0895802453160286, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08577831089496613, "qparams": { "group_size": { "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": 0.054584141820669174, "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": 0.047459639608860016, "qparams": { "group_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": 0.04658953472971916, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.046376507729291916, "qparams": { "group_size": { "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": 0.02719694748520851, "qparams": { "group_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": 0.023652615025639534, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02360653132200241, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02196880802512169, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0218412596732378, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014042286202311516, "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": 0.01376638188958168, "qparams": { "group_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": 0.01374923624098301, "qparams": { "group_size": { "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": 0.008549295365810394, "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": 0.014042286202311516, "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": 0.014042286202311516, "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": 0.2040105164051056, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18240801990032196, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17424601316452026, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15475478768348694, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09284515678882599, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08471821248531342, "qparams": { "group_size": { "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": 0.10830719769001007, "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": 0.09967831522226334, "qparams": { "group_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": 0.09564696997404099, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07984994351863861, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0754353329539299, "qparams": { "group_size": { "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": 0.05509929358959198, "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": 0.04762896150350571, "qparams": { "group_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": 0.04457453265786171, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.043821316212415695, "qparams": { "group_size": { "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": 0.027597514912486076, "qparams": { "group_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": 0.023301851004362106, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02305229753255844, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020776165649294853, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02028409205377102, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014860262162983418, "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": 0.014944595284759998, "qparams": { "group_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": 0.013883034698665142, "qparams": { "group_size": { "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": 0.010398630984127522, "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": 0.014860262162983418, "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": 0.014860262162983418, "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": 0.06631660461425781, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06114873290061951, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05896655470132828, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.05298031121492386, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.030353378504514694, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.028305286541581154, "qparams": { "group_size": { "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": 0.035136736929416656, "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": 0.03243778645992279, "qparams": { "group_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": 0.030940067023038864, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.026923302561044693, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.02552686259150505, "qparams": { "group_size": { "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": 0.017796626314520836, "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": 0.01549053005874157, "qparams": { "group_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": 0.014558765105903149, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.014333512634038925, "qparams": { "group_size": { "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": 0.00890420749783516, "qparams": { "group_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": 0.007588673382997513, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.007494232151657343, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.006914245430380106, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006771196611225605, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.004695301875472069, "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": 0.004823355469852686, "qparams": { "group_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": 0.004386282525956631, "qparams": { "group_size": { "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": 0.0033230639528483152, "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": 0.01549053005874157, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:5b_32g/0.9:4b_32g s4", "bpw": 4.22529296875, "total_bits": 110763520.0, "err": 0.014558765105903149, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.05654130503535271, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.052100490778684616, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05009734630584717, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.045046351850032806, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.02588750794529915, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.024086344987154007, "qparams": { "group_size": { "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": 0.03038131073117256, "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": 0.027835484594106674, "qparams": { "group_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": 0.026396002620458603, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0229656919836998, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.021902678534388542, "qparams": { "group_size": { "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": 0.015385407954454422, "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": 0.013274047523736954, "qparams": { "group_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": 0.01241080928593874, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.012200313620269299, "qparams": { "group_size": { "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": 0.007697210181504488, "qparams": { "group_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": 0.006444528233259916, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006353827193379402, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.00586804049089551, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.005735238082706928, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.004046664573252201, "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": 0.004075102508068085, "qparams": { "group_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": 0.0037563121877610683, "qparams": { "group_size": { "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": 0.0027560763992369175, "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": 0.015385407954454422, "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": 0.015385407954454422, "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": 0.17567096650600433, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16398940980434418, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1600937843322754, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1449539065361023, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0815177634358406, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07730316370725632, "qparams": { "group_size": { "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": 0.09161058813333511, "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": 0.08442226052284241, "qparams": { "group_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": 0.08279649168252945, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07309997826814651, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0693056508898735, "qparams": { "group_size": { "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": 0.04646537825465202, "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": 0.040232978761196136, "qparams": { "group_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": 0.03892156854271889, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.038609206676483154, "qparams": { "group_size": { "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": 0.02318304218351841, "qparams": { "group_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": 0.019819628447294235, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019719282165169716, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01808953657746315, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017884092405438423, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011967850849032402, "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": 0.011720314621925354, "qparams": { "group_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": 0.011474831029772758, "qparams": { "group_size": { "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": 0.007332487031817436, "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": 0.011967850849032402, "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": 0.011967850849032402, "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": 0.18391479551792145, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.159373939037323, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15082195401191711, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1232982948422432, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08510881662368774, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07578008621931076, "qparams": { "group_size": { "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": 0.1006983295083046, "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": 0.09159854054450989, "qparams": { "group_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": 0.08812689036130905, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06827206164598465, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0600595660507679, "qparams": { "group_size": { "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": 0.05196385830640793, "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": 0.044745925813913345, "qparams": { "group_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": 0.0418071523308754, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04106603562831879, "qparams": { "group_size": { "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": 0.026395728811621666, "qparams": { "group_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": 0.02310175821185112, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022885331884026527, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020056670531630516, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019607078284025192, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015013090334832668, "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": 0.016323445364832878, "qparams": { "group_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": 0.01412767730653286, "qparams": { "group_size": { "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": 0.012976562604308128, "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": 0.015013090334832668, "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": 0.015013090334832668, "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": 0.16769473254680634, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1585605889558792, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15583783388137817, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1427205502986908, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07814617455005646, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07507283985614777, "qparams": { "group_size": { "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": 0.08639676868915558, "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": 0.07992513477802277, "qparams": { "group_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": 0.07906932383775711, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07126820832490921, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06821408122777939, "qparams": { "group_size": { "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": 0.0439726822078228, "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": 0.03822141885757446, "qparams": { "group_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": 0.03742853179574013, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.037250347435474396, "qparams": { "group_size": { "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": 0.02195991575717926, "qparams": { "group_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": 0.019444627687335014, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019391460344195366, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018078574910759926, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01796684041619301, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011662664823234081, "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": 0.011991957202553749, "qparams": { "group_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": 0.011409273371100426, "qparams": { "group_size": { "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": 0.008326204493641853, "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": 0.011662664823234081, "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": 0.011662664823234081, "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": 0.21750330924987793, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20575878024101257, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2022586166858673, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18523621559143066, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10130727291107178, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09735129773616791, "qparams": { "group_size": { "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": 0.11190547049045563, "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": 0.10352659970521927, "qparams": { "group_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": 0.10251770913600922, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09242793172597885, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0883280485868454, "qparams": { "group_size": { "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": 0.056696683168411255, "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": 0.049310196191072464, "qparams": { "group_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": 0.048312168568372726, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04807545244693756, "qparams": { "group_size": { "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": 0.028275562450289726, "qparams": { "group_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": 0.024553194642066956, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02448784001171589, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0227286908775568, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022580068558454514, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014621028676629066, "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": 0.014327104203402996, "qparams": { "group_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": 0.014283438213169575, "qparams": { "group_size": { "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": 0.008930827490985394, "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": 0.014621028676629066, "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": 0.014621028676629066, "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": 0.20960582792758942, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18703262507915497, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17824453115463257, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15784084796905518, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09553647041320801, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08686435967683792, "qparams": { "group_size": { "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": 0.11215080320835114, "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": 0.10317990928888321, "qparams": { "group_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": 0.09851346909999847, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0819660872220993, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07725952565670013, "qparams": { "group_size": { "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": 0.05718836188316345, "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": 0.049425940960645676, "qparams": { "group_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": 0.046011101454496384, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04518240690231323, "qparams": { "group_size": { "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": 0.028739197179675102, "qparams": { "group_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": 0.02426089346408844, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02397198975086212, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021646184846758842, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021091697737574577, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015669429674744606, "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": 0.015861522406339645, "qparams": { "group_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": 0.014596120454370975, "qparams": { "group_size": { "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": 0.011307422071695328, "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": 0.014596120454370975, "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": 0.014596120454370975, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.08263138681650162, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0769897997379303, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07478007674217224, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06763485074043274, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03816717863082886, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.036015406250953674, "qparams": { "group_size": { "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": 0.043474312871694565, "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": 0.040148038417100906, "qparams": { "group_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": 0.038775525987148285, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.034211769700050354, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.032510895282030106, "qparams": { "group_size": { "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": 0.022059597074985504, "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": 0.019186336547136307, "qparams": { "group_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": 0.018307805061340332, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.01810351572930813, "qparams": { "group_size": { "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": 0.011034375056624413, "qparams": { "group_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": 0.009530781768262386, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009449823759496212, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0087475860491395, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.00861752312630415, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005830992944538593, "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": 0.0059850807301700115, "qparams": { "group_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": 0.005541675724089146, "qparams": { "group_size": { "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": 0.004135757684707642, "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": 0.011034375056624413, "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": 0.011034375056624413, "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": 0.06749206781387329, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06290165334939957, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06098015978932381, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.05521443858742714, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.031189044937491417, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.029368124902248383, "qparams": { "group_size": { "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": 0.03597268834710121, "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": 0.03299522399902344, "qparams": { "group_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": 0.03169631212949753, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.027985962107777596, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.026735913008451462, "qparams": { "group_size": { "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": 0.018244437873363495, "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": 0.015744579955935478, "qparams": { "group_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": 0.014944027177989483, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.014751414768397808, "qparams": { "group_size": { "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": 0.009123657830059528, "qparams": { "group_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": 0.0077255950309336185, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.007645983714610338, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.007088365964591503, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006966865621507168, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00478371512144804, "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": 0.0047981315292418, "qparams": { "group_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": 0.004511734936386347, "qparams": { "group_size": { "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": 0.0032126745209097862, "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": 0.014944027177989483, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 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": 0.014751414768397808, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 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": 0.19031234085559845, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.178382009267807, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17453064024448395, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15849456191062927, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08855656534433365, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08432173728942871, "qparams": { "group_size": { "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": 0.09911730885505676, "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": 0.09134015440940857, "qparams": { "group_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": 0.08984105288982391, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07977257668972015, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07576321810483932, "qparams": { "group_size": { "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": 0.050248634070158005, "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": 0.04353207349777222, "qparams": { "group_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": 0.04226510599255562, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.041957251727581024, "qparams": { "group_size": { "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": 0.02504747174680233, "qparams": { "group_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": 0.02147826738655567, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021379463374614716, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01966819353401661, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01946972869336605, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012882007285952568, "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": 0.0126060014590621, "qparams": { "group_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": 0.012405951507389545, "qparams": { "group_size": { "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": 0.007808849681168795, "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": 0.012882007285952568, "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": 0.012882007285952568, "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": 0.17534226179122925, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1524054855108261, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1430620402097702, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11853646486997604, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0812138095498085, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07176144421100616, "qparams": { "group_size": { "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": 0.0980842262506485, "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": 0.08918411284685135, "qparams": { "group_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": 0.08384811878204346, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06380020081996918, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.059720106422901154, "qparams": { "group_size": { "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": 0.05062625929713249, "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": 0.04341164603829384, "qparams": { "group_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": 0.0397997610270977, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03888535499572754, "qparams": { "group_size": { "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": 0.025575250387191772, "qparams": { "group_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": 0.021766582503914833, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02143951691687107, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018579361960291862, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01800806261599064, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01444377563893795, "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": 0.01529706735163927, "qparams": { "group_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": 0.013339297845959663, "qparams": { "group_size": { "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": 0.011874504387378693, "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": 0.01444377563893795, "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": 0.01444377563893795, "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": 0.15739145874977112, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14858362078666687, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14590565860271454, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13341347873210907, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07352536171674728, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0704871118068695, "qparams": { "group_size": { "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": 0.08131464570760727, "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": 0.07525917142629623, "qparams": { "group_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": 0.0744265466928482, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06686164438724518, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06377572566270828, "qparams": { "group_size": { "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": 0.04129713401198387, "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": 0.035929229110479355, "qparams": { "group_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": 0.035142213106155396, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03496207669377327, "qparams": { "group_size": { "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": 0.02060500904917717, "qparams": { "group_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": 0.018061580136418343, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01801333762705326, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.016717348247766495, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01660318486392498, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.010785351507365704, "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": 0.010866506025195122, "qparams": { "group_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": 0.010529423132538795, "qparams": { "group_size": { "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": 0.007200127467513084, "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": 0.010785351507365704, "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": 0.010785351507365704, "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": 0.21762898564338684, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20561067759990692, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2019951194524765, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1847510039806366, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10162442922592163, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09749191254377365, "qparams": { "group_size": { "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": 0.11236221343278885, "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": 0.10394348949193954, "qparams": { "group_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": 0.10286252945661545, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09246955066919327, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08831362426280975, "qparams": { "group_size": { "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": 0.05705024302005768, "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": 0.04954652488231659, "qparams": { "group_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": 0.04850394278764725, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04824790358543396, "qparams": { "group_size": { "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": 0.028458377346396446, "qparams": { "group_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": 0.024682670831680298, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024608131498098373, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0227943304926157, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022639404982328415, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014781338162720203, "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": 0.014465782791376114, "qparams": { "group_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": 0.014423400163650513, "qparams": { "group_size": { "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": 0.009071163833141327, "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": 0.014781338162720203, "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": 0.014781338162720203, "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": 0.20912273228168488, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.187406986951828, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1789555847644806, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15905125439167023, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0954720601439476, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08715903759002686, "qparams": { "group_size": { "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": 0.11196266114711761, "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": 0.10296901315450668, "qparams": { "group_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": 0.0983693078160286, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0823904424905777, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07778695970773697, "qparams": { "group_size": { "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": 0.05701151862740517, "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": 0.049339406192302704, "qparams": { "group_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": 0.04596063867211342, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04513132944703102, "qparams": { "group_size": { "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": 0.028612229973077774, "qparams": { "group_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": 0.024211199954152107, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023918529972434044, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021684812381863594, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02114252559840679, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015537399798631668, "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": 0.01579062081873417, "qparams": { "group_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": 0.014476464129984379, "qparams": { "group_size": { "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": 0.011219371110200882, "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": 0.015537399798631668, "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": 0.015537399798631668, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "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": 0.07997661828994751, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07466303557157516, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07249946892261505, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06549984216690063, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03696850687265396, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03490719199180603, "qparams": { "group_size": { "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": 0.042158495634794235, "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": 0.038890935480594635, "qparams": { "group_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": 0.03755975142121315, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03316167742013931, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03147264942526817, "qparams": { "group_size": { "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": 0.021375449374318123, "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": 0.018563931807875633, "qparams": { "group_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": 0.017701800912618637, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.017493989318609238, "qparams": { "group_size": { "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": 0.010686873458325863, "qparams": { "group_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": 0.0091421939432621, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009061228483915329, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008378788828849792, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008249114267528057, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00560764642432332, "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": 0.005650678649544716, "qparams": { "group_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": 0.005313404835760593, "qparams": { "group_size": { "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": 0.0037768555339425802, "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": 0.010686873458325863, "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": 0.010686873458325863, "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": 0.06650491058826447, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0621153898537159, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06017699092626572, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.05443815886974335, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.030780967324972153, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.028986047953367233, "qparams": { "group_size": { "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": 0.03540593013167381, "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": 0.03262975811958313, "qparams": { "group_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": 0.03125269338488579, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.027612928301095963, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.026319555938243866, "qparams": { "group_size": { "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": 0.017957111820578575, "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": 0.015564782544970512, "qparams": { "group_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": 0.014736847952008247, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.014535808935761452, "qparams": { "group_size": { "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": 0.008968534879386425, "qparams": { "group_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": 0.007603778969496489, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.007519189268350601, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.006972406059503555, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.006845106370747089, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00469293724745512, "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": 0.0047085462138056755, "qparams": { "group_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": 0.004416426178067923, "qparams": { "group_size": { "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": 0.0031209741719067097, "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": 0.015564782544970512, "qparams": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1.0 ], "scale_bits": 4 } }, "best_option": { "desc": "0.1:6b_32g/0.9:4b_32g s4", "bpw": 4.32529296875, "total_bits": 113384960.00000001, "err": 0.014535808935761452, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 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": 0.18831214308738708, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1765647977590561, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1727255880832672, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15669409930706024, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0876869410276413, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08350197970867157, "qparams": { "group_size": { "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": 0.09841330349445343, "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": 0.09054971486330032, "qparams": { "group_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": 0.08894455432891846, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07896939665079117, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07499624043703079, "qparams": { "group_size": { "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": 0.049944937229156494, "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": 0.04316340759396553, "qparams": { "group_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": 0.04186885431408882, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.041560012847185135, "qparams": { "group_size": { "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": 0.02491474710404873, "qparams": { "group_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": 0.021279828622937202, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021179502829909325, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01947356015443802, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019275620579719543, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012821688316762447, "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": 0.01249830424785614, "qparams": { "group_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": 0.012308752164244652, "qparams": { "group_size": { "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": 0.007725475821644068, "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": 0.012821688316762447, "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": 0.012821688316762447, "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": 0.20249402523040771, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16953502595424652, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15787449479103088, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12953852117061615, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09245772659778595, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.079719178378582, "qparams": { "group_size": { "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": 0.11100053787231445, "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": 0.1011943370103836, "qparams": { "group_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": 0.09653200209140778, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0713387206196785, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06536207348108292, "qparams": { "group_size": { "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": 0.057226646691560745, "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": 0.04951636493206024, "qparams": { "group_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": 0.045574840158224106, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04464035481214523, "qparams": { "group_size": { "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": 0.029115378856658936, "qparams": { "group_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": 0.025528842583298683, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025259600952267647, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02184218168258667, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021240826696157455, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016762983053922653, "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": 0.01849951595067978, "qparams": { "group_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": 0.01558807771652937, "qparams": { "group_size": { "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": 0.01495568547397852, "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": 0.01558807771652937, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, "best_option": { "desc": "1.0:8b_32g s4", "bpw": 8.12529296875, "total_bits": 212999679.99999997, "err": 0.01495568547397852, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "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": 0.15417391061782837, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14554615318775177, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14287330210208893, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13046085834503174, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07210931181907654, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06908639520406723, "qparams": { "group_size": { "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": 0.07993580400943756, "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": 0.07394130527973175, "qparams": { "group_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": 0.07302116602659225, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06554978340864182, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06252230703830719, "qparams": { "group_size": { "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": 0.040640607476234436, "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": 0.035344239324331284, "qparams": { "group_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": 0.03453007712960243, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.034333691000938416, "qparams": { "group_size": { "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": 0.020305929705500603, "qparams": { "group_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": 0.017836201936006546, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01778326742351055, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.016517776995897293, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016397882252931595, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.010716243647038937, "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": 0.010873463936150074, "qparams": { "group_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": 0.01044623926281929, "qparams": { "group_size": { "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": 0.007373521104454994, "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": 0.010716243647038937, "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": 0.010716243647038937, "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": 0.21840687096118927, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20631326735019684, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20259694755077362, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18519392609596252, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10230053961277008, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09810744971036911, "qparams": { "group_size": { "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": 0.11324860155582428, "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": 0.1047239601612091, "qparams": { "group_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": 0.1035500317811966, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09302067756652832, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08876092731952667, "qparams": { "group_size": { "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": 0.057632796466350555, "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": 0.05004275217652321, "qparams": { "group_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": 0.04894537478685379, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048687633126974106, "qparams": { "group_size": { "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": 0.02877134457230568, "qparams": { "group_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": 0.02524176426231861, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025177353993058205, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023377664387226105, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02321792207658291, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015150469727814198, "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": 0.015313557349145412, "qparams": { "group_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": 0.014779928140342236, "qparams": { "group_size": { "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": 0.010310078039765358, "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": 0.015150469727814198, "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": 0.014779928140342236, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.20362979173660278, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18128304183483124, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17273175716400146, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1530074179172516, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09275951236486435, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08422926068305969, "qparams": { "group_size": { "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": 0.1102830022573471, "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": 0.1003766879439354, "qparams": { "group_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": 0.09573273360729218, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07965688407421112, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07524874061346054, "qparams": { "group_size": { "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": 0.05598774924874306, "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": 0.04829439893364906, "qparams": { "group_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": 0.04489995539188385, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.044065698981285095, "qparams": { "group_size": { "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": 0.028371021151542664, "qparams": { "group_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": 0.024058539420366287, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02376674674451351, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021589230746030807, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021043498069047928, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015934791415929794, "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": 0.016208592802286148, "qparams": { "group_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": 0.01487103383988142, "qparams": { "group_size": { "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": 0.012058530002832413, "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": 0.01487103383988142, "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": 0.01487103383988142, "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": 0.0913994237780571, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08533370494842529, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08296939730644226, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07506226748228073, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.042279090732336044, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.039969105273485184, "qparams": { "group_size": { "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": 0.04814298450946808, "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": 0.04440569877624512, "qparams": { "group_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": 0.04293321073055267, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03794577717781067, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03610333055257797, "qparams": { "group_size": { "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": 0.02439308352768421, "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": 0.021183257922530174, "qparams": { "group_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": 0.020243018865585327, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.020018817856907845, "qparams": { "group_size": { "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": 0.012198574841022491, "qparams": { "group_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": 0.010457418859004974, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010365537367761135, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009589292109012604, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009444687515497208, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006395319476723671, "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": 0.006453932728618383, "qparams": { "group_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": 0.006077175959944725, "qparams": { "group_size": { "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": 0.004314494784921408, "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": 0.012198574841022491, "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": 0.012198574841022491, "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": 0.07388726621866226, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06903160363435745, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06692677736282349, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06056680902838707, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.034185826778411865, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.032238248735666275, "qparams": { "group_size": { "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": 0.0392426922917366, "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": 0.0361800454556942, "qparams": { "group_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": 0.03470661863684654, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.030710995197296143, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.029292307794094086, "qparams": { "group_size": { "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": 0.019886188209056854, "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": 0.01725573092699051, "qparams": { "group_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": 0.01637331023812294, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.01616102270781994, "qparams": { "group_size": { "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": 0.009941917844116688, "qparams": { "group_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": 0.008446836844086647, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008356206119060516, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.007747523952275515, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.007612589746713638, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005211750511080027, "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": 0.005216646008193493, "qparams": { "group_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": 0.004923857282847166, "qparams": { "group_size": { "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": 0.0034546246752142906, "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": 0.009941917844116688, "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": 0.009941917844116688, "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": 0.19911299645900726, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18681414425373077, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18278877437114716, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1658487319946289, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09278092533349991, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08833073079586029, "qparams": { "group_size": { "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": 0.10433278232812881, "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": 0.09575442224740982, "qparams": { "group_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": 0.09405618906021118, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08354464918375015, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0795566737651825, "qparams": { "group_size": { "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": 0.05293537676334381, "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": 0.04563726484775543, "qparams": { "group_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": 0.04427503049373627, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04395756497979164, "qparams": { "group_size": { "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": 0.026409106329083443, "qparams": { "group_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": 0.02250373736023903, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022409403696656227, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020602943375706673, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02039877325296402, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01360857393592596, "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": 0.013209995813667774, "qparams": { "group_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": 0.013053997419774532, "qparams": { "group_size": { "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": 0.008184962905943394, "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": 0.01360857393592596, "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": 0.01360857393592596, "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": 0.19153155386447906, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16917380690574646, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1616418957710266, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1371663212776184, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0889260545372963, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08073215186595917, "qparams": { "group_size": { "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": 0.10289756953716278, "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": 0.09455319494009018, "qparams": { "group_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": 0.0916893482208252, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07329671084880829, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0662662535905838, "qparams": { "group_size": { "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": 0.05336280167102814, "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": 0.04612269252538681, "qparams": { "group_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": 0.043629854917526245, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04301374405622482, "qparams": { "group_size": { "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": 0.02709147147834301, "qparams": { "group_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": 0.02393452264368534, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023766782134771347, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021054822951555252, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02067861333489418, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015488790348172188, "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": 0.016665898263454437, "qparams": { "group_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": 0.014758313074707985, "qparams": { "group_size": { "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": 0.013177549466490746, "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": 0.015488790348172188, "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": 0.014758313074707985, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.1591685265302658, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1502242535352707, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14738735556602478, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13458284735679626, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07466183602809906, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0715077817440033, "qparams": { "group_size": { "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": 0.08282018452882767, "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": 0.07656591385602951, "qparams": { "group_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": 0.07559245824813843, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06781604886054993, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06475263833999634, "qparams": { "group_size": { "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": 0.0422581322491169, "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": 0.036701761186122894, "qparams": { "group_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": 0.03584524989128113, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03564897179603577, "qparams": { "group_size": { "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": 0.021132497116923332, "qparams": { "group_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": 0.01869548112154007, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018635503947734833, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017338624224066734, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017219185829162598, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01129672396928072, "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": 0.011654000729322433, "qparams": { "group_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": 0.011014983989298344, "qparams": { "group_size": { "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": 0.008214421570301056, "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": 0.01129672396928072, "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": 0.01129672396928072, "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": 0.21926483511924744, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20709599554538727, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20335935056209564, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18578849732875824, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10258948802947998, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09834897518157959, "qparams": { "group_size": { "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": 0.11391996592283249, "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": 0.10509218275547028, "qparams": { "group_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": 0.10385587811470032, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09324363619089127, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08906936645507812, "qparams": { "group_size": { "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": 0.057798098772764206, "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": 0.05013773590326309, "qparams": { "group_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": 0.049006544053554535, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048747748136520386, "qparams": { "group_size": { "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": 0.02888457477092743, "qparams": { "group_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": 0.025008130818605423, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024927450343966484, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02308421954512596, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02291759103536606, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015134995803236961, "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": 0.014752657152712345, "qparams": { "group_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": 0.014756331220269203, "qparams": { "group_size": { "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": 0.009373093023896217, "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": 0.015134995803236961, "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": 0.014752657152712345, "qparams": { "group_size": { "6": 32 }, "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": 0.21666887402534485, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19285283982753754, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1835518479347229, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16253456473350525, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09909956902265549, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08975227922201157, "qparams": { "group_size": { "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": 0.11736395955085754, "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": 0.10726352035999298, "qparams": { "group_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": 0.10224354267120361, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08468431979417801, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07985924184322357, "qparams": { "group_size": { "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": 0.05972813814878464, "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": 0.051605597138404846, "qparams": { "group_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": 0.04794136434793472, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.047033779323101044, "qparams": { "group_size": { "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": 0.030125968158245087, "qparams": { "group_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": 0.025622757151722908, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025304317474365234, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022889679297804832, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02230292372405529, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01666833646595478, "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": 0.017186392098665237, "qparams": { "group_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": 0.01550954021513462, "qparams": { "group_size": { "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": 0.012704921886324883, "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": 0.01550954021513462, "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": 0.01550954021513462, "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": 0.09823460876941681, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0917985588312149, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08923868089914322, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08084531873464584, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.045592647045850754, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.043111495673656464, "qparams": { "group_size": { "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": 0.051838066428899765, "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": 0.04785216972231865, "qparams": { "group_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": 0.04628641530871391, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.040968816727399826, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.039000604301691055, "qparams": { "group_size": { "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": 0.026376958936452866, "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": 0.022959159687161446, "qparams": { "group_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": 0.021946461871266365, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021697871387004852, "qparams": { "group_size": { "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": 0.013223973102867603, "qparams": { "group_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": 0.011528156697750092, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011436319909989834, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010627125389873981, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010479030199348927, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0070502576418221, "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": 0.007384344004094601, "qparams": { "group_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": 0.006711822468787432, "qparams": { "group_size": { "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": 0.005282258149236441, "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": 0.013223973102867603, "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": 0.013223973102867603, "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": 0.08293703198432922, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07759023457765579, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07533261179924011, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06828731298446655, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03849683701992035, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03636477142572403, "qparams": { "group_size": { "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": 0.044048890471458435, "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": 0.0405893437564373, "qparams": { "group_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": 0.039074186235666275, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03461996093392372, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.033006928861141205, "qparams": { "group_size": { "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": 0.022363873198628426, "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": 0.01938590407371521, "qparams": { "group_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": 0.018445847555994987, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.01821933314204216, "qparams": { "group_size": { "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": 0.011181461624801159, "qparams": { "group_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": 0.009535547345876694, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009443591348826885, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008759214542806149, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008618964813649654, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005876001436263323, "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": 0.005905920173972845, "qparams": { "group_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": 0.005562817677855492, "qparams": { "group_size": { "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": 0.003957738634198904, "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": 0.011181461624801159, "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": 0.011181461624801159, "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": 0.19855834543704987, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18642769753932953, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18219029903411865, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1653296798467636, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09259714186191559, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08809898048639297, "qparams": { "group_size": { "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": 0.10405239462852478, "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": 0.09581455588340759, "qparams": { "group_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": 0.09388623386621475, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08343998342752457, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07933978736400604, "qparams": { "group_size": { "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": 0.05283971503376961, "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": 0.04570837691426277, "qparams": { "group_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": 0.044224921613931656, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.043873220682144165, "qparams": { "group_size": { "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": 0.02635665237903595, "qparams": { "group_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": 0.022532377392053604, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022409670054912567, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020657559856772423, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020431000739336014, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013639788143336773, "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": 0.013336271978914738, "qparams": { "group_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": 0.013090052641928196, "qparams": { "group_size": { "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": 0.008369720540940762, "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": 0.013639788143336773, "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": 0.013639788143336773, "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": 0.20082610845565796, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1721791923046112, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16102972626686096, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13068680465221405, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09267229586839676, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08165508508682251, "qparams": { "group_size": { "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": 0.11178670823574066, "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": 0.10203583538532257, "qparams": { "group_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": 0.09605756402015686, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07332175225019455, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06470084935426712, "qparams": { "group_size": { "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": 0.0578218474984169, "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": 0.050046615302562714, "qparams": { "group_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": 0.04584823548793793, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04480186477303505, "qparams": { "group_size": { "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": 0.029515191912651062, "qparams": { "group_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": 0.025818176567554474, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025465548038482666, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02244570292532444, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021821459755301476, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017147952690720558, "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": 0.018877128139138222, "qparams": { "group_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": 0.015913797542452812, "qparams": { "group_size": { "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": 0.015347938984632492, "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": 0.015347938984632492, "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": 0.015347938984632492, "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": 0.15860144793987274, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14967213571071625, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14686338603496552, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1341562569141388, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07438428699970245, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07123441249132156, "qparams": { "group_size": { "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": 0.08248969167470932, "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": 0.07632514834403992, "qparams": { "group_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": 0.0753384530544281, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06754317134618759, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06445638090372086, "qparams": { "group_size": { "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": 0.042035896331071854, "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": 0.036572568118572235, "qparams": { "group_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": 0.03570002317428589, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.035491958260536194, "qparams": { "group_size": { "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": 0.021001318469643593, "qparams": { "group_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": 0.01857399381697178, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01851402223110199, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017213748767971992, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01708832010626793, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011139974929392338, "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": 0.01152110192924738, "qparams": { "group_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": 0.010852634906768799, "qparams": { "group_size": { "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": 0.008051999844610691, "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": 0.011139974929392338, "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": 0.011139974929392338, "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": 0.2197127491235733, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20740187168121338, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20365794003009796, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1859809011220932, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10284675657749176, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09854689985513687, "qparams": { "group_size": { "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": 0.1141095757484436, "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": 0.10539098083972931, "qparams": { "group_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": 0.10413561761379242, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09342984855175018, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08912738412618637, "qparams": { "group_size": { "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": 0.05789988487958908, "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": 0.05027507245540619, "qparams": { "group_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": 0.049127060920000076, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048853375017642975, "qparams": { "group_size": { "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": 0.028884459286928177, "qparams": { "group_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": 0.025024916976690292, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0249472763389349, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023088490590453148, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022913383319973946, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015009934082627296, "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": 0.014720391482114792, "qparams": { "group_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": 0.014616211876273155, "qparams": { "group_size": { "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": 0.009281788021326065, "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": 0.015009934082627296, "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": 0.014616211876273155, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2218710482120514, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19836315512657166, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18914662301540375, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16773588955402374, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10167787224054337, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09251505136489868, "qparams": { "group_size": { "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": 0.1197233647108078, "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": 0.10983539372682571, "qparams": { "group_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": 0.10482349991798401, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08736130595207214, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08248469233512878, "qparams": { "group_size": { "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": 0.061259906738996506, "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": 0.05293148756027222, "qparams": { "group_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": 0.049246031790971756, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04834802448749542, "qparams": { "group_size": { "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": 0.030888719484210014, "qparams": { "group_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": 0.026439543813467026, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026127537712454796, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023736225441098213, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02317112125456333, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017178842797875404, "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": 0.017852934077382088, "qparams": { "group_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": 0.016025377437472343, "qparams": { "group_size": { "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": 0.013377208262681961, "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": 0.013377208262681961, "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": 0.013377208262681961, "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": 0.1034196987748146, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09662704914808273, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09406889975070953, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08509764075279236, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04793822392821312, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0453791506588459, "qparams": { "group_size": { "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": 0.05435623973608017, "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": 0.0502038411796093, "qparams": { "group_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": 0.048668600618839264, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04306980222463608, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04090934619307518, "qparams": { "group_size": { "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": 0.027612991631031036, "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": 0.024013206362724304, "qparams": { "group_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": 0.02299216017127037, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022749101743102074, "qparams": { "group_size": { "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": 0.013811349868774414, "qparams": { "group_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": 0.011936329305171967, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011840368621051311, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010970587842166424, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010817532427608967, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0072769662365317345, "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": 0.00744676124304533, "qparams": { "group_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": 0.00693373940885067, "qparams": { "group_size": { "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": 0.005100132431834936, "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": 0.013811349868774414, "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": 0.013811349868774414, "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": 0.08333383500576019, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07792524248361588, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07563605159521103, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06845735758543015, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.038627028465270996, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.036447808146476746, "qparams": { "group_size": { "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": 0.04419469088315964, "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": 0.040805548429489136, "qparams": { "group_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": 0.03919230401515961, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03472518548369408, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.033043988049030304, "qparams": { "group_size": { "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": 0.022383319213986397, "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": 0.019472256302833557, "qparams": { "group_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": 0.01849748194217682, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.018264858052134514, "qparams": { "group_size": { "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": 0.011193000711500645, "qparams": { "group_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": 0.009558453224599361, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009457592852413654, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008777927607297897, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008630163036286831, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00587055878713727, "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": 0.005922368261963129, "qparams": { "group_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": 0.005558731034398079, "qparams": { "group_size": { "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": 0.003953054081648588, "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": 0.011193000711500645, "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": 0.011193000711500645, "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": 0.19962167739868164, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18716590106487274, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18286313116550446, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16573797166347504, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09297686815261841, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08839292824268341, "qparams": { "group_size": { "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": 0.10455913841724396, "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": 0.09636907279491425, "qparams": { "group_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": 0.09428349882364273, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08373631536960602, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07950717210769653, "qparams": { "group_size": { "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": 0.053051698952913284, "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": 0.04596449434757233, "qparams": { "group_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": 0.04441925883293152, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.044052854180336, "qparams": { "group_size": { "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": 0.026472045108675957, "qparams": { "group_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": 0.022657115012407303, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022523587569594383, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020761506631970406, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020523976534605026, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013701719231903553, "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": 0.013462605886161327, "qparams": { "group_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": 0.013137217611074448, "qparams": { "group_size": { "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": 0.008502848446369171, "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": 0.013701719231903553, "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": 0.013701719231903553, "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": 0.21578755974769592, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19596411287784576, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18733516335487366, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16047459840774536, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10143311321735382, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09328560531139374, "qparams": { "group_size": { "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": 0.1190950870513916, "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": 0.10932518541812897, "qparams": { "group_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": 0.10361930727958679, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08615009486675262, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07830455899238586, "qparams": { "group_size": { "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": 0.06156498193740845, "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": 0.053426869213581085, "qparams": { "group_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": 0.0497903898358345, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04889819398522377, "qparams": { "group_size": { "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": 0.031174583360552788, "qparams": { "group_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": 0.02748112380504608, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.027135251089930534, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024682706221938133, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024166366085410118, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017683694139122963, "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": 0.019393406808376312, "qparams": { "group_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": 0.016624795272946358, "qparams": { "group_size": { "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": 0.015395582653582096, "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": 0.015395582653582096, "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": 0.015395582653582096, "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": 0.15804092586040497, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14891625940799713, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14604060351848602, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1331639587879181, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07416924089193344, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07096115499734879, "qparams": { "group_size": { "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": 0.08243358880281448, "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": 0.07615090906620026, "qparams": { "group_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": 0.0751262903213501, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0672258511185646, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06408466398715973, "qparams": { "group_size": { "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": 0.042110901325941086, "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": 0.036575328558683395, "qparams": { "group_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": 0.035686880350112915, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.035468265414237976, "qparams": { "group_size": { "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": 0.021053258329629898, "qparams": { "group_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": 0.01870736852288246, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01864480972290039, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017343366518616676, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017216069623827934, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011312459595501423, "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": 0.011808428913354874, "qparams": { "group_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": 0.011019881814718246, "qparams": { "group_size": { "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": 0.008475406095385551, "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": 0.011312459595501423, "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": 0.011312459595501423, "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": 0.21991916000843048, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20739591121673584, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2035297006368637, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18565012514591217, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10289251804351807, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09849321842193604, "qparams": { "group_size": { "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": 0.11429184675216675, "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": 0.10548626631498337, "qparams": { "group_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": 0.10420230031013489, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09325564652681351, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0889580175280571, "qparams": { "group_size": { "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": 0.057916030287742615, "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": 0.05033189803361893, "qparams": { "group_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": 0.04916216805577278, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048881981521844864, "qparams": { "group_size": { "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": 0.028945207595825195, "qparams": { "group_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": 0.025095654651522636, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02501457929611206, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023119648918509483, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022943878546357155, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01511488389223814, "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": 0.014834437519311905, "qparams": { "group_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": 0.014717847108840942, "qparams": { "group_size": { "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": 0.009454570710659027, "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": 0.01511488389223814, "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": 0.014717847108840942, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.22952356934547424, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20600803196430206, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1970595121383667, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17484092712402344, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10534843802452087, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0963919386267662, "qparams": { "group_size": { "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": 0.12269645929336548, "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": 0.11294776201248169, "qparams": { "group_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": 0.10841536521911621, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09071540087461472, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08537622541189194, "qparams": { "group_size": { "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": 0.06274347007274628, "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": 0.054299768060445786, "qparams": { "group_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": 0.05088581517338753, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05005861073732376, "qparams": { "group_size": { "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": 0.03162232041358948, "qparams": { "group_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": 0.027077365666627884, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026787158101797104, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02428571879863739, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023734496906399727, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017527395859360695, "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": 0.017944928258657455, "qparams": { "group_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": 0.016492310911417007, "qparams": { "group_size": { "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": 0.013171046040952206, "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": 0.013171046040952206, "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": 0.013171046040952206, "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": 0.10323838889598846, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09651362895965576, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09384023398160934, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0848458781838417, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04786819964647293, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0452754944562912, "qparams": { "group_size": { "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": 0.05437334626913071, "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": 0.0502186194062233, "qparams": { "group_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": 0.048583582043647766, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0429847426712513, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04081220552325249, "qparams": { "group_size": { "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": 0.027589203789830208, "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": 0.023996377363801003, "qparams": { "group_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": 0.022929424419999123, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022673025727272034, "qparams": { "group_size": { "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": 0.013790175318717957, "qparams": { "group_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": 0.011844818480312824, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0117416400462389, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010867991484701633, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010702581144869328, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007231354713439941, "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": 0.0073069059289991856, "qparams": { "group_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": 0.0068724192678928375, "qparams": { "group_size": { "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": 0.004890850745141506, "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": 0.013790175318717957, "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": 0.013790175318717957, "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": 0.08490868657827377, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07944120466709137, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0770629420876503, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06975635141134262, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03940696269273758, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.037188127636909485, "qparams": { "group_size": { "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": 0.04507056623697281, "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": 0.04166809096932411, "qparams": { "group_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": 0.03998866304755211, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03541511669754982, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03368842229247093, "qparams": { "group_size": { "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": 0.022855233401060104, "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": 0.019882962107658386, "qparams": { "group_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": 0.018866324797272682, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0186251662671566, "qparams": { "group_size": { "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": 0.011418124660849571, "qparams": { "group_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": 0.009724773466587067, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009622585028409958, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008920771069824696, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008767922408878803, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005980123765766621, "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": 0.0059907445684075356, "qparams": { "group_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": 0.005655237939208746, "qparams": { "group_size": { "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": 0.003952941857278347, "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": 0.011418124660849571, "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": 0.011418124660849571, "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": 0.20186981558799744, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18915194272994995, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1846855729818344, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1672881692647934, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09404631704092026, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08938305079936981, "qparams": { "group_size": { "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": 0.10601945966482162, "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": 0.09757892787456512, "qparams": { "group_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": 0.09539823979139328, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0845629870891571, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08022710680961609, "qparams": { "group_size": { "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": 0.05373474955558777, "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": 0.04653850197792053, "qparams": { "group_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": 0.044942706823349, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04456410929560661, "qparams": { "group_size": { "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": 0.026832986623048782, "qparams": { "group_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": 0.022952351719141006, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02281014434993267, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021007955074310303, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020759332925081253, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013890315778553486, "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": 0.013683604076504707, "qparams": { "group_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": 0.01329300832003355, "qparams": { "group_size": { "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": 0.008669965900480747, "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": 0.013890315778553486, "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": 0.013890315778553486, "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": 0.21062375605106354, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18768422305583954, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17733675241470337, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1518186330795288, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09728959202766418, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08791815489530563, "qparams": { "group_size": { "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": 0.11873985081911087, "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": 0.10831757634878159, "qparams": { "group_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": 0.10114531219005585, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08235026150941849, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07630736380815506, "qparams": { "group_size": { "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": 0.061113644391298294, "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": 0.05272505432367325, "qparams": { "group_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": 0.04763077199459076, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04637714847922325, "qparams": { "group_size": { "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": 0.030743977054953575, "qparams": { "group_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": 0.026061473414301872, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025607217103242874, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023297958076000214, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02252097800374031, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016975264996290207, "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": 0.018405204638838768, "qparams": { "group_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": 0.015370078384876251, "qparams": { "group_size": { "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": 0.014179988764226437, "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": 0.015370078384876251, "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": 0.015370078384876251, "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": 0.15889254212379456, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14963550865650177, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14662185311317444, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13364870846271515, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07458967715501785, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0713140219449997, "qparams": { "group_size": { "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": 0.08294253796339035, "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": 0.0767119899392128, "qparams": { "group_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": 0.07556746155023575, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06755757331848145, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06439531594514847, "qparams": { "group_size": { "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": 0.042385123670101166, "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": 0.03685370460152626, "qparams": { "group_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": 0.035914018750190735, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03568496182560921, "qparams": { "group_size": { "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": 0.02119295485317707, "qparams": { "group_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": 0.018872462213039398, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018804047256708145, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01749577559530735, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01736096478998661, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01138925738632679, "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": 0.011979250237345695, "qparams": { "group_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": 0.011085351929068565, "qparams": { "group_size": { "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": 0.008665981702506542, "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": 0.01138925738632679, "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": 0.01138925738632679, "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": 0.21704186499118805, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20445694029331207, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2005244493484497, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18267971277236938, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10156001895666122, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09714098274707794, "qparams": { "group_size": { "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": 0.11278364807367325, "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": 0.10419917106628418, "qparams": { "group_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": 0.10287287086248398, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0919332355260849, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08745458722114563, "qparams": { "group_size": { "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": 0.057214830070734024, "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": 0.04969990998506546, "qparams": { "group_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": 0.048505380749702454, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04822392761707306, "qparams": { "group_size": { "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": 0.028532372787594795, "qparams": { "group_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": 0.02472497709095478, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0246463380753994, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022753175348043442, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022577622905373573, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014791306108236313, "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": 0.014574708417057991, "qparams": { "group_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": 0.014375230297446251, "qparams": { "group_size": { "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": 0.00921012181788683, "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": 0.014791306108236313, "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": 0.014574708417057991, "qparams": { "group_size": { "6": 32 }, "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": 0.23230671882629395, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2092689573764801, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2004241645336151, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1772404909133911, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10691063851118088, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09797824919223785, "qparams": { "group_size": { "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": 0.12465143203735352, "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": 0.11454039812088013, "qparams": { "group_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": 0.10990122705698013, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09213875979185104, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08663027733564377, "qparams": { "group_size": { "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": 0.06378553062677383, "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": 0.055082544684410095, "qparams": { "group_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": 0.051570966839790344, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.050745848566293716, "qparams": { "group_size": { "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": 0.03215254843235016, "qparams": { "group_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": 0.0273437537252903, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.027046477422118187, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024507611989974976, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02396027185022831, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0178036168217659, "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": 0.017999514937400818, "qparams": { "group_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": 0.01671653799712658, "qparams": { "group_size": { "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": 0.013087870553135872, "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": 0.013087870553135872, "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": 0.013087870553135872, "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": 0.10979528725147247, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1025683730840683, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09982659667730331, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0902380496263504, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05098921060562134, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04821714758872986, "qparams": { "group_size": { "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": 0.057780373841524124, "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": 0.053355854004621506, "qparams": { "group_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": 0.05175023153424263, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04571004956960678, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.043389081954956055, "qparams": { "group_size": { "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": 0.029319550842046738, "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": 0.025471046566963196, "qparams": { "group_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": 0.0244012288749218, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02414655312895775, "qparams": { "group_size": { "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": 0.014645514078438282, "qparams": { "group_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": 0.012547153048217297, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012445258907973766, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011482831090688705, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011323275044560432, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007648067083209753, "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": 0.007649254519492388, "qparams": { "group_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": 0.007284554652869701, "qparams": { "group_size": { "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": 0.005027711391448975, "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": 0.014645514078438282, "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": 0.014645514078438282, "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": 0.08930663764476776, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08352401107549667, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0810493603348732, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07336125522851944, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.041462838649749756, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03913639113306999, "qparams": { "group_size": { "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": 0.04736847057938576, "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": 0.043774597346782684, "qparams": { "group_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": 0.04207707569003105, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.037256501615047455, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.035439617931842804, "qparams": { "group_size": { "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": 0.024000223726034164, "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": 0.02088848501443863, "qparams": { "group_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": 0.01984834298491478, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.019601672887802124, "qparams": { "group_size": { "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": 0.011983544565737247, "qparams": { "group_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": 0.010209540836513042, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010106416419148445, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009357653558254242, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009199468418955803, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006260917522013187, "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": 0.006256602704524994, "qparams": { "group_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": 0.005929167848080397, "qparams": { "group_size": { "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": 0.00408721761777997, "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": 0.011983544565737247, "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": 0.011983544565737247, "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": 0.20637144148349762, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19317032396793365, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18854254484176636, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1706109642982483, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.096175417304039, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09127388894557953, "qparams": { "group_size": { "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": 0.10856496542692184, "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": 0.09989390522241592, "qparams": { "group_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": 0.09758980572223663, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08638149499893188, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08193790167570114, "qparams": { "group_size": { "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": 0.055065054446458817, "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": 0.047706134617328644, "qparams": { "group_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": 0.046001922339200974, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04559335485100746, "qparams": { "group_size": { "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": 0.02751726470887661, "qparams": { "group_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": 0.023536736145615578, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023394297808408737, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02153547666966915, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02127670682966709, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014323080889880657, "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": 0.014109991490840912, "qparams": { "group_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": 0.013714365661144257, "qparams": { "group_size": { "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": 0.0090550621971488, "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": 0.014323080889880657, "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": 0.014323080889880657, "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": 0.21871821582317352, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19667311012744904, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18869586288928986, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16437377035617828, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10209696739912033, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09377863258123398, "qparams": { "group_size": { "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": 0.11835309118032455, "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": 0.10838740319013596, "qparams": { "group_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": 0.10462190210819244, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08671023696660995, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07962800562381744, "qparams": { "group_size": { "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": 0.0607743039727211, "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": 0.0525590144097805, "qparams": { "group_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": 0.04969790577888489, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04901818186044693, "qparams": { "group_size": { "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": 0.03058101236820221, "qparams": { "group_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": 0.026866361498832703, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02663334645330906, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02397248148918152, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023543890565633774, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016940956935286522, "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": 0.018231384456157684, "qparams": { "group_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": 0.01606195792555809, "qparams": { "group_size": { "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": 0.0139686968177557, "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": 0.0139686968177557, "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": 0.0139686968177557, "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": 0.1605602204799652, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.151184543967247, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14814680814743042, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13493652641773224, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07542375475168228, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07210846990346909, "qparams": { "group_size": { "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": 0.08393123000860214, "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": 0.07755376398563385, "qparams": { "group_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": 0.07642846554517746, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0682569295167923, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06509383022785187, "qparams": { "group_size": { "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": 0.04287952184677124, "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": 0.037284910678863525, "qparams": { "group_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": 0.0363253615796566, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03610822558403015, "qparams": { "group_size": { "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": 0.021454667672514915, "qparams": { "group_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": 0.01912534236907959, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019060105085372925, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017730101943016052, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01759042963385582, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011563851498067379, "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": 0.012189502827823162, "qparams": { "group_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": 0.011249310337007046, "qparams": { "group_size": { "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": 0.008856515400111675, "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": 0.011563851498067379, "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": 0.011563851498067379, "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": 0.2183949202299118, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20570994913578033, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.201686292886734, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18366149067878723, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10225381702184677, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09777547419071198, "qparams": { "group_size": { "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": 0.1136455312371254, "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": 0.10495319962501526, "qparams": { "group_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": 0.10357207804918289, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09251802414655685, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08803904801607132, "qparams": { "group_size": { "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": 0.05772437900304794, "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": 0.050051137804985046, "qparams": { "group_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": 0.048832595348358154, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048545561730861664, "qparams": { "group_size": { "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": 0.028783757239580154, "qparams": { "group_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": 0.024866672232747078, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024780938401818275, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02286285161972046, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02267911471426487, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014937619678676128, "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": 0.0146248210221529, "qparams": { "group_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": 0.014512896537780762, "qparams": { "group_size": { "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": 0.009180675260722637, "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": 0.014937619678676128, "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": 0.0146248210221529, "qparams": { "group_size": { "6": 32 }, "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": 0.23349811136722565, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21039116382598877, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20169289410114288, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17896541953086853, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1076163575053215, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09871077537536621, "qparams": { "group_size": { "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": 0.12529964745044708, "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": 0.1149045079946518, "qparams": { "group_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": 0.11051872372627258, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09289940446615219, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08738456666469574, "qparams": { "group_size": { "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": 0.06419861316680908, "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": 0.05531865358352661, "qparams": { "group_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": 0.05200396478176117, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05119892582297325, "qparams": { "group_size": { "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": 0.03246380388736725, "qparams": { "group_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": 0.027677537873387337, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02740759216248989, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02487284317612648, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024352554231882095, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.018135150894522667, "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": 0.018325503915548325, "qparams": { "group_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": 0.01710367761552334, "qparams": { "group_size": { "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": 0.01350038405507803, "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": 0.01350038405507803, "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": 0.01350038405507803, "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": 0.1180255264043808, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11026567220687866, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10737142711877823, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09712659567594528, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05483328551054001, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0519142672419548, "qparams": { "group_size": { "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": 0.0620843730866909, "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": 0.05735868588089943, "qparams": { "group_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": 0.055659424513578415, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04922547936439514, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04671429097652435, "qparams": { "group_size": { "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": 0.0315326452255249, "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": 0.02743176370859146, "qparams": { "group_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": 0.026309434324502945, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026040732860565186, "qparams": { "group_size": { "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": 0.015768878161907196, "qparams": { "group_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": 0.01364701148122549, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013539859093725681, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012526587583124638, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012356613762676716, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008307693526148796, "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": 0.00848522037267685, "qparams": { "group_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": 0.007935627363622189, "qparams": { "group_size": { "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": 0.005784531123936176, "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": 0.01364701148122549, "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": 0.01364701148122549, "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": 0.09278237819671631, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08673162758350372, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08409383893013, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07606416940689087, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04303666949272156, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04056334123015404, "qparams": { "group_size": { "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": 0.04936417564749718, "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": 0.045602116733789444, "qparams": { "group_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": 0.04368693754076958, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.038679298013448715, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03677428513765335, "qparams": { "group_size": { "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": 0.02502206154167652, "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": 0.021764449775218964, "qparams": { "group_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": 0.020611686632037163, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02033240720629692, "qparams": { "group_size": { "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": 0.012506142258644104, "qparams": { "group_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": 0.01061773020774126, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010496383532881737, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.00973700825124979, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009563904255628586, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0065485164523124695, "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": 0.006544409319758415, "qparams": { "group_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": 0.00618629390373826, "qparams": { "group_size": { "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": 0.004309381823986769, "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": 0.012506142258644104, "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": 0.012506142258644104, "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": 0.2159755676984787, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20223698019981384, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19747844338417053, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17890822887420654, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10077839344739914, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09572488814592361, "qparams": { "group_size": { "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": 0.11342068761587143, "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": 0.10442269593477249, "qparams": { "group_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": 0.1022273525595665, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09057269990444183, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08593432605266571, "qparams": { "group_size": { "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": 0.05761643871665001, "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": 0.049888238310813904, "qparams": { "group_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": 0.04818198084831238, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04776277765631676, "qparams": { "group_size": { "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": 0.02878197655081749, "qparams": { "group_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": 0.024647612124681473, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024492958560585976, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02256651781499386, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02230624482035637, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014959370717406273, "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": 0.014756537042558193, "qparams": { "group_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": 0.014334792271256447, "qparams": { "group_size": { "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": 0.009456712752580643, "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": 0.014959370717406273, "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": 0.014756537042558193, "qparams": { "group_size": { "6": 32 }, "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": 0.21747709810733795, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1931176334619522, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1849847137928009, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15863510966300964, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10150986909866333, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0923510193824768, "qparams": { "group_size": { "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": 0.11763546615839005, "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": 0.10741035640239716, "qparams": { "group_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": 0.10457853972911835, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08483196794986725, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07775552570819855, "qparams": { "group_size": { "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": 0.06081746146082878, "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": 0.05236322060227394, "qparams": { "group_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": 0.04969141259789467, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.049052294343709946, "qparams": { "group_size": { "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": 0.030704809352755547, "qparams": { "group_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": 0.02708035334944725, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02691827528178692, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02398875541985035, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023588743060827255, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017282292246818542, "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": 0.01862400211393833, "qparams": { "group_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": 0.016501057893037796, "qparams": { "group_size": { "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": 0.01456683874130249, "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": 0.01456683874130249, "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": 0.01456683874130249, "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": 0.15855464339256287, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14927060902118683, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14630018174648285, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13318371772766113, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07454072684049606, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07123260945081711, "qparams": { "group_size": { "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": 0.0828809142112732, "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": 0.07664623111486435, "qparams": { "group_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": 0.07553084194660187, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06745026260614395, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06423570960760117, "qparams": { "group_size": { "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": 0.04235098883509636, "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": 0.036858294159173965, "qparams": { "group_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": 0.035913728177547455, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03569105267524719, "qparams": { "group_size": { "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": 0.021197784692049026, "qparams": { "group_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": 0.018890488892793655, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018826015293598175, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017505250871181488, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017371056601405144, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011400355026125908, "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": 0.012026166543364525, "qparams": { "group_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": 0.011098548769950867, "qparams": { "group_size": { "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": 0.00872170552611351, "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": 0.011400355026125908, "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": 0.011400355026125908, "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": 0.2179897129535675, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20526543259620667, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20136186480522156, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1833401322364807, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10209034383296967, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0976223424077034, "qparams": { "group_size": { "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": 0.11336932331323624, "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": 0.1047547310590744, "qparams": { "group_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": 0.10341061651706696, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09237021207809448, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08794303238391876, "qparams": { "group_size": { "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": 0.0576116144657135, "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": 0.049970630556344986, "qparams": { "group_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": 0.04876956343650818, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048488982021808624, "qparams": { "group_size": { "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": 0.028758179396390915, "qparams": { "group_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": 0.02485528402030468, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024775860831141472, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022856172174215317, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022677404806017876, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014970979653298855, "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": 0.014642598107457161, "qparams": { "group_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": 0.014568113721907139, "qparams": { "group_size": { "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": 0.009244423359632492, "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": 0.014970979653298855, "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": 0.014568113721907139, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2312341332435608, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20809970796108246, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19919513165950775, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17696741223335266, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10624001175165176, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09731045365333557, "qparams": { "group_size": { "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": 0.12438449263572693, "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": 0.11391349136829376, "qparams": { "group_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": 0.10929252207279205, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09168168157339096, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08651523292064667, "qparams": { "group_size": { "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": 0.06334865093231201, "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": 0.05467002093791962, "qparams": { "group_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": 0.05122830718755722, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05039896070957184, "qparams": { "group_size": { "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": 0.031828105449676514, "qparams": { "group_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": 0.027088887989521027, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026803450658917427, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02428828552365303, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023741308599710464, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017363661900162697, "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": 0.017752306535840034, "qparams": { "group_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": 0.016271891072392464, "qparams": { "group_size": { "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": 0.012805548496544361, "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": 0.012805548496544361, "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": 0.012805548496544361, "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": 0.11935172230005264, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11172492057085037, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10884404182434082, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0985117256641388, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.055505942553281784, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05261604115366936, "qparams": { "group_size": { "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": 0.06275392323732376, "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": 0.057948071509599686, "qparams": { "group_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": 0.05631691962480545, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04990023374557495, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.047403354197740555, "qparams": { "group_size": { "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": 0.031858861446380615, "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": 0.02767593041062355, "qparams": { "group_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": 0.026576604694128036, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026307864114642143, "qparams": { "group_size": { "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": 0.01590607315301895, "qparams": { "group_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": 0.013674231246113777, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013572223484516144, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012541632167994976, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012375637888908386, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008315394632518291, "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": 0.008342577144503593, "qparams": { "group_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": 0.00793762318789959, "qparams": { "group_size": { "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": 0.005503727588802576, "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": 0.013674231246113777, "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": 0.013674231246113777, "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": 0.09296581149101257, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08702775835990906, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08442747592926025, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0764925554394722, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04320300742983818, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04078548774123192, "qparams": { "group_size": { "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": 0.04937296360731125, "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": 0.04565348103642464, "qparams": { "group_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": 0.04383651167154312, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.038853228092193604, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03697727993130684, "qparams": { "group_size": { "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": 0.02500741556286812, "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": 0.021771103143692017, "qparams": { "group_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": 0.02066408097743988, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02039785124361515, "qparams": { "group_size": { "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": 0.012484661303460598, "qparams": { "group_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": 0.010616057552397251, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010499785654246807, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009738419204950333, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.00956958718597889, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006509232334792614, "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": 0.006490151863545179, "qparams": { "group_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": 0.006160824093967676, "qparams": { "group_size": { "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": 0.004209634382277727, "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": 0.012484661303460598, "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": 0.012484661303460598, "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": 0.21869362890720367, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20519676804542542, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2002430260181427, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18161632120609283, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10228795558214188, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09712635725736618, "qparams": { "group_size": { "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": 0.11551395058631897, "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": 0.1063031405210495, "qparams": { "group_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": 0.10373198986053467, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.092088483273983, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08748292177915573, "qparams": { "group_size": { "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": 0.05863812565803528, "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": 0.05080419406294823, "qparams": { "group_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": 0.048928529024124146, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048471976071596146, "qparams": { "group_size": { "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": 0.02929200418293476, "qparams": { "group_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": 0.0249954666942358, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024818431586027145, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02290268801152706, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022611470893025398, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015245272777974606, "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": 0.014943419024348259, "qparams": { "group_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": 0.014581020921468735, "qparams": { "group_size": { "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": 0.009493138641119003, "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": 0.015245272777974606, "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": 0.014581020921468735, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2205626368522644, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2004324197769165, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1942402571439743, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.168453648686409, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10290469974279404, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09548010677099228, "qparams": { "group_size": { "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": 0.11748414486646652, "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": 0.10732212662696838, "qparams": { "group_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": 0.10502339899539948, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08809854835271835, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08107319474220276, "qparams": { "group_size": { "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": 0.06063156574964523, "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": 0.05255727469921112, "qparams": { "group_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": 0.0505606010556221, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.050084248185157776, "qparams": { "group_size": { "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": 0.030887702479958534, "qparams": { "group_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": 0.027891863137483597, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.027761386707425117, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.025197235867381096, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024900073185563087, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017894843593239784, "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": 0.01946813054382801, "qparams": { "group_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": 0.017318205907940865, "qparams": { "group_size": { "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": 0.015636442229151726, "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": 0.015636442229151726, "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": 0.015636442229151726, "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": 0.16347143054008484, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15389029681682587, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15081985294818878, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13734370470046997, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07684691250324249, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0734262764453888, "qparams": { "group_size": { "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": 0.08541522920131683, "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": 0.07899678498506546, "qparams": { "group_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": 0.07784334570169449, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06952722370624542, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0661957636475563, "qparams": { "group_size": { "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": 0.04361758753657341, "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": 0.03792550787329674, "qparams": { "group_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": 0.036952048540115356, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0367259755730629, "qparams": { "group_size": { "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": 0.02179640159010887, "qparams": { "group_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": 0.01933296024799347, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01926484890282154, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017889251932501793, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017751149833202362, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011652193032205105, "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": 0.012153816409409046, "qparams": { "group_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": 0.011336913332343102, "qparams": { "group_size": { "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": 0.008652850985527039, "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": 0.011652193032205105, "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": 0.011652193032205105, "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": 0.21874932944774628, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20608246326446533, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2021171897649765, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18414939939975739, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10253489017486572, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09807280451059341, "qparams": { "group_size": { "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": 0.11378517001867294, "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": 0.10522045195102692, "qparams": { "group_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": 0.10388962179422379, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09279554337263107, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0882941484451294, "qparams": { "group_size": { "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": 0.057774998247623444, "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": 0.05019833892583847, "qparams": { "group_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": 0.048985060304403305, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048704344779253006, "qparams": { "group_size": { "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": 0.028813665732741356, "qparams": { "group_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": 0.024945467710494995, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0248565636575222, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022936144843697548, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022754237055778503, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014914541505277157, "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": 0.01466182991862297, "qparams": { "group_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": 0.014495661482214928, "qparams": { "group_size": { "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": 0.009203877300024033, "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": 0.014914541505277157, "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": 0.01466182991862297, "qparams": { "group_size": { "6": 32 }, "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": 0.23219355940818787, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2077765315771103, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1983470469713211, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1755652129650116, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10643792152404785, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0969504565000534, "qparams": { "group_size": { "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": 0.12483672797679901, "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": 0.11469001322984695, "qparams": { "group_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": 0.10972621291875839, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09142578393220901, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08592444658279419, "qparams": { "group_size": { "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": 0.06371396780014038, "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": 0.054990336298942566, "qparams": { "group_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": 0.0512380376458168, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05033378675580025, "qparams": { "group_size": { "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": 0.03208324313163757, "qparams": { "group_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": 0.02694527432322502, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026630010455846786, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024024268612265587, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02342597208917141, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017553970217704773, "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": 0.01749793440103531, "qparams": { "group_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": 0.016387205570936203, "qparams": { "group_size": { "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": 0.01236578356474638, "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": 0.01236578356474638, "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": 0.01236578356474638, "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": 0.11513505131006241, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10788670182228088, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10512227565050125, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09526921808719635, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05359466373920441, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.050816651433706284, "qparams": { "group_size": { "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": 0.06058506667613983, "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": 0.055926840752363205, "qparams": { "group_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": 0.05435923486948013, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04823870211839676, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.045908667147159576, "qparams": { "group_size": { "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": 0.03077918477356434, "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": 0.026712652295827866, "qparams": { "group_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": 0.02565060183405876, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.025401480495929718, "qparams": { "group_size": { "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": 0.01537415012717247, "qparams": { "group_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": 0.013185153715312481, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013086321763694286, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012096833437681198, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011942211538553238, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008026422001421452, "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": 0.008015734143555164, "qparams": { "group_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": 0.007674003019928932, "qparams": { "group_size": { "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": 0.005251645110547543, "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": 0.01537415012717247, "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": 0.01537415012717247, "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": 0.09265274554491043, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08678431808948517, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08426224440336227, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07639190554618835, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0430472232401371, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04068147391080856, "qparams": { "group_size": { "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": 0.04910045862197876, "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": 0.04543893039226532, "qparams": { "group_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": 0.043688032776117325, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.038745492696762085, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03686782717704773, "qparams": { "group_size": { "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": 0.0248748529702425, "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": 0.021669143810868263, "qparams": { "group_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": 0.02060825191438198, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.020350266247987747, "qparams": { "group_size": { "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": 0.012424595654010773, "qparams": { "group_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": 0.010601385496556759, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010489514097571373, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009730499237775803, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0095676863566041, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0064865038730204105, "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": 0.0064916969276964664, "qparams": { "group_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": 0.006146772764623165, "qparams": { "group_size": { "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": 0.004238844383507967, "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": 0.012424595654010773, "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": 0.012424595654010773, "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": 0.22385789453983307, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21030853688716888, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.205491840839386, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18655014038085938, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10478025674819946, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09966816753149033, "qparams": { "group_size": { "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": 0.117934949696064, "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": 0.10852140188217163, "qparams": { "group_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": 0.10622784495353699, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09441763162612915, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08974705636501312, "qparams": { "group_size": { "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": 0.059876374900341034, "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": 0.05181189998984337, "qparams": { "group_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": 0.050079647451639175, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04966232180595398, "qparams": { "group_size": { "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": 0.02990066073834896, "qparams": { "group_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": 0.025566542521119118, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02541498653590679, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02344188280403614, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023180030286312103, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015518520958721638, "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": 0.01521311141550541, "qparams": { "group_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": 0.014879297465085983, "qparams": { "group_size": { "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": 0.009630052372813225, "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": 0.015518520958721638, "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": 0.014879297465085983, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2111966460943222, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19016888737678528, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1812392622232437, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15923449397087097, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0985933244228363, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08947145938873291, "qparams": { "group_size": { "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": 0.11744941771030426, "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": 0.10726305097341537, "qparams": { "group_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": 0.10119419544935226, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08441059291362762, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0784023180603981, "qparams": { "group_size": { "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": 0.0605192594230175, "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": 0.051811665296554565, "qparams": { "group_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": 0.04785216972231865, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.046881821006536484, "qparams": { "group_size": { "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": 0.030312051996588707, "qparams": { "group_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": 0.025455810129642487, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02506827563047409, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022701913490891457, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022090239450335503, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016370834782719612, "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": 0.016979539766907692, "qparams": { "group_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": 0.015087440609931946, "qparams": { "group_size": { "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": 0.012375123798847198, "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": 0.015087440609931946, "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": 0.015087440609931946, "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": 0.1621202975511551, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15263278782367706, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14964768290519714, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1361636221408844, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07622050493955612, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07286020368337631, "qparams": { "group_size": { "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": 0.08471689373254776, "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": 0.07833580672740936, "qparams": { "group_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": 0.07723376899957657, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06896672397851944, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06566113978624344, "qparams": { "group_size": { "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": 0.043276719748973846, "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": 0.03765053674578667, "qparams": { "group_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": 0.03669605776667595, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03646931052207947, "qparams": { "group_size": { "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": 0.021654171869158745, "qparams": { "group_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": 0.019258176907896996, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019195333123207092, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017831426113843918, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017697906121611595, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011637461371719837, "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": 0.012186768464744091, "qparams": { "group_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": 0.011334633454680443, "qparams": { "group_size": { "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": 0.008775764144957066, "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": 0.011637461371719837, "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": 0.011637461371719837, "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": 0.21908950805664062, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20627978444099426, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20236223936080933, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18428391218185425, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10265935957431793, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09817023575305939, "qparams": { "group_size": { "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": 0.11393322050571442, "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": 0.10527051240205765, "qparams": { "group_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": 0.10397344827651978, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0928647592663765, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0883021280169487, "qparams": { "group_size": { "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": 0.05783940106630325, "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": 0.05022985488176346, "qparams": { "group_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": 0.04904009401798248, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04875611141324043, "qparams": { "group_size": { "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": 0.028867414221167564, "qparams": { "group_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": 0.02499992400407791, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024915792047977448, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022991983219981194, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022809188812971115, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015023083426058292, "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": 0.01473424956202507, "qparams": { "group_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": 0.014620293863117695, "qparams": { "group_size": { "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": 0.009317870251834393, "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": 0.015023083426058292, "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": 0.014620293863117695, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.23122946918010712, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20713679492473602, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19778169691562653, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17536196112632751, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10609609633684158, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0967826098203659, "qparams": { "group_size": { "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": 0.12445244938135147, "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": 0.11433668434619904, "qparams": { "group_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": 0.1093418151140213, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09126590192317963, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08599486947059631, "qparams": { "group_size": { "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": 0.0636255145072937, "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": 0.05483078956604004, "qparams": { "group_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": 0.0510992631316185, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05019892379641533, "qparams": { "group_size": { "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": 0.03204012289643288, "qparams": { "group_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": 0.026895489543676376, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026576442644000053, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02401186153292656, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023412426933646202, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017582042142748833, "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": 0.01750160939991474, "qparams": { "group_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": 0.016406500712037086, "qparams": { "group_size": { "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": 0.012399239465594292, "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": 0.012399239465594292, "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": 0.012399239465594292, "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": 0.12177224457263947, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11425122618675232, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11140135675668716, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1010560542345047, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.056818533688783646, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05394895002245903, "qparams": { "group_size": { "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": 0.06411070376634598, "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": 0.0591680184006691, "qparams": { "group_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": 0.05760613828897476, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05119180679321289, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04872053861618042, "qparams": { "group_size": { "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": 0.03260248154401779, "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": 0.02828594297170639, "qparams": { "group_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": 0.027235064655542374, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02698419615626335, "qparams": { "group_size": { "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": 0.016308264806866646, "qparams": { "group_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": 0.014075015671551228, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013981243595480919, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012946275994181633, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012788332998752594, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008576581254601479, "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": 0.008656637743115425, "qparams": { "group_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": 0.008218571543693542, "qparams": { "group_size": { "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": 0.005828286986798048, "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": 0.014075015671551228, "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": 0.014075015671551228, "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": 0.09657128155231476, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09052947908639908, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08796728402376175, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07975969463586807, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04493415355682373, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04252252355217934, "qparams": { "group_size": { "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": 0.05116261541843414, "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": 0.04730309173464775, "qparams": { "group_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": 0.04557270556688309, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04048933833837509, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.038512222468853, "qparams": { "group_size": { "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": 0.02591656520962715, "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": 0.02257879450917244, "qparams": { "group_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": 0.02151981182396412, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021262316033244133, "qparams": { "group_size": { "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": 0.012948152609169483, "qparams": { "group_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": 0.011068851687014103, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010960346087813377, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010168615728616714, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010007863864302635, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006764707155525684, "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": 0.006777131464332342, "qparams": { "group_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": 0.00642869807779789, "qparams": { "group_size": { "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": 0.004440870601683855, "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": 0.012948152609169483, "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": 0.012948152609169483, "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": 0.2342512458562851, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22031915187835693, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2155901938676834, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19592826068401337, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10987208038568497, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10461147129535675, "qparams": { "group_size": { "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": 0.1232287585735321, "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": 0.11353206634521484, "qparams": { "group_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": 0.11133582144975662, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09913560003042221, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09428294003009796, "qparams": { "group_size": { "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": 0.06255333125591278, "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": 0.05421912297606468, "qparams": { "group_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": 0.052497271448373795, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05208621919155121, "qparams": { "group_size": { "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": 0.03123771958053112, "qparams": { "group_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": 0.02676461450755596, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026615992188453674, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024560781195759773, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024301793426275253, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016143308952450752, "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": 0.015844669193029404, "qparams": { "group_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": 0.015508919954299927, "qparams": { "group_size": { "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": 0.009957239963114262, "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": 0.015508919954299927, "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": 0.015508919954299927, "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": 0.21491308510303497, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19025194644927979, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.182305708527565, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15327885746955872, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09872705489397049, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08977792412042618, "qparams": { "group_size": { "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": 0.1152084618806839, "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": 0.10499511659145355, "qparams": { "group_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": 0.10196571797132492, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0807422548532486, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07303531467914581, "qparams": { "group_size": { "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": 0.05919479578733444, "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": 0.051140762865543365, "qparams": { "group_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": 0.04832527041435242, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04763941466808319, "qparams": { "group_size": { "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": 0.030098117887973785, "qparams": { "group_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": 0.02652624435722828, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02634153515100479, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023223228752613068, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02278442494571209, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017124658450484276, "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": 0.018466472625732422, "qparams": { "group_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": 0.016300763934850693, "qparams": { "group_size": { "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": 0.01456499844789505, "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": 0.01456499844789505, "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": 0.01456499844789505, "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": 0.16102346777915955, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15172851085662842, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14879298210144043, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.13546310365200043, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07567577064037323, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07236514240503311, "qparams": { "group_size": { "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": 0.0840793326497078, "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": 0.07775933295488358, "qparams": { "group_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": 0.07668335735797882, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06849675625562668, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06531050056219101, "qparams": { "group_size": { "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": 0.04288100823760033, "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": 0.037314992398023605, "qparams": { "group_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": 0.036382947117090225, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.036162614822387695, "qparams": { "group_size": { "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": 0.021439023315906525, "qparams": { "group_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": 0.01903139427304268, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018964873626828194, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017614519223570824, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017477691173553467, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011459278874099255, "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": 0.011953797191381454, "qparams": { "group_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": 0.01115796621888876, "qparams": { "group_size": { "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": 0.008504287339746952, "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": 0.011459278874099255, "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": 0.011459278874099255, "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": 0.2171318382024765, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20460465550422668, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20078259706497192, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18294142186641693, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10171227902173996, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09731978923082352, "qparams": { "group_size": { "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": 0.11295703053474426, "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": 0.10431476682424545, "qparams": { "group_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": 0.10304608941078186, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09213176369667053, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0877915769815445, "qparams": { "group_size": { "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": 0.05738518387079239, "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": 0.049777716398239136, "qparams": { "group_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": 0.04860623553395271, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048326604068279266, "qparams": { "group_size": { "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": 0.02865682542324066, "qparams": { "group_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": 0.02477860450744629, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024702662602066994, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022801602259278297, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022624947130680084, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014964153990149498, "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": 0.014604044146835804, "qparams": { "group_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": 0.014566602185368538, "qparams": { "group_size": { "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": 0.009234500117599964, "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": 0.014964153990149498, "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": 0.014566602185368538, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2252340018749237, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20069658756256104, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19083961844444275, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16872794926166534, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10305023193359375, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09335052967071533, "qparams": { "group_size": { "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": 0.12213942408561707, "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": 0.11203040182590485, "qparams": { "group_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": 0.10645659267902374, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08815541863441467, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08309366554021835, "qparams": { "group_size": { "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": 0.06244679167866707, "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": 0.05371437594294548, "qparams": { "group_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": 0.049660902470350266, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048671651631593704, "qparams": { "group_size": { "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": 0.03143211454153061, "qparams": { "group_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": 0.026159638538956642, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025809762999415398, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023265613242983818, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022610517218708992, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01720636896789074, "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": 0.017125843092799187, "qparams": { "group_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": 0.015921685844659805, "qparams": { "group_size": { "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": 0.01211825106292963, "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": 0.01211825106292963, "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": 0.01211825106292963, "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": 0.11869264394044876, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11144040524959564, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10866419970989227, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09880580753087997, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05528460070490837, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.052548233419656754, "qparams": { "group_size": { "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": 0.0625077337026596, "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": 0.057626303285360336, "qparams": { "group_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": 0.05604662001132965, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04994702711701393, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.047647811472415924, "qparams": { "group_size": { "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": 0.031745944172143936, "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": 0.027527613565325737, "qparams": { "group_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": 0.02646288089454174, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026214849203824997, "qparams": { "group_size": { "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": 0.015861861407756805, "qparams": { "group_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": 0.013595569878816605, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013498412445187569, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012516961432993412, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012355274520814419, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008280050940811634, "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": 0.00825564470142126, "qparams": { "group_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": 0.007934151217341423, "qparams": { "group_size": { "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": 0.005404595751315355, "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": 0.013595569878816605, "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": 0.013595569878816605, "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": 0.09521899372339249, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08936624228954315, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08681788295507431, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07886835187673569, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.044313229620456696, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0419282391667366, "qparams": { "group_size": { "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": 0.05050261691212654, "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": 0.04670814424753189, "qparams": { "group_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": 0.04491254687309265, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.039989206939935684, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03813350200653076, "qparams": { "group_size": { "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": 0.025607991963624954, "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": 0.022285297513008118, "qparams": { "group_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": 0.0212111696600914, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.020951632410287857, "qparams": { "group_size": { "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": 0.012796486727893353, "qparams": { "group_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": 0.010916821658611298, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01080414280295372, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010050837881863117, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009887137450277805, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00669837137684226, "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": 0.006700864061713219, "qparams": { "group_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": 0.006359039340168238, "qparams": { "group_size": { "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": 0.00440309289842844, "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": 0.012796486727893353, "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": 0.012796486727893353, "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": 0.2284725457429886, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21510273218154907, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21066319942474365, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1918047070503235, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10703007131814957, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10215771198272705, "qparams": { "group_size": { "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": 0.11989373713731766, "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": 0.110451340675354, "qparams": { "group_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": 0.1084444522857666, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09679936617612839, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09219638258218765, "qparams": { "group_size": { "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": 0.06087200343608856, "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": 0.052726779133081436, "qparams": { "group_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": 0.051155753433704376, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05078626424074173, "qparams": { "group_size": { "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": 0.030376501381397247, "qparams": { "group_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": 0.026062317192554474, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02593168243765831, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023956486955285072, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023713331669569016, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015708228573203087, "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": 0.015388093888759613, "qparams": { "group_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": 0.015124664641916752, "qparams": { "group_size": { "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": 0.009657279588282108, "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": 0.015388093888759613, "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": 0.015124664641916752, "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": 0.21548834443092346, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19201481342315674, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1843968778848648, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15123489499092102, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10089244693517685, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09250934422016144, "qparams": { "group_size": { "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": 0.11556107550859451, "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": 0.10603442788124084, "qparams": { "group_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": 0.1033065915107727, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08220374584197998, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07157059013843536, "qparams": { "group_size": { "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": 0.059730224311351776, "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": 0.051505617797374725, "qparams": { "group_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": 0.04917747154831886, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04863040894269943, "qparams": { "group_size": { "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": 0.03008301369845867, "qparams": { "group_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": 0.026576314121484756, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026420364156365395, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023072127252817154, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022721558809280396, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016805175691843033, "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": 0.017943667247891426, "qparams": { "group_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": 0.016105087473988533, "qparams": { "group_size": { "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": 0.013818293809890747, "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": 0.013818293809890747, "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": 0.013818293809890747, "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": 0.16705745458602905, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15750524401664734, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1545003056526184, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1409224271774292, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07850729674100876, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0751178115606308, "qparams": { "group_size": { "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": 0.08722370862960815, "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": 0.080609031021595, "qparams": { "group_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": 0.07955172657966614, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07115722447633743, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06778987497091293, "qparams": { "group_size": { "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": 0.04444720223546028, "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": 0.038651395589113235, "qparams": { "group_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": 0.03771747276186943, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.037496160715818405, "qparams": { "group_size": { "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": 0.022258378565311432, "qparams": { "group_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": 0.01964123174548149, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019571561366319656, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018172290176153183, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018030492588877678, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011914274655282497, "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": 0.012210698798298836, "qparams": { "group_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": 0.011615321971476078, "qparams": { "group_size": { "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": 0.008546730503439903, "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": 0.011914274655282497, "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": 0.011914274655282497, "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": 0.2241048514842987, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21138736605644226, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2074471265077591, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18918657302856445, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10504027456045151, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10057232528924942, "qparams": { "group_size": { "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": 0.11659403890371323, "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": 0.10765231400728226, "qparams": { "group_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": 0.10637017339468002, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09520990401506424, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09070742875337601, "qparams": { "group_size": { "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": 0.05919408053159714, "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": 0.051378749310970306, "qparams": { "group_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": 0.050198350101709366, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04991410672664642, "qparams": { "group_size": { "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": 0.029635289683938026, "qparams": { "group_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": 0.025620384141802788, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025542180985212326, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023597275838255882, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023425307124853134, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015581101179122925, "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": 0.015134397894144058, "qparams": { "group_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": 0.015195086598396301, "qparams": { "group_size": { "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": 0.009629599750041962, "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": 0.015581101179122925, "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": 0.015134397894144058, "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": 0.2285468578338623, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20289596915245056, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1925816535949707, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17006267607212067, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10459382832050323, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09435156732797623, "qparams": { "group_size": { "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": 0.12365137040615082, "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": 0.11388258635997772, "qparams": { "group_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": 0.10809101909399033, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08911889791488647, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08399029076099396, "qparams": { "group_size": { "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": 0.0631166398525238, "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": 0.054826103150844574, "qparams": { "group_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": 0.050522588193416595, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.049463026225566864, "qparams": { "group_size": { "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": 0.031763073056936264, "qparams": { "group_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": 0.026826437562704086, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026443183422088623, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02385028451681137, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023159515112638474, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017134161666035652, "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": 0.017835697159171104, "qparams": { "group_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": 0.015720298513770103, "qparams": { "group_size": { "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": 0.012897091917693615, "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": 0.012897091917693615, "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": 0.012897091917693615, "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": 0.11891096085309982, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11181946098804474, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10913249850273132, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09926894307136536, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.055502258241176605, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05275201424956322, "qparams": { "group_size": { "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": 0.06263826042413712, "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": 0.05778731033205986, "qparams": { "group_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": 0.05622922256588936, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05016711726784706, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04789987951517105, "qparams": { "group_size": { "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": 0.031838033348321915, "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": 0.027614574879407883, "qparams": { "group_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": 0.026565298438072205, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02631472609937191, "qparams": { "group_size": { "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": 0.015910759568214417, "qparams": { "group_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": 0.013675319030880928, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013575204648077488, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012605675496160984, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012449041940271854, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008342040702700615, "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": 0.008335278369486332, "qparams": { "group_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": 0.008002141490578651, "qparams": { "group_size": { "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": 0.0055078077130019665, "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": 0.013675319030880928, "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": 0.013675319030880928, "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": 0.09795869886875153, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09204284846782684, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08964193612337112, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08151700347661972, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04565494507551193, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.043298423290252686, "qparams": { "group_size": { "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": 0.0518498532474041, "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": 0.04791552200913429, "qparams": { "group_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": 0.04629112407565117, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04126293957233429, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0393543541431427, "qparams": { "group_size": { "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": 0.026301516219973564, "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": 0.02287128195166588, "qparams": { "group_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": 0.021848736330866814, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021604737266898155, "qparams": { "group_size": { "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": 0.013135036453604698, "qparams": { "group_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": 0.011218197643756866, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011112555861473083, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010327496565878391, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01017305813729763, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0068550072610378265, "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": 0.0068270741030573845, "qparams": { "group_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": 0.006529275327920914, "qparams": { "group_size": { "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": 0.0044365618377923965, "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": 0.013135036453604698, "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": 0.013135036453604698, "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": 0.2316964864730835, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21820800006389618, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21382462978363037, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1947123408317566, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10851924121379852, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10365653038024902, "qparams": { "group_size": { "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": 0.12121471762657166, "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": 0.11176545917987823, "qparams": { "group_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": 0.10996273159980774, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09824252128601074, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09359058737754822, "qparams": { "group_size": { "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": 0.06153950095176697, "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": 0.05332864075899124, "qparams": { "group_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": 0.051839932799339294, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05148717015981674, "qparams": { "group_size": { "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": 0.030697334557771683, "qparams": { "group_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": 0.02633296698331833, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026211518794298172, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024201544001698494, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02397727407515049, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01580570638179779, "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": 0.015419847331941128, "qparams": { "group_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": 0.015244454145431519, "qparams": { "group_size": { "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": 0.009511674754321575, "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": 0.015419847331941128, "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": 0.015244454145431519, "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": 0.22413785755634308, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19266220927238464, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18248103559017181, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15160532295703888, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10369378328323364, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09143050014972687, "qparams": { "group_size": { "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": 0.1206488162279129, "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": 0.11075519025325775, "qparams": { "group_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": 0.10734008252620697, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08292199671268463, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07338327914476395, "qparams": { "group_size": { "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": 0.06232164800167084, "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": 0.054010238498449326, "qparams": { "group_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": 0.05083825811743736, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0500730462372303, "qparams": { "group_size": { "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": 0.031664203852415085, "qparams": { "group_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": 0.02793687954545021, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.027730777859687805, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024158215150237083, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02366694249212742, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.018140209838747978, "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": 0.01952190324664116, "qparams": { "group_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": 0.017216403037309647, "qparams": { "group_size": { "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": 0.01543383114039898, "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": 0.01543383114039898, "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": 0.01543383114039898, "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": 0.17527982592582703, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.165274977684021, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16218872368335724, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.14796367287635803, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0824151262640953, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07888640463352203, "qparams": { "group_size": { "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": 0.09130053222179413, "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": 0.08454746007919312, "qparams": { "group_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": 0.08347348868846893, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07472540438175201, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07122140377759933, "qparams": { "group_size": { "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": 0.0465669259428978, "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": 0.04050193354487419, "qparams": { "group_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": 0.039554547518491745, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03932749107480049, "qparams": { "group_size": { "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": 0.023245176300406456, "qparams": { "group_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": 0.020531803369522095, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02046540379524231, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01899144798517227, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018853522837162018, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012298394925892353, "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": 0.012665393762290478, "qparams": { "group_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": 0.011983303353190422, "qparams": { "group_size": { "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": 0.008757628500461578, "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": 0.012298394925892353, "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": 0.012298394925892353, "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": 0.22846853733062744, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21551522612571716, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21159270405769348, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1931227296590805, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10712900757789612, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10259560495615005, "qparams": { "group_size": { "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": 0.11859654635190964, "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": 0.10975703597068787, "qparams": { "group_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": 0.10849177837371826, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0971917137503624, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09257292002439499, "qparams": { "group_size": { "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": 0.06020869314670563, "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": 0.05235451087355614, "qparams": { "group_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": 0.051168955862522125, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05089125782251358, "qparams": { "group_size": { "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": 0.03001321665942669, "qparams": { "group_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": 0.026049060747027397, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02596800960600376, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023990953341126442, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0238130372017622, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015518547967076302, "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": 0.015283472836017609, "qparams": { "group_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": 0.01512063480913639, "qparams": { "group_size": { "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": 0.009590750560164452, "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": 0.015518547967076302, "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": 0.01512063480913639, "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": 0.2320854216814041, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2054763287305832, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1945493370294571, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17149074375629425, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10609012842178345, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09543067216873169, "qparams": { "group_size": { "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": 0.12613698840141296, "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": 0.11598563194274902, "qparams": { "group_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": 0.10978162288665771, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09003586322069168, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08475159853696823, "qparams": { "group_size": { "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": 0.0641871988773346, "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": 0.055566344410181046, "qparams": { "group_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": 0.0511094368994236, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05000464618206024, "qparams": { "group_size": { "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": 0.032134100794792175, "qparams": { "group_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": 0.02687089703977108, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02647252380847931, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023744020611047745, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023014312610030174, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01722143590450287, "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": 0.017566394060850143, "qparams": { "group_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": 0.015794305130839348, "qparams": { "group_size": { "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": 0.012309590354561806, "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": 0.012309590354561806, "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": 0.012309590354561806, "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": 0.11812770366668701, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11100509017705917, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.108275406062603, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09840768575668335, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05508861690759659, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05234581232070923, "qparams": { "group_size": { "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": 0.062312327325344086, "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": 0.057446353137493134, "qparams": { "group_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": 0.05582885444164276, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04978642612695694, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.047532226890325546, "qparams": { "group_size": { "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": 0.0316837802529335, "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": 0.02748074196279049, "qparams": { "group_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": 0.026402272284030914, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026143033057451248, "qparams": { "group_size": { "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": 0.01584988459944725, "qparams": { "group_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": 0.013620792888104916, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013518184423446655, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012557283975183964, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01239542942494154, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008319661021232605, "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": 0.008361087180674076, "qparams": { "group_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": 0.007971658371388912, "qparams": { "group_size": { "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": 0.005586773157119751, "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": 0.013620792888104916, "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": 0.013620792888104916, "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": 0.09843088686466217, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09244368970394135, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08987642079591751, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08166611939668655, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04585302248597145, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04341713339090347, "qparams": { "group_size": { "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": 0.05229084566235542, "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": 0.048314835876226425, "qparams": { "group_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": 0.04648908972740173, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.041426882147789, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.039502326399087906, "qparams": { "group_size": { "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": 0.02650546282529831, "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": 0.023062940686941147, "qparams": { "group_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": 0.021944763138890266, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021678153425455093, "qparams": { "group_size": { "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": 0.013236609287559986, "qparams": { "group_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": 0.01127818413078785, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011164584197103977, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010385064408183098, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01021470408886671, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006913917604833841, "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": 0.006892867386341095, "qparams": { "group_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": 0.006566897965967655, "qparams": { "group_size": { "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": 0.004492092877626419, "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": 0.013236609287559986, "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": 0.013236609287559986, "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": 0.23150266706943512, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21800358593463898, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21345427632331848, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19440020620822906, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10840781778097153, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10347999632358551, "qparams": { "group_size": { "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": 0.12129861861467361, "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": 0.11178790032863617, "qparams": { "group_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": 0.10981447249650955, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0980919599533081, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09352494776248932, "qparams": { "group_size": { "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": 0.06160532683134079, "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": 0.05331116542220116, "qparams": { "group_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": 0.05179167166352272, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05141715705394745, "qparams": { "group_size": { "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": 0.03072560578584671, "qparams": { "group_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": 0.026341047137975693, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026220453903079033, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02421380952000618, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023980483412742615, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015834294259548187, "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": 0.015486359596252441, "qparams": { "group_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": 0.015255456790328026, "qparams": { "group_size": { "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": 0.009643089957535267, "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": 0.015486359596252441, "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": 0.015255456790328026, "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": 0.20792704820632935, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18203650414943695, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17361769080162048, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.14407262206077576, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0964139848947525, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0866360142827034, "qparams": { "group_size": { "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": 0.11167094856500626, "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": 0.10213151574134827, "qparams": { "group_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": 0.09951692819595337, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07784850150346756, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06849812716245651, "qparams": { "group_size": { "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": 0.05759742483496666, "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": 0.04948896914720535, "qparams": { "group_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": 0.046911515295505524, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04628949612379074, "qparams": { "group_size": { "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": 0.02911382168531418, "qparams": { "group_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": 0.025243263691663742, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025092095136642456, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02174408733844757, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021344155073165894, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016327155753970146, "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": 0.01697741635143757, "qparams": { "group_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": 0.015540960244834423, "qparams": { "group_size": { "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": 0.01289004273712635, "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": 0.015540960244834423, "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": 0.015540960244834423, "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": 0.17863868176937103, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16861894726753235, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16546332836151123, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15104131400585175, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08397562056779861, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08040580153465271, "qparams": { "group_size": { "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": 0.09325302392244339, "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": 0.0861208364367485, "qparams": { "group_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": 0.08504107594490051, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07620024681091309, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07272335886955261, "qparams": { "group_size": { "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": 0.04751666262745857, "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": 0.04130951687693596, "qparams": { "group_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": 0.040344588458538055, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04012170806527138, "qparams": { "group_size": { "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": 0.023797163739800453, "qparams": { "group_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": 0.021036013960838318, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02097010239958763, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.019490763545036316, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019353142008185387, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012771358713507652, "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": 0.013113731518387794, "qparams": { "group_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": 0.012466088868677616, "qparams": { "group_size": { "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": 0.009236868470907211, "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": 0.012771358713507652, "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": 0.012771358713507652, "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": 0.227318674325943, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2145501673221588, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2106020450592041, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.192250058054924, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10659917443990707, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10212472826242447, "qparams": { "group_size": { "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": 0.11814511567354202, "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": 0.10921541601419449, "qparams": { "group_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": 0.10793289542198181, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09670151770114899, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09215424209833145, "qparams": { "group_size": { "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": 0.059971340000629425, "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": 0.05214196816086769, "qparams": { "group_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": 0.05095764622092247, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05068463459610939, "qparams": { "group_size": { "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": 0.02999102883040905, "qparams": { "group_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": 0.02600414678454399, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025925468653440475, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023975063115358353, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023799268528819084, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015711061656475067, "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": 0.01535316277295351, "qparams": { "group_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": 0.015326561406254768, "qparams": { "group_size": { "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": 0.00976920872926712, "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": 0.01535316277295351, "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": 6.231277126736111, "total_bits": 441042816.0, "err": 0.015326561406254768, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2294359803199768, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20206794142723083, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1905193030834198, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1677798181772232, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10466107726097107, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0935632660984993, "qparams": { "group_size": { "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": 0.1252349615097046, "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": 0.11528460681438446, "qparams": { "group_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": 0.1084701269865036, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.088424913585186, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08332040905952454, "qparams": { "group_size": { "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": 0.06376142799854279, "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": 0.055175162851810455, "qparams": { "group_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": 0.05040070414543152, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.049218542873859406, "qparams": { "group_size": { "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": 0.03191665560007095, "qparams": { "group_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": 0.02645149827003479, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026012957096099854, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02328384481370449, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022495096549391747, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01711026020348072, "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": 0.017296047881245613, "qparams": { "group_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": 0.015584918670356274, "qparams": { "group_size": { "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": 0.01200621947646141, "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": 0.015584918670356274, "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": 0.015584918670356274, "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": 0.1148228794336319, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1079217866063118, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10523294657468796, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09571917355060577, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.053506121039390564, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05081949010491371, "qparams": { "group_size": { "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": 0.06052912771701813, "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": 0.05589313060045242, "qparams": { "group_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": 0.054217662662267685, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.048380810767412186, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04618294909596443, "qparams": { "group_size": { "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": 0.030714625492691994, "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": 0.02667119912803173, "qparams": { "group_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": 0.025594959035515785, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.025337880477309227, "qparams": { "group_size": { "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": 0.015345865860581398, "qparams": { "group_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": 0.01314582396298647, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013039244338870049, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012111619114875793, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01194894127547741, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008021618239581585, "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": 0.007984871044754982, "qparams": { "group_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": 0.007674495689570904, "qparams": { "group_size": { "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": 0.005210723262280226, "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": 0.015345865860581398, "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": 0.015345865860581398, "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": 0.09626813977956772, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09041551500558853, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08788993209600449, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07992476969957352, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04479311406612396, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04244151711463928, "qparams": { "group_size": { "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": 0.05110354349017143, "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": 0.04721732810139656, "qparams": { "group_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": 0.045439619570970535, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04051869735121727, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.038675833493471146, "qparams": { "group_size": { "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": 0.025911178439855576, "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": 0.02253696881234646, "qparams": { "group_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": 0.021440735086798668, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021176835522055626, "qparams": { "group_size": { "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": 0.012940621934831142, "qparams": { "group_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": 0.011019488796591759, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010908200405538082, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010152886621654034, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009986824356019497, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00675996346399188, "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": 0.006740225479006767, "qparams": { "group_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": 0.006414222065359354, "qparams": { "group_size": { "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": 0.004395695403218269, "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": 0.012940621934831142, "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": 0.012940621934831142, "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": 0.22425396740436554, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2111717015504837, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20671141147613525, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18829457461833954, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10481102764606476, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10004448145627975, "qparams": { "group_size": { "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": 0.11739153414964676, "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": 0.1082848459482193, "qparams": { "group_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": 0.10618281364440918, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09487657248973846, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09042246639728546, "qparams": { "group_size": { "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": 0.05957190692424774, "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": 0.051620762795209885, "qparams": { "group_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": 0.05004170909523964, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.049676597118377686, "qparams": { "group_size": { "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": 0.02971735969185829, "qparams": { "group_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": 0.025465719401836395, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025338610634207726, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023413781076669693, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023177819326519966, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015349514782428741, "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": 0.014998546801507473, "qparams": { "group_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": 0.0147749874740839, "qparams": { "group_size": { "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": 0.009339884854853153, "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": 0.015349514782428741, "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": 0.0147749874740839, "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.o_proj", "numel": 26214400, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.17529296875, "total_bits": 57024000.0, "err": 0.21128441393375397, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18069423735141754, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17038388550281525, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.14309853315353394, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09735576063394547, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08597064763307571, "qparams": { "group_size": { "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": 0.11441962420940399, "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": 0.10469996929168701, "qparams": { "group_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": 0.10124586522579193, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07699687033891678, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07030686736106873, "qparams": { "group_size": { "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": 0.05887607857584953, "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": 0.05073453485965729, "qparams": { "group_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": 0.047441065311431885, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.046641625463962555, "qparams": { "group_size": { "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": 0.029768802225589752, "qparams": { "group_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": 0.02561606466770172, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0254160575568676, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021807726472616196, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02129809744656086, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016755957156419754, "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": 0.017444979399442673, "qparams": { "group_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": 0.01577160507440567, "qparams": { "group_size": { "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": 0.013287834823131561, "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": 0.013287834823131561, "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": 0.013287834823131561, "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": 0.17862524092197418, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16860103607177734, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16546505689620972, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1509900987148285, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08388222754001617, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08036528527736664, "qparams": { "group_size": { "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": 0.09315454214811325, "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": 0.08601558953523636, "qparams": { "group_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": 0.08495105057954788, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07610905915498734, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07265882194042206, "qparams": { "group_size": { "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": 0.047426603734493256, "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": 0.04121343791484833, "qparams": { "group_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": 0.040264714509248734, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04003734141588211, "qparams": { "group_size": { "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": 0.02376086823642254, "qparams": { "group_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": 0.02092844247817993, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02086314558982849, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01938283070921898, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019239749759435654, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012708460912108421, "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": 0.012942018918693066, "qparams": { "group_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": 0.01239586528390646, "qparams": { "group_size": { "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": 0.009002727456390858, "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": 0.012708460912108421, "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": 0.012708460912108421, "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": 0.22417937219142914, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21162542700767517, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20783153176307678, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18971531093120575, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1050523966550827, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10067187994718552, "qparams": { "group_size": { "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": 0.1165136843919754, "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": 0.10758987069129944, "qparams": { "group_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": 0.10636016726493835, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09533947706222534, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09095508605241776, "qparams": { "group_size": { "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": 0.0591123104095459, "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": 0.05136018246412277, "qparams": { "group_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": 0.05020494759082794, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.049931835383176804, "qparams": { "group_size": { "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": 0.02953784354031086, "qparams": { "group_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": 0.025641001760959625, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025563588365912437, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023650873452425003, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02347942814230919, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015461105853319168, "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": 0.015170019119977951, "qparams": { "group_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": 0.015084792859852314, "qparams": { "group_size": { "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": 0.009693575091660023, "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": 0.015461105853319168, "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": 0.015084792859852314, "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": 0.22921957075595856, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20139659941196442, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18960042297840118, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16686415672302246, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.104562908411026, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09330588579177856, "qparams": { "group_size": { "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": 0.12655286490917206, "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": 0.11550809442996979, "qparams": { "group_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": 0.10853557288646698, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08828980475664139, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08324068784713745, "qparams": { "group_size": { "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": 0.06432734429836273, "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": 0.05553687736392021, "qparams": { "group_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": 0.05061258003115654, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04939998313784599, "qparams": { "group_size": { "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": 0.03233005851507187, "qparams": { "group_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": 0.027012471109628677, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026564862579107285, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02392006851732731, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023113973438739777, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01764593832194805, "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": 0.018231526017189026, "qparams": { "group_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": 0.016064869239926338, "qparams": { "group_size": { "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": 0.013289720751345158, "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": 0.013289720751345158, "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": 0.013289720751345158, "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": 0.11707495898008347, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11002394556999207, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10727798938751221, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09761195629835129, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05453955754637718, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05183298885822296, "qparams": { "group_size": { "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": 0.06172072887420654, "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": 0.056924834847450256, "qparams": { "group_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": 0.05527997389435768, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.049312490969896317, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04716392606496811, "qparams": { "group_size": { "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": 0.031357280910015106, "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": 0.027195041999220848, "qparams": { "group_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": 0.026103632524609566, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0258504468947649, "qparams": { "group_size": { "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": 0.01568489335477352, "qparams": { "group_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": 0.013433581218123436, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013328542932868004, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012385972775518894, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012224778532981873, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008221031166613102, "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": 0.008193857036530972, "qparams": { "group_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": 0.007874065078794956, "qparams": { "group_size": { "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": 0.005402613431215286, "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": 0.013433581218123436, "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": 0.013433581218123436, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.09691258519887924, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09103009104728699, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08842552453279495, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08038898557424545, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.045056089758872986, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04263165220618248, "qparams": { "group_size": { "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": 0.051618486642837524, "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": 0.04767244681715965, "qparams": { "group_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": 0.04569285362958908, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04075571894645691, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03893482685089111, "qparams": { "group_size": { "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": 0.02615894190967083, "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": 0.02274056151509285, "qparams": { "group_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": 0.021572299301624298, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02129700779914856, "qparams": { "group_size": { "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": 0.013080264441668987, "qparams": { "group_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": 0.011110146529972553, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010984810069203377, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010240179486572742, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010063448920845985, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006861853878945112, "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": 0.006828424520790577, "qparams": { "group_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": 0.0064951092936098576, "qparams": { "group_size": { "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": 0.00447961175814271, "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": 0.013080264441668987, "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": 0.013080264441668987, "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": 0.23369169235229492, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22008036077022552, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21568793058395386, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19647155702114105, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10932939499616623, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10440953820943832, "qparams": { "group_size": { "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": 0.12217236310243607, "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": 0.11256274580955505, "qparams": { "group_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": 0.11076926440000534, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09894657880067825, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09431818872690201, "qparams": { "group_size": { "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": 0.06202373281121254, "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": 0.053688980638980865, "qparams": { "group_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": 0.052179355174303055, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.051819123327732086, "qparams": { "group_size": { "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": 0.030928267166018486, "qparams": { "group_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": 0.02651744708418846, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026397433131933212, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02437446266412735, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024142632260918617, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015941327437758446, "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": 0.01552465371787548, "qparams": { "group_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": 0.015376264229416847, "qparams": { "group_size": { "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": 0.009579835459589958, "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": 0.01552465371787548, "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": 0.015376264229416847, "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": 0.19101154804229736, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1644974648952484, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1560475081205368, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1285889446735382, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08802292495965958, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07866579294204712, "qparams": { "group_size": { "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": 0.10199769586324692, "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": 0.0937962457537651, "qparams": { "group_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": 0.09098227322101593, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0700814425945282, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06137755513191223, "qparams": { "group_size": { "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": 0.05262557789683342, "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": 0.0458032451570034, "qparams": { "group_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": 0.043227821588516235, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04259256646037102, "qparams": { "group_size": { "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": 0.026720723137259483, "qparams": { "group_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": 0.02389858290553093, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02372295968234539, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020659463480114937, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020255370065569878, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015276825055480003, "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": 0.01684006303548813, "qparams": { "group_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": 0.014501964673399925, "qparams": { "group_size": { "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": 0.013441714458167553, "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": 0.015276825055480003, "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": 0.015276825055480003, "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": 0.1708148866891861, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16133826971054077, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15839232504367828, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.14467836916446686, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08021989464759827, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07687096297740936, "qparams": { "group_size": { "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": 0.08895346522331238, "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": 0.08224103599786758, "qparams": { "group_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": 0.08123457431793213, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07285190373659134, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06954402476549149, "qparams": { "group_size": { "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": 0.04528440907597542, "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": 0.039361026138067245, "qparams": { "group_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": 0.03846203163266182, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03824925795197487, "qparams": { "group_size": { "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": 0.022623134776949883, "qparams": { "group_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": 0.01990456134080887, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01984584331512451, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018425747752189636, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018290262669324875, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011953091248869896, "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": 0.012188374064862728, "qparams": { "group_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": 0.011658640578389168, "qparams": { "group_size": { "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": 0.008329242467880249, "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": 0.011953091248869896, "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": 0.011953091248869896, "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": 0.21700569987297058, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20499511063098907, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2012641280889511, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18387965857982635, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1017283946275711, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0975056067109108, "qparams": { "group_size": { "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": 0.11256968975067139, "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": 0.10417490452528, "qparams": { "group_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": 0.10297583788633347, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09238302707672119, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08809718489646912, "qparams": { "group_size": { "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": 0.057153038680553436, "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": 0.04968053102493286, "qparams": { "group_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": 0.04857315123081207, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.048312339931726456, "qparams": { "group_size": { "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": 0.028498852625489235, "qparams": { "group_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": 0.024715816602110863, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024641089141368866, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022791415452957153, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022626472637057304, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014747379347682, "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": 0.014491215348243713, "qparams": { "group_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": 0.014376712962985039, "qparams": { "group_size": { "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": 0.00907787587493658, "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": 0.014747379347682, "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": 0.014747379347682, "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": 0.21952226758003235, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19173115491867065, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1798284947872162, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15852990746498108, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0996900275349617, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08836887031793594, "qparams": { "group_size": { "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": 0.12091369181871414, "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": 0.11078326404094696, "qparams": { "group_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": 0.10364051163196564, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08391446620225906, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07942094653844833, "qparams": { "group_size": { "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": 0.06160333380103111, "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": 0.05320418253540993, "qparams": { "group_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": 0.04818131774663925, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04693904146552086, "qparams": { "group_size": { "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": 0.03097669780254364, "qparams": { "group_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": 0.025610236451029778, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02516039088368416, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02259591780602932, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021781256422400475, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01692088507115841, "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": 0.01721728779375553, "qparams": { "group_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": 0.01535826176404953, "qparams": { "group_size": { "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": 0.012398742139339447, "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": 0.01535826176404953, "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": 0.01535826176404953, "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": 0.11993512511253357, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11305629462003708, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11045607924461365, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10079536586999893, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05598248541355133, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05337826907634735, "qparams": { "group_size": { "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": 0.06290201097726822, "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": 0.05815514549612999, "qparams": { "group_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": 0.056709446012973785, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.050772592425346375, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04847552627325058, "qparams": { "group_size": { "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": 0.03190179914236069, "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": 0.027746178209781647, "qparams": { "group_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": 0.026754172518849373, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02652556635439396, "qparams": { "group_size": { "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": 0.015938349068164825, "qparams": { "group_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": 0.013673622161149979, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01358354277908802, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01260728295892477, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012460249476134777, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00829386804252863, "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": 0.008172377943992615, "qparams": { "group_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": 0.007968738675117493, "qparams": { "group_size": { "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": 0.005206773057579994, "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": 0.013673622161149979, "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": 0.013673622161149979, "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": 0.10366872698068619, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09768833220005035, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.095316082239151, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08695932477712631, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04839685186743736, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.046054448932409286, "qparams": { "group_size": { "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": 0.05475849285721779, "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": 0.05050777643918991, "qparams": { "group_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": 0.04902986064553261, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04390639066696167, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04201927036046982, "qparams": { "group_size": { "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": 0.02778550423681736, "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": 0.02409445308148861, "qparams": { "group_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": 0.02312951162457466, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02290443889796734, "qparams": { "group_size": { "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": 0.013863575644791126, "qparams": { "group_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": 0.011821289546787739, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011728538200259209, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010899391025304794, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010757211595773697, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007199018727988005, "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": 0.007084526587277651, "qparams": { "group_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": 0.006883518770337105, "qparams": { "group_size": { "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": 0.004500663839280605, "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": 0.013863575644791126, "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": 0.013863575644791126, "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": 0.23216626048088074, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21927927434444427, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2151067554950714, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19639191031455994, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10878272354602814, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10412650555372238, "qparams": { "group_size": { "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": 0.12114786356687546, "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": 0.11179719865322113, "qparams": { "group_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": 0.11015192419290543, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09873686730861664, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09430142492055893, "qparams": { "group_size": { "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": 0.06144530326128006, "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": 0.05328470095992088, "qparams": { "group_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": 0.0519205741584301, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.051601532846689224, "qparams": { "group_size": { "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": 0.030651982873678207, "qparams": { "group_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": 0.026374567300081253, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026271220296621323, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024297036230564117, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024100536480545998, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015787627547979355, "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": 0.01541487779468298, "qparams": { "group_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": 0.015289923176169395, "qparams": { "group_size": { "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": 0.009516018442809582, "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": 0.01541487779468298, "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": 0.015289923176169395, "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": 0.20963597297668457, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17063161730766296, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15676097571849823, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12262146174907684, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09520181268453598, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0803755521774292, "qparams": { "group_size": { "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": 0.11418371647596359, "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": 0.1048346534371376, "qparams": { "group_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": 0.10032296180725098, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06839641183614731, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06253477931022644, "qparams": { "group_size": { "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": 0.05871806666254997, "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": 0.05057480186223984, "qparams": { "group_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": 0.04623103514313698, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04517355188727379, "qparams": { "group_size": { "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": 0.029550781473517418, "qparams": { "group_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": 0.02477714791893959, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024493129923939705, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.019779549911618233, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019012030214071274, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016311997547745705, "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": 0.016844648867845535, "qparams": { "group_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": 0.014961592853069305, "qparams": { "group_size": { "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": 0.012438842095434666, "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": 0.014961592853069305, "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": 0.014961592853069305, "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": 0.1801735907793045, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17027486860752106, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1671709418296814, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15281611680984497, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0846274122595787, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08112578094005585, "qparams": { "group_size": { "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": 0.09373730421066284, "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": 0.08673077076673508, "qparams": { "group_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": 0.08570209890604019, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07691308110952377, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07346590608358383, "qparams": { "group_size": { "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": 0.047729723155498505, "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": 0.04150509089231491, "qparams": { "group_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": 0.04056305065751076, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04033902660012245, "qparams": { "group_size": { "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": 0.02383853867650032, "qparams": { "group_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": 0.02097334899008274, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020906437188386917, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01942364126443863, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019285695627331734, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012572670355439186, "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": 0.012818332761526108, "qparams": { "group_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": 0.01226461585611105, "qparams": { "group_size": { "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": 0.008735112845897675, "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": 0.012572670355439186, "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": 0.012572670355439186, "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": 0.22373120486736298, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2114328294992447, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20763444900512695, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1898609846830368, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10481463372707367, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10050684958696365, "qparams": { "group_size": { "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": 0.11626225709915161, "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": 0.10731739550828934, "qparams": { "group_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": 0.10612653195858002, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09531655162572861, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09112425893545151, "qparams": { "group_size": { "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": 0.059071484953165054, "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": 0.05121491104364395, "qparams": { "group_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": 0.05007346346974373, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04981182515621185, "qparams": { "group_size": { "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": 0.02951442077755928, "qparams": { "group_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": 0.02555367350578308, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02547418512403965, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023594200611114502, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023426171392202377, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015456806868314743, "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": 0.015078598633408546, "qparams": { "group_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": 0.015089184045791626, "qparams": { "group_size": { "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": 0.009588838554918766, "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": 0.015456806868314743, "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": 0.015078598633408546, "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": 0.22131790220737457, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19357937574386597, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1809234321117401, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15969428420066833, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1006845086812973, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08894917368888855, "qparams": { "group_size": { "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": 0.12285780906677246, "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": 0.11287803202867508, "qparams": { "group_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": 0.10471627861261368, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0849294438958168, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08036413788795471, "qparams": { "group_size": { "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": 0.06249075382947922, "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": 0.054206155240535736, "qparams": { "group_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": 0.04869479313492775, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.047315437346696854, "qparams": { "group_size": { "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": 0.03138722851872444, "qparams": { "group_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": 0.02588469348847866, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02535414695739746, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022864973172545433, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02196052484214306, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016941694542765617, "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": 0.017469877377152443, "qparams": { "group_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": 0.015194671228528023, "qparams": { "group_size": { "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": 0.01250724121928215, "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": 0.015194671228528023, "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": 0.015194671228528023, "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": 0.1184176504611969, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11144912242889404, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10879936814308167, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09916379302740097, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05527270585298538, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05262264981865883, "qparams": { "group_size": { "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": 0.06229100376367569, "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": 0.057568635791540146, "qparams": { "group_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": 0.056006867438554764, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.050087977200746536, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04778745770454407, "qparams": { "group_size": { "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": 0.031591419130563736, "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": 0.02748795598745346, "qparams": { "group_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": 0.026427801698446274, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026178482919931412, "qparams": { "group_size": { "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": 0.015784822404384613, "qparams": { "group_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": 0.013537689112126827, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0134401461109519, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012479047290980816, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012324515730142593, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008234264329075813, "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": 0.008149142377078533, "qparams": { "group_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": 0.007889575324952602, "qparams": { "group_size": { "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": 0.005249804817140102, "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": 0.013537689112126827, "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": 0.013537689112126827, "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": 0.09938713163137436, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09355676919221878, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09109923988580704, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08302617073059082, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04637032374739647, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.044017210602760315, "qparams": { "group_size": { "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": 0.052780091762542725, "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": 0.04870149493217468, "qparams": { "group_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": 0.04700347036123276, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.042038947343826294, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.040204089134931564, "qparams": { "group_size": { "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": 0.02676883339881897, "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": 0.023243002593517303, "qparams": { "group_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": 0.02218284271657467, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021927960216999054, "qparams": { "group_size": { "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": 0.01337120309472084, "qparams": { "group_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": 0.011361517012119293, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011254099197685719, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010480268858373165, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010320531204342842, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006961680483072996, "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": 0.006870738230645657, "qparams": { "group_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": 0.006620931439101696, "qparams": { "group_size": { "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": 0.004406890831887722, "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": 0.01337120309472084, "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": 0.01337120309472084, "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": 0.23104742169380188, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21787123382091522, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21359233558177948, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19489282369613647, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10814130306243896, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10343152284622192, "qparams": { "group_size": { "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": 0.12048623710870743, "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": 0.11122754216194153, "qparams": { "group_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": 0.10956394672393799, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09808614104986191, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09358338266611099, "qparams": { "group_size": { "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": 0.06114766374230385, "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": 0.05303611978888512, "qparams": { "group_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": 0.05163886025547981, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05130811035633087, "qparams": { "group_size": { "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": 0.03048338182270527, "qparams": { "group_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": 0.02622416615486145, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02611134573817253, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02414628677070141, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023932114243507385, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01568378321826458, "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": 0.015333211980760098, "qparams": { "group_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": 0.015161766670644283, "qparams": { "group_size": { "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": 0.009466776624321938, "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": 0.015333211980760098, "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": 0.015161766670644283, "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": 0.18946966528892517, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15035103261470795, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1356620192527771, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11023501306772232, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08612748980522156, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07056307047605515, "qparams": { "group_size": { "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": 0.1046781912446022, "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": 0.09615612030029297, "qparams": { "group_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": 0.09050960093736649, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06119995564222336, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05744050815701485, "qparams": { "group_size": { "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": 0.05395076796412468, "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": 0.04655655845999718, "qparams": { "group_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": 0.042044397443532944, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04091477394104004, "qparams": { "group_size": { "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": 0.02720690704882145, "qparams": { "group_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": 0.022835973650217056, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022478055208921432, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01829085312783718, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01750226877629757, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015154791064560413, "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": 0.01593085192143917, "qparams": { "group_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": 0.013791441917419434, "qparams": { "group_size": { "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": 0.012057843618094921, "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": 0.015154791064560413, "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": 0.015154791064560413, "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": 0.18424969911575317, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17415079474449158, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17105428874492645, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15644051134586334, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08654194325208664, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0829707682132721, "qparams": { "group_size": { "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": 0.09580311924219131, "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": 0.0886203721165657, "qparams": { "group_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": 0.08760713040828705, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07865756750106812, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07511141896247864, "qparams": { "group_size": { "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": 0.04877142608165741, "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": 0.04238254576921463, "qparams": { "group_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": 0.041449546813964844, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04122650995850563, "qparams": { "group_size": { "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": 0.024354469031095505, "qparams": { "group_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": 0.021407008171081543, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021339507773518562, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01982142962515354, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019682250916957855, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0128452954813838, "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": 0.013033709488809109, "qparams": { "group_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": 0.012540619820356369, "qparams": { "group_size": { "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": 0.008832900784909725, "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": 0.0128452954813838, "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": 0.0128452954813838, "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": 0.22535237669944763, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21296754479408264, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20917147397994995, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19135737419128418, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10561061650514603, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10124897211790085, "qparams": { "group_size": { "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": 0.11678124964237213, "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": 0.10809052735567093, "qparams": { "group_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": 0.10687921941280365, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09603408724069595, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0916605070233345, "qparams": { "group_size": { "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": 0.05936377868056297, "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": 0.051556579768657684, "qparams": { "group_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": 0.050421372056007385, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.050160590559244156, "qparams": { "group_size": { "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": 0.029582269489765167, "qparams": { "group_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": 0.025639362633228302, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025566795840859413, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023666638880968094, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023494092747569084, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015327871777117252, "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": 0.014994516968727112, "qparams": { "group_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": 0.014953216537833214, "qparams": { "group_size": { "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": 0.009359663352370262, "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": 0.015327871777117252, "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": 0.014953216537833214, "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": 0.2232251614332199, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1959487497806549, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1832566112279892, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16212749481201172, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10156232118606567, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09000203013420105, "qparams": { "group_size": { "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": 0.12439863383769989, "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": 0.11403355002403259, "qparams": { "group_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": 0.10557080805301666, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08609110116958618, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.081589475274086, "qparams": { "group_size": { "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": 0.06325719505548477, "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": 0.0547090508043766, "qparams": { "group_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": 0.049141671508550644, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0477389357984066, "qparams": { "group_size": { "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": 0.03208913654088974, "qparams": { "group_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": 0.026145722717046738, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02560119330883026, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02318590320646763, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02227865718305111, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017962247133255005, "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": 0.01767653599381447, "qparams": { "group_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": 0.016310155391693115, "qparams": { "group_size": { "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": 0.01268927101045847, "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": 0.01268927101045847, "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": 0.01268927101045847, "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": 0.1195513978600502, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11265336722135544, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10994277149438858, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10027801990509033, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.055823877453804016, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0531490221619606, "qparams": { "group_size": { "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": 0.06322851777076721, "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": 0.05819234251976013, "qparams": { "group_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": 0.056569747626781464, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.050626300275325775, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04842260479927063, "qparams": { "group_size": { "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": 0.03205963224172592, "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": 0.02778429538011551, "qparams": { "group_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": 0.026716182008385658, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026460954919457436, "qparams": { "group_size": { "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": 0.01604999415576458, "qparams": { "group_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": 0.013692334294319153, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01358844619244337, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012633733451366425, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012475324794650078, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008398151956498623, "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": 0.00825821328908205, "qparams": { "group_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": 0.00803415384143591, "qparams": { "group_size": { "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": 0.005328319501131773, "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": 0.013692334294319153, "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": 0.013692334294319153, "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": 0.10196754336357117, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09604029357433319, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09369567781686783, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08539425581693649, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04765617847442627, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.045302338898181915, "qparams": { "group_size": { "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": 0.05401848256587982, "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": 0.04984872788190842, "qparams": { "group_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": 0.04828788340091705, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.043194837868213654, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04130808264017105, "qparams": { "group_size": { "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": 0.027387982234358788, "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": 0.023780427873134613, "qparams": { "group_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": 0.02278904989361763, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022550150752067566, "qparams": { "group_size": { "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": 0.013676533475518227, "qparams": { "group_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": 0.011649062857031822, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01155009213835001, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010738151147961617, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010589756071567535, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007097463123500347, "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": 0.006993760354816914, "qparams": { "group_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": 0.00677026342600584, "qparams": { "group_size": { "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": 0.004443841520696878, "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": 0.013676533475518227, "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": 0.013676533475518227, "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": 0.22707240283489227, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21431657671928406, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21013858914375305, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19175000488758087, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10634229332208633, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.101758673787117, "qparams": { "group_size": { "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": 0.11852490156888962, "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": 0.10939846932888031, "qparams": { "group_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": 0.1077122688293457, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09649845957756042, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09210189431905746, "qparams": { "group_size": { "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": 0.06013020873069763, "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": 0.052159108221530914, "qparams": { "group_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": 0.05076044052839279, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05043698847293854, "qparams": { "group_size": { "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": 0.02998347021639347, "qparams": { "group_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": 0.025778621435165405, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025674819946289062, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023744693025946617, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02353314496576786, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01544014923274517, "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": 0.01507329847663641, "qparams": { "group_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": 0.014934570528566837, "qparams": { "group_size": { "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": 0.009287657216191292, "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": 0.01544014923274517, "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": 0.014934570528566837, "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": 0.1952149122953415, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1560494750738144, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14371778070926666, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12471280992031097, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08716060221195221, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07205286622047424, "qparams": { "group_size": { "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": 0.10528864711523056, "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": 0.09489179402589798, "qparams": { "group_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": 0.09200551360845566, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06670327484607697, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05993904918432236, "qparams": { "group_size": { "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": 0.05374487489461899, "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": 0.046409375965595245, "qparams": { "group_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": 0.04296021908521652, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04212384670972824, "qparams": { "group_size": { "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": 0.027963412925601006, "qparams": { "group_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": 0.02394373156130314, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023760655894875526, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020326875150203705, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019782962277531624, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016811804845929146, "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": 0.01716727763414383, "qparams": { "group_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": 0.015909910202026367, "qparams": { "group_size": { "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": 0.013772360049188137, "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": 0.013772360049188137, "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": 0.013772360049188137, "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": 0.191115602850914, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18061742186546326, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1773638278245926, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16220274567604065, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08978594094514847, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08607572317123413, "qparams": { "group_size": { "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": 0.09937521815299988, "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": 0.09193460643291473, "qparams": { "group_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": 0.0908985584974289, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08162643760442734, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07790733873844147, "qparams": { "group_size": { "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": 0.050586361438035965, "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": 0.04397628456354141, "qparams": { "group_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": 0.04299577698111534, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.042771365493535995, "qparams": { "group_size": { "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": 0.025262365117669106, "qparams": { "group_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": 0.022167092189192772, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02209840714931488, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02051394432783127, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02037038467824459, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013280206359922886, "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": 0.013429897837340832, "qparams": { "group_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": 0.012964742258191109, "qparams": { "group_size": { "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": 0.009011048823595047, "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": 0.013280206359922886, "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": 0.013280206359922886, "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": 0.230477973818779, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21790845692157745, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21402239799499512, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19585639238357544, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10814379900693893, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10368844866752625, "qparams": { "group_size": { "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": 0.11982766538858414, "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": 0.11066153645515442, "qparams": { "group_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": 0.10944269597530365, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09837029874324799, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09397849440574646, "qparams": { "group_size": { "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": 0.06098746135830879, "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": 0.05285518243908882, "qparams": { "group_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": 0.05169796198606491, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05142468214035034, "qparams": { "group_size": { "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": 0.030470140278339386, "qparams": { "group_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": 0.02641245722770691, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02633713185787201, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024404292926192284, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024234944954514503, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016008714213967323, "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": 0.015635015442967415, "qparams": { "group_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": 0.0156348068267107, "qparams": { "group_size": { "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": 0.010034274309873581, "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": 0.015635015442967415, "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": 6.231277126736111, "total_bits": 441042816.0, "err": 0.0156348068267107, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.22786355018615723, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20101869106292725, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18878334760665894, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1673079878091812, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10373244434595108, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09245797991752625, "qparams": { "group_size": { "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": 0.12569913268089294, "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": 0.11576321721076965, "qparams": { "group_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": 0.1076335534453392, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0883733481168747, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08371983468532562, "qparams": { "group_size": { "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": 0.06392835825681686, "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": 0.05532858893275261, "qparams": { "group_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": 0.049910176545381546, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0485658198595047, "qparams": { "group_size": { "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": 0.03202858194708824, "qparams": { "group_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": 0.02605002000927925, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025520432740449905, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023010283708572388, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022111015394330025, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01713053323328495, "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": 0.01695595495402813, "qparams": { "group_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": 0.015408330596983433, "qparams": { "group_size": { "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": 0.011448226869106293, "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": 0.015408330596983433, "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": 0.015408330596983433, "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": 0.12216578423976898, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11512391269207001, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11244111508131027, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1025143414735794, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05707284063100815, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05435001105070114, "qparams": { "group_size": { "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": 0.06430254876613617, "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": 0.05936538428068161, "qparams": { "group_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": 0.057835545390844345, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05172891542315483, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0494658388197422, "qparams": { "group_size": { "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": 0.032619841396808624, "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": 0.028330042958259583, "qparams": { "group_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": 0.027296386659145355, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02704889141023159, "qparams": { "group_size": { "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": 0.016297580674290657, "qparams": { "group_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": 0.013974550180137157, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01387694664299488, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012886599637567997, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012731668539345264, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008502805605530739, "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": 0.00839486625045538, "qparams": { "group_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": 0.008162942714989185, "qparams": { "group_size": { "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": 0.005399538204073906, "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": 0.013974550180137157, "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": 0.013974550180137157, "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": 0.10330401360988617, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09731965512037277, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0949055626988411, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08653287589550018, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04827486723661423, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.045888204127550125, "qparams": { "group_size": { "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": 0.05477892979979515, "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": 0.05049514397978783, "qparams": { "group_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": 0.04891780763864517, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0437655933201313, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.041887830942869186, "qparams": { "group_size": { "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": 0.02779323235154152, "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": 0.024092352017760277, "qparams": { "group_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": 0.0230817012488842, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022839868441224098, "qparams": { "group_size": { "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": 0.013873358257114887, "qparams": { "group_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": 0.01179509423673153, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011693780310451984, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010873797349631786, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010720022022724152, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007200716063380241, "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": 0.007074249442666769, "qparams": { "group_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": 0.006869391072541475, "qparams": { "group_size": { "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": 0.004489735700190067, "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": 0.013873358257114887, "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": 0.013873358257114887, "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": 0.2325068712234497, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21938055753707886, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2152509093284607, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19638818502426147, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10888697952032089, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10420291870832443, "qparams": { "group_size": { "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": 0.12118128687143326, "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": 0.11187094449996948, "qparams": { "group_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": 0.11028467118740082, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09877385199069977, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09428045898675919, "qparams": { "group_size": { "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": 0.06145670264959335, "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": 0.05331730842590332, "qparams": { "group_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": 0.051969461143016815, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05165945366024971, "qparams": { "group_size": { "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": 0.030650459229946136, "qparams": { "group_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": 0.02637527324259281, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02627270668745041, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024282274767756462, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02408038266003132, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015741903334856033, "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": 0.015377974137663841, "qparams": { "group_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": 0.015236163511872292, "qparams": { "group_size": { "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": 0.009452572092413902, "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": 0.015377974137663841, "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": 0.015236163511872292, "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": 0.20084048807621002, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15598689019680023, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.13958317041397095, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11699319630861282, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08935996890068054, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.068650022149086, "qparams": { "group_size": { "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": 0.11000216752290726, "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": 0.10117853432893753, "qparams": { "group_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": 0.09604497253894806, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06627179682254791, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05973239615559578, "qparams": { "group_size": { "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": 0.056618303060531616, "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": 0.04873328283429146, "qparams": { "group_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": 0.04337754473090172, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04203455522656441, "qparams": { "group_size": { "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": 0.02850980870425701, "qparams": { "group_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": 0.02313171699643135, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02279866486787796, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018769381567835808, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01783355325460434, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01564808003604412, "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": 0.015790142118930817, "qparams": { "group_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": 0.014033980667591095, "qparams": { "group_size": { "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": 0.011368881911039352, "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": 0.014033980667591095, "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": 0.014033980667591095, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.19052654504776, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18013030290603638, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1769452840089798, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1618431806564331, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08963301032781601, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08593849092721939, "qparams": { "group_size": { "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": 0.09938275814056396, "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": 0.091769739985466, "qparams": { "group_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": 0.0907452404499054, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08148370683193207, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0779326781630516, "qparams": { "group_size": { "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": 0.05064751207828522, "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": 0.043929532170295715, "qparams": { "group_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": 0.04296398162841797, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04273107647895813, "qparams": { "group_size": { "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": 0.025336788967251778, "qparams": { "group_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": 0.022203847765922546, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02213510312139988, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020562347024679184, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02041599154472351, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01345920655876398, "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": 0.013545699417591095, "qparams": { "group_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": 0.013146058656275272, "qparams": { "group_size": { "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": 0.009206950664520264, "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": 0.01345920655876398, "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": 0.01345920655876398, "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": 0.23220476508140564, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21955205500125885, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.215664342045784, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19737458229064941, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10902509838342667, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10456699877977371, "qparams": { "group_size": { "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": 0.12087012082338333, "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": 0.1115800067782402, "qparams": { "group_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": 0.11035338044166565, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09916462004184723, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09488032013177872, "qparams": { "group_size": { "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": 0.06144437938928604, "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": 0.053308356553316116, "qparams": { "group_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": 0.052154187113046646, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.051884688436985016, "qparams": { "group_size": { "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": 0.030709441751241684, "qparams": { "group_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": 0.026683365926146507, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026610348373651505, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024662543088197708, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024494249373674393, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01616526022553444, "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": 0.015862295404076576, "qparams": { "group_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": 0.015790605917572975, "qparams": { "group_size": { "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": 0.010256389155983925, "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": 0.010256389155983925, "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": 0.010256389155983925, "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": 0.22548192739486694, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19965259730815887, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18714402616024017, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16623109579086304, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10257546603679657, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09141542762517929, "qparams": { "group_size": { "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": 0.12538635730743408, "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": 0.11552596092224121, "qparams": { "group_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": 0.10649663209915161, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0879693254828453, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08356763422489166, "qparams": { "group_size": { "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": 0.0636359304189682, "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": 0.05517829209566116, "qparams": { "group_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": 0.049360089004039764, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04791581258177757, "qparams": { "group_size": { "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": 0.031807564198970795, "qparams": { "group_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": 0.02579004503786564, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025203775614500046, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022910529747605324, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021954528987407684, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016864709556102753, "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": 0.01690259762108326, "qparams": { "group_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": 0.014991775155067444, "qparams": { "group_size": { "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": 0.011404911987483501, "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": 0.014991775155067444, "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": 0.014991775155067444, "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": 0.11926151067018509, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1123838871717453, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10977362096309662, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10011056810617447, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05576455593109131, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05307979881763458, "qparams": { "group_size": { "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": 0.06306381523609161, "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": 0.05808448791503906, "qparams": { "group_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": 0.05649878829717636, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05056626349687576, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.048424702137708664, "qparams": { "group_size": { "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": 0.03199617937207222, "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": 0.02773059345781803, "qparams": { "group_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": 0.026665491983294487, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02641177549958229, "qparams": { "group_size": { "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": 0.015990186482667923, "qparams": { "group_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": 0.013632796704769135, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013530869968235493, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012566603720188141, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012407134287059307, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008323589339852333, "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": 0.008162942714989185, "qparams": { "group_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": 0.007974958047270775, "qparams": { "group_size": { "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": 0.005199331324547529, "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": 0.013632796704769135, "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": 0.013632796704769135, "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": 0.10264293849468231, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09675808995962143, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09430745989084244, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08607636392116547, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04796970263123512, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0456024594604969, "qparams": { "group_size": { "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": 0.05470235273241997, "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": 0.05017761141061783, "qparams": { "group_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": 0.04860474541783333, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04349944740533829, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.041761282831430435, "qparams": { "group_size": { "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": 0.027730830013751984, "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": 0.023954633623361588, "qparams": { "group_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": 0.022961029782891273, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02272285521030426, "qparams": { "group_size": { "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": 0.013848350383341312, "qparams": { "group_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": 0.011745457537472248, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011644361540675163, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010829686187207699, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010679984465241432, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007222762797027826, "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": 0.007062830496579409, "qparams": { "group_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": 0.00690086092799902, "qparams": { "group_size": { "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": 0.004504044074565172, "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": 0.013848350383341312, "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": 0.013848350383341312, "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": 0.23293502628803253, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21979045867919922, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2156248390674591, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19689665734767914, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10923784971237183, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10452587157487869, "qparams": { "group_size": { "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": 0.12162306904792786, "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": 0.1122136041522026, "qparams": { "group_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": 0.11063408851623535, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09912131726741791, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09461913257837296, "qparams": { "group_size": { "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": 0.06181648001074791, "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": 0.053545549511909485, "qparams": { "group_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": 0.05218014493584633, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.051854588091373444, "qparams": { "group_size": { "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": 0.030828841030597687, "qparams": { "group_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": 0.026569921523332596, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02646825648844242, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024474594742059708, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024273162707686424, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01592566817998886, "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": 0.015636412426829338, "qparams": { "group_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": 0.015417229384183884, "qparams": { "group_size": { "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": 0.009814541786909103, "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": 0.015636412426829338, "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": 0.015417229384183884, "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": 0.18696622550487518, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15445588529109955, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14224328100681305, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11391570419073105, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08483023196458817, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07132407277822495, "qparams": { "group_size": { "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": 0.10347040742635727, "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": 0.09452234953641891, "qparams": { "group_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": 0.08944149315357208, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06443764269351959, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05706513300538063, "qparams": { "group_size": { "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": 0.053221363574266434, "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": 0.045874908566474915, "qparams": { "group_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": 0.041531287133693695, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04044324904680252, "qparams": { "group_size": { "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": 0.026765653863549232, "qparams": { "group_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": 0.022631846368312836, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02231474220752716, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018881160765886307, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018163885921239853, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014819735661149025, "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": 0.015875017270445824, "qparams": { "group_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": 0.013446495868265629, "qparams": { "group_size": { "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": 0.012143590487539768, "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": 0.014819735661149025, "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": 0.014819735661149025, "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": 0.19195400178432465, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18145610392093658, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17815732955932617, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16307643055915833, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09028833359479904, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08652714639902115, "qparams": { "group_size": { "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": 0.10000863671302795, "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": 0.09247331321239471, "qparams": { "group_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": 0.09140057861804962, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08205083012580872, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07839825004339218, "qparams": { "group_size": { "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": 0.05096709355711937, "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": 0.04423123598098755, "qparams": { "group_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": 0.04323991760611534, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04300733655691147, "qparams": { "group_size": { "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": 0.025437884032726288, "qparams": { "group_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": 0.022299064323306084, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02223636955022812, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020634982734918594, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020487980917096138, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013419123366475105, "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": 0.013520401902496815, "qparams": { "group_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": 0.013095893897116184, "qparams": { "group_size": { "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": 0.00908749271184206, "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": 0.013419123366475105, "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": 0.013419123366475105, "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": 0.23259775340557098, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21986165642738342, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21601031720638275, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1976940780878067, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10915403068065643, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10463405400514603, "qparams": { "group_size": { "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": 0.1208869069814682, "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": 0.11173884570598602, "qparams": { "group_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": 0.1104818731546402, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09924694150686264, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09483117610216141, "qparams": { "group_size": { "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": 0.06138190999627113, "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": 0.05331628769636154, "qparams": { "group_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": 0.0521451011300087, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05185966566205025, "qparams": { "group_size": { "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": 0.030634645372629166, "qparams": { "group_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": 0.026548592373728752, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026470912620425224, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024506719782948494, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02432839758694172, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01593499444425106, "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": 0.01558273658156395, "qparams": { "group_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": 0.015551875345408916, "qparams": { "group_size": { "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": 0.009800790809094906, "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": 0.01558273658156395, "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": 6.231277126736111, "total_bits": 441042816.0, "err": 0.015551875345408916, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.22476579248905182, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19916687905788422, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18693704903125763, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16643086075782776, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10243634879589081, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09146319329738617, "qparams": { "group_size": { "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": 0.12511421740055084, "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": 0.1150251179933548, "qparams": { "group_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": 0.10635383427143097, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08807962387800217, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08378211408853531, "qparams": { "group_size": { "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": 0.06390541791915894, "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": 0.055164702236652374, "qparams": { "group_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": 0.04941888898611069, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04799308255314827, "qparams": { "group_size": { "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": 0.03223205357789993, "qparams": { "group_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": 0.02601214312016964, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02544625662267208, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023207195103168488, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022281331941485405, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017520461231470108, "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": 0.017262570559978485, "qparams": { "group_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": 0.01569712534546852, "qparams": { "group_size": { "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": 0.011983683332800865, "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": 0.011983683332800865, "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": 0.011983683332800865, "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": 0.12878069281578064, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12136729061603546, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.118661068379879, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10827606171369553, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.060231808573007584, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.057430848479270935, "qparams": { "group_size": { "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": 0.06759648770093918, "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": 0.06252900511026382, "qparams": { "group_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": 0.06101961061358452, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.054613642394542694, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.052087437361478806, "qparams": { "group_size": { "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": 0.034255918115377426, "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": 0.02982965111732483, "qparams": { "group_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": 0.028792642056941986, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.028544392436742783, "qparams": { "group_size": { "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": 0.017097195610404015, "qparams": { "group_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": 0.014673800207674503, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014581335708498955, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0135167445987463, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013358982279896736, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008847350254654884, "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": 0.008715363219380379, "qparams": { "group_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": 0.008499943651258945, "qparams": { "group_size": { "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": 0.005482120905071497, "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": 0.014673800207674503, "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": 0.014581335708498955, "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.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": 0.1091785579919815, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10294509679079056, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10049550980329514, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09173735976219177, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05111154541373253, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04868502914905548, "qparams": { "group_size": { "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": 0.05775643140077591, "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": 0.05322940647602081, "qparams": { "group_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": 0.05177735537290573, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.046373769640922546, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04436855390667915, "qparams": { "group_size": { "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": 0.029309356585144997, "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": 0.025408565998077393, "qparams": { "group_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": 0.024456709623336792, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.024219904094934464, "qparams": { "group_size": { "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": 0.01463855430483818, "qparams": { "group_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": 0.012481080368161201, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01239327248185873, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01150671299546957, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0113620450720191, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00759613374248147, "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": 0.007450253702700138, "qparams": { "group_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": 0.007278898730874062, "qparams": { "group_size": { "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": 0.004716535564512014, "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": 0.01463855430483818, "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": 0.01463855430483818, "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": 0.23699918389320374, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22363629937171936, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2194158434867859, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20046751201152802, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11111751943826675, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.1063397079706192, "qparams": { "group_size": { "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": 0.12356395274400711, "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": 0.11404917389154434, "qparams": { "group_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": 0.11253128200769424, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.1008196547627449, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09623794257640839, "qparams": { "group_size": { "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": 0.06267649680376053, "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": 0.054394517093896866, "qparams": { "group_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": 0.053051937371492386, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05273575708270073, "qparams": { "group_size": { "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": 0.031280238181352615, "qparams": { "group_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": 0.02694196254014969, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02685198001563549, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02480386570096016, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024600207805633545, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016088899224996567, "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": 0.015731245279312134, "qparams": { "group_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": 0.015600892715156078, "qparams": { "group_size": { "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": 0.00971923116594553, "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": 0.015600892715156078, "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": 0.015600892715156078, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.17576001584529877, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1358279436826706, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1191195622086525, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09609098732471466, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07705257087945938, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06165752932429314, "qparams": { "group_size": { "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": 0.09923604130744934, "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": 0.09131504595279694, "qparams": { "group_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": 0.0838567391037941, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.055486682802438736, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05271251127123833, "qparams": { "group_size": { "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": 0.05090872570872307, "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": 0.043826501816511154, "qparams": { "group_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": 0.03730005398392677, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.035624388605356216, "qparams": { "group_size": { "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": 0.025517644360661507, "qparams": { "group_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": 0.01965949870646, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01914929784834385, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.015520213171839714, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014279473572969437, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013673552311956882, "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": 0.013472612015902996, "qparams": { "group_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": 0.011608442291617393, "qparams": { "group_size": { "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": 0.009131302125751972, "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": 0.015520213171839714, "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": 0.015520213171839714, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.4, 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": 0.19269947707653046, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18205863237380981, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1787981390953064, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16347075998783112, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09064116328954697, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08686775714159012, "qparams": { "group_size": { "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": 0.1004854142665863, "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": 0.09285824000835419, "qparams": { "group_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": 0.091769739985466, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08233743906021118, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07857529819011688, "qparams": { "group_size": { "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": 0.051156725734472275, "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": 0.044410526752471924, "qparams": { "group_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": 0.043409328907728195, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.043178778141736984, "qparams": { "group_size": { "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": 0.02555854804813862, "qparams": { "group_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": 0.022335028275847435, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02226925455033779, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020649360492825508, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0204994585365057, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013470090925693512, "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": 0.013474242761731148, "qparams": { "group_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": 0.013143316842615604, "qparams": { "group_size": { "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": 0.008968919515609741, "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": 0.013470090925693512, "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": 0.013470090925693512, "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": 0.23396345973014832, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22101639211177826, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2171136885881424, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1985500156879425, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10977882891893387, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10524293035268784, "qparams": { "group_size": { "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": 0.12151844054460526, "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": 0.11240670830011368, "qparams": { "group_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": 0.11116738617420197, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09978266060352325, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09523050487041473, "qparams": { "group_size": { "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": 0.06173831969499588, "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": 0.053635843098163605, "qparams": { "group_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": 0.05244334414601326, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05216066166758537, "qparams": { "group_size": { "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": 0.030781030654907227, "qparams": { "group_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": 0.02666018344461918, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02658245526254177, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0245863925665617, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02440682426095009, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015943285077810287, "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": 0.015595689415931702, "qparams": { "group_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": 0.015552014112472534, "qparams": { "group_size": { "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": 0.00971982628107071, "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": 0.015595689415931702, "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": 6.231277126736111, "total_bits": 441042816.0, "err": 0.015552014112472534, "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": 0.22440767288208008, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19957572221755981, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.187491774559021, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1671353280544281, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10235929489135742, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0915655568242073, "qparams": { "group_size": { "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": 0.12470963597297668, "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": 0.11488964408636093, "qparams": { "group_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": 0.10609712451696396, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08817103505134583, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08380059897899628, "qparams": { "group_size": { "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": 0.06331876665353775, "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": 0.05486731976270676, "qparams": { "group_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": 0.04922250658273697, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04781423136591911, "qparams": { "group_size": { "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": 0.031683407723903656, "qparams": { "group_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": 0.025614524260163307, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025036398321390152, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02280067652463913, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02186141349375248, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016799915581941605, "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": 0.016631001606583595, "qparams": { "group_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": 0.01497088186442852, "qparams": { "group_size": { "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": 0.011054286733269691, "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": 0.01497088186442852, "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": 0.01497088186442852, "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": 0.12701910734176636, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1196383535861969, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11686813086271286, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10649372637271881, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0594797320663929, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.056596577167510986, "qparams": { "group_size": { "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": 0.06697437167167664, "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": 0.06188283860683441, "qparams": { "group_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": 0.06025199592113495, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05381951853632927, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05137144774198532, "qparams": { "group_size": { "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": 0.03400333970785141, "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": 0.029568085446953773, "qparams": { "group_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": 0.02844918519258499, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02819058671593666, "qparams": { "group_size": { "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": 0.016984427347779274, "qparams": { "group_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": 0.014561435207724571, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01445773895829916, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013411005027592182, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013243203051388264, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00885497871786356, "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": 0.008748509921133518, "qparams": { "group_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": 0.008483772166073322, "qparams": { "group_size": { "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": 0.005608329549431801, "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": 0.014561435207724571, "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": 0.014561435207724571, "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": 0.10781118273735046, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10153287649154663, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09902732074260712, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0903095006942749, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.050436832010746, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04795841872692108, "qparams": { "group_size": { "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": 0.0571279413998127, "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": 0.05268001928925514, "qparams": { "group_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": 0.051121264696121216, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04568011686205864, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04368126019835472, "qparams": { "group_size": { "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": 0.028992753475904465, "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": 0.025137100368738174, "qparams": { "group_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": 0.024128971621394157, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.023887939751148224, "qparams": { "group_size": { "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": 0.014476316049695015, "qparams": { "group_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": 0.012320246547460556, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012223871424794197, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011345171369612217, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011188678443431854, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007510256953537464, "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": 0.007372081279754639, "qparams": { "group_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": 0.007174524012953043, "qparams": { "group_size": { "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": 0.004659418482333422, "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": 0.014476316049695015, "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": 0.014476316049695015, "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": 0.23213587701320648, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2188679724931717, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21453046798706055, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.195770725607872, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10872216522693634, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10389271378517151, "qparams": { "group_size": { "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": 0.12105760723352432, "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": 0.11178537458181381, "qparams": { "group_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": 0.11015856266021729, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09846781194210052, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09398464858531952, "qparams": { "group_size": { "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": 0.061444997787475586, "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": 0.053281333297491074, "qparams": { "group_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": 0.05190010741353035, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.051560595631599426, "qparams": { "group_size": { "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": 0.030639421194791794, "qparams": { "group_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": 0.026329733431339264, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026224078610539436, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024206416681408882, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02399439364671707, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01574418880045414, "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": 0.015343580394983292, "qparams": { "group_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": 0.015229170210659504, "qparams": { "group_size": { "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": 0.009404277428984642, "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": 0.015343580394983292, "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": 0.015229170210659504, "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": 0.19418838620185852, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.140753835439682, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.118857242166996, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09474141150712967, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0820087194442749, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.058647263795137405, "qparams": { "group_size": { "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": 0.10946749150753021, "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": 0.0994262769818306, "qparams": { "group_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": 0.09176965802907944, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.057295188307762146, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05559943988919258, "qparams": { "group_size": { "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": 0.055857930332422256, "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": 0.0482214093208313, "qparams": { "group_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": 0.04046645388007164, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03843877092003822, "qparams": { "group_size": { "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": 0.028396159410476685, "qparams": { "group_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": 0.022445574402809143, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021948186680674553, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018085500225424767, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016718106344342232, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016050046309828758, "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": 0.016651835292577744, "qparams": { "group_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": 0.01374353189021349, "qparams": { "group_size": { "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": 0.012706518173217773, "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": 0.01374353189021349, "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": 0.01374353189021349, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.19351781904697418, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1827496886253357, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1794222891330719, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16402707993984222, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09112613648176193, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08728977292776108, "qparams": { "group_size": { "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": 0.10117917507886887, "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": 0.09335850924253464, "qparams": { "group_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": 0.09227541834115982, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08271101117134094, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07898469269275665, "qparams": { "group_size": { "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": 0.051580771803855896, "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": 0.04471233859658241, "qparams": { "group_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": 0.04369771480560303, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04345434159040451, "qparams": { "group_size": { "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": 0.02586038038134575, "qparams": { "group_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": 0.022567780688405037, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022493278607726097, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020868655294179916, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020717274397611618, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013853120617568493, "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": 0.013735941611230373, "qparams": { "group_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": 0.013525987043976784, "qparams": { "group_size": { "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": 0.009299129247665405, "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": 0.013853120617568493, "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": 0.013853120617568493, "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": 0.23615802824497223, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22305424511432648, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21898473799228668, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20018963515758514, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11089310795068741, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10626158863306046, "qparams": { "group_size": { "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": 0.12305319309234619, "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": 0.11357197910547256, "qparams": { "group_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": 0.1122952252626419, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10068628937005997, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09617435187101364, "qparams": { "group_size": { "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": 0.06255722045898438, "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": 0.05426858365535736, "qparams": { "group_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": 0.053052689880132675, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0527653731405735, "qparams": { "group_size": { "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": 0.03132639080286026, "qparams": { "group_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": 0.027121849358081818, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02704611048102379, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02502688206732273, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024844206869602203, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01654948852956295, "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": 0.01608491688966751, "qparams": { "group_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": 0.016159556806087494, "qparams": { "group_size": { "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": 0.010343726724386215, "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": 0.010343726724386215, "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": 0.010343726724386215, "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": 0.22913886606693268, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20499399304389954, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19286465644836426, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17222437262535095, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10498219728469849, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09425251185894012, "qparams": { "group_size": { "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": 0.1275111585855484, "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": 0.11769124865531921, "qparams": { "group_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": 0.10852666944265366, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.090985506772995, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08652035146951675, "qparams": { "group_size": { "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": 0.06491854041814804, "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": 0.056541454046964645, "qparams": { "group_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": 0.050751153379678726, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04931160435080528, "qparams": { "group_size": { "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": 0.032638393342494965, "qparams": { "group_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": 0.02690228819847107, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026311365887522697, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02419591136276722, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023274105042219162, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.017445754259824753, "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": 0.018054768443107605, "qparams": { "group_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": 0.015546786598861217, "qparams": { "group_size": { "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": 0.012818210758268833, "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": 0.015546786598861217, "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": 0.015546786598861217, "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": 0.12954790890216827, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12210850417613983, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11938467621803284, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10887011885643005, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.060723185539245605, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.057893332093954086, "qparams": { "group_size": { "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": 0.06811666488647461, "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": 0.06295686215162277, "qparams": { "group_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": 0.061536338180303574, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05501125380396843, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05244524031877518, "qparams": { "group_size": { "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": 0.034516461193561554, "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": 0.030052561312913895, "qparams": { "group_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": 0.029019460082054138, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.028776563704013824, "qparams": { "group_size": { "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": 0.017227202653884888, "qparams": { "group_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": 0.014781788922846317, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014692208729684353, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013603532686829567, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013449911959469318, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008901895023882389, "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": 0.008751443587243557, "qparams": { "group_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": 0.008558200672268867, "qparams": { "group_size": { "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": 0.0054804603569209576, "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": 0.014781788922846317, "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": 0.014692208729684353, "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.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": 0.11085300147533417, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10443099588155746, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10201143473386765, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09303870797157288, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05192791670560837, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04943683370947838, "qparams": { "group_size": { "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": 0.058567557483911514, "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": 0.0539931058883667, "qparams": { "group_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": 0.05260747671127319, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04703402519226074, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.044949691742658615, "qparams": { "group_size": { "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": 0.02970639429986477, "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": 0.025755412876605988, "qparams": { "group_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": 0.02482679672539234, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.024601206183433533, "qparams": { "group_size": { "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": 0.014829776249825954, "qparams": { "group_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": 0.012654637917876244, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012565527111291885, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011648941785097122, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011507018469274044, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007679651491343975, "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": 0.007511540316045284, "qparams": { "group_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": 0.007367512211203575, "qparams": { "group_size": { "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": 0.004705808125436306, "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": 0.014829776249825954, "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": 0.014829776249825954, "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": 0.2369992583990097, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22352559864521027, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21924929320812225, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20010298490524292, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11118492484092712, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10635396093130112, "qparams": { "group_size": { "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": 0.12360873818397522, "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": 0.11411338299512863, "qparams": { "group_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": 0.1126222014427185, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10074542462825775, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09604746848344803, "qparams": { "group_size": { "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": 0.06270290166139603, "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": 0.05443032085895538, "qparams": { "group_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": 0.05306810140609741, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05274876952171326, "qparams": { "group_size": { "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": 0.03126997500658035, "qparams": { "group_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": 0.026918239891529083, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026816170662641525, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024748625233769417, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02454698085784912, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01605713926255703, "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": 0.01565217413008213, "qparams": { "group_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": 0.015558946877717972, "qparams": { "group_size": { "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": 0.009570715017616749, "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": 0.015558946877717972, "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": 0.015558946877717972, "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.o_proj", "numel": 26214400, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.17529296875, "total_bits": 57024000.0, "err": 0.20410597324371338, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14712122082710266, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12070908397436142, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09873905777931213, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08603765070438385, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0646277666091919, "qparams": { "group_size": { "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": 0.11867810785770416, "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": 0.10911842435598373, "qparams": { "group_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": 0.0978950709104538, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06152309477329254, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06032470613718033, "qparams": { "group_size": { "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": 0.061144739389419556, "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": 0.05273108929395676, "qparams": { "group_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": 0.04223534092307091, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03939865902066231, "qparams": { "group_size": { "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": 0.03082113154232502, "qparams": { "group_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": 0.023153619840741158, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022379497066140175, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018755368888378143, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016826050356030464, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01701373979449272, "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": 0.017336437478661537, "qparams": { "group_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": 0.0138488644734025, "qparams": { "group_size": { "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": 0.012691011652350426, "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": 0.0138488644734025, "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": 0.0138488644734025, "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": 0.19386643171310425, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18308572471141815, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.179754376411438, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16415083408355713, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09124868363142014, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.087418332695961, "qparams": { "group_size": { "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": 0.10136078298091888, "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": 0.09355504810810089, "qparams": { "group_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": 0.09243478626012802, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08281321078538895, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07912169396877289, "qparams": { "group_size": { "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": 0.05169358476996422, "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": 0.04476671293377876, "qparams": { "group_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": 0.0437367744743824, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.043491803109645844, "qparams": { "group_size": { "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": 0.025861753150820732, "qparams": { "group_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": 0.022558365017175674, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022486208006739616, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020851856097579002, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020696938037872314, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013752739876508713, "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": 0.013698550872504711, "qparams": { "group_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": 0.01341722346842289, "qparams": { "group_size": { "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": 0.009213758632540703, "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": 0.013752739876508713, "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": 0.013752739876508713, "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": 0.23652249574661255, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2233547866344452, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21930120885372162, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20036594569683075, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11107922345399857, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10641884058713913, "qparams": { "group_size": { "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": 0.12301968038082123, "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": 0.11380349099636078, "qparams": { "group_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": 0.1124904453754425, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10083053261041641, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09617866575717926, "qparams": { "group_size": { "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": 0.06254784762859344, "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": 0.0543162003159523, "qparams": { "group_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": 0.053086407482624054, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.052798643708229065, "qparams": { "group_size": { "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": 0.03119577094912529, "qparams": { "group_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": 0.027019251137971878, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02693977579474449, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024898353964090347, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024715052917599678, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016215113922953606, "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": 0.015859579667448997, "qparams": { "group_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": 0.0158074963837862, "qparams": { "group_size": { "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": 0.009951029904186726, "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": 0.009951029904186726, "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": 0.009951029904186726, "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": 0.22513094544410706, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2017795741558075, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18942289054393768, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16909432411193848, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10308198630809784, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09249545633792877, "qparams": { "group_size": { "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": 0.12706778943538666, "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": 0.11644642055034637, "qparams": { "group_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": 0.10657133907079697, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08955328166484833, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0854891762137413, "qparams": { "group_size": { "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": 0.06491124629974365, "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": 0.05577767267823219, "qparams": { "group_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": 0.04973649978637695, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0482163205742836, "qparams": { "group_size": { "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": 0.03291484713554382, "qparams": { "group_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": 0.026158330962061882, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025509938597679138, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023510226979851723, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022526908665895462, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.018164757639169693, "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": 0.017387591302394867, "qparams": { "group_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": 0.01623910292983055, "qparams": { "group_size": { "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": 0.011988045647740364, "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": 0.011988045647740364, "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": 0.011988045647740364, "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": 0.13443177938461304, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.126692533493042, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12385853379964828, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11292057484388351, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0630279928445816, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06005684286355972, "qparams": { "group_size": { "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": 0.07078997045755386, "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": 0.06540456414222717, "qparams": { "group_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": 0.06386265903711319, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05706740915775299, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05443922057747841, "qparams": { "group_size": { "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": 0.03594009578227997, "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": 0.03121691569685936, "qparams": { "group_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": 0.03013872168958187, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.029884323477745056, "qparams": { "group_size": { "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": 0.01794656179845333, "qparams": { "group_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": 0.015368819236755371, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01527561154216528, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.014142274856567383, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013980467803776264, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00931928213685751, "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": 0.009130267426371574, "qparams": { "group_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": 0.008955669589340687, "qparams": { "group_size": { "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": 0.005755279678851366, "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": 0.015368819236755371, "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": 0.01527561154216528, "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": 0.11473400145769119, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10810170322656631, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.105622798204422, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09633392095565796, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05381272733211517, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05125501751899719, "qparams": { "group_size": { "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": 0.0607832707464695, "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": 0.05594492331147194, "qparams": { "group_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": 0.05452919378876686, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04873485118150711, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04661397263407707, "qparams": { "group_size": { "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": 0.030857425183057785, "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": 0.0267078448086977, "qparams": { "group_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": 0.02574077621102333, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02551240660250187, "qparams": { "group_size": { "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": 0.01540906261652708, "qparams": { "group_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": 0.013135199435055256, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013046552427113056, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012092992663383484, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01194439921528101, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008006464689970016, "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": 0.007826326414942741, "qparams": { "group_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": 0.0076838526874780655, "qparams": { "group_size": { "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": 0.00494012888520956, "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": 0.01540906261652708, "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": 0.01540906261652708, "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": 0.23973247408866882, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.226060688495636, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2217993289232254, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20241957902908325, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11246828734874725, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.1075887456536293, "qparams": { "group_size": { "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": 0.12497391551733017, "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": 0.11541768163442612, "qparams": { "group_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": 0.11391040682792664, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10187786817550659, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09712786227464676, "qparams": { "group_size": { "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": 0.06342565268278122, "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": 0.055053550750017166, "qparams": { "group_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": 0.05369747057557106, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0533762127161026, "qparams": { "group_size": { "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": 0.031627852469682693, "qparams": { "group_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": 0.027254778891801834, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02715807408094406, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.0250544436275959, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02485041320323944, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016274815425276756, "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": 0.015884744003415108, "qparams": { "group_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": 0.015773998573422432, "qparams": { "group_size": { "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": 0.009771538898348808, "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": 0.009771538898348808, "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": 0.009771538898348808, "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": 0.2186415195465088, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1647917628288269, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14110510051250458, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11313688009977341, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09714899957180023, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07492668926715851, "qparams": { "group_size": { "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": 0.1254958212375641, "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": 0.11520253866910934, "qparams": { "group_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": 0.10492511093616486, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06430713832378387, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06484650075435638, "qparams": { "group_size": { "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": 0.06447536498308182, "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": 0.055443018674850464, "qparams": { "group_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": 0.04732148349285126, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04522906243801117, "qparams": { "group_size": { "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": 0.03231959417462349, "qparams": { "group_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": 0.025434881448745728, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024751830846071243, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01941077411174774, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017835887148976326, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0174417644739151, "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": 0.01796034909784794, "qparams": { "group_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": 0.014877174980938435, "qparams": { "group_size": { "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": 0.012909923680126667, "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": 0.014877174980938435, "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": 0.014877174980938435, "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": 0.1895413100719452, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17888657748699188, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17556075751781464, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16041991114616394, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08924752473831177, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08545277267694473, "qparams": { "group_size": { "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": 0.09893615543842316, "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": 0.09145819395780563, "qparams": { "group_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": 0.09037454426288605, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08094263076782227, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0772114098072052, "qparams": { "group_size": { "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": 0.05038151144981384, "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": 0.043760765343904495, "qparams": { "group_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": 0.04275514557957649, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.042513176798820496, "qparams": { "group_size": { "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": 0.025170739740133286, "qparams": { "group_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": 0.022018415853381157, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021952450275421143, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02033371478319168, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02017892524600029, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013250870630145073, "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": 0.013317872770130634, "qparams": { "group_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": 0.01292251143604517, "qparams": { "group_size": { "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": 0.008896388113498688, "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": 0.013250870630145073, "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": 0.013250870630145073, "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": 0.23633329570293427, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2230612337589264, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21897996962070465, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20007920265197754, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11102215945720673, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10633835941553116, "qparams": { "group_size": { "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": 0.12307160347700119, "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": 0.11374011635780334, "qparams": { "group_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": 0.11244969069957733, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.1007152870297432, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09606105834245682, "qparams": { "group_size": { "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": 0.06259388476610184, "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": 0.05431722104549408, "qparams": { "group_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": 0.05307924747467041, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05278927460312843, "qparams": { "group_size": { "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": 0.03122761845588684, "qparams": { "group_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": 0.027064377442002296, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02698574960231781, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024937452748417854, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024751901626586914, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01626361720263958, "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": 0.015948927029967308, "qparams": { "group_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": 0.015858719125390053, "qparams": { "group_size": { "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": 0.010095268487930298, "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": 0.010095268487930298, "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": 0.010095268487930298, "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": 0.22764617204666138, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20413656532764435, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19213928282260895, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17172296345233917, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10429057478904724, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09383750706911087, "qparams": { "group_size": { "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": 0.12713386118412018, "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": 0.11720135807991028, "qparams": { "group_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": 0.10781863331794739, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0907556489109993, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08624445647001266, "qparams": { "group_size": { "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": 0.06458059698343277, "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": 0.056210506707429886, "qparams": { "group_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": 0.05037476122379303, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0489196702837944, "qparams": { "group_size": { "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": 0.032421767711639404, "qparams": { "group_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": 0.026591263711452484, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025983605533838272, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023957351222634315, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023023685440421104, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01736438274383545, "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": 0.01774054951965809, "qparams": { "group_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": 0.015515157952904701, "qparams": { "group_size": { "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": 0.012425323948264122, "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": 0.015515157952904701, "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": 0.015515157952904701, "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": 0.12984225153923035, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12225769460201263, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11945244669914246, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10883186757564545, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.06088056415319443, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.057952795177698135, "qparams": { "group_size": { "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": 0.06835426390171051, "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": 0.0632466971874237, "qparams": { "group_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": 0.061701688915491104, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05507480353116989, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.052459873259067535, "qparams": { "group_size": { "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": 0.03465983271598816, "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": 0.030173959210515022, "qparams": { "group_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": 0.029111869633197784, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.028861887753009796, "qparams": { "group_size": { "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": 0.017301063984632492, "qparams": { "group_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": 0.014845729805529118, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014747949317097664, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013642469421029091, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013483419083058834, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008960245177149773, "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": 0.008818983100354671, "qparams": { "group_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": 0.00860685296356678, "qparams": { "group_size": { "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": 0.0055489917285740376, "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": 0.014845729805529118, "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": 0.014747949317097664, "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.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": 0.1099461019039154, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10348565131425858, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10106152296066284, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09211961179971695, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05153203755617142, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.049038782715797424, "qparams": { "group_size": { "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": 0.05817960947751999, "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": 0.05363790690898895, "qparams": { "group_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": 0.052242159843444824, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04663655906915665, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04456707090139389, "qparams": { "group_size": { "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": 0.029510557651519775, "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": 0.02560908906161785, "qparams": { "group_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": 0.02465357631444931, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02442769519984722, "qparams": { "group_size": { "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": 0.014748649671673775, "qparams": { "group_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": 0.012591177597641945, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01249887328594923, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011579261161386967, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01143847219645977, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0076635051518678665, "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": 0.007521223276853561, "qparams": { "group_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": 0.007346896920353174, "qparams": { "group_size": { "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": 0.004765493329614401, "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": 0.014748649671673775, "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": 0.014748649671673775, "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": 0.2407146394252777, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22683461010456085, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.2225644737482071, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20276713371276855, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11293333023786545, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10796660929918289, "qparams": { "group_size": { "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": 0.1254996806383133, "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": 0.11594151705503464, "qparams": { "group_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": 0.11442124843597412, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10220057517290115, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09738021343946457, "qparams": { "group_size": { "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": 0.06372786313295364, "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": 0.0552898608148098, "qparams": { "group_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": 0.05392463877797127, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05359703302383423, "qparams": { "group_size": { "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": 0.03175634518265724, "qparams": { "group_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": 0.02733570896089077, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02724219113588333, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02510290965437889, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024904189631342888, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01630820706486702, "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": 0.015882711857557297, "qparams": { "group_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": 0.01580725610256195, "qparams": { "group_size": { "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": 0.00969433318823576, "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": 0.00969433318823576, "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": 0.00969433318823576, "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": 0.18530340492725372, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1397499144077301, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11872652173042297, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10229755192995071, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0841292068362236, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06438503414392471, "qparams": { "group_size": { "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": 0.1084364652633667, "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": 0.09902551025152206, "qparams": { "group_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": 0.0889337956905365, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05926324054598808, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05666268616914749, "qparams": { "group_size": { "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": 0.05594823136925697, "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": 0.047713153064250946, "qparams": { "group_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": 0.040891554206609726, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03913387656211853, "qparams": { "group_size": { "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": 0.027966836467385292, "qparams": { "group_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": 0.021801074966788292, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021120302379131317, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017111994326114655, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.015855632722377777, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014992063865065575, "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": 0.015146358869969845, "qparams": { "group_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": 0.012717941775918007, "qparams": { "group_size": { "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": 0.010706944391131401, "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": 0.014992063865065575, "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": 0.014992063865065575, "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": 0.18757647275924683, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17695023119449615, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17361952364444733, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15847665071487427, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08833441138267517, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0845254510641098, "qparams": { "group_size": { "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": 0.09793386608362198, "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": 0.0905693843960762, "qparams": { "group_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": 0.08948789536952972, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08003611862659454, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0763232409954071, "qparams": { "group_size": { "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": 0.049884945154190063, "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": 0.04337428882718086, "qparams": { "group_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": 0.04235708341002464, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04211825504899025, "qparams": { "group_size": { "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": 0.02494160458445549, "qparams": { "group_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": 0.02186710014939308, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02180047146975994, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020188985392451286, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02004302106797695, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013169333338737488, "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": 0.013311413116753101, "qparams": { "group_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": 0.012840638868510723, "qparams": { "group_size": { "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": 0.008997407741844654, "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": 0.013169333338737488, "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": 0.013169333338737488, "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": 0.23593540489673615, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22264115512371063, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21857361495494843, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19957567751407623, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11086587607860565, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10613057762384415, "qparams": { "group_size": { "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": 0.12272366881370544, "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": 0.11361072957515717, "qparams": { "group_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": 0.11230389028787613, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10051244497299194, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0958293080329895, "qparams": { "group_size": { "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": 0.06233031675219536, "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": 0.05423856899142265, "qparams": { "group_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": 0.0529916025698185, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05269799754023552, "qparams": { "group_size": { "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": 0.031087545678019524, "qparams": { "group_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": 0.02697618305683136, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026892513036727905, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024832691997289658, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024645207449793816, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016077952459454536, "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": 0.015840210020542145, "qparams": { "group_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": 0.015665341168642044, "qparams": { "group_size": { "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": 0.009942199103534222, "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": 0.009942199103534222, "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": 0.009942199103534222, "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": 0.22926661372184753, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.205442875623703, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19291074573993683, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17241890728473663, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10500558465719223, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09428223222494125, "qparams": { "group_size": { "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": 0.13081791996955872, "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": 0.1186414510011673, "qparams": { "group_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": 0.10869373381137848, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0913572609424591, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08709844201803207, "qparams": { "group_size": { "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": 0.06618180125951767, "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": 0.0569574236869812, "qparams": { "group_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": 0.05080518126487732, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04926128312945366, "qparams": { "group_size": { "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": 0.033577729016542435, "qparams": { "group_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": 0.02696753665804863, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02632821537554264, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024332942441105843, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02335519902408123, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.018696114420890808, "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": 0.018204694613814354, "qparams": { "group_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": 0.01676434837281704, "qparams": { "group_size": { "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": 0.012936626560986042, "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": 0.012936626560986042, "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": 0.012936626560986042, "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": 0.13158151507377625, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12369339913129807, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12082066386938095, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11003471910953522, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.06166842207312584, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05865529924631119, "qparams": { "group_size": { "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": 0.06932087242603302, "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": 0.06411609053611755, "qparams": { "group_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": 0.06249357759952545, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05569785460829735, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0530734583735466, "qparams": { "group_size": { "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": 0.035158149898052216, "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": 0.03059866465628147, "qparams": { "group_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": 0.029494663700461388, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.029225921258330345, "qparams": { "group_size": { "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": 0.017549028620123863, "qparams": { "group_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": 0.015048607252538204, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01495057251304388, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013823071494698524, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013653598725795746, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.009091641753911972, "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": 0.008964424021542072, "qparams": { "group_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": 0.008718286640942097, "qparams": { "group_size": { "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": 0.005662041250616312, "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": 0.015048607252538204, "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": 0.01495057251304388, "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.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": 0.11028677225112915, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10378435254096985, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10124010592699051, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09221695363521576, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.051711149513721466, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04914235696196556, "qparams": { "group_size": { "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": 0.058535441756248474, "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": 0.05391961336135864, "qparams": { "group_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": 0.052426766604185104, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04674453288316727, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04465595632791519, "qparams": { "group_size": { "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": 0.02972141094505787, "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": 0.025741424411535263, "qparams": { "group_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": 0.024739034473896027, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.024498648941516876, "qparams": { "group_size": { "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": 0.014845207333564758, "qparams": { "group_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": 0.012636560946702957, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012541131116449833, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011613543145358562, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011463134549558163, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0077120596542954445, "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": 0.007560080382972956, "qparams": { "group_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": 0.007379457354545593, "qparams": { "group_size": { "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": 0.004793899599462748, "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": 0.014845207333564758, "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": 0.014845207333564758, "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": 0.24559493362903595, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.23128487169742584, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22678007185459137, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.2066962867975235, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11526473611593246, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.1101486012339592, "qparams": { "group_size": { "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": 0.12818914651870728, "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": 0.11835967004299164, "qparams": { "group_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": 0.116839200258255, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10424628108739853, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09920773655176163, "qparams": { "group_size": { "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": 0.06510335206985474, "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": 0.056465402245521545, "qparams": { "group_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": 0.05504129081964493, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05470149219036102, "qparams": { "group_size": { "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": 0.032455675303936005, "qparams": { "group_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": 0.027936171740293503, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.027833983302116394, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02564195543527603, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02542579174041748, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01668456941843033, "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": 0.016279445961117744, "qparams": { "group_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": 0.016163986176252365, "qparams": { "group_size": { "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": 0.009992030449211597, "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": 0.009992030449211597, "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": 0.009992030449211597, "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": 0.20181941986083984, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14590638875961304, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11912627518177032, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09711287915706635, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0848267525434494, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.062294021248817444, "qparams": { "group_size": { "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": 0.11900509893894196, "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": 0.10879692435264587, "qparams": { "group_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": 0.09661657363176346, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05929519981145859, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.060319188982248306, "qparams": { "group_size": { "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": 0.06124038249254227, "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": 0.05228923633694649, "qparams": { "group_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": 0.041264455765485764, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03824923187494278, "qparams": { "group_size": { "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": 0.030619030818343163, "qparams": { "group_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": 0.021983083337545395, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02106579951941967, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017168840393424034, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014921816065907478, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016315868124365807, "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": 0.01596187986433506, "qparams": { "group_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": 0.01273279543966055, "qparams": { "group_size": { "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": 0.010698753409087658, "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": 0.014921816065907478, "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": 0.014921816065907478, "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": 0.18369947373867035, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17310859262943268, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16982367634773254, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1550094485282898, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08650518208742142, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08276057243347168, "qparams": { "group_size": { "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": 0.09594225138425827, "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": 0.08874078840017319, "qparams": { "group_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": 0.08765524625778198, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07833150029182434, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07465463131666183, "qparams": { "group_size": { "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": 0.04891595616936684, "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": 0.042479678988456726, "qparams": { "group_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": 0.041472211480140686, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04123879596590996, "qparams": { "group_size": { "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": 0.024443743750452995, "qparams": { "group_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": 0.02140367589890957, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021335631608963013, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.019750632345676422, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01960103027522564, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01290902215987444, "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": 0.013017025776207447, "qparams": { "group_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": 0.012580107897520065, "qparams": { "group_size": { "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": 0.008774537593126297, "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": 0.01290902215987444, "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": 0.01290902215987444, "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": 0.2335682362318039, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22019651532173157, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21612688899040222, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19726769626140594, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10981758683919907, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10507048666477203, "qparams": { "group_size": { "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": 0.12167254090309143, "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": 0.11253286898136139, "qparams": { "group_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": 0.11123104393482208, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09946910291910172, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09477216750383377, "qparams": { "group_size": { "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": 0.061918843537569046, "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": 0.05375276878476143, "qparams": { "group_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": 0.05250151827931404, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05220292508602142, "qparams": { "group_size": { "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": 0.030883610248565674, "qparams": { "group_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": 0.02677176333963871, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026689039543271065, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02464209869503975, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02445237711071968, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016075683757662773, "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": 0.015783706679940224, "qparams": { "group_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": 0.009997373446822166, "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": 0.009997373446822166, "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": 0.009997373446822166, "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": 0.22210875153541565, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1990455538034439, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18662689626216888, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16693206131458282, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10161957144737244, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09109726548194885, "qparams": { "group_size": { "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": 0.12600207328796387, "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": 0.11536473035812378, "qparams": { "group_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": 0.10524741560220718, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08850864320993423, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0844479650259018, "qparams": { "group_size": { "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": 0.06435629725456238, "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": 0.055214136838912964, "qparams": { "group_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": 0.04901823773980141, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.047449443489313126, "qparams": { "group_size": { "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": 0.032699570059776306, "qparams": { "group_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": 0.02571985311806202, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02505694516003132, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023146918043494225, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022137532010674477, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.018226586282253265, "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": 0.01706933043897152, "qparams": { "group_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": 0.01629301719367504, "qparams": { "group_size": { "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": 0.011656684800982475, "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": 0.011656684800982475, "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": 0.011656684800982475, "qparams": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1.0 ], "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": 0.12684392929077148, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11930840462446213, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11643058061599731, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10597915947437286, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05947529897093773, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.056513164192438126, "qparams": { "group_size": { "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": 0.06704146414995193, "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": 0.06193198263645172, "qparams": { "group_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": 0.06028437986969948, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05371218919754028, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.051153674721717834, "qparams": { "group_size": { "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": 0.03403014317154884, "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": 0.02956709824502468, "qparams": { "group_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": 0.028439855203032494, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02817348763346672, "qparams": { "group_size": { "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": 0.0169928427785635, "qparams": { "group_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": 0.014521750621497631, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014413418248295784, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013336334377527237, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013162833638489246, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008814713917672634, "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": 0.008666628040373325, "qparams": { "group_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": 0.00844193808734417, "qparams": { "group_size": { "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": 0.0054739429615437984, "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": 0.014521750621497631, "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": 0.014521750621497631, "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": 0.10818953067064285, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10178602486848831, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09928789734840393, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09039606153964996, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05071886256337166, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0482114814221859, "qparams": { "group_size": { "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": 0.05728723853826523, "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": 0.05287434533238411, "qparams": { "group_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": 0.05142094939947128, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.045805223286151886, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04371284693479538, "qparams": { "group_size": { "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": 0.029093073680996895, "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": 0.025247223675251007, "qparams": { "group_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": 0.024270152673125267, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.024034926667809486, "qparams": { "group_size": { "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": 0.0145293939858675, "qparams": { "group_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": 0.012397664599120617, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012299849651753902, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01138851698487997, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011240567080676556, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007540025748312473, "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": 0.007414994295686483, "qparams": { "group_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": 0.007215441670268774, "qparams": { "group_size": { "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": 0.004703862592577934, "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": 0.0145293939858675, "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": 0.0145293939858675, "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": 0.2382349967956543, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22431638836860657, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.219955176115036, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20053468644618988, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11184398084878922, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10682304203510284, "qparams": { "group_size": { "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": 0.12438271194696426, "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": 0.11486486345529556, "qparams": { "group_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": 0.11334408074617386, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10109542310237885, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09623781591653824, "qparams": { "group_size": { "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": 0.0631081610918045, "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": 0.05478550121188164, "qparams": { "group_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": 0.05339212715625763, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.053063444793224335, "qparams": { "group_size": { "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": 0.03146781772375107, "qparams": { "group_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": 0.0270707905292511, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026967080309987068, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02483639493584633, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02462165243923664, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01615751162171364, "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": 0.015730956569314003, "qparams": { "group_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": 0.015644598752260208, "qparams": { "group_size": { "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": 0.009581511840224266, "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": 0.009581511840224266, "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": 0.009581511840224266, "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": 0.18916891515254974, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1465279906988144, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12744185328483582, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10379550606012344, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08424469083547592, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06434980034828186, "qparams": { "group_size": { "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": 0.11038479954004288, "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": 0.09966839849948883, "qparams": { "group_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": 0.09024389833211899, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06066526100039482, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.058046791702508926, "qparams": { "group_size": { "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": 0.056328851729631424, "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": 0.04846090078353882, "qparams": { "group_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": 0.0415215864777565, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.039707839488983154, "qparams": { "group_size": { "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": 0.028659913688898087, "qparams": { "group_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": 0.02311497926712036, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022501695901155472, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018973253667354584, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017776384949684143, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01642960123717785, "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": 0.01708993688225746, "qparams": { "group_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": 0.014430688694119453, "qparams": { "group_size": { "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": 0.013236213475465775, "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": 0.014430688694119453, "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": 0.014430688694119453, "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": 0.18155185878276825, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1711346060037613, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16788853704929352, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1531095802783966, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08552181720733643, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08180305361747742, "qparams": { "group_size": { "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": 0.09487038105726242, "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": 0.08774211257696152, "qparams": { "group_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": 0.08664213120937347, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07740316540002823, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07378337532281876, "qparams": { "group_size": { "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": 0.048356544226408005, "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": 0.04200061038136482, "qparams": { "group_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": 0.04100029543042183, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0407656654715538, "qparams": { "group_size": { "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": 0.024153707548975945, "qparams": { "group_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": 0.02114996314048767, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02108117938041687, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.019516246393322945, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01936209388077259, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012743809260427952, "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": 0.012855341657996178, "qparams": { "group_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": 0.012415844947099686, "qparams": { "group_size": { "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": 0.008647552691400051, "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": 0.012743809260427952, "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": 0.012743809260427952, "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": 0.23214028775691986, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21888591349124908, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21479232609272003, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19598661363124847, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1091875433921814, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10446298867464066, "qparams": { "group_size": { "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": 0.12117965519428253, "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": 0.11195672303438187, "qparams": { "group_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": 0.11064065992832184, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09886348992586136, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09421724081039429, "qparams": { "group_size": { "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": 0.061633262783288956, "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": 0.053493063896894455, "qparams": { "group_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": 0.05223247408866882, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05193129926919937, "qparams": { "group_size": { "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": 0.030796783044934273, "qparams": { "group_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": 0.02670016698539257, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02661861851811409, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024576621130108833, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024388067424297333, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016142893582582474, "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": 0.015837781131267548, "qparams": { "group_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": 0.015733681619167328, "qparams": { "group_size": { "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": 0.010166504420340061, "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": 0.010166504420340061, "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": 0.010166504420340061, "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": 0.2300148904323578, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20599065721035004, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19356603920459747, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17293144762516022, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10531333833932877, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0945485383272171, "qparams": { "group_size": { "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": 0.1298852562904358, "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": 0.11876540631055832, "qparams": { "group_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": 0.10907073318958282, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.091544009745121, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08712649345397949, "qparams": { "group_size": { "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": 0.06588541716337204, "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": 0.05689096823334694, "qparams": { "group_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": 0.05079364776611328, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0492582805454731, "qparams": { "group_size": { "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": 0.033445440232753754, "qparams": { "group_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": 0.026667699217796326, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026030724868178368, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023965513333678246, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02297145500779152, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.018494773656129837, "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": 0.01765206642448902, "qparams": { "group_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": 0.0166223905980587, "qparams": { "group_size": { "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": 0.01210356317460537, "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": 0.01210356317460537, "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": 0.01210356317460537, "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": 0.1288144439458847, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12106956541538239, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1181226521730423, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10742956399917603, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.060384150594472885, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05738651379942894, "qparams": { "group_size": { "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": 0.06803260743618011, "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": 0.06295860558748245, "qparams": { "group_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": 0.061221592128276825, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05448857322335243, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.051845841109752655, "qparams": { "group_size": { "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": 0.03448218107223511, "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": 0.03003651462495327, "qparams": { "group_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": 0.028890900313854218, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.028614724054932594, "qparams": { "group_size": { "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": 0.01722881942987442, "qparams": { "group_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": 0.014765461906790733, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014656593091785908, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013549755327403545, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013376078568398952, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008947926573455334, "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": 0.008840566501021385, "qparams": { "group_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": 0.008571014739573002, "qparams": { "group_size": { "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": 0.005612025503069162, "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": 0.014765461906790733, "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": 0.014656593091785908, "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.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": 0.10737495869398117, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10089022666215897, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09833577275276184, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08943884074687958, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0503392368555069, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04776649549603462, "qparams": { "group_size": { "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": 0.05709407478570938, "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": 0.05266003683209419, "qparams": { "group_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": 0.05104487016797066, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.045444898307323456, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04333409294486046, "qparams": { "group_size": { "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": 0.029004856944084167, "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": 0.025158436968922615, "qparams": { "group_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": 0.02410733699798584, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.023860719054937363, "qparams": { "group_size": { "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": 0.014488544315099716, "qparams": { "group_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": 0.012337828055024147, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012233748100697994, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011330886743962765, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011171851307153702, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007544994819909334, "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": 0.0074273389764130116, "qparams": { "group_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": 0.0071997796185314655, "qparams": { "group_size": { "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": 0.0047475905157625675, "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": 0.014488544315099716, "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": 0.014488544315099716, "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": 0.24310700595378876, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22880330681800842, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22432726621627808, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20420247316360474, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11416004598140717, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.109044149518013, "qparams": { "group_size": { "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": 0.12710367143154144, "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": 0.11733506619930267, "qparams": { "group_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": 0.11570701748132706, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10313022881746292, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09818954765796661, "qparams": { "group_size": { "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": 0.06448739767074585, "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": 0.055960316210985184, "qparams": { "group_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": 0.05451676622033119, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.054181523621082306, "qparams": { "group_size": { "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": 0.03216736391186714, "qparams": { "group_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": 0.027673978358507156, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02756873331964016, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.025384197011590004, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02515936829149723, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01652049832046032, "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": 0.016124311834573746, "qparams": { "group_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": 0.015993772074580193, "qparams": { "group_size": { "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": 0.009886009618639946, "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": 0.009886009618639946, "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": 0.009886009618639946, "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": 0.19323715567588806, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15048947930335999, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.13315673172473907, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1000504121184349, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08695286512374878, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06871511787176132, "qparams": { "group_size": { "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": 0.10808644443750381, "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": 0.09967262297868729, "qparams": { "group_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": 0.09234553575515747, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05946751683950424, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05562444031238556, "qparams": { "group_size": { "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": 0.05529224872589111, "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": 0.04774077609181404, "qparams": { "group_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": 0.0419788621366024, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0404965840280056, "qparams": { "group_size": { "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": 0.027634810656309128, "qparams": { "group_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": 0.02199827879667282, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021507499739527702, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.016722887754440308, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.015608656220138073, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014657153747975826, "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": 0.01467018760740757, "qparams": { "group_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": 0.012791529297828674, "qparams": { "group_size": { "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": 0.009927371516823769, "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": 0.015608656220138073, "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": "1.0:6b_128g s4", "bpw": 6.0313232421875, "total_bits": 158107520.0, "err": 0.014657153747975826, "qparams": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1.0 ], "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": 0.17980356514453888, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16946418583393097, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16619397699832916, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1515357941389084, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08473780751228333, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08102238178253174, "qparams": { "group_size": { "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": 0.09398210048675537, "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": 0.08695600926876068, "qparams": { "group_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": 0.08587737381458282, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07667160779237747, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0730544924736023, "qparams": { "group_size": { "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": 0.047927334904670715, "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": 0.04164743423461914, "qparams": { "group_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": 0.04065077751874924, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04041268303990364, "qparams": { "group_size": { "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": 0.023943185806274414, "qparams": { "group_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": 0.020994288846850395, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02092919498682022, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01936146430671215, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019214892759919167, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012642478570342064, "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": 0.012797828763723373, "qparams": { "group_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": 0.012318269349634647, "qparams": { "group_size": { "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": 0.00866234302520752, "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": 0.012642478570342064, "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": 0.012642478570342064, "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": 0.22451338171958923, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21162356436252594, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20763429999351501, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18936799466609955, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10566983371973038, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10106736421585083, "qparams": { "group_size": { "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": 0.11721218377351761, "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": 0.10833904147148132, "qparams": { "group_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": 0.10704845190048218, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09560588747262955, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09114779531955719, "qparams": { "group_size": { "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": 0.059666089713573456, "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": 0.05178660899400711, "qparams": { "group_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": 0.05056130886077881, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05027037486433983, "qparams": { "group_size": { "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": 0.02978621982038021, "qparams": { "group_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": 0.025856448337435722, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025772208347916603, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023792756721377373, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023607291281223297, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.0155717134475708, "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": 0.015358999371528625, "qparams": { "group_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": 0.01516673807054758, "qparams": { "group_size": { "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": 0.009880103170871735, "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": 0.0155717134475708, "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": 0.01516673807054758, "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": 0.22127269208431244, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19698955118656158, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1847093403339386, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16489280760288239, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10098341107368469, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09031195938587189, "qparams": { "group_size": { "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": 0.12414342910051346, "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": 0.1143270954489708, "qparams": { "group_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": 0.10496506094932556, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08746356517076492, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08325899392366409, "qparams": { "group_size": { "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": 0.06322319805622101, "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": 0.05468221381306648, "qparams": { "group_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": 0.048678986728191376, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04717289283871651, "qparams": { "group_size": { "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": 0.03168396279215813, "qparams": { "group_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": 0.025502437725663185, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02489454485476017, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022841256111860275, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02186312898993492, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01696004718542099, "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": 0.016840502619743347, "qparams": { "group_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": 0.015033145435154438, "qparams": { "group_size": { "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": 0.011444130912423134, "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": 0.015033145435154438, "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": 0.015033145435154438, "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": 0.12482133507728577, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1173185482621193, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11442749947309494, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10404200106859207, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05851542577147484, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.055583298206329346, "qparams": { "group_size": { "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": 0.06590639799833298, "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": 0.06098724901676178, "qparams": { "group_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": 0.059326767921447754, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05275820940732956, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05020558461546898, "qparams": { "group_size": { "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": 0.03342242166399956, "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": 0.029106847941875458, "qparams": { "group_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": 0.027986709028482437, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.027722474187612534, "qparams": { "group_size": { "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": 0.016689563170075417, "qparams": { "group_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": 0.014302820898592472, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01419619657099247, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013123219832777977, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012954709120094776, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.008661753498017788, "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": 0.00855918787419796, "qparams": { "group_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": 0.008294216357171535, "qparams": { "group_size": { "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": 0.005429795477539301, "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": 0.014302820898592472, "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": 0.014302820898592472, "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": 0.10346879810094833, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09721051156520844, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09478520601987839, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08622022718191147, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.048514075577259064, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04604676365852356, "qparams": { "group_size": { "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": 0.054927758872509, "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": 0.05069027468562126, "qparams": { "group_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": 0.04918604716658592, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04375443607568741, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04173039644956589, "qparams": { "group_size": { "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": 0.027882494032382965, "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": 0.024200627580285072, "qparams": { "group_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": 0.023218069225549698, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022981343790888786, "qparams": { "group_size": { "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": 0.013935577124357224, "qparams": { "group_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": 0.011881353333592415, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011786329559981823, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010910598561167717, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010761079378426075, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007256083190441132, "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": 0.00714879622682929, "qparams": { "group_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": 0.006934419274330139, "qparams": { "group_size": { "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": 0.0045770262368023396, "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": 0.013935577124357224, "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": 0.013935577124357224, "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": 0.24026010930538177, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22612346708774567, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22161559760570526, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.20180468261241913, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.11284935474395752, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10783509165048599, "qparams": { "group_size": { "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": 0.12551626563072205, "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": 0.11597036570310593, "qparams": { "group_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": 0.11437761783599854, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10192688554525375, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09696807712316513, "qparams": { "group_size": { "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": 0.06372667849063873, "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": 0.05531159043312073, "qparams": { "group_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": 0.05389486625790596, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.053563524037599564, "qparams": { "group_size": { "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": 0.03179209679365158, "qparams": { "group_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": 0.027362234890460968, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.027260754257440567, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02508794330060482, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024871373549103737, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016339225694537163, "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": 0.01594563201069832, "qparams": { "group_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": 0.015827452763915062, "qparams": { "group_size": { "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": 0.009791185148060322, "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": 0.009791185148060322, "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": 0.009791185148060322, "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": 0.18154531717300415, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14874623715877533, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1374669224023819, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10704904049634933, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08228439092636108, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06989660859107971, "qparams": { "group_size": { "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": 0.0984291210770607, "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": 0.0904373973608017, "qparams": { "group_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": 0.08686283230781555, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06101172789931297, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.052441250532865524, "qparams": { "group_size": { "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": 0.05046189948916435, "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": 0.043456751853227615, "qparams": { "group_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": 0.039792198687791824, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.038889240473508835, "qparams": { "group_size": { "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": 0.02526881918311119, "qparams": { "group_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": 0.0209516491740942, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02072557434439659, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01684390753507614, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016200479120016098, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013592247851192951, "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": 0.013777279295027256, "qparams": { "group_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": 0.012437342666089535, "qparams": { "group_size": { "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": 0.009678566828370094, "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": 0.013592247851192951, "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": 0.013592247851192951, "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": 0.17781347036361694, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16747546195983887, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1642642617225647, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1496969610452652, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0837780237197876, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08009492605924606, "qparams": { "group_size": { "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": 0.09292332082986832, "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": 0.08597716689109802, "qparams": { "group_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": 0.08489803969860077, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07573853433132172, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07213129103183746, "qparams": { "group_size": { "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": 0.04734078422188759, "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": 0.041158273816108704, "qparams": { "group_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": 0.040164850652217865, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.039922863245010376, "qparams": { "group_size": { "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": 0.023651134222745895, "qparams": { "group_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": 0.020702732726931572, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.020634159445762634, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01907607726752758, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018924960866570473, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012433690950274467, "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": 0.012558636255562305, "qparams": { "group_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": 0.012111491523683071, "qparams": { "group_size": { "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": 0.00841611996293068, "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": 0.012433690950274467, "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": 0.012433690950274467, "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": 0.2174234241247177, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20493589341640472, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20109310746192932, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.18322575092315674, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10229691118001938, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09783391654491425, "qparams": { "group_size": { "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": 0.11343052238225937, "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": 0.10494031757116318, "qparams": { "group_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": 0.10367244482040405, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0925084725022316, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08810978382825851, "qparams": { "group_size": { "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": 0.05770544335246086, "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": 0.050114165991544724, "qparams": { "group_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": 0.04891367629170418, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04863617569208145, "qparams": { "group_size": { "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": 0.028777139261364937, "qparams": { "group_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": 0.024927031248807907, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024843957275152206, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022905299440026283, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.022725434973835945, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014944138005375862, "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": 0.014669690281152725, "qparams": { "group_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": 0.014543880708515644, "qparams": { "group_size": { "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": 0.009244324639439583, "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": 0.014944138005375862, "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": 0.014543880708515644, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.2222495973110199, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19623252749443054, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18396282196044922, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.16427527368068695, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1009560078382492, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0899796262383461, "qparams": { "group_size": { "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": 0.12378399074077606, "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": 0.11395115405321121, "qparams": { "group_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": 0.10544101893901825, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08705289661884308, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08281844109296799, "qparams": { "group_size": { "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": 0.0629534050822258, "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": 0.05445701256394386, "qparams": { "group_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": 0.04857659712433815, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.047112252563238144, "qparams": { "group_size": { "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": 0.031498294323682785, "qparams": { "group_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": 0.025298647582530975, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02473476715385914, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022542251273989677, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021568477153778076, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01670379936695099, "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": 0.016497217118740082, "qparams": { "group_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": 0.014809218235313892, "qparams": { "group_size": { "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": 0.010959832929074764, "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": 0.014809218235313892, "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": 0.014809218235313892, "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": 0.12029596418142319, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11290999501943588, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11005201935768127, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09998639672994614, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.056329648941755295, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05340912565588951, "qparams": { "group_size": { "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": 0.06372453272342682, "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": 0.05895271897315979, "qparams": { "group_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": 0.057148732244968414, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.050764478743076324, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0483260340988636, "qparams": { "group_size": { "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": 0.03234259411692619, "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": 0.028148602694272995, "qparams": { "group_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": 0.02696177177131176, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.026676498353481293, "qparams": { "group_size": { "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": 0.016168219968676567, "qparams": { "group_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": 0.013793811202049255, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01367843896150589, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.012655334547162056, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01247325073927641, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00841327104717493, "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": 0.008304067887365818, "qparams": { "group_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": 0.00803183950483799, "qparams": { "group_size": { "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": 0.005295384209603071, "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": 0.013793811202049255, "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": 0.013793811202049255, "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": 0.09779976308345795, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09174191206693649, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08928490430116653, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08113560825586319, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04578564316034317, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.043345268815755844, "qparams": { "group_size": { "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": 0.05204504728317261, "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": 0.04808984696865082, "qparams": { "group_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": 0.046447817236185074, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.041248295456171036, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03926965966820717, "qparams": { "group_size": { "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": 0.026398388668894768, "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": 0.022956551983952522, "qparams": { "group_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": 0.021917151287198067, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.021669967100024223, "qparams": { "group_size": { "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": 0.013192502781748772, "qparams": { "group_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": 0.011225180700421333, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011119470931589603, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.010296685621142387, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010139270685613155, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006873179227113724, "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": 0.006784756667912006, "qparams": { "group_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": 0.006537461653351784, "qparams": { "group_size": { "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": 0.004341823048889637, "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": 0.013192502781748772, "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": 0.013192502781748772, "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": 0.23589332401752472, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.22191986441612244, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21739472448825836, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1979457139968872, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.1107800230383873, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10573898255825043, "qparams": { "group_size": { "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": 0.12339331209659576, "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": 0.11387092620134354, "qparams": { "group_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": 0.11229652911424637, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.10000849515199661, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09510663151741028, "qparams": { "group_size": { "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": 0.06269664317369461, "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": 0.05436094477772713, "qparams": { "group_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": 0.052947331219911575, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.052607566118240356, "qparams": { "group_size": { "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": 0.0312536358833313, "qparams": { "group_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": 0.026903530582785606, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.026803186163306236, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.024670423939824104, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024453703314065933, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016092920675873756, "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": 0.015754334628582, "qparams": { "group_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": 0.015564917586743832, "qparams": { "group_size": { "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": 0.009739190340042114, "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": 0.015564917586743832, "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": 0.015564917586743832, "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": 0.18007737398147583, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14252620935440063, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12938004732131958, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10002515465021133, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08115506917238235, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06688880920410156, "qparams": { "group_size": { "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": 0.09891603887081146, "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": 0.09043456614017487, "qparams": { "group_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": 0.08674298226833344, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05692548304796219, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05161997303366661, "qparams": { "group_size": { "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": 0.05099933221936226, "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": 0.0436139851808548, "qparams": { "group_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": 0.03936593234539032, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03832128271460533, "qparams": { "group_size": { "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": 0.025629686191678047, "qparams": { "group_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": 0.020959291607141495, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02072232961654663, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.016386697068810463, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.015623900108039379, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013959023170173168, "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": 0.014154445379972458, "qparams": { "group_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": 0.012599091976881027, "qparams": { "group_size": { "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": 0.010219455696642399, "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": 0.015623900108039379, "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": 0.015623900108039379, "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": 0.1706148236989975, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16060151159763336, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15748217701911926, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.14347463846206665, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08034010976552963, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07675590366125107, "qparams": { "group_size": { "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": 0.08914779126644135, "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": 0.08249378949403763, "qparams": { "group_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": 0.08144834637641907, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07260178029537201, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06914075464010239, "qparams": { "group_size": { "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": 0.04543376341462135, "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": 0.039490289986133575, "qparams": { "group_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": 0.03852144256234169, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03828402981162071, "qparams": { "group_size": { "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": 0.022702554240822792, "qparams": { "group_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": 0.01988081820309162, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01981305330991745, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018314026296138763, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018170170485973358, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011961695738136768, "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": 0.012097852304577827, "qparams": { "group_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": 0.011640192940831184, "qparams": { "group_size": { "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": 0.008153430186212063, "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": 0.011961695738136768, "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": 0.011961695738136768, "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": 0.20421725511550903, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.19230814278125763, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.188653364777565, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1719357669353485, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09608633071184158, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0918227806687355, "qparams": { "group_size": { "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": 0.10654784739017487, "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": 0.09857559949159622, "qparams": { "group_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": 0.09736184030771255, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08684312552213669, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08265932649374008, "qparams": { "group_size": { "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": 0.054256971925497055, "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": 0.047089237719774246, "qparams": { "group_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": 0.04594288766384125, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.045680269598960876, "qparams": { "group_size": { "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": 0.027058664709329605, "qparams": { "group_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": 0.023434260860085487, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023361261934041977, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021534403786063194, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021359719336032867, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.014077908359467983, "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": 0.01383458636701107, "qparams": { "group_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": 0.0136974286288023, "qparams": { "group_size": { "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": 0.008769504725933075, "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": 0.014077908359467983, "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": 0.014077908359467983, "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": 0.21570101380348206, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.18873031437397003, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.176243856549263, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15722033381462097, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09790763258934021, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08656986057758331, "qparams": { "group_size": { "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": 0.1212335154414177, "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": 0.11075533926486969, "qparams": { "group_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": 0.10247232019901276, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08374632894992828, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07974498718976974, "qparams": { "group_size": { "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": 0.06163046881556511, "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": 0.05319199711084366, "qparams": { "group_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": 0.04740079864859581, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04594491794705391, "qparams": { "group_size": { "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": 0.031061891466379166, "qparams": { "group_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": 0.025188740342855453, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024650979787111282, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02246309071779251, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02152078226208687, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016908777877688408, "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": 0.01704627275466919, "qparams": { "group_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": 0.0150492824614048, "qparams": { "group_size": { "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": 0.012123122811317444, "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": 0.0150492824614048, "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": 0.0150492824614048, "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": 0.11229933798313141, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10531925410032272, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10247547924518585, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.09304413944482803, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05250558257102966, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04973535239696503, "qparams": { "group_size": { "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": 0.059677042067050934, "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": 0.055068593472242355, "qparams": { "group_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": 0.05328832194209099, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.047286733984947205, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04500677436590195, "qparams": { "group_size": { "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": 0.03031567856669426, "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": 0.026319555938243866, "qparams": { "group_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": 0.025153720751404762, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02488255873322487, "qparams": { "group_size": { "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": 0.01516167726367712, "qparams": { "group_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": 0.012940121814608574, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012829095125198364, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011877425946295261, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011702659539878368, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00795708503574133, "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": 0.00789911113679409, "qparams": { "group_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": 0.007583580911159515, "qparams": { "group_size": { "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": 0.005175837315618992, "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": 0.01516167726367712, "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": 0.01516167726367712, "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": 0.09448615461587906, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08863457292318344, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08613507449626923, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07818422466516495, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04419270530343056, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04177960008382797, "qparams": { "group_size": { "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": 0.05052275210618973, "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": 0.04660334810614586, "qparams": { "group_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": 0.04482947289943695, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03981826454401016, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03788890317082405, "qparams": { "group_size": { "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": 0.025649582967162132, "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": 0.022273996844887733, "qparams": { "group_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": 0.02117675170302391, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.020911891013383865, "qparams": { "group_size": { "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": 0.012828955426812172, "qparams": { "group_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": 0.010884695686399937, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01077301800251007, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009995495900511742, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0098262345418334, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006722983438521624, "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": 0.006655856501311064, "qparams": { "group_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": 0.006365718320012093, "qparams": { "group_size": { "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": 0.004332184791564941, "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": 0.012828955426812172, "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": 0.012828955426812172, "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": 0.2220975011587143, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20873896777629852, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20446175336837769, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1857803761959076, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10422838479280472, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09938914328813553, "qparams": { "group_size": { "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": 0.11631452292203903, "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": 0.1072869822382927, "qparams": { "group_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": 0.10568291693925858, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09392594546079636, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08935084193944931, "qparams": { "group_size": { "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": 0.05911729857325554, "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": 0.05118204653263092, "qparams": { "group_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": 0.049794070422649384, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04945698007941246, "qparams": { "group_size": { "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": 0.029491644352674484, "qparams": { "group_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": 0.025330286473035812, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02522956393659115, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.02320893295109272, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.02299625612795353, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015226267278194427, "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": 0.014877264387905598, "qparams": { "group_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": 0.014726398512721062, "qparams": { "group_size": { "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": 0.009268095716834068, "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": 0.015226267278194427, "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": 0.014726398512721062, "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.o_proj", "numel": 26214400, "options": [ { "desc": "0.05:3b_32g/0.95:2b_32g s4", "bpw": 2.17529296875, "total_bits": 57024000.0, "err": 0.12440725415945053, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10680769383907318, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09909786283969879, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08018098026514053, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.057189732789993286, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0498114638030529, "qparams": { "group_size": { "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": 0.07040571421384811, "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": 0.06417139619588852, "qparams": { "group_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": 0.058975931257009506, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.044552870094776154, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04049618914723396, "qparams": { "group_size": { "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": 0.0360650010406971, "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": 0.03164626285433769, "qparams": { "group_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": 0.02858482301235199, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.027848953381180763, "qparams": { "group_size": { "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": 0.018429560586810112, "qparams": { "group_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": 0.01655648648738861, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016284633427858353, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01445941161364317, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014017648994922638, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.010757171548902988, "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": 0.012594809755682945, "qparams": { "group_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": 0.00987238995730877, "qparams": { "group_size": { "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": 0.010548129677772522, "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": 0.01445941161364317, "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": 0.01445941161364317, "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": 0.1620086133480072, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.15245547890663147, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14948485791683197, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.1361178457736969, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07628127187490463, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0728723481297493, "qparams": { "group_size": { "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": 0.08477578312158585, "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": 0.07833320647478104, "qparams": { "group_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": 0.07732028514146805, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06891946494579315, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06562294065952301, "qparams": { "group_size": { "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": 0.04324110597372055, "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": 0.03755141794681549, "qparams": { "group_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": 0.03661390766501427, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.036387912929058075, "qparams": { "group_size": { "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": 0.02161944843828678, "qparams": { "group_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": 0.018975283950567245, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.018912028521299362, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017493169754743576, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017351901158690453, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011476765386760235, "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": 0.011664684861898422, "qparams": { "group_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": 0.01116617675870657, "qparams": { "group_size": { "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": 0.007999643683433533, "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": 0.011476765386760235, "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": 0.011476765386760235, "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": 0.18961957097053528, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.17850446701049805, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1750485599040985, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15948835015296936, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.08923114836215973, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08525529503822327, "qparams": { "group_size": { "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": 0.09911192208528519, "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": 0.09161935746669769, "qparams": { "group_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": 0.09045380353927612, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08062287420034409, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07673673331737518, "qparams": { "group_size": { "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": 0.05052028223872185, "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": 0.043801307678222656, "qparams": { "group_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": 0.042724933475255966, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04246820881962776, "qparams": { "group_size": { "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": 0.02522844821214676, "qparams": { "group_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": 0.021887222304940224, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021813610568642616, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.020122114568948746, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.019960740581154823, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.01324128732085228, "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": 0.013061542995274067, "qparams": { "group_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": 0.012886223383247852, "qparams": { "group_size": { "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": 0.008479935117065907, "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": 0.01324128732085228, "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": 0.01324128732085228, "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": 0.20848912000656128, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1809784322977066, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.16918113827705383, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.15100958943367004, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.09462404996156693, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08338942378759384, "qparams": { "group_size": { "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": 0.1165044754743576, "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": 0.10612131655216217, "qparams": { "group_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": 0.09915654361248016, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.080325648188591, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.07644861936569214, "qparams": { "group_size": { "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": 0.05953438580036163, "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": 0.05110165849328041, "qparams": { "group_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": 0.04591705650091171, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.0446203388273716, "qparams": { "group_size": { "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": 0.030075587332248688, "qparams": { "group_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": 0.024568766355514526, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024113310500979424, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.021844536066055298, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.021007420495152473, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.016638070344924927, "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": 0.016758114099502563, "qparams": { "group_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": 0.015026682987809181, "qparams": { "group_size": { "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": 0.012207216583192348, "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": 0.015026682987809181, "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": 0.015026682987809181, "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": 0.10992715507745743, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10322270542383194, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10046768188476562, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.0913061797618866, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.0514434352517128, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04875757917761803, "qparams": { "group_size": { "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": 0.05872815474867821, "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": 0.05401892587542534, "qparams": { "group_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": 0.05218281224370003, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04638892039656639, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04430689290165901, "qparams": { "group_size": { "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": 0.029819780960679054, "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": 0.025822199881076813, "qparams": { "group_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": 0.024668091908097267, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02439023181796074, "qparams": { "group_size": { "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": 0.014932944439351559, "qparams": { "group_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": 0.012670200318098068, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01255657896399498, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011635659262537956, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011461552232503891, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.007822149433195591, "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": 0.007704109884798527, "qparams": { "group_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": 0.007456101011484861, "qparams": { "group_size": { "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": 0.005004198290407658, "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": 0.014932944439351559, "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": 0.014932944439351559, "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": 0.0905926376581192, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08502881973981857, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.08259175717830658, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.07506130635738373, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04239695519208908, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.040057893842458725, "qparams": { "group_size": { "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": 0.04853637143969536, "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": 0.044767700135707855, "qparams": { "group_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": 0.04300018027424812, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03820812329649925, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03638247400522232, "qparams": { "group_size": { "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": 0.024631697684526443, "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": 0.02139926888048649, "qparams": { "group_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": 0.020320957526564598, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.020057877525687218, "qparams": { "group_size": { "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": 0.012329380959272385, "qparams": { "group_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": 0.010452413000166416, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010338925756514072, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.009602140635251999, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.00943596288561821, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006486647762358189, "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": 0.006408591754734516, "qparams": { "group_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": 0.006139567121863365, "qparams": { "group_size": { "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": 0.004185652360320091, "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": 0.012329380959272385, "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": 0.012329380959272385, "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": 0.2278243452310562, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21436230838298798, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.21009880304336548, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.19103066623210907, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10703609883785248, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.1021445170044899, "qparams": { "group_size": { "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": 0.11916086077690125, "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": 0.11001115292310715, "qparams": { "group_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": 0.10850728303194046, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09654755145311356, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09178728610277176, "qparams": { "group_size": { "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": 0.06055167317390442, "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": 0.0524914525449276, "qparams": { "group_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": 0.05113788694143295, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.05080674588680267, "qparams": { "group_size": { "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": 0.030200131237506866, "qparams": { "group_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": 0.026023944839835167, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.025924092158675194, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.023846013471484184, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.023643938824534416, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015557296574115753, "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": 0.015268518589437008, "qparams": { "group_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": 0.0150537621229887, "qparams": { "group_size": { "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": 0.009511199779808521, "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": 0.015557296574115753, "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": 0.0150537621229887, "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": 0.13637849688529968, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11317898333072662, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.10381937026977539, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08298437297344208, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.061427369713783264, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05305328965187073, "qparams": { "group_size": { "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": 0.07706134766340256, "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": 0.07022549211978912, "qparams": { "group_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": 0.06527788937091827, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04693355783820152, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04239984601736069, "qparams": { "group_size": { "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": 0.03978729248046875, "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": 0.0341806598007679, "qparams": { "group_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": 0.030296068638563156, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.029309365898370743, "qparams": { "group_size": { "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": 0.020229129120707512, "qparams": { "group_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": 0.016866248100996017, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016569482162594795, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.014300612732768059, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013676001690328121, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.011625437997281551, "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": 0.012323044240474701, "qparams": { "group_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": 0.010496996343135834, "qparams": { "group_size": { "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": 0.009715359658002853, "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": 0.014300612732768059, "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": 0.014300612732768059, "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": 0.1359107345342636, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12783588469028473, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12541525065898895, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11426962167024612, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.06426665931940079, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.061381760984659195, "qparams": { "group_size": { "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": 0.07146678864955902, "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": 0.06599479168653488, "qparams": { "group_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": 0.06512749940156937, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.058118052780628204, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.055420517921447754, "qparams": { "group_size": { "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": 0.036858633160591125, "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": 0.0322447195649147, "qparams": { "group_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": 0.03146406263113022, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03128302842378616, "qparams": { "group_size": { "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": 0.01859574392437935, "qparams": { "group_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": 0.017335819080471992, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.017286458984017372, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01621982641518116, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016115328297019005, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.010559950955212116, "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": 0.012018888257443905, "qparams": { "group_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": 0.010329273529350758, "qparams": { "group_size": { "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": 0.009689771570265293, "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": 0.010559950955212116, "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": 0.010559950955212116, "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": 0.1529221534729004, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14397278428077698, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.14114081859588623, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.12882167100906372, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07236886024475098, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06914446502923965, "qparams": { "group_size": { "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": 0.0805879533290863, "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": 0.07428545504808426, "qparams": { "group_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": 0.07330814749002457, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06550082564353943, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06255950033664703, "qparams": { "group_size": { "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": 0.041587721556425095, "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": 0.0362810380756855, "qparams": { "group_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": 0.03541838750243187, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.03521564602851868, "qparams": { "group_size": { "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": 0.021013235673308372, "qparams": { "group_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": 0.019520696252584457, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01946723833680153, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.018255271017551422, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01814386062324047, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.012004125863313675, "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": 0.013517535291612148, "qparams": { "group_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": 0.011751928366720676, "qparams": { "group_size": { "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": 0.010901267640292645, "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": 0.012004125863313675, "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": 0.012004125863313675, "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": 0.15567637979984283, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1336648315191269, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.12459005415439606, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.11074596643447876, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.07059961557388306, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.06179233267903328, "qparams": { "group_size": { "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": 0.0881650373339653, "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": 0.07928311824798584, "qparams": { "group_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": 0.07400727272033691, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.059504471719264984, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05686921626329422, "qparams": { "group_size": { "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": 0.045177262276411057, "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": 0.038724057376384735, "qparams": { "group_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": 0.0347728431224823, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.033823177218437195, "qparams": { "group_size": { "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": 0.0232890322804451, "qparams": { "group_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": 0.01943979784846306, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01909380778670311, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.017429189756512642, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.016803888604044914, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.013711432926356792, "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": 0.014120745472609997, "qparams": { "group_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": 0.01251068152487278, "qparams": { "group_size": { "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": 0.011187792755663395, "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": 0.013711432926356792, "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": 0.013711432926356792, "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": 0.10402284562587738, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09760479629039764, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.09484101086854935, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.08620788156986237, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.04852648824453354, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04592062905430794, "qparams": { "group_size": { "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": 0.0558420829474926, "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": 0.051187679171562195, "qparams": { "group_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": 0.04926322400569916, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04378395527601242, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.04189547523856163, "qparams": { "group_size": { "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": 0.028336206451058388, "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": 0.024463215842843056, "qparams": { "group_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": 0.023260409012436867, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.022973960265517235, "qparams": { "group_size": { "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": 0.01418638601899147, "qparams": { "group_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": 0.011967826634645462, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.01184788066893816, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011010177433490753, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.010827305726706982, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00746067613363266, "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": 0.007340132724493742, "qparams": { "group_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": 0.007067234721034765, "qparams": { "group_size": { "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": 0.004807380959391594, "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": 0.01418638601899147, "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": 0.01418638601899147, "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": 0.08110543340444565, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07615718990564346, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.0738440454006195, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.06707435101270676, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.037877053022384644, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03574122488498688, "qparams": { "group_size": { "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": 0.04356268793344498, "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": 0.04018799588084221, "qparams": { "group_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": 0.03842747211456299, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.034149643033742905, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03253582864999771, "qparams": { "group_size": { "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": 0.022126315161585808, "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": 0.019219376146793365, "qparams": { "group_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": 0.01815355010330677, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.017898958176374435, "qparams": { "group_size": { "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": 0.011080958880484104, "qparams": { "group_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": 0.009366171434521675, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.009254288859665394, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008612746372818947, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008453793823719025, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.005846511106938124, "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": 0.0057978201657533646, "qparams": { "group_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": 0.005505243316292763, "qparams": { "group_size": { "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": 0.0038297404535114765, "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": 0.011080958880484104, "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": 0.011080958880484104, "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": 0.2134602665901184, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.20063307881355286, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.1965043842792511, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.17893262207508087, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.10020314157009125, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09559960663318634, "qparams": { "group_size": { "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": 0.11319180577993393, "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": 0.10319299250841141, "qparams": { "group_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": 0.10159973800182343, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.09054001420736313, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.08660712838172913, "qparams": { "group_size": { "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": 0.057661157101392746, "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": 0.049537092447280884, "qparams": { "group_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": 0.04808579012751579, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.04770456999540329, "qparams": { "group_size": { "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": 0.029172565788030624, "qparams": { "group_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": 0.02475259266793728, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.024636387825012207, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.022792819887399673, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.0225560050457716, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.015317348763346672, "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": 0.014941075816750526, "qparams": { "group_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": 0.014731282368302345, "qparams": { "group_size": { "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": 0.009753829799592495, "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": 0.015317348763346672, "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": 0.014731282368302345, "qparams": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "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": 0.10525704175233841, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.07811636477708817, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.06799637526273727, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.060983072966337204, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.046730976551771164, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0353124737739563, "qparams": { "group_size": { "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": 0.060424432158470154, "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": 0.05289187654852867, "qparams": { "group_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": 0.05016818270087242, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03320539742708206, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.03297945857048035, "qparams": { "group_size": { "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": 0.03141693398356438, "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": 0.026432594284415245, "qparams": { "group_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": 0.023708410561084747, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.02303098514676094, "qparams": { "group_size": { "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": 0.016566522419452667, "qparams": { "group_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": 0.014062127098441124, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013913734816014767, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.011898266151547432, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.011495969258248806, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.010283200070261955, "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": 0.01096876710653305, "qparams": { "group_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": 0.009503896348178387, "qparams": { "group_size": { "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": 0.009354663081467152, "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": 0.014062127098441124, "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": 0.014062127098441124, "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": 0.12120886892080307, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11404703557491302, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11181706935167313, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10189990699291229, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.05691583827137947, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.054368868470191956, "qparams": { "group_size": { "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": 0.06335721909999847, "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": 0.05845890939235687, "qparams": { "group_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": 0.05769854784011841, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05141424015164375, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.0490160696208477, "qparams": { "group_size": { "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": 0.03226924687623978, "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": 0.027988962829113007, "qparams": { "group_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": 0.027287350967526436, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.027123279869556427, "qparams": { "group_size": { "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": 0.016136616468429565, "qparams": { "group_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": 0.014125737361609936, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.014078466221690178, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.013021789491176605, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.012918158434331417, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.00854200404137373, "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": 0.008657736703753471, "qparams": { "group_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": 0.008307383395731449, "qparams": { "group_size": { "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": 0.005909500177949667, "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": 0.014125737361609936, "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": 0.014125737361609936, "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": 0.12118949741125107, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11402478069067001, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.11179500818252563, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.10188253223896027, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.057258181273937225, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.054710060358047485, "qparams": { "group_size": { "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": 0.06379316747188568, "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": 0.05879659950733185, "qparams": { "group_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": 0.05804779753088951, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.05178450420498848, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.049459874629974365, "qparams": { "group_size": { "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": 0.03286599740386009, "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": 0.028552785515785217, "qparams": { "group_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": 0.02786431647837162, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.027705682441592216, "qparams": { "group_size": { "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": 0.016568923369050026, "qparams": { "group_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": 0.015095295384526253, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.015041311271488667, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.01405760645866394, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.013956149108707905, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.009301932528614998, "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": 0.010139411315321922, "qparams": { "group_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": 0.009079544804990292, "qparams": { "group_size": { "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": 0.007917718030512333, "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": 0.015095295384526253, "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": 5.325108506944445, "total_bits": 376905216.0, "err": 0.015041311271488667, "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.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": 0.06599714607000351, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05617176368832588, "qparams": { "group_size": { "3": 32, "2": 32 }, "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": 0.05135984718799591, "qparams": { "group_size": { "4": 32, "2": 32 }, "bits": [ 4, 2 ], "bits_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": 0.04553810507059097, "qparams": { "group_size": { "4": 32, "3": 32, "2": 32 }, "bits": [ 4, 3, 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": 0.03002185747027397, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.02564265765249729, "qparams": { "group_size": { "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": 0.0396818183362484, "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": 0.034976813942193985, "qparams": { "group_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": 0.031643982976675034, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.02523069642484188, "qparams": { "group_size": { "4": 32, "3": 32 }, "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": 0.024312473833560944, "qparams": { "group_size": { "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": 0.020424949005246162, "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": 0.017403345555067062, "qparams": { "group_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": 0.015180864371359348, "qparams": { "group_size": { "5": 32, "4": 32 }, "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": 0.014617918990552425, "qparams": { "group_size": { "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": 0.010556859895586967, "qparams": { "group_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": 0.009029464796185493, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.008827369660139084, "qparams": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 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": 0.008241817355155945, "qparams": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_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": 0.007921379990875721, "qparams": { "group_size": { "8": 32, "6": 32, "5": 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": 0.006303707603365183, "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": 0.007129846140742302, "qparams": { "group_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": 0.005586730316281319, "qparams": { "group_size": { "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": 0.0060434360057115555, "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": 0.015180864371359348, "qparams": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 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": 0.014617918990552425, "qparams": { "group_size": { "6": 32, "4": 32 }, "bits": [ 6, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } } } ], "last_module_idx": 82, "base_perplexity": 7.984414475324483 }