{ "best_metric": null, "best_model_checkpoint": null, "epoch": 0.2336, "eval_steps": 500, "global_step": 2920, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "epoch": 0.0, "grad_norm": 62.10089111328125, "learning_rate": 1e-06, "loss": 4.5777, "step": 20 }, { "epoch": 0.0, "grad_norm": 39.39016342163086, "learning_rate": 2e-06, "loss": 4.4077, "step": 40 }, { "epoch": 0.0, "grad_norm": 54.24020767211914, "learning_rate": 3e-06, "loss": 4.4807, "step": 60 }, { "epoch": 0.01, "grad_norm": 30.161609649658203, "learning_rate": 4e-06, "loss": 4.5756, "step": 80 }, { "epoch": 0.01, "grad_norm": 40.131675720214844, "learning_rate": 4.9999999999999996e-06, "loss": 4.4352, "step": 100 }, { "epoch": 0.01, "grad_norm": 52.3621940612793, "learning_rate": 6e-06, "loss": 4.5096, "step": 120 }, { "epoch": 0.01, "grad_norm": 49.86561584472656, "learning_rate": 7e-06, "loss": 4.493, "step": 140 }, { "epoch": 0.01, "grad_norm": 20.034923553466797, "learning_rate": 8e-06, "loss": 4.4088, "step": 160 }, { "epoch": 0.01, "grad_norm": 50.790679931640625, "learning_rate": 9e-06, "loss": 4.4901, "step": 180 }, { "epoch": 0.02, "grad_norm": 48.5693473815918, "learning_rate": 9.999999999999999e-06, "loss": 4.3628, "step": 200 }, { "epoch": 0.02, "grad_norm": 37.95353698730469, "learning_rate": 1.1e-05, "loss": 4.3298, "step": 220 }, { "epoch": 0.02, "grad_norm": 35.7153434753418, "learning_rate": 1.2e-05, "loss": 4.2839, "step": 240 }, { "epoch": 0.02, "grad_norm": 91.47773742675781, "learning_rate": 1.3000000000000001e-05, "loss": 4.1238, "step": 260 }, { "epoch": 0.02, "grad_norm": 23.16193389892578, "learning_rate": 1.4e-05, "loss": 4.1245, "step": 280 }, { "epoch": 0.02, "grad_norm": 28.304485321044922, "learning_rate": 1.5e-05, "loss": 4.2198, "step": 300 }, { "epoch": 0.03, "grad_norm": 34.03230285644531, "learning_rate": 1.6e-05, "loss": 4.0958, "step": 320 }, { "epoch": 0.03, "grad_norm": 29.786975860595703, "learning_rate": 1.7e-05, "loss": 4.024, "step": 340 }, { "epoch": 0.03, "grad_norm": 33.04754638671875, "learning_rate": 1.8e-05, "loss": 4.0832, "step": 360 }, { "epoch": 0.03, "grad_norm": 28.68460464477539, "learning_rate": 1.9e-05, "loss": 3.9827, "step": 380 }, { "epoch": 0.03, "grad_norm": 26.463253021240234, "learning_rate": 1.9999999999999998e-05, "loss": 3.9454, "step": 400 }, { "epoch": 0.03, "grad_norm": 19.407127380371094, "learning_rate": 2.1e-05, "loss": 4.0119, "step": 420 }, { "epoch": 0.04, "grad_norm": 26.383380889892578, "learning_rate": 2.2e-05, "loss": 3.9554, "step": 440 }, { "epoch": 0.04, "grad_norm": 33.225223541259766, "learning_rate": 2.3000000000000003e-05, "loss": 3.8172, "step": 460 }, { "epoch": 0.04, "grad_norm": 26.000978469848633, "learning_rate": 2.4e-05, "loss": 3.8934, "step": 480 }, { "epoch": 0.04, "grad_norm": 28.714366912841797, "learning_rate": 2.5e-05, "loss": 3.9194, "step": 500 }, { "epoch": 0.04, "grad_norm": 28.721248626708984, "learning_rate": 2.6000000000000002e-05, "loss": 3.8144, "step": 520 }, { "epoch": 0.04, "grad_norm": 24.934555053710938, "learning_rate": 2.7000000000000002e-05, "loss": 3.9166, "step": 540 }, { "epoch": 0.04, "grad_norm": 23.113840103149414, "learning_rate": 2.8e-05, "loss": 3.8248, "step": 560 }, { "epoch": 0.05, "grad_norm": 21.58758544921875, "learning_rate": 2.9e-05, "loss": 3.7538, "step": 580 }, { "epoch": 0.05, "grad_norm": 22.339618682861328, "learning_rate": 3e-05, "loss": 3.726, "step": 600 }, { "epoch": 0.05, "grad_norm": 49.25693893432617, "learning_rate": 2.999992132854894e-05, "loss": 3.8692, "step": 620 }, { "epoch": 0.05, "grad_norm": 45.1494026184082, "learning_rate": 2.999968531502098e-05, "loss": 3.7374, "step": 640 }, { "epoch": 0.05, "grad_norm": 72.25853729248047, "learning_rate": 2.99992919618918e-05, "loss": 3.7735, "step": 660 }, { "epoch": 0.05, "grad_norm": 39.445220947265625, "learning_rate": 2.999874127328748e-05, "loss": 3.759, "step": 680 }, { "epoch": 0.06, "grad_norm": 21.18370246887207, "learning_rate": 2.9998033254984483e-05, "loss": 3.7841, "step": 700 }, { "epoch": 0.06, "grad_norm": 24.310373306274414, "learning_rate": 2.999716791440959e-05, "loss": 3.679, "step": 720 }, { "epoch": 0.06, "grad_norm": 36.432350158691406, "learning_rate": 2.9996145260639812e-05, "loss": 3.6796, "step": 740 }, { "epoch": 0.06, "grad_norm": 32.12275314331055, "learning_rate": 2.9994965304402304e-05, "loss": 3.7613, "step": 760 }, { "epoch": 0.06, "grad_norm": 38.32442855834961, "learning_rate": 2.999362805807425e-05, "loss": 3.7586, "step": 780 }, { "epoch": 0.06, "grad_norm": 30.289432525634766, "learning_rate": 2.9992133535682725e-05, "loss": 3.6919, "step": 800 }, { "epoch": 0.07, "grad_norm": 32.69138717651367, "learning_rate": 2.9990481752904566e-05, "loss": 3.6855, "step": 820 }, { "epoch": 0.07, "grad_norm": 46.554874420166016, "learning_rate": 2.9988672727066197e-05, "loss": 3.7201, "step": 840 }, { "epoch": 0.07, "grad_norm": 28.671123504638672, "learning_rate": 2.9986706477143436e-05, "loss": 3.6594, "step": 860 }, { "epoch": 0.07, "grad_norm": 49.44480895996094, "learning_rate": 2.9984583023761318e-05, "loss": 3.7271, "step": 880 }, { "epoch": 0.07, "grad_norm": 26.61457061767578, "learning_rate": 2.998230238919386e-05, "loss": 3.7376, "step": 900 }, { "epoch": 0.07, "grad_norm": 27.453275680541992, "learning_rate": 2.9979864597363846e-05, "loss": 3.6716, "step": 920 }, { "epoch": 0.08, "grad_norm": 22.791175842285156, "learning_rate": 2.9977269673842554e-05, "loss": 3.6172, "step": 940 }, { "epoch": 0.08, "grad_norm": 58.2718620300293, "learning_rate": 2.997451764584951e-05, "loss": 3.7494, "step": 960 }, { "epoch": 0.08, "grad_norm": 33.610286712646484, "learning_rate": 2.9971608542252175e-05, "loss": 3.7077, "step": 980 }, { "epoch": 0.08, "grad_norm": 25.48147201538086, "learning_rate": 2.9968542393565674e-05, "loss": 3.6721, "step": 1000 }, { "epoch": 0.08, "grad_norm": 27.07135581970215, "learning_rate": 2.996531923195246e-05, "loss": 3.7106, "step": 1020 }, { "epoch": 0.08, "grad_norm": 37.24673843383789, "learning_rate": 2.996193909122197e-05, "loss": 3.7447, "step": 1040 }, { "epoch": 0.08, "grad_norm": 26.41890525817871, "learning_rate": 2.995840200683028e-05, "loss": 3.5839, "step": 1060 }, { "epoch": 0.09, "grad_norm": 32.88002014160156, "learning_rate": 2.995470801587973e-05, "loss": 3.6606, "step": 1080 }, { "epoch": 0.09, "grad_norm": 32.0895881652832, "learning_rate": 2.9950857157118544e-05, "loss": 3.677, "step": 1100 }, { "epoch": 0.09, "grad_norm": 37.726783752441406, "learning_rate": 2.9946849470940395e-05, "loss": 3.5546, "step": 1120 }, { "epoch": 0.09, "grad_norm": 56.246299743652344, "learning_rate": 2.9942684999384034e-05, "loss": 3.6391, "step": 1140 }, { "epoch": 0.09, "grad_norm": 35.675662994384766, "learning_rate": 2.993836378613278e-05, "loss": 3.5918, "step": 1160 }, { "epoch": 0.09, "grad_norm": 26.685134887695312, "learning_rate": 2.993388587651412e-05, "loss": 3.6331, "step": 1180 }, { "epoch": 0.1, "grad_norm": 27.400333404541016, "learning_rate": 2.992925131749921e-05, "loss": 3.6214, "step": 1200 }, { "epoch": 0.1, "grad_norm": 28.501314163208008, "learning_rate": 2.9924460157702378e-05, "loss": 3.6619, "step": 1220 }, { "epoch": 0.1, "grad_norm": 30.773778915405273, "learning_rate": 2.991951244738063e-05, "loss": 3.6453, "step": 1240 }, { "epoch": 0.1, "grad_norm": 24.701374053955078, "learning_rate": 2.9914408238433095e-05, "loss": 3.7282, "step": 1260 }, { "epoch": 0.1, "grad_norm": 27.605117797851562, "learning_rate": 2.990914758440052e-05, "loss": 3.6635, "step": 1280 }, { "epoch": 0.1, "grad_norm": 27.829086303710938, "learning_rate": 2.9903730540464668e-05, "loss": 3.5293, "step": 1300 }, { "epoch": 0.11, "grad_norm": 40.916263580322266, "learning_rate": 2.9898157163447767e-05, "loss": 3.6976, "step": 1320 }, { "epoch": 0.11, "grad_norm": 33.31068420410156, "learning_rate": 2.9892427511811912e-05, "loss": 3.548, "step": 1340 }, { "epoch": 0.11, "grad_norm": 29.932533264160156, "learning_rate": 2.9886541645658435e-05, "loss": 3.7486, "step": 1360 }, { "epoch": 0.11, "grad_norm": 35.59455490112305, "learning_rate": 2.9880499626727284e-05, "loss": 3.6342, "step": 1380 }, { "epoch": 0.11, "grad_norm": 29.93869400024414, "learning_rate": 2.9874301518396377e-05, "loss": 3.6615, "step": 1400 }, { "epoch": 0.11, "grad_norm": 43.417213439941406, "learning_rate": 2.986794738568094e-05, "loss": 3.607, "step": 1420 }, { "epoch": 0.12, "grad_norm": 52.483917236328125, "learning_rate": 2.9861437295232825e-05, "loss": 3.5937, "step": 1440 }, { "epoch": 0.12, "grad_norm": 30.312334060668945, "learning_rate": 2.9854771315339787e-05, "loss": 3.5991, "step": 1460 }, { "epoch": 0.12, "grad_norm": 49.459136962890625, "learning_rate": 2.984794951592481e-05, "loss": 3.5261, "step": 1480 }, { "epoch": 0.12, "grad_norm": 34.81111526489258, "learning_rate": 2.984097196854534e-05, "loss": 3.6818, "step": 1500 }, { "epoch": 0.12, "grad_norm": 34.721946716308594, "learning_rate": 2.9833838746392544e-05, "loss": 3.5636, "step": 1520 }, { "epoch": 0.12, "grad_norm": 31.46621322631836, "learning_rate": 2.982654992429056e-05, "loss": 3.5597, "step": 1540 }, { "epoch": 0.12, "grad_norm": 38.78512191772461, "learning_rate": 2.981910557869566e-05, "loss": 3.661, "step": 1560 }, { "epoch": 0.13, "grad_norm": 27.38837432861328, "learning_rate": 2.981150578769553e-05, "loss": 3.6173, "step": 1580 }, { "epoch": 0.13, "grad_norm": 45.619632720947266, "learning_rate": 2.980375063100836e-05, "loss": 3.6632, "step": 1600 }, { "epoch": 0.13, "grad_norm": 30.708433151245117, "learning_rate": 2.979584018998209e-05, "loss": 3.5165, "step": 1620 }, { "epoch": 0.13, "grad_norm": 35.472938537597656, "learning_rate": 2.97877745475935e-05, "loss": 3.5157, "step": 1640 }, { "epoch": 0.13, "grad_norm": 39.029415130615234, "learning_rate": 2.9779553788447358e-05, "loss": 3.6259, "step": 1660 }, { "epoch": 0.13, "grad_norm": 57.90769577026367, "learning_rate": 2.977117799877554e-05, "loss": 3.6378, "step": 1680 }, { "epoch": 0.14, "grad_norm": 36.95255661010742, "learning_rate": 2.9762647266436115e-05, "loss": 3.5845, "step": 1700 }, { "epoch": 0.14, "grad_norm": 27.456787109375, "learning_rate": 2.9753961680912432e-05, "loss": 3.6647, "step": 1720 }, { "epoch": 0.14, "grad_norm": 27.383285522460938, "learning_rate": 2.9745121333312166e-05, "loss": 3.6668, "step": 1740 }, { "epoch": 0.14, "grad_norm": 26.555049896240234, "learning_rate": 2.9736126316366385e-05, "loss": 3.6617, "step": 1760 }, { "epoch": 0.14, "grad_norm": 34.009620666503906, "learning_rate": 2.9726976724428563e-05, "loss": 3.572, "step": 1780 }, { "epoch": 0.14, "grad_norm": 45.44181823730469, "learning_rate": 2.9717672653473588e-05, "loss": 3.6354, "step": 1800 }, { "epoch": 0.15, "grad_norm": 30.79588508605957, "learning_rate": 2.9708214201096758e-05, "loss": 3.6953, "step": 1820 }, { "epoch": 0.15, "grad_norm": 46.61872482299805, "learning_rate": 2.9698601466512767e-05, "loss": 3.5373, "step": 1840 }, { "epoch": 0.15, "grad_norm": 42.86500930786133, "learning_rate": 2.9688834550554647e-05, "loss": 3.5982, "step": 1860 }, { "epoch": 0.15, "grad_norm": 33.480289459228516, "learning_rate": 2.9678913555672733e-05, "loss": 3.6024, "step": 1880 }, { "epoch": 0.15, "grad_norm": 36.41415786743164, "learning_rate": 2.966883858593356e-05, "loss": 3.4843, "step": 1900 }, { "epoch": 0.15, "grad_norm": 41.39873123168945, "learning_rate": 2.9658609747018796e-05, "loss": 3.5257, "step": 1920 }, { "epoch": 0.16, "grad_norm": 31.24024200439453, "learning_rate": 2.964822714622412e-05, "loss": 3.5927, "step": 1940 }, { "epoch": 0.16, "grad_norm": 52.78026580810547, "learning_rate": 2.9637690892458103e-05, "loss": 3.4678, "step": 1960 }, { "epoch": 0.16, "grad_norm": 27.40117835998535, "learning_rate": 2.962700109624106e-05, "loss": 3.5541, "step": 1980 }, { "epoch": 0.16, "grad_norm": 23.172683715820312, "learning_rate": 2.961615786970389e-05, "loss": 3.5713, "step": 2000 }, { "epoch": 0.16, "grad_norm": 24.177541732788086, "learning_rate": 2.960516132658692e-05, "loss": 3.585, "step": 2020 }, { "epoch": 0.16, "grad_norm": 44.673912048339844, "learning_rate": 2.9594011582238672e-05, "loss": 3.5035, "step": 2040 }, { "epoch": 0.16, "grad_norm": 35.91664505004883, "learning_rate": 2.95827087536147e-05, "loss": 3.6404, "step": 2060 }, { "epoch": 0.17, "grad_norm": 27.3450870513916, "learning_rate": 2.9571252959276313e-05, "loss": 3.5121, "step": 2080 }, { "epoch": 0.17, "grad_norm": 25.66405487060547, "learning_rate": 2.955964431938939e-05, "loss": 3.5009, "step": 2100 }, { "epoch": 0.17, "grad_norm": 24.1674861907959, "learning_rate": 2.9547882955723052e-05, "loss": 3.5482, "step": 2120 }, { "epoch": 0.17, "grad_norm": 49.72268295288086, "learning_rate": 2.953596899164846e-05, "loss": 3.4969, "step": 2140 }, { "epoch": 0.17, "grad_norm": 26.238168716430664, "learning_rate": 2.9523902552137436e-05, "loss": 3.5541, "step": 2160 }, { "epoch": 0.17, "grad_norm": 30.524545669555664, "learning_rate": 2.951168376376124e-05, "loss": 3.6343, "step": 2180 }, { "epoch": 0.18, "grad_norm": 38.179908752441406, "learning_rate": 2.9499312754689168e-05, "loss": 3.4795, "step": 2200 }, { "epoch": 0.18, "grad_norm": 32.98453903198242, "learning_rate": 2.9486789654687256e-05, "loss": 3.6333, "step": 2220 }, { "epoch": 0.18, "grad_norm": 26.77848243713379, "learning_rate": 2.94741145951169e-05, "loss": 3.5654, "step": 2240 }, { "epoch": 0.18, "grad_norm": 27.737852096557617, "learning_rate": 2.9461287708933475e-05, "loss": 3.5044, "step": 2260 }, { "epoch": 0.18, "grad_norm": 34.2584342956543, "learning_rate": 2.9448309130684944e-05, "loss": 3.5979, "step": 2280 }, { "epoch": 0.18, "grad_norm": 57.86616897583008, "learning_rate": 2.9435178996510456e-05, "loss": 3.5726, "step": 2300 }, { "epoch": 0.19, "grad_norm": 37.64597702026367, "learning_rate": 2.9421897444138902e-05, "loss": 3.5913, "step": 2320 }, { "epoch": 0.19, "grad_norm": 35.22037124633789, "learning_rate": 2.9408464612887484e-05, "loss": 3.5959, "step": 2340 }, { "epoch": 0.19, "grad_norm": 31.878395080566406, "learning_rate": 2.9394880643660242e-05, "loss": 3.5974, "step": 2360 }, { "epoch": 0.19, "grad_norm": 43.614994049072266, "learning_rate": 2.938114567894659e-05, "loss": 3.4834, "step": 2380 }, { "epoch": 0.19, "grad_norm": 27.587766647338867, "learning_rate": 2.9367259862819805e-05, "loss": 3.6154, "step": 2400 }, { "epoch": 0.19, "grad_norm": 30.223772048950195, "learning_rate": 2.9353223340935533e-05, "loss": 3.4871, "step": 2420 }, { "epoch": 0.2, "grad_norm": 34.057884216308594, "learning_rate": 2.933903626053024e-05, "loss": 3.605, "step": 2440 }, { "epoch": 0.2, "grad_norm": 39.219242095947266, "learning_rate": 2.932469877041969e-05, "loss": 3.6091, "step": 2460 }, { "epoch": 0.2, "grad_norm": 33.33955001831055, "learning_rate": 2.931021102099737e-05, "loss": 3.4862, "step": 2480 }, { "epoch": 0.2, "grad_norm": 37.07484436035156, "learning_rate": 2.9295573164232913e-05, "loss": 3.5267, "step": 2500 }, { "epoch": 0.2, "grad_norm": 27.145864486694336, "learning_rate": 2.9280785353670514e-05, "loss": 3.4369, "step": 2520 }, { "epoch": 0.2, "grad_norm": 30.31035041809082, "learning_rate": 2.9265847744427305e-05, "loss": 3.6056, "step": 2540 }, { "epoch": 0.2, "grad_norm": 40.823490142822266, "learning_rate": 2.925076049319174e-05, "loss": 3.5916, "step": 2560 }, { "epoch": 0.21, "grad_norm": 44.224796295166016, "learning_rate": 2.9235523758221944e-05, "loss": 3.5881, "step": 2580 }, { "epoch": 0.21, "grad_norm": 33.34773254394531, "learning_rate": 2.922013769934406e-05, "loss": 3.5315, "step": 2600 }, { "epoch": 0.21, "grad_norm": 25.755775451660156, "learning_rate": 2.920460247795056e-05, "loss": 3.621, "step": 2620 }, { "epoch": 0.21, "grad_norm": 29.75731086730957, "learning_rate": 2.918891825699857e-05, "loss": 3.6067, "step": 2640 }, { "epoch": 0.21, "grad_norm": 40.04263687133789, "learning_rate": 2.9173085201008144e-05, "loss": 3.5365, "step": 2660 }, { "epoch": 0.21, "grad_norm": 26.394916534423828, "learning_rate": 2.9157103476060547e-05, "loss": 3.5944, "step": 2680 }, { "epoch": 0.22, "grad_norm": 23.64137077331543, "learning_rate": 2.914097324979651e-05, "loss": 3.5048, "step": 2700 }, { "epoch": 0.22, "grad_norm": 35.18840408325195, "learning_rate": 2.9124694691414485e-05, "loss": 3.5769, "step": 2720 }, { "epoch": 0.22, "grad_norm": 25.55812644958496, "learning_rate": 2.9108267971668828e-05, "loss": 3.5082, "step": 2740 }, { "epoch": 0.22, "grad_norm": 26.25998878479004, "learning_rate": 2.909169326286807e-05, "loss": 3.4967, "step": 2760 }, { "epoch": 0.22, "grad_norm": 41.78467559814453, "learning_rate": 2.9074970738873054e-05, "loss": 3.6409, "step": 2780 }, { "epoch": 0.22, "grad_norm": 27.53672218322754, "learning_rate": 2.9058100575095156e-05, "loss": 3.5903, "step": 2800 }, { "epoch": 0.23, "grad_norm": 27.329477310180664, "learning_rate": 2.90410829484944e-05, "loss": 3.5148, "step": 2820 }, { "epoch": 0.23, "grad_norm": 34.077083587646484, "learning_rate": 2.902391803757764e-05, "loss": 3.5561, "step": 2840 }, { "epoch": 0.23, "grad_norm": 33.75153732299805, "learning_rate": 2.900660602239667e-05, "loss": 3.498, "step": 2860 }, { "epoch": 0.23, "grad_norm": 28.681190490722656, "learning_rate": 2.8989147084546335e-05, "loss": 3.501, "step": 2880 }, { "epoch": 0.23, "grad_norm": 41.04441833496094, "learning_rate": 2.8971541407162637e-05, "loss": 3.5707, "step": 2900 }, { "epoch": 0.23, "grad_norm": 31.303403854370117, "learning_rate": 2.8953789174920795e-05, "loss": 3.5748, "step": 2920 } ], "logging_steps": 20, "max_steps": 20000, "num_input_tokens_seen": 0, "num_train_epochs": 2, "save_steps": 20, "total_flos": 6907678563827712.0, "train_batch_size": 8, "trial_name": null, "trial_params": null }