diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,5391 @@ +{ + "best_metric": 1.437897801399231, + "best_model_checkpoint": "runs/deepseek_lora_20240422-165831/checkpoint-7500", + "epoch": 0.1875, + "eval_steps": 500, + "global_step": 7500, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0, + "grad_norm": 2.7229208946228027, + "learning_rate": 4.0000000000000003e-07, + "loss": 2.2066, + "step": 10 + }, + { + "epoch": 0.0, + "grad_norm": 5.007352828979492, + "learning_rate": 8.000000000000001e-07, + "loss": 2.1122, + "step": 20 + }, + { + "epoch": 0.0, + "grad_norm": 3.8536276817321777, + "learning_rate": 1.2000000000000002e-06, + "loss": 2.2073, + "step": 30 + }, + { + "epoch": 0.0, + "grad_norm": 6.411635875701904, + "learning_rate": 1.6000000000000001e-06, + "loss": 2.3019, + "step": 40 + }, + { + "epoch": 0.0, + "grad_norm": 8.24316692352295, + "learning_rate": 2.0000000000000003e-06, + "loss": 2.1321, + "step": 50 + }, + { + "epoch": 0.0, + "grad_norm": 4.362759113311768, + "learning_rate": 2.4000000000000003e-06, + "loss": 2.1916, + "step": 60 + }, + { + "epoch": 0.0, + "grad_norm": 4.222783088684082, + "learning_rate": 2.8000000000000003e-06, + "loss": 2.2857, + "step": 70 + }, + { + "epoch": 0.0, + "grad_norm": 3.4690635204315186, + "learning_rate": 3.2000000000000003e-06, + "loss": 2.2253, + "step": 80 + }, + { + "epoch": 0.0, + "grad_norm": 4.878602981567383, + "learning_rate": 3.6000000000000003e-06, + "loss": 2.1578, + "step": 90 + }, + { + "epoch": 0.0, + "grad_norm": 4.750248908996582, + "learning_rate": 4.000000000000001e-06, + "loss": 2.0044, + "step": 100 + }, + { + "epoch": 0.0, + "grad_norm": 7.198167324066162, + "learning_rate": 4.4e-06, + "loss": 1.8984, + "step": 110 + }, + { + "epoch": 0.0, + "grad_norm": 3.500101327896118, + "learning_rate": 4.800000000000001e-06, + "loss": 1.9501, + "step": 120 + }, + { + "epoch": 0.0, + "grad_norm": 7.465259552001953, + "learning_rate": 5.2e-06, + "loss": 1.8413, + "step": 130 + }, + { + "epoch": 0.0, + "grad_norm": 3.6927084922790527, + "learning_rate": 5.600000000000001e-06, + "loss": 2.0101, + "step": 140 + }, + { + "epoch": 0.0, + "grad_norm": 8.518953323364258, + "learning_rate": 6e-06, + "loss": 1.8778, + "step": 150 + }, + { + "epoch": 0.0, + "grad_norm": 3.7254889011383057, + "learning_rate": 6.4000000000000006e-06, + "loss": 1.8361, + "step": 160 + }, + { + "epoch": 0.0, + "grad_norm": 4.634269714355469, + "learning_rate": 6.800000000000001e-06, + "loss": 1.6214, + "step": 170 + }, + { + "epoch": 0.0, + "grad_norm": 8.053196907043457, + "learning_rate": 7.2000000000000005e-06, + "loss": 1.7552, + "step": 180 + }, + { + "epoch": 0.0, + "grad_norm": 3.407431125640869, + "learning_rate": 7.600000000000001e-06, + "loss": 1.8447, + "step": 190 + }, + { + "epoch": 0.01, + "grad_norm": 8.015592575073242, + "learning_rate": 8.000000000000001e-06, + "loss": 1.736, + "step": 200 + }, + { + "epoch": 0.01, + "grad_norm": 10.115213394165039, + "learning_rate": 8.400000000000001e-06, + "loss": 1.5497, + "step": 210 + }, + { + "epoch": 0.01, + "grad_norm": 6.565807819366455, + "learning_rate": 8.8e-06, + "loss": 1.6854, + "step": 220 + }, + { + "epoch": 0.01, + "grad_norm": 4.260842800140381, + "learning_rate": 9.200000000000002e-06, + "loss": 1.986, + "step": 230 + }, + { + "epoch": 0.01, + "grad_norm": 4.239616394042969, + "learning_rate": 9.600000000000001e-06, + "loss": 1.5284, + "step": 240 + }, + { + "epoch": 0.01, + "grad_norm": 3.048999786376953, + "learning_rate": 1e-05, + "loss": 1.7342, + "step": 250 + }, + { + "epoch": 0.01, + "grad_norm": 3.2346177101135254, + "learning_rate": 1.04e-05, + "loss": 1.8604, + "step": 260 + }, + { + "epoch": 0.01, + "grad_norm": 4.236307144165039, + "learning_rate": 1.0800000000000002e-05, + "loss": 1.6303, + "step": 270 + }, + { + "epoch": 0.01, + "grad_norm": 3.784558057785034, + "learning_rate": 1.1200000000000001e-05, + "loss": 1.6309, + "step": 280 + }, + { + "epoch": 0.01, + "grad_norm": 3.2512731552124023, + "learning_rate": 1.16e-05, + "loss": 1.8278, + "step": 290 + }, + { + "epoch": 0.01, + "grad_norm": 4.695291996002197, + "learning_rate": 1.2e-05, + "loss": 1.6175, + "step": 300 + }, + { + "epoch": 0.01, + "grad_norm": 5.554133892059326, + "learning_rate": 1.2400000000000002e-05, + "loss": 1.6657, + "step": 310 + }, + { + "epoch": 0.01, + "grad_norm": 24.56378936767578, + "learning_rate": 1.2800000000000001e-05, + "loss": 1.663, + "step": 320 + }, + { + "epoch": 0.01, + "grad_norm": 2.6168453693389893, + "learning_rate": 1.3200000000000002e-05, + "loss": 1.811, + "step": 330 + }, + { + "epoch": 0.01, + "grad_norm": 3.714534282684326, + "learning_rate": 1.3600000000000002e-05, + "loss": 1.7414, + "step": 340 + }, + { + "epoch": 0.01, + "grad_norm": 4.220114231109619, + "learning_rate": 1.4e-05, + "loss": 1.69, + "step": 350 + }, + { + "epoch": 0.01, + "grad_norm": 2.1215527057647705, + "learning_rate": 1.4400000000000001e-05, + "loss": 1.6113, + "step": 360 + }, + { + "epoch": 0.01, + "grad_norm": 2.8771586418151855, + "learning_rate": 1.48e-05, + "loss": 1.6141, + "step": 370 + }, + { + "epoch": 0.01, + "grad_norm": 5.036810398101807, + "learning_rate": 1.5200000000000002e-05, + "loss": 1.7124, + "step": 380 + }, + { + "epoch": 0.01, + "grad_norm": 1.9079569578170776, + "learning_rate": 1.5600000000000003e-05, + "loss": 1.629, + "step": 390 + }, + { + "epoch": 0.01, + "grad_norm": 9.102452278137207, + "learning_rate": 1.6000000000000003e-05, + "loss": 1.659, + "step": 400 + }, + { + "epoch": 0.01, + "grad_norm": 6.5280866622924805, + "learning_rate": 1.64e-05, + "loss": 1.6147, + "step": 410 + }, + { + "epoch": 0.01, + "grad_norm": 6.27349853515625, + "learning_rate": 1.6800000000000002e-05, + "loss": 1.5316, + "step": 420 + }, + { + "epoch": 0.01, + "grad_norm": 21.586790084838867, + "learning_rate": 1.72e-05, + "loss": 1.7693, + "step": 430 + }, + { + "epoch": 0.01, + "grad_norm": 3.198234796524048, + "learning_rate": 1.76e-05, + "loss": 1.465, + "step": 440 + }, + { + "epoch": 0.01, + "grad_norm": 5.132129669189453, + "learning_rate": 1.8e-05, + "loss": 1.6176, + "step": 450 + }, + { + "epoch": 0.01, + "grad_norm": 5.825719833374023, + "learning_rate": 1.8400000000000003e-05, + "loss": 1.6984, + "step": 460 + }, + { + "epoch": 0.01, + "grad_norm": 6.724545478820801, + "learning_rate": 1.88e-05, + "loss": 1.7011, + "step": 470 + }, + { + "epoch": 0.01, + "grad_norm": 3.723080635070801, + "learning_rate": 1.9200000000000003e-05, + "loss": 1.6338, + "step": 480 + }, + { + "epoch": 0.01, + "grad_norm": 2.8718888759613037, + "learning_rate": 1.9600000000000002e-05, + "loss": 1.5002, + "step": 490 + }, + { + "epoch": 0.01, + "grad_norm": 11.207817077636719, + "learning_rate": 2e-05, + "loss": 1.6829, + "step": 500 + }, + { + "epoch": 0.01, + "eval_loss": 1.6066228151321411, + "eval_runtime": 66.1997, + "eval_samples_per_second": 15.106, + "eval_steps_per_second": 15.106, + "step": 500 + }, + { + "epoch": 0.01, + "grad_norm": 5.459104537963867, + "learning_rate": 1.999322033898305e-05, + "loss": 1.6216, + "step": 510 + }, + { + "epoch": 0.01, + "grad_norm": 2.6112444400787354, + "learning_rate": 1.9986440677966104e-05, + "loss": 1.5761, + "step": 520 + }, + { + "epoch": 0.01, + "grad_norm": 4.285440921783447, + "learning_rate": 1.9979661016949154e-05, + "loss": 1.6097, + "step": 530 + }, + { + "epoch": 0.01, + "grad_norm": 2.781306505203247, + "learning_rate": 1.9972881355932204e-05, + "loss": 1.6967, + "step": 540 + }, + { + "epoch": 0.01, + "grad_norm": 7.55482292175293, + "learning_rate": 1.9966101694915257e-05, + "loss": 1.5974, + "step": 550 + }, + { + "epoch": 0.01, + "grad_norm": 5.083601474761963, + "learning_rate": 1.9959322033898307e-05, + "loss": 1.6287, + "step": 560 + }, + { + "epoch": 0.01, + "grad_norm": 3.7552945613861084, + "learning_rate": 1.995254237288136e-05, + "loss": 1.4225, + "step": 570 + }, + { + "epoch": 0.01, + "grad_norm": 3.149674892425537, + "learning_rate": 1.994576271186441e-05, + "loss": 1.5439, + "step": 580 + }, + { + "epoch": 0.01, + "grad_norm": 5.270935535430908, + "learning_rate": 1.993898305084746e-05, + "loss": 1.8439, + "step": 590 + }, + { + "epoch": 0.01, + "grad_norm": 2.552823543548584, + "learning_rate": 1.9932203389830512e-05, + "loss": 1.5678, + "step": 600 + }, + { + "epoch": 0.02, + "grad_norm": 2.6956074237823486, + "learning_rate": 1.992542372881356e-05, + "loss": 1.6002, + "step": 610 + }, + { + "epoch": 0.02, + "grad_norm": 3.0761303901672363, + "learning_rate": 1.991864406779661e-05, + "loss": 1.7211, + "step": 620 + }, + { + "epoch": 0.02, + "grad_norm": 3.1744227409362793, + "learning_rate": 1.991186440677966e-05, + "loss": 1.6327, + "step": 630 + }, + { + "epoch": 0.02, + "grad_norm": 4.774084091186523, + "learning_rate": 1.990508474576271e-05, + "loss": 1.6871, + "step": 640 + }, + { + "epoch": 0.02, + "grad_norm": 3.4767563343048096, + "learning_rate": 1.9898305084745764e-05, + "loss": 1.6038, + "step": 650 + }, + { + "epoch": 0.02, + "grad_norm": 2.3587417602539062, + "learning_rate": 1.9891525423728814e-05, + "loss": 1.8443, + "step": 660 + }, + { + "epoch": 0.02, + "grad_norm": 3.177602529525757, + "learning_rate": 1.9884745762711867e-05, + "loss": 1.5371, + "step": 670 + }, + { + "epoch": 0.02, + "grad_norm": 2.579742193222046, + "learning_rate": 1.9877966101694917e-05, + "loss": 1.4555, + "step": 680 + }, + { + "epoch": 0.02, + "grad_norm": 3.631253242492676, + "learning_rate": 1.9871186440677966e-05, + "loss": 1.6794, + "step": 690 + }, + { + "epoch": 0.02, + "grad_norm": 2.278071641921997, + "learning_rate": 1.986440677966102e-05, + "loss": 1.477, + "step": 700 + }, + { + "epoch": 0.02, + "grad_norm": 3.924114227294922, + "learning_rate": 1.985762711864407e-05, + "loss": 1.6933, + "step": 710 + }, + { + "epoch": 0.02, + "grad_norm": 3.5087924003601074, + "learning_rate": 1.985084745762712e-05, + "loss": 1.5713, + "step": 720 + }, + { + "epoch": 0.02, + "grad_norm": 2.0184319019317627, + "learning_rate": 1.9844067796610172e-05, + "loss": 1.6689, + "step": 730 + }, + { + "epoch": 0.02, + "grad_norm": 2.6397581100463867, + "learning_rate": 1.9837288135593222e-05, + "loss": 1.5636, + "step": 740 + }, + { + "epoch": 0.02, + "grad_norm": 3.12296986579895, + "learning_rate": 1.9830508474576275e-05, + "loss": 1.6867, + "step": 750 + }, + { + "epoch": 0.02, + "grad_norm": 2.04766583442688, + "learning_rate": 1.9823728813559324e-05, + "loss": 1.6027, + "step": 760 + }, + { + "epoch": 0.02, + "grad_norm": 1.8128042221069336, + "learning_rate": 1.9816949152542374e-05, + "loss": 1.6354, + "step": 770 + }, + { + "epoch": 0.02, + "grad_norm": 2.708834171295166, + "learning_rate": 1.9810169491525427e-05, + "loss": 1.5484, + "step": 780 + }, + { + "epoch": 0.02, + "grad_norm": 1.9796117544174194, + "learning_rate": 1.9803389830508477e-05, + "loss": 1.4511, + "step": 790 + }, + { + "epoch": 0.02, + "grad_norm": 4.195634365081787, + "learning_rate": 1.9796610169491527e-05, + "loss": 1.6435, + "step": 800 + }, + { + "epoch": 0.02, + "grad_norm": 3.132399320602417, + "learning_rate": 1.978983050847458e-05, + "loss": 1.5196, + "step": 810 + }, + { + "epoch": 0.02, + "grad_norm": 2.9280669689178467, + "learning_rate": 1.9783050847457626e-05, + "loss": 1.522, + "step": 820 + }, + { + "epoch": 0.02, + "grad_norm": 2.2013494968414307, + "learning_rate": 1.977627118644068e-05, + "loss": 1.6589, + "step": 830 + }, + { + "epoch": 0.02, + "grad_norm": 9.277434349060059, + "learning_rate": 1.976949152542373e-05, + "loss": 1.5828, + "step": 840 + }, + { + "epoch": 0.02, + "grad_norm": 6.09869384765625, + "learning_rate": 1.9762711864406782e-05, + "loss": 1.6683, + "step": 850 + }, + { + "epoch": 0.02, + "grad_norm": 1.7093573808670044, + "learning_rate": 1.9755932203389832e-05, + "loss": 1.6881, + "step": 860 + }, + { + "epoch": 0.02, + "grad_norm": 1.834920048713684, + "learning_rate": 1.974915254237288e-05, + "loss": 1.6169, + "step": 870 + }, + { + "epoch": 0.02, + "grad_norm": 3.802130699157715, + "learning_rate": 1.9742372881355935e-05, + "loss": 1.5526, + "step": 880 + }, + { + "epoch": 0.02, + "grad_norm": 2.086676836013794, + "learning_rate": 1.9735593220338984e-05, + "loss": 1.661, + "step": 890 + }, + { + "epoch": 0.02, + "grad_norm": 1.9825890064239502, + "learning_rate": 1.9728813559322034e-05, + "loss": 1.6302, + "step": 900 + }, + { + "epoch": 0.02, + "grad_norm": 1.9660803079605103, + "learning_rate": 1.9722033898305087e-05, + "loss": 1.6146, + "step": 910 + }, + { + "epoch": 0.02, + "grad_norm": 3.5572495460510254, + "learning_rate": 1.9715254237288137e-05, + "loss": 1.6143, + "step": 920 + }, + { + "epoch": 0.02, + "grad_norm": 5.440743923187256, + "learning_rate": 1.970847457627119e-05, + "loss": 1.4907, + "step": 930 + }, + { + "epoch": 0.02, + "grad_norm": 2.432176351547241, + "learning_rate": 1.970169491525424e-05, + "loss": 1.6272, + "step": 940 + }, + { + "epoch": 0.02, + "grad_norm": 1.2422198057174683, + "learning_rate": 1.969491525423729e-05, + "loss": 1.6404, + "step": 950 + }, + { + "epoch": 0.02, + "grad_norm": 3.0598487854003906, + "learning_rate": 1.9688135593220342e-05, + "loss": 1.5811, + "step": 960 + }, + { + "epoch": 0.02, + "grad_norm": 3.399864673614502, + "learning_rate": 1.9681355932203392e-05, + "loss": 1.6221, + "step": 970 + }, + { + "epoch": 0.02, + "grad_norm": 1.3376494646072388, + "learning_rate": 1.9674576271186442e-05, + "loss": 1.5384, + "step": 980 + }, + { + "epoch": 0.02, + "grad_norm": 4.062708854675293, + "learning_rate": 1.9667796610169495e-05, + "loss": 1.5448, + "step": 990 + }, + { + "epoch": 0.03, + "grad_norm": 4.200625419616699, + "learning_rate": 1.9661016949152545e-05, + "loss": 1.6656, + "step": 1000 + }, + { + "epoch": 0.03, + "eval_loss": 1.5288029909133911, + "eval_runtime": 66.2062, + "eval_samples_per_second": 15.104, + "eval_steps_per_second": 15.104, + "step": 1000 + }, + { + "epoch": 0.03, + "grad_norm": 4.217132091522217, + "learning_rate": 1.9654237288135594e-05, + "loss": 1.6834, + "step": 1010 + }, + { + "epoch": 0.03, + "grad_norm": 1.8246206045150757, + "learning_rate": 1.9647457627118644e-05, + "loss": 1.7242, + "step": 1020 + }, + { + "epoch": 0.03, + "grad_norm": 3.0907254219055176, + "learning_rate": 1.9640677966101697e-05, + "loss": 1.6493, + "step": 1030 + }, + { + "epoch": 0.03, + "grad_norm": 2.1235597133636475, + "learning_rate": 1.9633898305084747e-05, + "loss": 1.6628, + "step": 1040 + }, + { + "epoch": 0.03, + "grad_norm": 4.815682411193848, + "learning_rate": 1.9627118644067796e-05, + "loss": 1.5552, + "step": 1050 + }, + { + "epoch": 0.03, + "grad_norm": 2.7466652393341064, + "learning_rate": 1.962033898305085e-05, + "loss": 1.7091, + "step": 1060 + }, + { + "epoch": 0.03, + "grad_norm": 1.9031518697738647, + "learning_rate": 1.96135593220339e-05, + "loss": 1.4593, + "step": 1070 + }, + { + "epoch": 0.03, + "grad_norm": 7.788964748382568, + "learning_rate": 1.960677966101695e-05, + "loss": 1.7782, + "step": 1080 + }, + { + "epoch": 0.03, + "grad_norm": 3.2768681049346924, + "learning_rate": 1.9600000000000002e-05, + "loss": 1.6339, + "step": 1090 + }, + { + "epoch": 0.03, + "grad_norm": 2.7264978885650635, + "learning_rate": 1.9593220338983052e-05, + "loss": 1.7666, + "step": 1100 + }, + { + "epoch": 0.03, + "grad_norm": 4.865339279174805, + "learning_rate": 1.95864406779661e-05, + "loss": 1.7127, + "step": 1110 + }, + { + "epoch": 0.03, + "grad_norm": 2.9894134998321533, + "learning_rate": 1.9579661016949155e-05, + "loss": 1.5671, + "step": 1120 + }, + { + "epoch": 0.03, + "grad_norm": 3.2608160972595215, + "learning_rate": 1.9572881355932204e-05, + "loss": 1.5396, + "step": 1130 + }, + { + "epoch": 0.03, + "grad_norm": 2.3654749393463135, + "learning_rate": 1.9566101694915257e-05, + "loss": 1.514, + "step": 1140 + }, + { + "epoch": 0.03, + "grad_norm": 1.9910569190979004, + "learning_rate": 1.9559322033898307e-05, + "loss": 1.6519, + "step": 1150 + }, + { + "epoch": 0.03, + "grad_norm": 3.1228861808776855, + "learning_rate": 1.9552542372881357e-05, + "loss": 1.5683, + "step": 1160 + }, + { + "epoch": 0.03, + "grad_norm": 2.8690907955169678, + "learning_rate": 1.954576271186441e-05, + "loss": 1.5795, + "step": 1170 + }, + { + "epoch": 0.03, + "grad_norm": 7.847595691680908, + "learning_rate": 1.953898305084746e-05, + "loss": 1.4718, + "step": 1180 + }, + { + "epoch": 0.03, + "grad_norm": 3.4904606342315674, + "learning_rate": 1.953220338983051e-05, + "loss": 1.6105, + "step": 1190 + }, + { + "epoch": 0.03, + "grad_norm": 2.352919340133667, + "learning_rate": 1.9525423728813562e-05, + "loss": 1.6323, + "step": 1200 + }, + { + "epoch": 0.03, + "grad_norm": 1.9066511392593384, + "learning_rate": 1.9518644067796612e-05, + "loss": 1.5846, + "step": 1210 + }, + { + "epoch": 0.03, + "grad_norm": 2.2066802978515625, + "learning_rate": 1.9511864406779665e-05, + "loss": 1.4947, + "step": 1220 + }, + { + "epoch": 0.03, + "grad_norm": 2.038661003112793, + "learning_rate": 1.950508474576271e-05, + "loss": 1.5167, + "step": 1230 + }, + { + "epoch": 0.03, + "grad_norm": 4.544153690338135, + "learning_rate": 1.9498305084745765e-05, + "loss": 1.7429, + "step": 1240 + }, + { + "epoch": 0.03, + "grad_norm": 4.264445781707764, + "learning_rate": 1.9491525423728814e-05, + "loss": 1.7164, + "step": 1250 + }, + { + "epoch": 0.03, + "grad_norm": 7.955377101898193, + "learning_rate": 1.9484745762711864e-05, + "loss": 1.4268, + "step": 1260 + }, + { + "epoch": 0.03, + "grad_norm": 3.6983351707458496, + "learning_rate": 1.9477966101694917e-05, + "loss": 1.5497, + "step": 1270 + }, + { + "epoch": 0.03, + "grad_norm": 2.3593084812164307, + "learning_rate": 1.9471186440677967e-05, + "loss": 1.4209, + "step": 1280 + }, + { + "epoch": 0.03, + "grad_norm": 5.2454938888549805, + "learning_rate": 1.9464406779661017e-05, + "loss": 1.7169, + "step": 1290 + }, + { + "epoch": 0.03, + "grad_norm": 4.33056116104126, + "learning_rate": 1.945762711864407e-05, + "loss": 1.491, + "step": 1300 + }, + { + "epoch": 0.03, + "grad_norm": 3.095102071762085, + "learning_rate": 1.945084745762712e-05, + "loss": 1.5366, + "step": 1310 + }, + { + "epoch": 0.03, + "grad_norm": 1.7006287574768066, + "learning_rate": 1.9444067796610172e-05, + "loss": 1.5085, + "step": 1320 + }, + { + "epoch": 0.03, + "grad_norm": 2.643651008605957, + "learning_rate": 1.9437288135593222e-05, + "loss": 1.4597, + "step": 1330 + }, + { + "epoch": 0.03, + "grad_norm": 1.9585516452789307, + "learning_rate": 1.9430508474576272e-05, + "loss": 1.6975, + "step": 1340 + }, + { + "epoch": 0.03, + "grad_norm": 2.669740676879883, + "learning_rate": 1.9423728813559325e-05, + "loss": 1.4074, + "step": 1350 + }, + { + "epoch": 0.03, + "grad_norm": 3.0379438400268555, + "learning_rate": 1.9416949152542375e-05, + "loss": 1.5256, + "step": 1360 + }, + { + "epoch": 0.03, + "grad_norm": 2.81781005859375, + "learning_rate": 1.9410169491525424e-05, + "loss": 1.5329, + "step": 1370 + }, + { + "epoch": 0.03, + "grad_norm": 1.2218108177185059, + "learning_rate": 1.9403389830508477e-05, + "loss": 1.5283, + "step": 1380 + }, + { + "epoch": 0.03, + "grad_norm": 2.4585816860198975, + "learning_rate": 1.9396610169491527e-05, + "loss": 1.4265, + "step": 1390 + }, + { + "epoch": 0.04, + "grad_norm": 6.117985725402832, + "learning_rate": 1.938983050847458e-05, + "loss": 1.6968, + "step": 1400 + }, + { + "epoch": 0.04, + "grad_norm": 1.7194918394088745, + "learning_rate": 1.938305084745763e-05, + "loss": 1.5565, + "step": 1410 + }, + { + "epoch": 0.04, + "grad_norm": 3.65537691116333, + "learning_rate": 1.937627118644068e-05, + "loss": 1.7526, + "step": 1420 + }, + { + "epoch": 0.04, + "grad_norm": 3.0874905586242676, + "learning_rate": 1.9369491525423733e-05, + "loss": 1.4787, + "step": 1430 + }, + { + "epoch": 0.04, + "grad_norm": 3.945019006729126, + "learning_rate": 1.936271186440678e-05, + "loss": 1.5238, + "step": 1440 + }, + { + "epoch": 0.04, + "grad_norm": 2.7200796604156494, + "learning_rate": 1.9355932203389832e-05, + "loss": 1.4571, + "step": 1450 + }, + { + "epoch": 0.04, + "grad_norm": 8.368704795837402, + "learning_rate": 1.9349152542372882e-05, + "loss": 1.5916, + "step": 1460 + }, + { + "epoch": 0.04, + "grad_norm": 3.0204453468322754, + "learning_rate": 1.934237288135593e-05, + "loss": 1.6997, + "step": 1470 + }, + { + "epoch": 0.04, + "grad_norm": 2.3051364421844482, + "learning_rate": 1.9335593220338985e-05, + "loss": 1.5315, + "step": 1480 + }, + { + "epoch": 0.04, + "grad_norm": 1.246631145477295, + "learning_rate": 1.9328813559322034e-05, + "loss": 1.478, + "step": 1490 + }, + { + "epoch": 0.04, + "grad_norm": 1.8742200136184692, + "learning_rate": 1.9322033898305087e-05, + "loss": 1.5419, + "step": 1500 + }, + { + "epoch": 0.04, + "eval_loss": 1.5387661457061768, + "eval_runtime": 66.1849, + "eval_samples_per_second": 15.109, + "eval_steps_per_second": 15.109, + "step": 1500 + }, + { + "epoch": 0.04, + "grad_norm": 1.624708652496338, + "learning_rate": 1.9315254237288137e-05, + "loss": 1.5002, + "step": 1510 + }, + { + "epoch": 0.04, + "grad_norm": 5.5269856452941895, + "learning_rate": 1.9308474576271187e-05, + "loss": 1.6267, + "step": 1520 + }, + { + "epoch": 0.04, + "grad_norm": 1.879159688949585, + "learning_rate": 1.930169491525424e-05, + "loss": 1.4951, + "step": 1530 + }, + { + "epoch": 0.04, + "grad_norm": 4.727252006530762, + "learning_rate": 1.929491525423729e-05, + "loss": 1.5438, + "step": 1540 + }, + { + "epoch": 0.04, + "grad_norm": 1.9965201616287231, + "learning_rate": 1.928813559322034e-05, + "loss": 1.6525, + "step": 1550 + }, + { + "epoch": 0.04, + "grad_norm": 2.61816143989563, + "learning_rate": 1.9281355932203392e-05, + "loss": 1.4746, + "step": 1560 + }, + { + "epoch": 0.04, + "grad_norm": 2.8157131671905518, + "learning_rate": 1.9274576271186442e-05, + "loss": 1.5454, + "step": 1570 + }, + { + "epoch": 0.04, + "grad_norm": 2.738640546798706, + "learning_rate": 1.9267796610169492e-05, + "loss": 1.4876, + "step": 1580 + }, + { + "epoch": 0.04, + "grad_norm": 3.118277072906494, + "learning_rate": 1.9261016949152545e-05, + "loss": 1.5587, + "step": 1590 + }, + { + "epoch": 0.04, + "grad_norm": 6.730972766876221, + "learning_rate": 1.9254237288135595e-05, + "loss": 1.6752, + "step": 1600 + }, + { + "epoch": 0.04, + "grad_norm": 9.461782455444336, + "learning_rate": 1.9247457627118648e-05, + "loss": 1.5409, + "step": 1610 + }, + { + "epoch": 0.04, + "grad_norm": 4.566100120544434, + "learning_rate": 1.9240677966101698e-05, + "loss": 1.4724, + "step": 1620 + }, + { + "epoch": 0.04, + "grad_norm": 3.8972136974334717, + "learning_rate": 1.9233898305084747e-05, + "loss": 1.5706, + "step": 1630 + }, + { + "epoch": 0.04, + "grad_norm": 2.0403244495391846, + "learning_rate": 1.92271186440678e-05, + "loss": 1.6694, + "step": 1640 + }, + { + "epoch": 0.04, + "grad_norm": 3.0590033531188965, + "learning_rate": 1.9220338983050847e-05, + "loss": 1.5711, + "step": 1650 + }, + { + "epoch": 0.04, + "grad_norm": 0.9862588047981262, + "learning_rate": 1.92135593220339e-05, + "loss": 1.486, + "step": 1660 + }, + { + "epoch": 0.04, + "grad_norm": 2.9184350967407227, + "learning_rate": 1.920677966101695e-05, + "loss": 1.5526, + "step": 1670 + }, + { + "epoch": 0.04, + "grad_norm": 3.7024717330932617, + "learning_rate": 1.9200000000000003e-05, + "loss": 1.6754, + "step": 1680 + }, + { + "epoch": 0.04, + "grad_norm": 4.887876510620117, + "learning_rate": 1.9193220338983052e-05, + "loss": 1.6843, + "step": 1690 + }, + { + "epoch": 0.04, + "grad_norm": 2.6079001426696777, + "learning_rate": 1.9186440677966102e-05, + "loss": 1.5116, + "step": 1700 + }, + { + "epoch": 0.04, + "grad_norm": 3.407667875289917, + "learning_rate": 1.9179661016949155e-05, + "loss": 1.581, + "step": 1710 + }, + { + "epoch": 0.04, + "grad_norm": 2.201183795928955, + "learning_rate": 1.9172881355932205e-05, + "loss": 1.6449, + "step": 1720 + }, + { + "epoch": 0.04, + "grad_norm": 1.4738363027572632, + "learning_rate": 1.9166101694915254e-05, + "loss": 1.3959, + "step": 1730 + }, + { + "epoch": 0.04, + "grad_norm": 1.745976209640503, + "learning_rate": 1.9159322033898308e-05, + "loss": 1.394, + "step": 1740 + }, + { + "epoch": 0.04, + "grad_norm": 3.694169759750366, + "learning_rate": 1.9152542372881357e-05, + "loss": 1.6542, + "step": 1750 + }, + { + "epoch": 0.04, + "grad_norm": 2.3387959003448486, + "learning_rate": 1.9145762711864407e-05, + "loss": 1.6928, + "step": 1760 + }, + { + "epoch": 0.04, + "grad_norm": 2.8310043811798096, + "learning_rate": 1.913898305084746e-05, + "loss": 1.5955, + "step": 1770 + }, + { + "epoch": 0.04, + "grad_norm": 2.731762647628784, + "learning_rate": 1.913220338983051e-05, + "loss": 1.7997, + "step": 1780 + }, + { + "epoch": 0.04, + "grad_norm": 2.896686553955078, + "learning_rate": 1.9125423728813563e-05, + "loss": 1.5137, + "step": 1790 + }, + { + "epoch": 0.04, + "grad_norm": 1.8352593183517456, + "learning_rate": 1.9118644067796613e-05, + "loss": 1.5224, + "step": 1800 + }, + { + "epoch": 0.05, + "grad_norm": 4.474410057067871, + "learning_rate": 1.9111864406779662e-05, + "loss": 1.6659, + "step": 1810 + }, + { + "epoch": 0.05, + "grad_norm": 3.063828468322754, + "learning_rate": 1.9105084745762715e-05, + "loss": 1.4191, + "step": 1820 + }, + { + "epoch": 0.05, + "grad_norm": 1.9058163166046143, + "learning_rate": 1.9098305084745765e-05, + "loss": 1.5386, + "step": 1830 + }, + { + "epoch": 0.05, + "grad_norm": 2.915477991104126, + "learning_rate": 1.9091525423728815e-05, + "loss": 1.5756, + "step": 1840 + }, + { + "epoch": 0.05, + "grad_norm": 1.423071265220642, + "learning_rate": 1.9084745762711868e-05, + "loss": 1.5377, + "step": 1850 + }, + { + "epoch": 0.05, + "grad_norm": 2.5523242950439453, + "learning_rate": 1.9077966101694914e-05, + "loss": 1.5199, + "step": 1860 + }, + { + "epoch": 0.05, + "grad_norm": 2.609219789505005, + "learning_rate": 1.9071186440677967e-05, + "loss": 1.5402, + "step": 1870 + }, + { + "epoch": 0.05, + "grad_norm": 7.032884120941162, + "learning_rate": 1.9064406779661017e-05, + "loss": 1.4348, + "step": 1880 + }, + { + "epoch": 0.05, + "grad_norm": 3.192718029022217, + "learning_rate": 1.905762711864407e-05, + "loss": 1.5971, + "step": 1890 + }, + { + "epoch": 0.05, + "grad_norm": 2.0084311962127686, + "learning_rate": 1.905084745762712e-05, + "loss": 1.459, + "step": 1900 + }, + { + "epoch": 0.05, + "grad_norm": 1.646857738494873, + "learning_rate": 1.904406779661017e-05, + "loss": 1.507, + "step": 1910 + }, + { + "epoch": 0.05, + "grad_norm": 3.855058431625366, + "learning_rate": 1.9037288135593223e-05, + "loss": 1.5667, + "step": 1920 + }, + { + "epoch": 0.05, + "grad_norm": 1.520700216293335, + "learning_rate": 1.9030508474576272e-05, + "loss": 1.573, + "step": 1930 + }, + { + "epoch": 0.05, + "grad_norm": 2.6929845809936523, + "learning_rate": 1.9023728813559322e-05, + "loss": 1.4966, + "step": 1940 + }, + { + "epoch": 0.05, + "grad_norm": 2.283622980117798, + "learning_rate": 1.9016949152542375e-05, + "loss": 1.5018, + "step": 1950 + }, + { + "epoch": 0.05, + "grad_norm": 1.3686760663986206, + "learning_rate": 1.9010169491525425e-05, + "loss": 1.3841, + "step": 1960 + }, + { + "epoch": 0.05, + "grad_norm": 2.02280592918396, + "learning_rate": 1.9003389830508478e-05, + "loss": 1.3902, + "step": 1970 + }, + { + "epoch": 0.05, + "grad_norm": 2.312986135482788, + "learning_rate": 1.8996610169491528e-05, + "loss": 1.6016, + "step": 1980 + }, + { + "epoch": 0.05, + "grad_norm": 2.497074604034424, + "learning_rate": 1.8989830508474577e-05, + "loss": 1.6256, + "step": 1990 + }, + { + "epoch": 0.05, + "grad_norm": 8.351258277893066, + "learning_rate": 1.898305084745763e-05, + "loss": 1.5034, + "step": 2000 + }, + { + "epoch": 0.05, + "eval_loss": 1.555788516998291, + "eval_runtime": 66.1337, + "eval_samples_per_second": 15.121, + "eval_steps_per_second": 15.121, + "step": 2000 + }, + { + "epoch": 0.05, + "grad_norm": 10.036495208740234, + "learning_rate": 1.897627118644068e-05, + "loss": 1.6968, + "step": 2010 + }, + { + "epoch": 0.05, + "grad_norm": 5.859747886657715, + "learning_rate": 1.896949152542373e-05, + "loss": 1.655, + "step": 2020 + }, + { + "epoch": 0.05, + "grad_norm": 2.809723138809204, + "learning_rate": 1.8962711864406783e-05, + "loss": 1.5661, + "step": 2030 + }, + { + "epoch": 0.05, + "grad_norm": 3.2033305168151855, + "learning_rate": 1.8955932203389833e-05, + "loss": 1.5528, + "step": 2040 + }, + { + "epoch": 0.05, + "grad_norm": 2.821791887283325, + "learning_rate": 1.8949152542372882e-05, + "loss": 1.5215, + "step": 2050 + }, + { + "epoch": 0.05, + "grad_norm": 3.9140360355377197, + "learning_rate": 1.8942372881355932e-05, + "loss": 1.6541, + "step": 2060 + }, + { + "epoch": 0.05, + "grad_norm": 1.9218617677688599, + "learning_rate": 1.8935593220338985e-05, + "loss": 1.6119, + "step": 2070 + }, + { + "epoch": 0.05, + "grad_norm": 1.7994463443756104, + "learning_rate": 1.8928813559322035e-05, + "loss": 1.5222, + "step": 2080 + }, + { + "epoch": 0.05, + "grad_norm": 2.291020631790161, + "learning_rate": 1.8922033898305085e-05, + "loss": 1.5315, + "step": 2090 + }, + { + "epoch": 0.05, + "grad_norm": 2.737337589263916, + "learning_rate": 1.8915254237288138e-05, + "loss": 1.4785, + "step": 2100 + }, + { + "epoch": 0.05, + "grad_norm": 3.1229870319366455, + "learning_rate": 1.8908474576271187e-05, + "loss": 1.6181, + "step": 2110 + }, + { + "epoch": 0.05, + "grad_norm": 2.7481417655944824, + "learning_rate": 1.8901694915254237e-05, + "loss": 1.4123, + "step": 2120 + }, + { + "epoch": 0.05, + "grad_norm": 3.3576602935791016, + "learning_rate": 1.889491525423729e-05, + "loss": 1.5839, + "step": 2130 + }, + { + "epoch": 0.05, + "grad_norm": 3.292433261871338, + "learning_rate": 1.888813559322034e-05, + "loss": 1.5668, + "step": 2140 + }, + { + "epoch": 0.05, + "grad_norm": 5.039631366729736, + "learning_rate": 1.8881355932203393e-05, + "loss": 1.3932, + "step": 2150 + }, + { + "epoch": 0.05, + "grad_norm": 2.142812490463257, + "learning_rate": 1.8874576271186443e-05, + "loss": 1.6613, + "step": 2160 + }, + { + "epoch": 0.05, + "grad_norm": 1.2970080375671387, + "learning_rate": 1.8867796610169492e-05, + "loss": 1.4982, + "step": 2170 + }, + { + "epoch": 0.05, + "grad_norm": 7.887810230255127, + "learning_rate": 1.8861016949152545e-05, + "loss": 1.6085, + "step": 2180 + }, + { + "epoch": 0.05, + "grad_norm": 3.8233675956726074, + "learning_rate": 1.8854237288135595e-05, + "loss": 1.668, + "step": 2190 + }, + { + "epoch": 0.06, + "grad_norm": 2.357754707336426, + "learning_rate": 1.8847457627118645e-05, + "loss": 1.6456, + "step": 2200 + }, + { + "epoch": 0.06, + "grad_norm": 5.471128463745117, + "learning_rate": 1.8840677966101698e-05, + "loss": 1.5198, + "step": 2210 + }, + { + "epoch": 0.06, + "grad_norm": 1.850195050239563, + "learning_rate": 1.8833898305084748e-05, + "loss": 1.5557, + "step": 2220 + }, + { + "epoch": 0.06, + "grad_norm": 2.1556105613708496, + "learning_rate": 1.8827118644067797e-05, + "loss": 1.6411, + "step": 2230 + }, + { + "epoch": 0.06, + "grad_norm": 3.6570167541503906, + "learning_rate": 1.882033898305085e-05, + "loss": 1.474, + "step": 2240 + }, + { + "epoch": 0.06, + "grad_norm": 2.989947557449341, + "learning_rate": 1.88135593220339e-05, + "loss": 1.5413, + "step": 2250 + }, + { + "epoch": 0.06, + "grad_norm": 1.8367825746536255, + "learning_rate": 1.8806779661016953e-05, + "loss": 1.5744, + "step": 2260 + }, + { + "epoch": 0.06, + "grad_norm": 2.2120065689086914, + "learning_rate": 1.88e-05, + "loss": 1.5589, + "step": 2270 + }, + { + "epoch": 0.06, + "grad_norm": 3.29278302192688, + "learning_rate": 1.8793220338983053e-05, + "loss": 1.4468, + "step": 2280 + }, + { + "epoch": 0.06, + "grad_norm": 2.6164047718048096, + "learning_rate": 1.8786440677966102e-05, + "loss": 1.5143, + "step": 2290 + }, + { + "epoch": 0.06, + "grad_norm": 3.008776903152466, + "learning_rate": 1.8779661016949152e-05, + "loss": 1.5692, + "step": 2300 + }, + { + "epoch": 0.06, + "grad_norm": 2.0924556255340576, + "learning_rate": 1.8772881355932205e-05, + "loss": 1.4559, + "step": 2310 + }, + { + "epoch": 0.06, + "grad_norm": 2.1632914543151855, + "learning_rate": 1.8766101694915255e-05, + "loss": 1.5876, + "step": 2320 + }, + { + "epoch": 0.06, + "grad_norm": 2.367845058441162, + "learning_rate": 1.8759322033898305e-05, + "loss": 1.5093, + "step": 2330 + }, + { + "epoch": 0.06, + "grad_norm": 2.618950843811035, + "learning_rate": 1.8752542372881358e-05, + "loss": 1.4839, + "step": 2340 + }, + { + "epoch": 0.06, + "grad_norm": 2.7340502738952637, + "learning_rate": 1.8745762711864407e-05, + "loss": 1.7354, + "step": 2350 + }, + { + "epoch": 0.06, + "grad_norm": 3.4295051097869873, + "learning_rate": 1.873898305084746e-05, + "loss": 1.4937, + "step": 2360 + }, + { + "epoch": 0.06, + "grad_norm": 1.078732967376709, + "learning_rate": 1.873220338983051e-05, + "loss": 1.7125, + "step": 2370 + }, + { + "epoch": 0.06, + "grad_norm": 3.103182554244995, + "learning_rate": 1.872542372881356e-05, + "loss": 1.4019, + "step": 2380 + }, + { + "epoch": 0.06, + "grad_norm": 1.760866641998291, + "learning_rate": 1.8718644067796613e-05, + "loss": 1.4383, + "step": 2390 + }, + { + "epoch": 0.06, + "grad_norm": 1.4431618452072144, + "learning_rate": 1.8711864406779663e-05, + "loss": 1.5302, + "step": 2400 + }, + { + "epoch": 0.06, + "grad_norm": 3.5087249279022217, + "learning_rate": 1.8705084745762712e-05, + "loss": 1.5693, + "step": 2410 + }, + { + "epoch": 0.06, + "grad_norm": 3.648737668991089, + "learning_rate": 1.8698305084745765e-05, + "loss": 1.2027, + "step": 2420 + }, + { + "epoch": 0.06, + "grad_norm": 4.0298357009887695, + "learning_rate": 1.8691525423728815e-05, + "loss": 1.4772, + "step": 2430 + }, + { + "epoch": 0.06, + "grad_norm": 4.11863899230957, + "learning_rate": 1.8684745762711868e-05, + "loss": 1.402, + "step": 2440 + }, + { + "epoch": 0.06, + "grad_norm": 1.1432223320007324, + "learning_rate": 1.8677966101694918e-05, + "loss": 1.5907, + "step": 2450 + }, + { + "epoch": 0.06, + "grad_norm": 2.0923776626586914, + "learning_rate": 1.8671186440677968e-05, + "loss": 1.569, + "step": 2460 + }, + { + "epoch": 0.06, + "grad_norm": 1.987396001815796, + "learning_rate": 1.866440677966102e-05, + "loss": 1.6504, + "step": 2470 + }, + { + "epoch": 0.06, + "grad_norm": 3.959685802459717, + "learning_rate": 1.8657627118644067e-05, + "loss": 1.4029, + "step": 2480 + }, + { + "epoch": 0.06, + "grad_norm": 3.1348440647125244, + "learning_rate": 1.865084745762712e-05, + "loss": 1.5063, + "step": 2490 + }, + { + "epoch": 0.06, + "grad_norm": 2.788222074508667, + "learning_rate": 1.864406779661017e-05, + "loss": 1.5971, + "step": 2500 + }, + { + "epoch": 0.06, + "eval_loss": 1.5344786643981934, + "eval_runtime": 66.1387, + "eval_samples_per_second": 15.12, + "eval_steps_per_second": 15.12, + "step": 2500 + }, + { + "epoch": 0.06, + "grad_norm": 1.9007914066314697, + "learning_rate": 1.863728813559322e-05, + "loss": 1.5292, + "step": 2510 + }, + { + "epoch": 0.06, + "grad_norm": 2.1382060050964355, + "learning_rate": 1.8630508474576273e-05, + "loss": 1.516, + "step": 2520 + }, + { + "epoch": 0.06, + "grad_norm": 0.914665162563324, + "learning_rate": 1.8623728813559322e-05, + "loss": 1.3994, + "step": 2530 + }, + { + "epoch": 0.06, + "grad_norm": 4.413509368896484, + "learning_rate": 1.8616949152542376e-05, + "loss": 1.5696, + "step": 2540 + }, + { + "epoch": 0.06, + "grad_norm": 1.9687520265579224, + "learning_rate": 1.8610169491525425e-05, + "loss": 1.6693, + "step": 2550 + }, + { + "epoch": 0.06, + "grad_norm": 3.4700605869293213, + "learning_rate": 1.8603389830508475e-05, + "loss": 1.6217, + "step": 2560 + }, + { + "epoch": 0.06, + "grad_norm": 2.2257847785949707, + "learning_rate": 1.8596610169491528e-05, + "loss": 1.65, + "step": 2570 + }, + { + "epoch": 0.06, + "grad_norm": 15.683956146240234, + "learning_rate": 1.8589830508474578e-05, + "loss": 1.5663, + "step": 2580 + }, + { + "epoch": 0.06, + "grad_norm": 2.2584922313690186, + "learning_rate": 1.8583050847457627e-05, + "loss": 1.5146, + "step": 2590 + }, + { + "epoch": 0.07, + "grad_norm": 1.900083065032959, + "learning_rate": 1.857627118644068e-05, + "loss": 1.5062, + "step": 2600 + }, + { + "epoch": 0.07, + "grad_norm": 2.6166887283325195, + "learning_rate": 1.856949152542373e-05, + "loss": 1.3145, + "step": 2610 + }, + { + "epoch": 0.07, + "grad_norm": 0.9894976019859314, + "learning_rate": 1.856271186440678e-05, + "loss": 1.6037, + "step": 2620 + }, + { + "epoch": 0.07, + "grad_norm": 5.897536277770996, + "learning_rate": 1.8555932203389833e-05, + "loss": 1.5846, + "step": 2630 + }, + { + "epoch": 0.07, + "grad_norm": 3.8714261054992676, + "learning_rate": 1.8549152542372883e-05, + "loss": 1.5008, + "step": 2640 + }, + { + "epoch": 0.07, + "grad_norm": 3.062786817550659, + "learning_rate": 1.8542372881355936e-05, + "loss": 1.5885, + "step": 2650 + }, + { + "epoch": 0.07, + "grad_norm": 4.548590660095215, + "learning_rate": 1.8535593220338986e-05, + "loss": 1.4718, + "step": 2660 + }, + { + "epoch": 0.07, + "grad_norm": 1.9573743343353271, + "learning_rate": 1.8528813559322035e-05, + "loss": 1.5926, + "step": 2670 + }, + { + "epoch": 0.07, + "grad_norm": 1.7771761417388916, + "learning_rate": 1.852203389830509e-05, + "loss": 1.5136, + "step": 2680 + }, + { + "epoch": 0.07, + "grad_norm": 3.312812328338623, + "learning_rate": 1.8515254237288135e-05, + "loss": 1.4674, + "step": 2690 + }, + { + "epoch": 0.07, + "grad_norm": 2.288461685180664, + "learning_rate": 1.8508474576271188e-05, + "loss": 1.576, + "step": 2700 + }, + { + "epoch": 0.07, + "grad_norm": 3.608473062515259, + "learning_rate": 1.8501694915254237e-05, + "loss": 1.5574, + "step": 2710 + }, + { + "epoch": 0.07, + "grad_norm": 2.6018731594085693, + "learning_rate": 1.849491525423729e-05, + "loss": 1.473, + "step": 2720 + }, + { + "epoch": 0.07, + "grad_norm": 2.2333261966705322, + "learning_rate": 1.848813559322034e-05, + "loss": 1.4562, + "step": 2730 + }, + { + "epoch": 0.07, + "grad_norm": 6.693967342376709, + "learning_rate": 1.848135593220339e-05, + "loss": 1.433, + "step": 2740 + }, + { + "epoch": 0.07, + "grad_norm": 2.6075336933135986, + "learning_rate": 1.8474576271186443e-05, + "loss": 1.2995, + "step": 2750 + }, + { + "epoch": 0.07, + "grad_norm": 2.2900924682617188, + "learning_rate": 1.8467796610169493e-05, + "loss": 1.4313, + "step": 2760 + }, + { + "epoch": 0.07, + "grad_norm": 5.270084381103516, + "learning_rate": 1.8461016949152542e-05, + "loss": 1.4531, + "step": 2770 + }, + { + "epoch": 0.07, + "grad_norm": 2.3377845287323, + "learning_rate": 1.8454237288135596e-05, + "loss": 1.4393, + "step": 2780 + }, + { + "epoch": 0.07, + "grad_norm": 1.9369862079620361, + "learning_rate": 1.8447457627118645e-05, + "loss": 1.5269, + "step": 2790 + }, + { + "epoch": 0.07, + "grad_norm": 3.107175350189209, + "learning_rate": 1.8440677966101695e-05, + "loss": 1.4581, + "step": 2800 + }, + { + "epoch": 0.07, + "grad_norm": 2.7586755752563477, + "learning_rate": 1.8433898305084748e-05, + "loss": 1.4577, + "step": 2810 + }, + { + "epoch": 0.07, + "grad_norm": 1.3124182224273682, + "learning_rate": 1.8427118644067798e-05, + "loss": 1.6085, + "step": 2820 + }, + { + "epoch": 0.07, + "grad_norm": 1.8919298648834229, + "learning_rate": 1.842033898305085e-05, + "loss": 1.3827, + "step": 2830 + }, + { + "epoch": 0.07, + "grad_norm": 2.49289870262146, + "learning_rate": 1.84135593220339e-05, + "loss": 1.4628, + "step": 2840 + }, + { + "epoch": 0.07, + "grad_norm": 2.832827568054199, + "learning_rate": 1.840677966101695e-05, + "loss": 1.4701, + "step": 2850 + }, + { + "epoch": 0.07, + "grad_norm": 1.2280476093292236, + "learning_rate": 1.8400000000000003e-05, + "loss": 1.5576, + "step": 2860 + }, + { + "epoch": 0.07, + "grad_norm": 2.703963279724121, + "learning_rate": 1.8393220338983053e-05, + "loss": 1.3686, + "step": 2870 + }, + { + "epoch": 0.07, + "grad_norm": 1.554801106452942, + "learning_rate": 1.8386440677966103e-05, + "loss": 1.5867, + "step": 2880 + }, + { + "epoch": 0.07, + "grad_norm": 1.4219539165496826, + "learning_rate": 1.8379661016949153e-05, + "loss": 1.696, + "step": 2890 + }, + { + "epoch": 0.07, + "grad_norm": 7.199713706970215, + "learning_rate": 1.8372881355932202e-05, + "loss": 1.5041, + "step": 2900 + }, + { + "epoch": 0.07, + "grad_norm": 2.7661526203155518, + "learning_rate": 1.8366101694915255e-05, + "loss": 1.4472, + "step": 2910 + }, + { + "epoch": 0.07, + "grad_norm": 6.59897518157959, + "learning_rate": 1.8359322033898305e-05, + "loss": 1.4166, + "step": 2920 + }, + { + "epoch": 0.07, + "grad_norm": 1.927669882774353, + "learning_rate": 1.8352542372881358e-05, + "loss": 1.4674, + "step": 2930 + }, + { + "epoch": 0.07, + "grad_norm": 1.0190672874450684, + "learning_rate": 1.8345762711864408e-05, + "loss": 1.4591, + "step": 2940 + }, + { + "epoch": 0.07, + "grad_norm": 3.63718843460083, + "learning_rate": 1.8338983050847458e-05, + "loss": 1.7469, + "step": 2950 + }, + { + "epoch": 0.07, + "grad_norm": 1.5168763399124146, + "learning_rate": 1.833220338983051e-05, + "loss": 1.5691, + "step": 2960 + }, + { + "epoch": 0.07, + "grad_norm": 1.323123574256897, + "learning_rate": 1.832542372881356e-05, + "loss": 1.385, + "step": 2970 + }, + { + "epoch": 0.07, + "grad_norm": 3.590506076812744, + "learning_rate": 1.831864406779661e-05, + "loss": 1.429, + "step": 2980 + }, + { + "epoch": 0.07, + "grad_norm": 4.562034606933594, + "learning_rate": 1.8311864406779663e-05, + "loss": 1.4169, + "step": 2990 + }, + { + "epoch": 0.07, + "grad_norm": 3.3229966163635254, + "learning_rate": 1.8305084745762713e-05, + "loss": 1.5123, + "step": 3000 + }, + { + "epoch": 0.07, + "eval_loss": 1.5320124626159668, + "eval_runtime": 66.1653, + "eval_samples_per_second": 15.114, + "eval_steps_per_second": 15.114, + "step": 3000 + }, + { + "epoch": 0.08, + "grad_norm": 3.8690502643585205, + "learning_rate": 1.8298305084745766e-05, + "loss": 1.2614, + "step": 3010 + }, + { + "epoch": 0.08, + "grad_norm": 2.8421173095703125, + "learning_rate": 1.8291525423728816e-05, + "loss": 1.3572, + "step": 3020 + }, + { + "epoch": 0.08, + "grad_norm": 3.3784098625183105, + "learning_rate": 1.8284745762711865e-05, + "loss": 1.4465, + "step": 3030 + }, + { + "epoch": 0.08, + "grad_norm": 3.611422061920166, + "learning_rate": 1.827796610169492e-05, + "loss": 1.4531, + "step": 3040 + }, + { + "epoch": 0.08, + "grad_norm": 2.3969950675964355, + "learning_rate": 1.8271186440677968e-05, + "loss": 1.3137, + "step": 3050 + }, + { + "epoch": 0.08, + "grad_norm": 4.92153263092041, + "learning_rate": 1.8264406779661018e-05, + "loss": 1.6209, + "step": 3060 + }, + { + "epoch": 0.08, + "grad_norm": 3.952179431915283, + "learning_rate": 1.825762711864407e-05, + "loss": 1.4923, + "step": 3070 + }, + { + "epoch": 0.08, + "grad_norm": 4.00267219543457, + "learning_rate": 1.825084745762712e-05, + "loss": 1.5688, + "step": 3080 + }, + { + "epoch": 0.08, + "grad_norm": 2.1664321422576904, + "learning_rate": 1.824406779661017e-05, + "loss": 1.4545, + "step": 3090 + }, + { + "epoch": 0.08, + "grad_norm": 3.1541237831115723, + "learning_rate": 1.823728813559322e-05, + "loss": 1.6623, + "step": 3100 + }, + { + "epoch": 0.08, + "grad_norm": 2.577681064605713, + "learning_rate": 1.8230508474576273e-05, + "loss": 1.6301, + "step": 3110 + }, + { + "epoch": 0.08, + "grad_norm": 2.2123563289642334, + "learning_rate": 1.8223728813559323e-05, + "loss": 1.5614, + "step": 3120 + }, + { + "epoch": 0.08, + "grad_norm": 1.7630640268325806, + "learning_rate": 1.8216949152542373e-05, + "loss": 1.5482, + "step": 3130 + }, + { + "epoch": 0.08, + "grad_norm": 1.5110881328582764, + "learning_rate": 1.8210169491525426e-05, + "loss": 1.4324, + "step": 3140 + }, + { + "epoch": 0.08, + "grad_norm": 2.9992825984954834, + "learning_rate": 1.8203389830508475e-05, + "loss": 1.5376, + "step": 3150 + }, + { + "epoch": 0.08, + "grad_norm": 2.825230836868286, + "learning_rate": 1.8196610169491525e-05, + "loss": 1.3925, + "step": 3160 + }, + { + "epoch": 0.08, + "grad_norm": 5.502832412719727, + "learning_rate": 1.8189830508474578e-05, + "loss": 1.5082, + "step": 3170 + }, + { + "epoch": 0.08, + "grad_norm": 1.258467197418213, + "learning_rate": 1.8183050847457628e-05, + "loss": 1.512, + "step": 3180 + }, + { + "epoch": 0.08, + "grad_norm": 6.260064601898193, + "learning_rate": 1.817627118644068e-05, + "loss": 1.5949, + "step": 3190 + }, + { + "epoch": 0.08, + "grad_norm": 1.3906792402267456, + "learning_rate": 1.816949152542373e-05, + "loss": 1.4887, + "step": 3200 + }, + { + "epoch": 0.08, + "grad_norm": 6.833672523498535, + "learning_rate": 1.816271186440678e-05, + "loss": 1.6286, + "step": 3210 + }, + { + "epoch": 0.08, + "grad_norm": 7.004729747772217, + "learning_rate": 1.8155932203389833e-05, + "loss": 1.6974, + "step": 3220 + }, + { + "epoch": 0.08, + "grad_norm": 3.688082695007324, + "learning_rate": 1.8149152542372883e-05, + "loss": 1.577, + "step": 3230 + }, + { + "epoch": 0.08, + "grad_norm": 1.2561973333358765, + "learning_rate": 1.8142372881355933e-05, + "loss": 1.5154, + "step": 3240 + }, + { + "epoch": 0.08, + "grad_norm": 2.1860358715057373, + "learning_rate": 1.8135593220338986e-05, + "loss": 1.5281, + "step": 3250 + }, + { + "epoch": 0.08, + "grad_norm": 1.4622399806976318, + "learning_rate": 1.8128813559322036e-05, + "loss": 1.4103, + "step": 3260 + }, + { + "epoch": 0.08, + "grad_norm": 3.036226511001587, + "learning_rate": 1.8122033898305085e-05, + "loss": 1.4003, + "step": 3270 + }, + { + "epoch": 0.08, + "grad_norm": 4.228930473327637, + "learning_rate": 1.811525423728814e-05, + "loss": 1.6889, + "step": 3280 + }, + { + "epoch": 0.08, + "grad_norm": 4.7660322189331055, + "learning_rate": 1.8108474576271188e-05, + "loss": 1.5185, + "step": 3290 + }, + { + "epoch": 0.08, + "grad_norm": 2.857886552810669, + "learning_rate": 1.810169491525424e-05, + "loss": 1.4166, + "step": 3300 + }, + { + "epoch": 0.08, + "grad_norm": 2.357570171356201, + "learning_rate": 1.8094915254237288e-05, + "loss": 1.6341, + "step": 3310 + }, + { + "epoch": 0.08, + "grad_norm": 2.1648855209350586, + "learning_rate": 1.808813559322034e-05, + "loss": 1.6001, + "step": 3320 + }, + { + "epoch": 0.08, + "grad_norm": 2.5360536575317383, + "learning_rate": 1.808135593220339e-05, + "loss": 1.5059, + "step": 3330 + }, + { + "epoch": 0.08, + "grad_norm": 3.741981267929077, + "learning_rate": 1.807457627118644e-05, + "loss": 1.5082, + "step": 3340 + }, + { + "epoch": 0.08, + "grad_norm": 3.4550766944885254, + "learning_rate": 1.8067796610169493e-05, + "loss": 1.6378, + "step": 3350 + }, + { + "epoch": 0.08, + "grad_norm": 1.9454740285873413, + "learning_rate": 1.8061016949152543e-05, + "loss": 1.6602, + "step": 3360 + }, + { + "epoch": 0.08, + "grad_norm": 3.708360195159912, + "learning_rate": 1.8054237288135593e-05, + "loss": 1.4584, + "step": 3370 + }, + { + "epoch": 0.08, + "grad_norm": 2.4115514755249023, + "learning_rate": 1.8047457627118646e-05, + "loss": 1.4053, + "step": 3380 + }, + { + "epoch": 0.08, + "grad_norm": 2.075867176055908, + "learning_rate": 1.8040677966101695e-05, + "loss": 1.46, + "step": 3390 + }, + { + "epoch": 0.09, + "grad_norm": 3.1672651767730713, + "learning_rate": 1.803389830508475e-05, + "loss": 1.4964, + "step": 3400 + }, + { + "epoch": 0.09, + "grad_norm": 4.5812578201293945, + "learning_rate": 1.8027118644067798e-05, + "loss": 1.4005, + "step": 3410 + }, + { + "epoch": 0.09, + "grad_norm": 2.583771228790283, + "learning_rate": 1.8020338983050848e-05, + "loss": 1.6024, + "step": 3420 + }, + { + "epoch": 0.09, + "grad_norm": 4.464291095733643, + "learning_rate": 1.80135593220339e-05, + "loss": 1.5075, + "step": 3430 + }, + { + "epoch": 0.09, + "grad_norm": 2.609657049179077, + "learning_rate": 1.800677966101695e-05, + "loss": 1.4527, + "step": 3440 + }, + { + "epoch": 0.09, + "grad_norm": 1.844227910041809, + "learning_rate": 1.8e-05, + "loss": 1.471, + "step": 3450 + }, + { + "epoch": 0.09, + "grad_norm": 3.5781702995300293, + "learning_rate": 1.7993220338983054e-05, + "loss": 1.5221, + "step": 3460 + }, + { + "epoch": 0.09, + "grad_norm": 1.8293837308883667, + "learning_rate": 1.7986440677966103e-05, + "loss": 1.4831, + "step": 3470 + }, + { + "epoch": 0.09, + "grad_norm": 5.273003101348877, + "learning_rate": 1.7979661016949156e-05, + "loss": 1.5131, + "step": 3480 + }, + { + "epoch": 0.09, + "grad_norm": 3.1019296646118164, + "learning_rate": 1.7972881355932206e-05, + "loss": 1.4346, + "step": 3490 + }, + { + "epoch": 0.09, + "grad_norm": 3.419081687927246, + "learning_rate": 1.7966101694915256e-05, + "loss": 1.496, + "step": 3500 + }, + { + "epoch": 0.09, + "eval_loss": 1.5301283597946167, + "eval_runtime": 66.1385, + "eval_samples_per_second": 15.12, + "eval_steps_per_second": 15.12, + "step": 3500 + }, + { + "epoch": 0.09, + "grad_norm": 4.49799919128418, + "learning_rate": 1.795932203389831e-05, + "loss": 1.5304, + "step": 3510 + }, + { + "epoch": 0.09, + "grad_norm": 3.267191171646118, + "learning_rate": 1.7952542372881355e-05, + "loss": 1.5103, + "step": 3520 + }, + { + "epoch": 0.09, + "grad_norm": 3.4655425548553467, + "learning_rate": 1.7945762711864408e-05, + "loss": 1.3131, + "step": 3530 + }, + { + "epoch": 0.09, + "grad_norm": 2.289358615875244, + "learning_rate": 1.7938983050847458e-05, + "loss": 1.2731, + "step": 3540 + }, + { + "epoch": 0.09, + "grad_norm": 1.528286337852478, + "learning_rate": 1.7932203389830508e-05, + "loss": 1.6249, + "step": 3550 + }, + { + "epoch": 0.09, + "grad_norm": 1.5927760601043701, + "learning_rate": 1.792542372881356e-05, + "loss": 1.5177, + "step": 3560 + }, + { + "epoch": 0.09, + "grad_norm": 1.5520817041397095, + "learning_rate": 1.791864406779661e-05, + "loss": 1.4523, + "step": 3570 + }, + { + "epoch": 0.09, + "grad_norm": 3.531754970550537, + "learning_rate": 1.7911864406779664e-05, + "loss": 1.5342, + "step": 3580 + }, + { + "epoch": 0.09, + "grad_norm": 2.6139304637908936, + "learning_rate": 1.7905084745762713e-05, + "loss": 1.3377, + "step": 3590 + }, + { + "epoch": 0.09, + "grad_norm": 5.242190837860107, + "learning_rate": 1.7898305084745763e-05, + "loss": 1.592, + "step": 3600 + }, + { + "epoch": 0.09, + "grad_norm": 3.10337495803833, + "learning_rate": 1.7891525423728816e-05, + "loss": 1.4516, + "step": 3610 + }, + { + "epoch": 0.09, + "grad_norm": 2.615224599838257, + "learning_rate": 1.7884745762711866e-05, + "loss": 1.4325, + "step": 3620 + }, + { + "epoch": 0.09, + "grad_norm": 5.353131294250488, + "learning_rate": 1.7877966101694916e-05, + "loss": 1.5772, + "step": 3630 + }, + { + "epoch": 0.09, + "grad_norm": 1.1678059101104736, + "learning_rate": 1.787118644067797e-05, + "loss": 1.572, + "step": 3640 + }, + { + "epoch": 0.09, + "grad_norm": 3.2298007011413574, + "learning_rate": 1.7864406779661018e-05, + "loss": 1.5214, + "step": 3650 + }, + { + "epoch": 0.09, + "grad_norm": 3.405092477798462, + "learning_rate": 1.785762711864407e-05, + "loss": 1.4677, + "step": 3660 + }, + { + "epoch": 0.09, + "grad_norm": 3.6669983863830566, + "learning_rate": 1.785084745762712e-05, + "loss": 1.2853, + "step": 3670 + }, + { + "epoch": 0.09, + "grad_norm": 3.2381701469421387, + "learning_rate": 1.784406779661017e-05, + "loss": 1.3501, + "step": 3680 + }, + { + "epoch": 0.09, + "grad_norm": 2.3621437549591064, + "learning_rate": 1.7837288135593224e-05, + "loss": 1.5105, + "step": 3690 + }, + { + "epoch": 0.09, + "grad_norm": 1.1474658250808716, + "learning_rate": 1.7830508474576274e-05, + "loss": 1.5107, + "step": 3700 + }, + { + "epoch": 0.09, + "grad_norm": 9.769930839538574, + "learning_rate": 1.7823728813559323e-05, + "loss": 1.8114, + "step": 3710 + }, + { + "epoch": 0.09, + "grad_norm": 1.6212892532348633, + "learning_rate": 1.7816949152542376e-05, + "loss": 1.4948, + "step": 3720 + }, + { + "epoch": 0.09, + "grad_norm": 3.909820795059204, + "learning_rate": 1.7810169491525423e-05, + "loss": 1.4145, + "step": 3730 + }, + { + "epoch": 0.09, + "grad_norm": 1.870615005493164, + "learning_rate": 1.7803389830508476e-05, + "loss": 1.4856, + "step": 3740 + }, + { + "epoch": 0.09, + "grad_norm": 2.5912163257598877, + "learning_rate": 1.7796610169491526e-05, + "loss": 1.5021, + "step": 3750 + }, + { + "epoch": 0.09, + "grad_norm": 2.303807020187378, + "learning_rate": 1.778983050847458e-05, + "loss": 1.4348, + "step": 3760 + }, + { + "epoch": 0.09, + "grad_norm": 2.5953946113586426, + "learning_rate": 1.778305084745763e-05, + "loss": 1.4073, + "step": 3770 + }, + { + "epoch": 0.09, + "grad_norm": 2.129793643951416, + "learning_rate": 1.7776271186440678e-05, + "loss": 1.4827, + "step": 3780 + }, + { + "epoch": 0.09, + "grad_norm": 8.186966896057129, + "learning_rate": 1.776949152542373e-05, + "loss": 1.5887, + "step": 3790 + }, + { + "epoch": 0.1, + "grad_norm": 2.2956864833831787, + "learning_rate": 1.776271186440678e-05, + "loss": 1.4925, + "step": 3800 + }, + { + "epoch": 0.1, + "grad_norm": 1.3646732568740845, + "learning_rate": 1.775593220338983e-05, + "loss": 1.5814, + "step": 3810 + }, + { + "epoch": 0.1, + "grad_norm": 5.3318071365356445, + "learning_rate": 1.7749152542372884e-05, + "loss": 1.6139, + "step": 3820 + }, + { + "epoch": 0.1, + "grad_norm": 2.53743314743042, + "learning_rate": 1.7742372881355933e-05, + "loss": 1.5198, + "step": 3830 + }, + { + "epoch": 0.1, + "grad_norm": 7.048913478851318, + "learning_rate": 1.7735593220338983e-05, + "loss": 1.4417, + "step": 3840 + }, + { + "epoch": 0.1, + "grad_norm": 2.367314100265503, + "learning_rate": 1.7728813559322036e-05, + "loss": 1.4633, + "step": 3850 + }, + { + "epoch": 0.1, + "grad_norm": 1.619931697845459, + "learning_rate": 1.7722033898305086e-05, + "loss": 1.3719, + "step": 3860 + }, + { + "epoch": 0.1, + "grad_norm": 3.264190435409546, + "learning_rate": 1.771525423728814e-05, + "loss": 1.3079, + "step": 3870 + }, + { + "epoch": 0.1, + "grad_norm": 4.68599796295166, + "learning_rate": 1.770847457627119e-05, + "loss": 1.6278, + "step": 3880 + }, + { + "epoch": 0.1, + "grad_norm": 4.21970272064209, + "learning_rate": 1.770169491525424e-05, + "loss": 1.4814, + "step": 3890 + }, + { + "epoch": 0.1, + "grad_norm": 2.3647706508636475, + "learning_rate": 1.769491525423729e-05, + "loss": 1.5973, + "step": 3900 + }, + { + "epoch": 0.1, + "grad_norm": 3.2886738777160645, + "learning_rate": 1.768813559322034e-05, + "loss": 1.5224, + "step": 3910 + }, + { + "epoch": 0.1, + "grad_norm": 2.6817705631256104, + "learning_rate": 1.768135593220339e-05, + "loss": 1.566, + "step": 3920 + }, + { + "epoch": 0.1, + "grad_norm": 1.6243913173675537, + "learning_rate": 1.767457627118644e-05, + "loss": 1.6011, + "step": 3930 + }, + { + "epoch": 0.1, + "grad_norm": 1.8161165714263916, + "learning_rate": 1.7667796610169494e-05, + "loss": 1.5303, + "step": 3940 + }, + { + "epoch": 0.1, + "grad_norm": 2.486697196960449, + "learning_rate": 1.7661016949152543e-05, + "loss": 1.4032, + "step": 3950 + }, + { + "epoch": 0.1, + "grad_norm": 1.8153432607650757, + "learning_rate": 1.7654237288135593e-05, + "loss": 1.5167, + "step": 3960 + }, + { + "epoch": 0.1, + "grad_norm": 2.186868190765381, + "learning_rate": 1.7647457627118646e-05, + "loss": 1.6066, + "step": 3970 + }, + { + "epoch": 0.1, + "grad_norm": 2.1040332317352295, + "learning_rate": 1.7640677966101696e-05, + "loss": 1.4142, + "step": 3980 + }, + { + "epoch": 0.1, + "grad_norm": 3.989042282104492, + "learning_rate": 1.7633898305084746e-05, + "loss": 1.4877, + "step": 3990 + }, + { + "epoch": 0.1, + "grad_norm": 3.7532989978790283, + "learning_rate": 1.76271186440678e-05, + "loss": 1.4016, + "step": 4000 + }, + { + "epoch": 0.1, + "eval_loss": 1.5089428424835205, + "eval_runtime": 66.2329, + "eval_samples_per_second": 15.098, + "eval_steps_per_second": 15.098, + "step": 4000 + }, + { + "epoch": 0.1, + "grad_norm": 6.605846405029297, + "learning_rate": 1.762033898305085e-05, + "loss": 1.6163, + "step": 4010 + }, + { + "epoch": 0.1, + "grad_norm": 3.1976850032806396, + "learning_rate": 1.7613559322033898e-05, + "loss": 1.4298, + "step": 4020 + }, + { + "epoch": 0.1, + "grad_norm": 5.412558555603027, + "learning_rate": 1.760677966101695e-05, + "loss": 1.3316, + "step": 4030 + }, + { + "epoch": 0.1, + "grad_norm": 5.241945266723633, + "learning_rate": 1.76e-05, + "loss": 1.6258, + "step": 4040 + }, + { + "epoch": 0.1, + "grad_norm": 11.647382736206055, + "learning_rate": 1.7593220338983054e-05, + "loss": 1.5479, + "step": 4050 + }, + { + "epoch": 0.1, + "grad_norm": 2.091214656829834, + "learning_rate": 1.7586440677966104e-05, + "loss": 1.4385, + "step": 4060 + }, + { + "epoch": 0.1, + "grad_norm": 2.492511034011841, + "learning_rate": 1.7579661016949153e-05, + "loss": 1.6785, + "step": 4070 + }, + { + "epoch": 0.1, + "grad_norm": 1.9180244207382202, + "learning_rate": 1.7572881355932206e-05, + "loss": 1.6234, + "step": 4080 + }, + { + "epoch": 0.1, + "grad_norm": 2.256627082824707, + "learning_rate": 1.7566101694915256e-05, + "loss": 1.3143, + "step": 4090 + }, + { + "epoch": 0.1, + "grad_norm": 2.2864432334899902, + "learning_rate": 1.7559322033898306e-05, + "loss": 1.4961, + "step": 4100 + }, + { + "epoch": 0.1, + "grad_norm": 6.091422080993652, + "learning_rate": 1.755254237288136e-05, + "loss": 1.27, + "step": 4110 + }, + { + "epoch": 0.1, + "grad_norm": 1.6744223833084106, + "learning_rate": 1.754576271186441e-05, + "loss": 1.6859, + "step": 4120 + }, + { + "epoch": 0.1, + "grad_norm": 3.741482973098755, + "learning_rate": 1.753898305084746e-05, + "loss": 1.5382, + "step": 4130 + }, + { + "epoch": 0.1, + "grad_norm": 3.2129416465759277, + "learning_rate": 1.7532203389830508e-05, + "loss": 1.548, + "step": 4140 + }, + { + "epoch": 0.1, + "grad_norm": 1.9851702451705933, + "learning_rate": 1.752542372881356e-05, + "loss": 1.5777, + "step": 4150 + }, + { + "epoch": 0.1, + "grad_norm": 2.7135250568389893, + "learning_rate": 1.751864406779661e-05, + "loss": 1.6374, + "step": 4160 + }, + { + "epoch": 0.1, + "grad_norm": 6.881992340087891, + "learning_rate": 1.751186440677966e-05, + "loss": 1.5171, + "step": 4170 + }, + { + "epoch": 0.1, + "grad_norm": 3.8673460483551025, + "learning_rate": 1.7505084745762714e-05, + "loss": 1.3488, + "step": 4180 + }, + { + "epoch": 0.1, + "grad_norm": 3.7248244285583496, + "learning_rate": 1.7498305084745763e-05, + "loss": 1.4269, + "step": 4190 + }, + { + "epoch": 0.1, + "grad_norm": 4.191452503204346, + "learning_rate": 1.7491525423728813e-05, + "loss": 1.591, + "step": 4200 + }, + { + "epoch": 0.11, + "grad_norm": 1.9477959871292114, + "learning_rate": 1.7484745762711866e-05, + "loss": 1.5194, + "step": 4210 + }, + { + "epoch": 0.11, + "grad_norm": 4.151371002197266, + "learning_rate": 1.7477966101694916e-05, + "loss": 1.4644, + "step": 4220 + }, + { + "epoch": 0.11, + "grad_norm": 3.285062551498413, + "learning_rate": 1.747118644067797e-05, + "loss": 1.3697, + "step": 4230 + }, + { + "epoch": 0.11, + "grad_norm": 5.2172136306762695, + "learning_rate": 1.746440677966102e-05, + "loss": 1.3542, + "step": 4240 + }, + { + "epoch": 0.11, + "grad_norm": 2.321295738220215, + "learning_rate": 1.745762711864407e-05, + "loss": 1.6265, + "step": 4250 + }, + { + "epoch": 0.11, + "grad_norm": 3.2060482501983643, + "learning_rate": 1.745084745762712e-05, + "loss": 1.5214, + "step": 4260 + }, + { + "epoch": 0.11, + "grad_norm": 2.697052240371704, + "learning_rate": 1.744406779661017e-05, + "loss": 1.5446, + "step": 4270 + }, + { + "epoch": 0.11, + "grad_norm": 2.639937162399292, + "learning_rate": 1.743728813559322e-05, + "loss": 1.5329, + "step": 4280 + }, + { + "epoch": 0.11, + "grad_norm": 3.5502707958221436, + "learning_rate": 1.7430508474576274e-05, + "loss": 1.4997, + "step": 4290 + }, + { + "epoch": 0.11, + "grad_norm": 5.514580726623535, + "learning_rate": 1.7423728813559324e-05, + "loss": 1.4641, + "step": 4300 + }, + { + "epoch": 0.11, + "grad_norm": 2.264512062072754, + "learning_rate": 1.7416949152542373e-05, + "loss": 1.5588, + "step": 4310 + }, + { + "epoch": 0.11, + "grad_norm": 4.865510940551758, + "learning_rate": 1.7410169491525427e-05, + "loss": 1.5777, + "step": 4320 + }, + { + "epoch": 0.11, + "grad_norm": 2.872544050216675, + "learning_rate": 1.7403389830508476e-05, + "loss": 1.4683, + "step": 4330 + }, + { + "epoch": 0.11, + "grad_norm": 2.1891725063323975, + "learning_rate": 1.739661016949153e-05, + "loss": 1.4572, + "step": 4340 + }, + { + "epoch": 0.11, + "grad_norm": 5.604435443878174, + "learning_rate": 1.7389830508474576e-05, + "loss": 1.3685, + "step": 4350 + }, + { + "epoch": 0.11, + "grad_norm": 1.9309135675430298, + "learning_rate": 1.738305084745763e-05, + "loss": 1.6086, + "step": 4360 + }, + { + "epoch": 0.11, + "grad_norm": 1.2651278972625732, + "learning_rate": 1.737627118644068e-05, + "loss": 1.5768, + "step": 4370 + }, + { + "epoch": 0.11, + "grad_norm": 3.4745676517486572, + "learning_rate": 1.7369491525423728e-05, + "loss": 1.4553, + "step": 4380 + }, + { + "epoch": 0.11, + "grad_norm": 5.904428958892822, + "learning_rate": 1.736271186440678e-05, + "loss": 1.654, + "step": 4390 + }, + { + "epoch": 0.11, + "grad_norm": 1.6250745058059692, + "learning_rate": 1.735593220338983e-05, + "loss": 1.5443, + "step": 4400 + }, + { + "epoch": 0.11, + "grad_norm": 3.289626121520996, + "learning_rate": 1.734915254237288e-05, + "loss": 1.4691, + "step": 4410 + }, + { + "epoch": 0.11, + "grad_norm": 5.800081729888916, + "learning_rate": 1.7342372881355934e-05, + "loss": 1.543, + "step": 4420 + }, + { + "epoch": 0.11, + "grad_norm": 9.145909309387207, + "learning_rate": 1.7335593220338983e-05, + "loss": 1.4814, + "step": 4430 + }, + { + "epoch": 0.11, + "grad_norm": 2.22583270072937, + "learning_rate": 1.7328813559322037e-05, + "loss": 1.548, + "step": 4440 + }, + { + "epoch": 0.11, + "grad_norm": 4.693650722503662, + "learning_rate": 1.7322033898305086e-05, + "loss": 1.5602, + "step": 4450 + }, + { + "epoch": 0.11, + "grad_norm": 2.1424403190612793, + "learning_rate": 1.7315254237288136e-05, + "loss": 1.5298, + "step": 4460 + }, + { + "epoch": 0.11, + "grad_norm": 2.8103983402252197, + "learning_rate": 1.730847457627119e-05, + "loss": 1.435, + "step": 4470 + }, + { + "epoch": 0.11, + "grad_norm": 2.2972493171691895, + "learning_rate": 1.730169491525424e-05, + "loss": 1.4906, + "step": 4480 + }, + { + "epoch": 0.11, + "grad_norm": 2.437459707260132, + "learning_rate": 1.729491525423729e-05, + "loss": 1.5637, + "step": 4490 + }, + { + "epoch": 0.11, + "grad_norm": 3.561988115310669, + "learning_rate": 1.728813559322034e-05, + "loss": 1.572, + "step": 4500 + }, + { + "epoch": 0.11, + "eval_loss": 1.4887409210205078, + "eval_runtime": 66.1778, + "eval_samples_per_second": 15.111, + "eval_steps_per_second": 15.111, + "step": 4500 + }, + { + "epoch": 0.11, + "grad_norm": 0.4782548248767853, + "learning_rate": 1.728135593220339e-05, + "loss": 1.606, + "step": 4510 + }, + { + "epoch": 0.11, + "grad_norm": 4.1626386642456055, + "learning_rate": 1.7274576271186444e-05, + "loss": 1.6311, + "step": 4520 + }, + { + "epoch": 0.11, + "grad_norm": 3.057555675506592, + "learning_rate": 1.7267796610169494e-05, + "loss": 1.495, + "step": 4530 + }, + { + "epoch": 0.11, + "grad_norm": 4.714250087738037, + "learning_rate": 1.7261016949152544e-05, + "loss": 1.5219, + "step": 4540 + }, + { + "epoch": 0.11, + "grad_norm": 1.1815061569213867, + "learning_rate": 1.7254237288135597e-05, + "loss": 1.5227, + "step": 4550 + }, + { + "epoch": 0.11, + "grad_norm": 3.602353811264038, + "learning_rate": 1.7247457627118643e-05, + "loss": 1.6756, + "step": 4560 + }, + { + "epoch": 0.11, + "grad_norm": 1.7794777154922485, + "learning_rate": 1.7240677966101696e-05, + "loss": 1.3716, + "step": 4570 + }, + { + "epoch": 0.11, + "grad_norm": 2.1229944229125977, + "learning_rate": 1.7233898305084746e-05, + "loss": 1.4612, + "step": 4580 + }, + { + "epoch": 0.11, + "grad_norm": 4.656471252441406, + "learning_rate": 1.7227118644067796e-05, + "loss": 1.6358, + "step": 4590 + }, + { + "epoch": 0.12, + "grad_norm": 2.0882911682128906, + "learning_rate": 1.722033898305085e-05, + "loss": 1.5147, + "step": 4600 + }, + { + "epoch": 0.12, + "grad_norm": 3.070713996887207, + "learning_rate": 1.72135593220339e-05, + "loss": 1.5253, + "step": 4610 + }, + { + "epoch": 0.12, + "grad_norm": 1.7038753032684326, + "learning_rate": 1.720677966101695e-05, + "loss": 1.4375, + "step": 4620 + }, + { + "epoch": 0.12, + "grad_norm": 2.5298070907592773, + "learning_rate": 1.72e-05, + "loss": 1.5301, + "step": 4630 + }, + { + "epoch": 0.12, + "grad_norm": 2.2928740978240967, + "learning_rate": 1.719322033898305e-05, + "loss": 1.2571, + "step": 4640 + }, + { + "epoch": 0.12, + "grad_norm": 5.660508155822754, + "learning_rate": 1.7186440677966104e-05, + "loss": 1.5168, + "step": 4650 + }, + { + "epoch": 0.12, + "grad_norm": 5.104097366333008, + "learning_rate": 1.7179661016949154e-05, + "loss": 1.4134, + "step": 4660 + }, + { + "epoch": 0.12, + "grad_norm": 4.954836368560791, + "learning_rate": 1.7172881355932204e-05, + "loss": 1.4909, + "step": 4670 + }, + { + "epoch": 0.12, + "grad_norm": 4.392303466796875, + "learning_rate": 1.7166101694915257e-05, + "loss": 1.5639, + "step": 4680 + }, + { + "epoch": 0.12, + "grad_norm": 3.9031317234039307, + "learning_rate": 1.7159322033898306e-05, + "loss": 1.4521, + "step": 4690 + }, + { + "epoch": 0.12, + "grad_norm": 4.280984401702881, + "learning_rate": 1.715254237288136e-05, + "loss": 1.3877, + "step": 4700 + }, + { + "epoch": 0.12, + "grad_norm": 1.2300760746002197, + "learning_rate": 1.714576271186441e-05, + "loss": 1.5971, + "step": 4710 + }, + { + "epoch": 0.12, + "grad_norm": 3.845804452896118, + "learning_rate": 1.713898305084746e-05, + "loss": 1.5987, + "step": 4720 + }, + { + "epoch": 0.12, + "grad_norm": 5.861780643463135, + "learning_rate": 1.7132203389830512e-05, + "loss": 1.7391, + "step": 4730 + }, + { + "epoch": 0.12, + "grad_norm": 4.704226970672607, + "learning_rate": 1.712542372881356e-05, + "loss": 1.3629, + "step": 4740 + }, + { + "epoch": 0.12, + "grad_norm": 1.362378716468811, + "learning_rate": 1.711864406779661e-05, + "loss": 1.5607, + "step": 4750 + }, + { + "epoch": 0.12, + "grad_norm": 3.89095139503479, + "learning_rate": 1.711186440677966e-05, + "loss": 1.6382, + "step": 4760 + }, + { + "epoch": 0.12, + "grad_norm": 2.8033549785614014, + "learning_rate": 1.710508474576271e-05, + "loss": 1.6113, + "step": 4770 + }, + { + "epoch": 0.12, + "grad_norm": 3.045806884765625, + "learning_rate": 1.7098305084745764e-05, + "loss": 1.4223, + "step": 4780 + }, + { + "epoch": 0.12, + "grad_norm": 7.401500701904297, + "learning_rate": 1.7091525423728814e-05, + "loss": 1.4659, + "step": 4790 + }, + { + "epoch": 0.12, + "grad_norm": 1.490501880645752, + "learning_rate": 1.7084745762711867e-05, + "loss": 1.4543, + "step": 4800 + }, + { + "epoch": 0.12, + "grad_norm": 2.137923240661621, + "learning_rate": 1.7077966101694916e-05, + "loss": 1.4117, + "step": 4810 + }, + { + "epoch": 0.12, + "grad_norm": 3.254201889038086, + "learning_rate": 1.7071186440677966e-05, + "loss": 1.4404, + "step": 4820 + }, + { + "epoch": 0.12, + "grad_norm": 2.1423888206481934, + "learning_rate": 1.706440677966102e-05, + "loss": 1.5539, + "step": 4830 + }, + { + "epoch": 0.12, + "grad_norm": 1.927777647972107, + "learning_rate": 1.705762711864407e-05, + "loss": 1.5285, + "step": 4840 + }, + { + "epoch": 0.12, + "grad_norm": 3.6136608123779297, + "learning_rate": 1.705084745762712e-05, + "loss": 1.3503, + "step": 4850 + }, + { + "epoch": 0.12, + "grad_norm": 4.160021781921387, + "learning_rate": 1.704406779661017e-05, + "loss": 1.5243, + "step": 4860 + }, + { + "epoch": 0.12, + "grad_norm": 3.0722756385803223, + "learning_rate": 1.703728813559322e-05, + "loss": 1.5306, + "step": 4870 + }, + { + "epoch": 0.12, + "grad_norm": 6.304449558258057, + "learning_rate": 1.703050847457627e-05, + "loss": 1.5878, + "step": 4880 + }, + { + "epoch": 0.12, + "grad_norm": 1.935084581375122, + "learning_rate": 1.7023728813559324e-05, + "loss": 1.5446, + "step": 4890 + }, + { + "epoch": 0.12, + "grad_norm": 2.6202478408813477, + "learning_rate": 1.7016949152542374e-05, + "loss": 1.6461, + "step": 4900 + }, + { + "epoch": 0.12, + "grad_norm": 3.2252795696258545, + "learning_rate": 1.7010169491525427e-05, + "loss": 1.445, + "step": 4910 + }, + { + "epoch": 0.12, + "grad_norm": 4.115199565887451, + "learning_rate": 1.7003389830508477e-05, + "loss": 1.5513, + "step": 4920 + }, + { + "epoch": 0.12, + "grad_norm": 2.852402448654175, + "learning_rate": 1.6996610169491526e-05, + "loss": 1.496, + "step": 4930 + }, + { + "epoch": 0.12, + "grad_norm": 2.5148980617523193, + "learning_rate": 1.698983050847458e-05, + "loss": 1.4281, + "step": 4940 + }, + { + "epoch": 0.12, + "grad_norm": 1.3501824140548706, + "learning_rate": 1.698305084745763e-05, + "loss": 1.4301, + "step": 4950 + }, + { + "epoch": 0.12, + "grad_norm": 3.4159467220306396, + "learning_rate": 1.697627118644068e-05, + "loss": 1.5491, + "step": 4960 + }, + { + "epoch": 0.12, + "grad_norm": 4.547379970550537, + "learning_rate": 1.696949152542373e-05, + "loss": 1.5137, + "step": 4970 + }, + { + "epoch": 0.12, + "grad_norm": 3.4466910362243652, + "learning_rate": 1.6962711864406782e-05, + "loss": 1.4767, + "step": 4980 + }, + { + "epoch": 0.12, + "grad_norm": 6.4336395263671875, + "learning_rate": 1.695593220338983e-05, + "loss": 1.5213, + "step": 4990 + }, + { + "epoch": 0.12, + "grad_norm": 1.0396947860717773, + "learning_rate": 1.694915254237288e-05, + "loss": 1.4402, + "step": 5000 + }, + { + "epoch": 0.12, + "eval_loss": 1.4466354846954346, + "eval_runtime": 66.2264, + "eval_samples_per_second": 15.1, + "eval_steps_per_second": 15.1, + "step": 5000 + }, + { + "epoch": 0.13, + "grad_norm": 3.8836324214935303, + "learning_rate": 1.6942372881355934e-05, + "loss": 1.6772, + "step": 5010 + }, + { + "epoch": 0.13, + "grad_norm": 4.293881893157959, + "learning_rate": 1.6935593220338984e-05, + "loss": 1.4618, + "step": 5020 + }, + { + "epoch": 0.13, + "grad_norm": 3.111931800842285, + "learning_rate": 1.6928813559322034e-05, + "loss": 1.5564, + "step": 5030 + }, + { + "epoch": 0.13, + "grad_norm": 3.7768237590789795, + "learning_rate": 1.6922033898305087e-05, + "loss": 1.5239, + "step": 5040 + }, + { + "epoch": 0.13, + "grad_norm": 1.8081879615783691, + "learning_rate": 1.6915254237288136e-05, + "loss": 1.5257, + "step": 5050 + }, + { + "epoch": 0.13, + "grad_norm": 3.2274467945098877, + "learning_rate": 1.6908474576271186e-05, + "loss": 1.3785, + "step": 5060 + }, + { + "epoch": 0.13, + "grad_norm": 1.4177571535110474, + "learning_rate": 1.690169491525424e-05, + "loss": 1.4504, + "step": 5070 + }, + { + "epoch": 0.13, + "grad_norm": 4.527300834655762, + "learning_rate": 1.689491525423729e-05, + "loss": 1.4865, + "step": 5080 + }, + { + "epoch": 0.13, + "grad_norm": 2.1885902881622314, + "learning_rate": 1.6888135593220342e-05, + "loss": 1.5584, + "step": 5090 + }, + { + "epoch": 0.13, + "grad_norm": 2.6522934436798096, + "learning_rate": 1.6881355932203392e-05, + "loss": 1.4736, + "step": 5100 + }, + { + "epoch": 0.13, + "grad_norm": 2.8651719093322754, + "learning_rate": 1.687457627118644e-05, + "loss": 1.6247, + "step": 5110 + }, + { + "epoch": 0.13, + "grad_norm": 3.211879253387451, + "learning_rate": 1.6867796610169495e-05, + "loss": 1.4822, + "step": 5120 + }, + { + "epoch": 0.13, + "grad_norm": 4.725778579711914, + "learning_rate": 1.6861016949152544e-05, + "loss": 1.6364, + "step": 5130 + }, + { + "epoch": 0.13, + "grad_norm": 2.7316319942474365, + "learning_rate": 1.6854237288135594e-05, + "loss": 1.4218, + "step": 5140 + }, + { + "epoch": 0.13, + "grad_norm": 3.2881827354431152, + "learning_rate": 1.6847457627118647e-05, + "loss": 1.4729, + "step": 5150 + }, + { + "epoch": 0.13, + "grad_norm": 4.106601238250732, + "learning_rate": 1.6840677966101697e-05, + "loss": 1.5303, + "step": 5160 + }, + { + "epoch": 0.13, + "grad_norm": 1.9981019496917725, + "learning_rate": 1.683389830508475e-05, + "loss": 1.4583, + "step": 5170 + }, + { + "epoch": 0.13, + "grad_norm": 2.5599071979522705, + "learning_rate": 1.6827118644067796e-05, + "loss": 1.5817, + "step": 5180 + }, + { + "epoch": 0.13, + "grad_norm": 4.673508644104004, + "learning_rate": 1.682033898305085e-05, + "loss": 1.4967, + "step": 5190 + }, + { + "epoch": 0.13, + "grad_norm": 7.4875922203063965, + "learning_rate": 1.68135593220339e-05, + "loss": 1.4482, + "step": 5200 + }, + { + "epoch": 0.13, + "grad_norm": 5.208539962768555, + "learning_rate": 1.680677966101695e-05, + "loss": 1.5031, + "step": 5210 + }, + { + "epoch": 0.13, + "grad_norm": 5.5398664474487305, + "learning_rate": 1.6800000000000002e-05, + "loss": 1.3661, + "step": 5220 + }, + { + "epoch": 0.13, + "grad_norm": 4.9544267654418945, + "learning_rate": 1.679322033898305e-05, + "loss": 1.599, + "step": 5230 + }, + { + "epoch": 0.13, + "grad_norm": 3.121668815612793, + "learning_rate": 1.67864406779661e-05, + "loss": 1.4801, + "step": 5240 + }, + { + "epoch": 0.13, + "grad_norm": 4.368370532989502, + "learning_rate": 1.6779661016949154e-05, + "loss": 1.5737, + "step": 5250 + }, + { + "epoch": 0.13, + "grad_norm": 4.03584098815918, + "learning_rate": 1.6772881355932204e-05, + "loss": 1.4995, + "step": 5260 + }, + { + "epoch": 0.13, + "grad_norm": 5.6201090812683105, + "learning_rate": 1.6766101694915257e-05, + "loss": 1.3567, + "step": 5270 + }, + { + "epoch": 0.13, + "grad_norm": 4.009582996368408, + "learning_rate": 1.6759322033898307e-05, + "loss": 1.4241, + "step": 5280 + }, + { + "epoch": 0.13, + "grad_norm": 3.4552884101867676, + "learning_rate": 1.6752542372881357e-05, + "loss": 1.3472, + "step": 5290 + }, + { + "epoch": 0.13, + "grad_norm": 4.6100029945373535, + "learning_rate": 1.674576271186441e-05, + "loss": 1.3696, + "step": 5300 + }, + { + "epoch": 0.13, + "grad_norm": 6.752505779266357, + "learning_rate": 1.673898305084746e-05, + "loss": 1.5192, + "step": 5310 + }, + { + "epoch": 0.13, + "grad_norm": 1.4629709720611572, + "learning_rate": 1.673220338983051e-05, + "loss": 1.5774, + "step": 5320 + }, + { + "epoch": 0.13, + "grad_norm": 2.904249906539917, + "learning_rate": 1.6725423728813562e-05, + "loss": 1.4843, + "step": 5330 + }, + { + "epoch": 0.13, + "grad_norm": 3.3308334350585938, + "learning_rate": 1.6718644067796612e-05, + "loss": 1.6006, + "step": 5340 + }, + { + "epoch": 0.13, + "grad_norm": 2.405097484588623, + "learning_rate": 1.671186440677966e-05, + "loss": 1.5736, + "step": 5350 + }, + { + "epoch": 0.13, + "grad_norm": 3.4328196048736572, + "learning_rate": 1.6705084745762715e-05, + "loss": 1.4613, + "step": 5360 + }, + { + "epoch": 0.13, + "grad_norm": 1.5162967443466187, + "learning_rate": 1.6698305084745764e-05, + "loss": 1.4562, + "step": 5370 + }, + { + "epoch": 0.13, + "grad_norm": 2.473905563354492, + "learning_rate": 1.6691525423728817e-05, + "loss": 1.4899, + "step": 5380 + }, + { + "epoch": 0.13, + "grad_norm": 1.3600870370864868, + "learning_rate": 1.6684745762711864e-05, + "loss": 1.4197, + "step": 5390 + }, + { + "epoch": 0.14, + "grad_norm": 3.9598913192749023, + "learning_rate": 1.6677966101694917e-05, + "loss": 1.4268, + "step": 5400 + }, + { + "epoch": 0.14, + "grad_norm": 2.6608197689056396, + "learning_rate": 1.6671186440677967e-05, + "loss": 1.2926, + "step": 5410 + }, + { + "epoch": 0.14, + "grad_norm": 2.557682991027832, + "learning_rate": 1.6664406779661016e-05, + "loss": 1.6283, + "step": 5420 + }, + { + "epoch": 0.14, + "grad_norm": 2.969449281692505, + "learning_rate": 1.665762711864407e-05, + "loss": 1.4047, + "step": 5430 + }, + { + "epoch": 0.14, + "grad_norm": 2.7862861156463623, + "learning_rate": 1.665084745762712e-05, + "loss": 1.4233, + "step": 5440 + }, + { + "epoch": 0.14, + "grad_norm": 1.7756729125976562, + "learning_rate": 1.6644067796610172e-05, + "loss": 1.4485, + "step": 5450 + }, + { + "epoch": 0.14, + "grad_norm": 2.082773447036743, + "learning_rate": 1.6637288135593222e-05, + "loss": 1.3878, + "step": 5460 + }, + { + "epoch": 0.14, + "grad_norm": 5.100788116455078, + "learning_rate": 1.663050847457627e-05, + "loss": 1.5481, + "step": 5470 + }, + { + "epoch": 0.14, + "grad_norm": 3.218986749649048, + "learning_rate": 1.6623728813559325e-05, + "loss": 1.3906, + "step": 5480 + }, + { + "epoch": 0.14, + "grad_norm": 1.727528691291809, + "learning_rate": 1.6616949152542374e-05, + "loss": 1.4824, + "step": 5490 + }, + { + "epoch": 0.14, + "grad_norm": 4.321578502655029, + "learning_rate": 1.6610169491525424e-05, + "loss": 1.5586, + "step": 5500 + }, + { + "epoch": 0.14, + "eval_loss": 1.469241976737976, + "eval_runtime": 66.1281, + "eval_samples_per_second": 15.122, + "eval_steps_per_second": 15.122, + "step": 5500 + }, + { + "epoch": 0.14, + "grad_norm": 1.7050514221191406, + "learning_rate": 1.6603389830508477e-05, + "loss": 1.2704, + "step": 5510 + }, + { + "epoch": 0.14, + "grad_norm": 1.4852951765060425, + "learning_rate": 1.6596610169491527e-05, + "loss": 1.6239, + "step": 5520 + }, + { + "epoch": 0.14, + "grad_norm": 3.524787664413452, + "learning_rate": 1.6589830508474577e-05, + "loss": 1.45, + "step": 5530 + }, + { + "epoch": 0.14, + "grad_norm": 1.804272174835205, + "learning_rate": 1.658305084745763e-05, + "loss": 1.4039, + "step": 5540 + }, + { + "epoch": 0.14, + "grad_norm": 2.2891910076141357, + "learning_rate": 1.657627118644068e-05, + "loss": 1.3969, + "step": 5550 + }, + { + "epoch": 0.14, + "grad_norm": 2.775587797164917, + "learning_rate": 1.6569491525423732e-05, + "loss": 1.5571, + "step": 5560 + }, + { + "epoch": 0.14, + "grad_norm": 4.006053924560547, + "learning_rate": 1.6562711864406782e-05, + "loss": 1.4743, + "step": 5570 + }, + { + "epoch": 0.14, + "grad_norm": 2.258488416671753, + "learning_rate": 1.6555932203389832e-05, + "loss": 1.3094, + "step": 5580 + }, + { + "epoch": 0.14, + "grad_norm": 1.469382405281067, + "learning_rate": 1.654915254237288e-05, + "loss": 1.4247, + "step": 5590 + }, + { + "epoch": 0.14, + "grad_norm": 2.234476089477539, + "learning_rate": 1.654237288135593e-05, + "loss": 1.4024, + "step": 5600 + }, + { + "epoch": 0.14, + "grad_norm": 4.297300338745117, + "learning_rate": 1.6535593220338984e-05, + "loss": 1.3422, + "step": 5610 + }, + { + "epoch": 0.14, + "grad_norm": 2.316805362701416, + "learning_rate": 1.6528813559322034e-05, + "loss": 1.4723, + "step": 5620 + }, + { + "epoch": 0.14, + "grad_norm": 2.859321117401123, + "learning_rate": 1.6522033898305084e-05, + "loss": 1.4444, + "step": 5630 + }, + { + "epoch": 0.14, + "grad_norm": 5.158230304718018, + "learning_rate": 1.6515254237288137e-05, + "loss": 1.6279, + "step": 5640 + }, + { + "epoch": 0.14, + "grad_norm": 2.160407781600952, + "learning_rate": 1.6508474576271187e-05, + "loss": 1.4487, + "step": 5650 + }, + { + "epoch": 0.14, + "grad_norm": 2.0725409984588623, + "learning_rate": 1.650169491525424e-05, + "loss": 1.4906, + "step": 5660 + }, + { + "epoch": 0.14, + "grad_norm": 1.5659565925598145, + "learning_rate": 1.649491525423729e-05, + "loss": 1.4545, + "step": 5670 + }, + { + "epoch": 0.14, + "grad_norm": 8.198447227478027, + "learning_rate": 1.648813559322034e-05, + "loss": 1.5087, + "step": 5680 + }, + { + "epoch": 0.14, + "grad_norm": 3.8482344150543213, + "learning_rate": 1.6481355932203392e-05, + "loss": 1.5596, + "step": 5690 + }, + { + "epoch": 0.14, + "grad_norm": 3.828826665878296, + "learning_rate": 1.6474576271186442e-05, + "loss": 1.5452, + "step": 5700 + }, + { + "epoch": 0.14, + "grad_norm": 9.930662155151367, + "learning_rate": 1.646779661016949e-05, + "loss": 1.4667, + "step": 5710 + }, + { + "epoch": 0.14, + "grad_norm": 2.9717953205108643, + "learning_rate": 1.6461016949152545e-05, + "loss": 1.5651, + "step": 5720 + }, + { + "epoch": 0.14, + "grad_norm": 1.6129778623580933, + "learning_rate": 1.6454237288135594e-05, + "loss": 1.3477, + "step": 5730 + }, + { + "epoch": 0.14, + "grad_norm": 2.341064214706421, + "learning_rate": 1.6447457627118648e-05, + "loss": 1.5159, + "step": 5740 + }, + { + "epoch": 0.14, + "grad_norm": 4.6363205909729, + "learning_rate": 1.6440677966101697e-05, + "loss": 1.5524, + "step": 5750 + }, + { + "epoch": 0.14, + "grad_norm": 4.433897495269775, + "learning_rate": 1.6433898305084747e-05, + "loss": 1.46, + "step": 5760 + }, + { + "epoch": 0.14, + "grad_norm": 3.2978014945983887, + "learning_rate": 1.64271186440678e-05, + "loss": 1.5847, + "step": 5770 + }, + { + "epoch": 0.14, + "grad_norm": 3.325284004211426, + "learning_rate": 1.642033898305085e-05, + "loss": 1.3553, + "step": 5780 + }, + { + "epoch": 0.14, + "grad_norm": 1.9704495668411255, + "learning_rate": 1.64135593220339e-05, + "loss": 1.3661, + "step": 5790 + }, + { + "epoch": 0.14, + "grad_norm": 1.6186383962631226, + "learning_rate": 1.640677966101695e-05, + "loss": 1.5538, + "step": 5800 + }, + { + "epoch": 0.15, + "grad_norm": 2.1665966510772705, + "learning_rate": 1.64e-05, + "loss": 1.5188, + "step": 5810 + }, + { + "epoch": 0.15, + "grad_norm": 7.165323734283447, + "learning_rate": 1.6393220338983052e-05, + "loss": 1.627, + "step": 5820 + }, + { + "epoch": 0.15, + "grad_norm": 7.434396743774414, + "learning_rate": 1.63864406779661e-05, + "loss": 1.4751, + "step": 5830 + }, + { + "epoch": 0.15, + "grad_norm": 2.549182176589966, + "learning_rate": 1.6379661016949155e-05, + "loss": 1.5013, + "step": 5840 + }, + { + "epoch": 0.15, + "grad_norm": 2.3907253742218018, + "learning_rate": 1.6372881355932204e-05, + "loss": 1.5532, + "step": 5850 + }, + { + "epoch": 0.15, + "grad_norm": 2.3262014389038086, + "learning_rate": 1.6366101694915254e-05, + "loss": 1.4404, + "step": 5860 + }, + { + "epoch": 0.15, + "grad_norm": 2.387146234512329, + "learning_rate": 1.6359322033898307e-05, + "loss": 1.4057, + "step": 5870 + }, + { + "epoch": 0.15, + "grad_norm": 4.461369514465332, + "learning_rate": 1.6352542372881357e-05, + "loss": 1.4003, + "step": 5880 + }, + { + "epoch": 0.15, + "grad_norm": 5.3152384757995605, + "learning_rate": 1.6345762711864407e-05, + "loss": 1.6658, + "step": 5890 + }, + { + "epoch": 0.15, + "grad_norm": 3.839587450027466, + "learning_rate": 1.633898305084746e-05, + "loss": 1.6396, + "step": 5900 + }, + { + "epoch": 0.15, + "grad_norm": 1.5456769466400146, + "learning_rate": 1.633220338983051e-05, + "loss": 1.4543, + "step": 5910 + }, + { + "epoch": 0.15, + "grad_norm": 1.218180775642395, + "learning_rate": 1.6325423728813563e-05, + "loss": 1.6549, + "step": 5920 + }, + { + "epoch": 0.15, + "grad_norm": 3.9048516750335693, + "learning_rate": 1.6318644067796612e-05, + "loss": 1.4558, + "step": 5930 + }, + { + "epoch": 0.15, + "grad_norm": 2.0440735816955566, + "learning_rate": 1.6311864406779662e-05, + "loss": 1.392, + "step": 5940 + }, + { + "epoch": 0.15, + "grad_norm": 5.26229190826416, + "learning_rate": 1.6305084745762715e-05, + "loss": 1.5081, + "step": 5950 + }, + { + "epoch": 0.15, + "grad_norm": 2.3289761543273926, + "learning_rate": 1.6298305084745765e-05, + "loss": 1.3642, + "step": 5960 + }, + { + "epoch": 0.15, + "grad_norm": 2.584726333618164, + "learning_rate": 1.6291525423728814e-05, + "loss": 1.3108, + "step": 5970 + }, + { + "epoch": 0.15, + "grad_norm": 3.9245219230651855, + "learning_rate": 1.6284745762711868e-05, + "loss": 1.5209, + "step": 5980 + }, + { + "epoch": 0.15, + "grad_norm": 1.5690237283706665, + "learning_rate": 1.6277966101694917e-05, + "loss": 1.4554, + "step": 5990 + }, + { + "epoch": 0.15, + "grad_norm": 2.7308597564697266, + "learning_rate": 1.6271186440677967e-05, + "loss": 1.3758, + "step": 6000 + }, + { + "epoch": 0.15, + "eval_loss": 1.4649540185928345, + "eval_runtime": 66.131, + "eval_samples_per_second": 15.122, + "eval_steps_per_second": 15.122, + "step": 6000 + }, + { + "epoch": 0.15, + "grad_norm": 3.45599627494812, + "learning_rate": 1.6264406779661017e-05, + "loss": 1.4816, + "step": 6010 + }, + { + "epoch": 0.15, + "grad_norm": 2.358900785446167, + "learning_rate": 1.625762711864407e-05, + "loss": 1.5672, + "step": 6020 + }, + { + "epoch": 0.15, + "grad_norm": 1.577291488647461, + "learning_rate": 1.625084745762712e-05, + "loss": 1.536, + "step": 6030 + }, + { + "epoch": 0.15, + "grad_norm": 1.5293314456939697, + "learning_rate": 1.624406779661017e-05, + "loss": 1.3852, + "step": 6040 + }, + { + "epoch": 0.15, + "grad_norm": 3.6409552097320557, + "learning_rate": 1.6237288135593222e-05, + "loss": 1.3703, + "step": 6050 + }, + { + "epoch": 0.15, + "grad_norm": 1.245363712310791, + "learning_rate": 1.6230508474576272e-05, + "loss": 1.4489, + "step": 6060 + }, + { + "epoch": 0.15, + "grad_norm": 3.5935614109039307, + "learning_rate": 1.6223728813559322e-05, + "loss": 1.3177, + "step": 6070 + }, + { + "epoch": 0.15, + "grad_norm": 3.89214825630188, + "learning_rate": 1.6216949152542375e-05, + "loss": 1.6123, + "step": 6080 + }, + { + "epoch": 0.15, + "grad_norm": 2.6698529720306396, + "learning_rate": 1.6210169491525424e-05, + "loss": 1.4598, + "step": 6090 + }, + { + "epoch": 0.15, + "grad_norm": 4.818114757537842, + "learning_rate": 1.6203389830508474e-05, + "loss": 1.3712, + "step": 6100 + }, + { + "epoch": 0.15, + "grad_norm": 5.041522979736328, + "learning_rate": 1.6196610169491527e-05, + "loss": 1.5556, + "step": 6110 + }, + { + "epoch": 0.15, + "grad_norm": 2.3514511585235596, + "learning_rate": 1.6189830508474577e-05, + "loss": 1.5359, + "step": 6120 + }, + { + "epoch": 0.15, + "grad_norm": 1.9724153280258179, + "learning_rate": 1.618305084745763e-05, + "loss": 1.5191, + "step": 6130 + }, + { + "epoch": 0.15, + "grad_norm": 3.0398383140563965, + "learning_rate": 1.617627118644068e-05, + "loss": 1.4751, + "step": 6140 + }, + { + "epoch": 0.15, + "grad_norm": 1.3548771142959595, + "learning_rate": 1.616949152542373e-05, + "loss": 1.4159, + "step": 6150 + }, + { + "epoch": 0.15, + "grad_norm": 3.616636276245117, + "learning_rate": 1.6162711864406783e-05, + "loss": 1.4667, + "step": 6160 + }, + { + "epoch": 0.15, + "grad_norm": 3.348968744277954, + "learning_rate": 1.6155932203389832e-05, + "loss": 1.4056, + "step": 6170 + }, + { + "epoch": 0.15, + "grad_norm": 2.5771777629852295, + "learning_rate": 1.6149152542372882e-05, + "loss": 1.4256, + "step": 6180 + }, + { + "epoch": 0.15, + "grad_norm": 3.0773298740386963, + "learning_rate": 1.6142372881355935e-05, + "loss": 1.2936, + "step": 6190 + }, + { + "epoch": 0.15, + "grad_norm": 2.114321708679199, + "learning_rate": 1.6135593220338985e-05, + "loss": 1.5362, + "step": 6200 + }, + { + "epoch": 0.16, + "grad_norm": 3.142282485961914, + "learning_rate": 1.6128813559322038e-05, + "loss": 1.506, + "step": 6210 + }, + { + "epoch": 0.16, + "grad_norm": 5.1931562423706055, + "learning_rate": 1.6122033898305084e-05, + "loss": 1.581, + "step": 6220 + }, + { + "epoch": 0.16, + "grad_norm": 3.0442159175872803, + "learning_rate": 1.6115254237288137e-05, + "loss": 1.4553, + "step": 6230 + }, + { + "epoch": 0.16, + "grad_norm": 4.841310501098633, + "learning_rate": 1.6108474576271187e-05, + "loss": 1.5559, + "step": 6240 + }, + { + "epoch": 0.16, + "grad_norm": 4.651193141937256, + "learning_rate": 1.6101694915254237e-05, + "loss": 1.3535, + "step": 6250 + }, + { + "epoch": 0.16, + "grad_norm": 3.571608543395996, + "learning_rate": 1.609491525423729e-05, + "loss": 1.5062, + "step": 6260 + }, + { + "epoch": 0.16, + "grad_norm": 6.008980751037598, + "learning_rate": 1.608813559322034e-05, + "loss": 1.4696, + "step": 6270 + }, + { + "epoch": 0.16, + "grad_norm": 11.914450645446777, + "learning_rate": 1.608135593220339e-05, + "loss": 1.4643, + "step": 6280 + }, + { + "epoch": 0.16, + "grad_norm": 1.8239995241165161, + "learning_rate": 1.6074576271186442e-05, + "loss": 1.6175, + "step": 6290 + }, + { + "epoch": 0.16, + "grad_norm": 2.543314218521118, + "learning_rate": 1.6067796610169492e-05, + "loss": 1.4362, + "step": 6300 + }, + { + "epoch": 0.16, + "grad_norm": 3.3625097274780273, + "learning_rate": 1.6061016949152545e-05, + "loss": 1.5615, + "step": 6310 + }, + { + "epoch": 0.16, + "grad_norm": 4.757710933685303, + "learning_rate": 1.6054237288135595e-05, + "loss": 1.4382, + "step": 6320 + }, + { + "epoch": 0.16, + "grad_norm": 2.5102877616882324, + "learning_rate": 1.6047457627118645e-05, + "loss": 1.5367, + "step": 6330 + }, + { + "epoch": 0.16, + "grad_norm": 3.226902484893799, + "learning_rate": 1.6040677966101698e-05, + "loss": 1.7053, + "step": 6340 + }, + { + "epoch": 0.16, + "grad_norm": 2.9312872886657715, + "learning_rate": 1.6033898305084747e-05, + "loss": 1.5362, + "step": 6350 + }, + { + "epoch": 0.16, + "grad_norm": 1.0346354246139526, + "learning_rate": 1.6027118644067797e-05, + "loss": 1.6161, + "step": 6360 + }, + { + "epoch": 0.16, + "grad_norm": 1.9462825059890747, + "learning_rate": 1.602033898305085e-05, + "loss": 1.4455, + "step": 6370 + }, + { + "epoch": 0.16, + "grad_norm": 1.9478490352630615, + "learning_rate": 1.60135593220339e-05, + "loss": 1.5183, + "step": 6380 + }, + { + "epoch": 0.16, + "grad_norm": 2.781726598739624, + "learning_rate": 1.600677966101695e-05, + "loss": 1.3517, + "step": 6390 + }, + { + "epoch": 0.16, + "grad_norm": 2.242448091506958, + "learning_rate": 1.6000000000000003e-05, + "loss": 1.6257, + "step": 6400 + }, + { + "epoch": 0.16, + "grad_norm": 1.3541291952133179, + "learning_rate": 1.5993220338983052e-05, + "loss": 1.4108, + "step": 6410 + }, + { + "epoch": 0.16, + "grad_norm": 2.1203219890594482, + "learning_rate": 1.5986440677966105e-05, + "loss": 1.5363, + "step": 6420 + }, + { + "epoch": 0.16, + "grad_norm": 2.9840617179870605, + "learning_rate": 1.5979661016949152e-05, + "loss": 1.4282, + "step": 6430 + }, + { + "epoch": 0.16, + "grad_norm": 3.6846158504486084, + "learning_rate": 1.5972881355932205e-05, + "loss": 1.4535, + "step": 6440 + }, + { + "epoch": 0.16, + "grad_norm": 1.272042155265808, + "learning_rate": 1.5966101694915255e-05, + "loss": 1.5842, + "step": 6450 + }, + { + "epoch": 0.16, + "grad_norm": 2.9574079513549805, + "learning_rate": 1.5959322033898304e-05, + "loss": 1.3752, + "step": 6460 + }, + { + "epoch": 0.16, + "grad_norm": 2.928889751434326, + "learning_rate": 1.5952542372881357e-05, + "loss": 1.4675, + "step": 6470 + }, + { + "epoch": 0.16, + "grad_norm": 5.292194843292236, + "learning_rate": 1.5945762711864407e-05, + "loss": 1.5172, + "step": 6480 + }, + { + "epoch": 0.16, + "grad_norm": 3.3911092281341553, + "learning_rate": 1.593898305084746e-05, + "loss": 1.467, + "step": 6490 + }, + { + "epoch": 0.16, + "grad_norm": 4.30944299697876, + "learning_rate": 1.593220338983051e-05, + "loss": 1.3515, + "step": 6500 + }, + { + "epoch": 0.16, + "eval_loss": 1.465691089630127, + "eval_runtime": 66.1688, + "eval_samples_per_second": 15.113, + "eval_steps_per_second": 15.113, + "step": 6500 + }, + { + "epoch": 0.16, + "grad_norm": 2.5098283290863037, + "learning_rate": 1.592542372881356e-05, + "loss": 1.5554, + "step": 6510 + }, + { + "epoch": 0.16, + "grad_norm": 7.33607292175293, + "learning_rate": 1.5918644067796613e-05, + "loss": 1.6483, + "step": 6520 + }, + { + "epoch": 0.16, + "grad_norm": 3.0138282775878906, + "learning_rate": 1.5911864406779662e-05, + "loss": 1.5784, + "step": 6530 + }, + { + "epoch": 0.16, + "grad_norm": 3.527956008911133, + "learning_rate": 1.5905084745762712e-05, + "loss": 1.265, + "step": 6540 + }, + { + "epoch": 0.16, + "grad_norm": 3.4589829444885254, + "learning_rate": 1.5898305084745765e-05, + "loss": 1.6605, + "step": 6550 + }, + { + "epoch": 0.16, + "grad_norm": 5.758669853210449, + "learning_rate": 1.5891525423728815e-05, + "loss": 1.2722, + "step": 6560 + }, + { + "epoch": 0.16, + "grad_norm": 2.0331597328186035, + "learning_rate": 1.5884745762711865e-05, + "loss": 1.5016, + "step": 6570 + }, + { + "epoch": 0.16, + "grad_norm": 7.730234622955322, + "learning_rate": 1.5877966101694918e-05, + "loss": 1.6401, + "step": 6580 + }, + { + "epoch": 0.16, + "grad_norm": 1.977071762084961, + "learning_rate": 1.5871186440677967e-05, + "loss": 1.3554, + "step": 6590 + }, + { + "epoch": 0.17, + "grad_norm": 3.8177216053009033, + "learning_rate": 1.586440677966102e-05, + "loss": 1.6109, + "step": 6600 + }, + { + "epoch": 0.17, + "grad_norm": 4.232168197631836, + "learning_rate": 1.585762711864407e-05, + "loss": 1.3901, + "step": 6610 + }, + { + "epoch": 0.17, + "grad_norm": 6.004239082336426, + "learning_rate": 1.585084745762712e-05, + "loss": 1.648, + "step": 6620 + }, + { + "epoch": 0.17, + "grad_norm": 4.166013240814209, + "learning_rate": 1.584406779661017e-05, + "loss": 1.4462, + "step": 6630 + }, + { + "epoch": 0.17, + "grad_norm": 2.389451503753662, + "learning_rate": 1.583728813559322e-05, + "loss": 1.5163, + "step": 6640 + }, + { + "epoch": 0.17, + "grad_norm": 2.042405843734741, + "learning_rate": 1.5830508474576272e-05, + "loss": 1.4931, + "step": 6650 + }, + { + "epoch": 0.17, + "grad_norm": 2.3263092041015625, + "learning_rate": 1.5823728813559322e-05, + "loss": 1.4758, + "step": 6660 + }, + { + "epoch": 0.17, + "grad_norm": 5.392455577850342, + "learning_rate": 1.5816949152542372e-05, + "loss": 1.6324, + "step": 6670 + }, + { + "epoch": 0.17, + "grad_norm": 4.789013385772705, + "learning_rate": 1.5810169491525425e-05, + "loss": 1.5744, + "step": 6680 + }, + { + "epoch": 0.17, + "grad_norm": 2.284268617630005, + "learning_rate": 1.5803389830508475e-05, + "loss": 1.6875, + "step": 6690 + }, + { + "epoch": 0.17, + "grad_norm": 2.3472728729248047, + "learning_rate": 1.5796610169491528e-05, + "loss": 1.5855, + "step": 6700 + }, + { + "epoch": 0.17, + "grad_norm": 4.676074981689453, + "learning_rate": 1.5789830508474577e-05, + "loss": 1.4166, + "step": 6710 + }, + { + "epoch": 0.17, + "grad_norm": 3.129818916320801, + "learning_rate": 1.5783050847457627e-05, + "loss": 1.3874, + "step": 6720 + }, + { + "epoch": 0.17, + "grad_norm": 4.993739604949951, + "learning_rate": 1.577627118644068e-05, + "loss": 1.4709, + "step": 6730 + }, + { + "epoch": 0.17, + "grad_norm": 2.868453025817871, + "learning_rate": 1.576949152542373e-05, + "loss": 1.5804, + "step": 6740 + }, + { + "epoch": 0.17, + "grad_norm": 1.086580753326416, + "learning_rate": 1.576271186440678e-05, + "loss": 1.3879, + "step": 6750 + }, + { + "epoch": 0.17, + "grad_norm": 5.881350994110107, + "learning_rate": 1.5755932203389833e-05, + "loss": 1.4923, + "step": 6760 + }, + { + "epoch": 0.17, + "grad_norm": 3.4320054054260254, + "learning_rate": 1.5749152542372882e-05, + "loss": 1.401, + "step": 6770 + }, + { + "epoch": 0.17, + "grad_norm": 2.2193899154663086, + "learning_rate": 1.5742372881355936e-05, + "loss": 1.5497, + "step": 6780 + }, + { + "epoch": 0.17, + "grad_norm": 3.470163583755493, + "learning_rate": 1.5735593220338985e-05, + "loss": 1.4824, + "step": 6790 + }, + { + "epoch": 0.17, + "grad_norm": 3.5179107189178467, + "learning_rate": 1.5728813559322035e-05, + "loss": 1.5401, + "step": 6800 + }, + { + "epoch": 0.17, + "grad_norm": 4.4173359870910645, + "learning_rate": 1.5722033898305088e-05, + "loss": 1.3075, + "step": 6810 + }, + { + "epoch": 0.17, + "grad_norm": 4.17536735534668, + "learning_rate": 1.5715254237288138e-05, + "loss": 1.5135, + "step": 6820 + }, + { + "epoch": 0.17, + "grad_norm": 3.278881311416626, + "learning_rate": 1.5708474576271187e-05, + "loss": 1.5745, + "step": 6830 + }, + { + "epoch": 0.17, + "grad_norm": 2.718890905380249, + "learning_rate": 1.5701694915254237e-05, + "loss": 1.5397, + "step": 6840 + }, + { + "epoch": 0.17, + "grad_norm": 1.8117082118988037, + "learning_rate": 1.5694915254237287e-05, + "loss": 1.5904, + "step": 6850 + }, + { + "epoch": 0.17, + "grad_norm": 1.4601502418518066, + "learning_rate": 1.568813559322034e-05, + "loss": 1.5485, + "step": 6860 + }, + { + "epoch": 0.17, + "grad_norm": 2.0269432067871094, + "learning_rate": 1.568135593220339e-05, + "loss": 1.4487, + "step": 6870 + }, + { + "epoch": 0.17, + "grad_norm": 3.047973394393921, + "learning_rate": 1.5674576271186443e-05, + "loss": 1.2999, + "step": 6880 + }, + { + "epoch": 0.17, + "grad_norm": 2.7541468143463135, + "learning_rate": 1.5667796610169492e-05, + "loss": 1.4651, + "step": 6890 + }, + { + "epoch": 0.17, + "grad_norm": 5.5456767082214355, + "learning_rate": 1.5661016949152542e-05, + "loss": 1.4433, + "step": 6900 + }, + { + "epoch": 0.17, + "grad_norm": 3.4916276931762695, + "learning_rate": 1.5654237288135595e-05, + "loss": 1.5764, + "step": 6910 + }, + { + "epoch": 0.17, + "grad_norm": 2.6278750896453857, + "learning_rate": 1.5647457627118645e-05, + "loss": 1.3208, + "step": 6920 + }, + { + "epoch": 0.17, + "grad_norm": 2.9469902515411377, + "learning_rate": 1.5640677966101695e-05, + "loss": 1.6433, + "step": 6930 + }, + { + "epoch": 0.17, + "grad_norm": 5.235908031463623, + "learning_rate": 1.5633898305084748e-05, + "loss": 1.2608, + "step": 6940 + }, + { + "epoch": 0.17, + "grad_norm": 1.5849263668060303, + "learning_rate": 1.5627118644067798e-05, + "loss": 1.5088, + "step": 6950 + }, + { + "epoch": 0.17, + "grad_norm": 3.6014277935028076, + "learning_rate": 1.562033898305085e-05, + "loss": 1.3976, + "step": 6960 + }, + { + "epoch": 0.17, + "grad_norm": 3.220320701599121, + "learning_rate": 1.56135593220339e-05, + "loss": 1.5011, + "step": 6970 + }, + { + "epoch": 0.17, + "grad_norm": 3.434382915496826, + "learning_rate": 1.560677966101695e-05, + "loss": 1.4733, + "step": 6980 + }, + { + "epoch": 0.17, + "grad_norm": 4.612408638000488, + "learning_rate": 1.5600000000000003e-05, + "loss": 1.4618, + "step": 6990 + }, + { + "epoch": 0.17, + "grad_norm": 2.9301650524139404, + "learning_rate": 1.5593220338983053e-05, + "loss": 1.4165, + "step": 7000 + }, + { + "epoch": 0.17, + "eval_loss": 1.469406247138977, + "eval_runtime": 66.2443, + "eval_samples_per_second": 15.096, + "eval_steps_per_second": 15.096, + "step": 7000 + }, + { + "epoch": 0.18, + "grad_norm": 5.865874767303467, + "learning_rate": 1.5586440677966103e-05, + "loss": 1.3156, + "step": 7010 + }, + { + "epoch": 0.18, + "grad_norm": 5.411038875579834, + "learning_rate": 1.5579661016949156e-05, + "loss": 1.4303, + "step": 7020 + }, + { + "epoch": 0.18, + "grad_norm": 2.869673728942871, + "learning_rate": 1.5572881355932205e-05, + "loss": 1.5055, + "step": 7030 + }, + { + "epoch": 0.18, + "grad_norm": 1.0629090070724487, + "learning_rate": 1.5566101694915255e-05, + "loss": 1.6094, + "step": 7040 + }, + { + "epoch": 0.18, + "grad_norm": 2.271601676940918, + "learning_rate": 1.5559322033898305e-05, + "loss": 1.4648, + "step": 7050 + }, + { + "epoch": 0.18, + "grad_norm": 1.5404294729232788, + "learning_rate": 1.5552542372881358e-05, + "loss": 1.3793, + "step": 7060 + }, + { + "epoch": 0.18, + "grad_norm": 3.9599192142486572, + "learning_rate": 1.5545762711864408e-05, + "loss": 1.3563, + "step": 7070 + }, + { + "epoch": 0.18, + "grad_norm": 6.35668420791626, + "learning_rate": 1.5538983050847457e-05, + "loss": 1.4637, + "step": 7080 + }, + { + "epoch": 0.18, + "grad_norm": 7.315426349639893, + "learning_rate": 1.553220338983051e-05, + "loss": 1.3284, + "step": 7090 + }, + { + "epoch": 0.18, + "grad_norm": 1.8330402374267578, + "learning_rate": 1.552542372881356e-05, + "loss": 1.3762, + "step": 7100 + }, + { + "epoch": 0.18, + "grad_norm": 3.072326183319092, + "learning_rate": 1.551864406779661e-05, + "loss": 1.4194, + "step": 7110 + }, + { + "epoch": 0.18, + "grad_norm": 1.952779769897461, + "learning_rate": 1.5511864406779663e-05, + "loss": 1.4886, + "step": 7120 + }, + { + "epoch": 0.18, + "grad_norm": 4.332074165344238, + "learning_rate": 1.5505084745762713e-05, + "loss": 1.4912, + "step": 7130 + }, + { + "epoch": 0.18, + "grad_norm": 3.962344169616699, + "learning_rate": 1.5498305084745762e-05, + "loss": 1.5502, + "step": 7140 + }, + { + "epoch": 0.18, + "grad_norm": 7.930359363555908, + "learning_rate": 1.5491525423728815e-05, + "loss": 1.471, + "step": 7150 + }, + { + "epoch": 0.18, + "grad_norm": 4.925573825836182, + "learning_rate": 1.5484745762711865e-05, + "loss": 1.3155, + "step": 7160 + }, + { + "epoch": 0.18, + "grad_norm": 3.2943711280822754, + "learning_rate": 1.5477966101694918e-05, + "loss": 1.3094, + "step": 7170 + }, + { + "epoch": 0.18, + "grad_norm": 1.9433196783065796, + "learning_rate": 1.5471186440677968e-05, + "loss": 1.3453, + "step": 7180 + }, + { + "epoch": 0.18, + "grad_norm": 1.8384184837341309, + "learning_rate": 1.5464406779661018e-05, + "loss": 1.3788, + "step": 7190 + }, + { + "epoch": 0.18, + "grad_norm": 2.92781138420105, + "learning_rate": 1.545762711864407e-05, + "loss": 1.5704, + "step": 7200 + }, + { + "epoch": 0.18, + "grad_norm": 1.0172346830368042, + "learning_rate": 1.545084745762712e-05, + "loss": 1.5442, + "step": 7210 + }, + { + "epoch": 0.18, + "grad_norm": 1.0915580987930298, + "learning_rate": 1.544406779661017e-05, + "loss": 1.5226, + "step": 7220 + }, + { + "epoch": 0.18, + "grad_norm": 9.170607566833496, + "learning_rate": 1.5437288135593223e-05, + "loss": 1.5066, + "step": 7230 + }, + { + "epoch": 0.18, + "grad_norm": 4.84716796875, + "learning_rate": 1.5430508474576273e-05, + "loss": 1.4508, + "step": 7240 + }, + { + "epoch": 0.18, + "grad_norm": 5.949992656707764, + "learning_rate": 1.5423728813559326e-05, + "loss": 1.3393, + "step": 7250 + }, + { + "epoch": 0.18, + "grad_norm": 5.912881851196289, + "learning_rate": 1.5416949152542372e-05, + "loss": 1.5918, + "step": 7260 + }, + { + "epoch": 0.18, + "grad_norm": 1.8128445148468018, + "learning_rate": 1.5410169491525425e-05, + "loss": 1.5794, + "step": 7270 + }, + { + "epoch": 0.18, + "grad_norm": 3.9023091793060303, + "learning_rate": 1.5403389830508475e-05, + "loss": 1.345, + "step": 7280 + }, + { + "epoch": 0.18, + "grad_norm": 2.263958692550659, + "learning_rate": 1.5396610169491525e-05, + "loss": 1.3651, + "step": 7290 + }, + { + "epoch": 0.18, + "grad_norm": 8.103110313415527, + "learning_rate": 1.5389830508474578e-05, + "loss": 1.4784, + "step": 7300 + }, + { + "epoch": 0.18, + "grad_norm": 4.5859456062316895, + "learning_rate": 1.5383050847457628e-05, + "loss": 1.4087, + "step": 7310 + }, + { + "epoch": 0.18, + "grad_norm": 5.6322503089904785, + "learning_rate": 1.5376271186440677e-05, + "loss": 1.518, + "step": 7320 + }, + { + "epoch": 0.18, + "grad_norm": 2.4485771656036377, + "learning_rate": 1.536949152542373e-05, + "loss": 1.3471, + "step": 7330 + }, + { + "epoch": 0.18, + "grad_norm": 2.0619120597839355, + "learning_rate": 1.536271186440678e-05, + "loss": 1.2563, + "step": 7340 + }, + { + "epoch": 0.18, + "grad_norm": 4.037415027618408, + "learning_rate": 1.5355932203389833e-05, + "loss": 1.4414, + "step": 7350 + }, + { + "epoch": 0.18, + "grad_norm": 2.4679503440856934, + "learning_rate": 1.5349152542372883e-05, + "loss": 1.4021, + "step": 7360 + }, + { + "epoch": 0.18, + "grad_norm": 1.4774163961410522, + "learning_rate": 1.5342372881355933e-05, + "loss": 1.4331, + "step": 7370 + }, + { + "epoch": 0.18, + "grad_norm": 4.21893310546875, + "learning_rate": 1.5335593220338986e-05, + "loss": 1.4847, + "step": 7380 + }, + { + "epoch": 0.18, + "grad_norm": 4.261248588562012, + "learning_rate": 1.5328813559322035e-05, + "loss": 1.4098, + "step": 7390 + }, + { + "epoch": 0.18, + "grad_norm": 2.419386625289917, + "learning_rate": 1.5322033898305085e-05, + "loss": 1.4635, + "step": 7400 + }, + { + "epoch": 0.19, + "grad_norm": 3.3119351863861084, + "learning_rate": 1.5315254237288138e-05, + "loss": 1.3862, + "step": 7410 + }, + { + "epoch": 0.19, + "grad_norm": 4.350612163543701, + "learning_rate": 1.5308474576271188e-05, + "loss": 1.5094, + "step": 7420 + }, + { + "epoch": 0.19, + "grad_norm": 3.089937925338745, + "learning_rate": 1.530169491525424e-05, + "loss": 1.6249, + "step": 7430 + }, + { + "epoch": 0.19, + "grad_norm": 6.980156898498535, + "learning_rate": 1.529491525423729e-05, + "loss": 1.6102, + "step": 7440 + }, + { + "epoch": 0.19, + "grad_norm": 3.881622552871704, + "learning_rate": 1.528813559322034e-05, + "loss": 1.5149, + "step": 7450 + }, + { + "epoch": 0.19, + "grad_norm": 2.8510727882385254, + "learning_rate": 1.528135593220339e-05, + "loss": 1.4542, + "step": 7460 + }, + { + "epoch": 0.19, + "grad_norm": 4.4826130867004395, + "learning_rate": 1.527457627118644e-05, + "loss": 1.5085, + "step": 7470 + }, + { + "epoch": 0.19, + "grad_norm": 4.991846084594727, + "learning_rate": 1.5267796610169493e-05, + "loss": 1.4761, + "step": 7480 + }, + { + "epoch": 0.19, + "grad_norm": 4.136972904205322, + "learning_rate": 1.5261016949152543e-05, + "loss": 1.4851, + "step": 7490 + }, + { + "epoch": 0.19, + "grad_norm": 3.58878231048584, + "learning_rate": 1.5254237288135594e-05, + "loss": 1.4117, + "step": 7500 + }, + { + "epoch": 0.19, + "eval_loss": 1.437897801399231, + "eval_runtime": 66.1821, + "eval_samples_per_second": 15.11, + "eval_steps_per_second": 15.11, + "step": 7500 + } + ], + "logging_steps": 10, + "max_steps": 30000, + "num_input_tokens_seen": 0, + "num_train_epochs": 1, + "save_steps": 2500, + "total_flos": 1.2076594495488e+17, + "train_batch_size": 1, + "trial_name": null, + "trial_params": null +}