diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,488 +1,4637 @@
 {
-  "best_metric": 0.9297520661157025,
-  "best_model_checkpoint": "swin-tiny-patch4-window7-224-finetuned-wsdmhar\\checkpoint-530",
-  "epoch": 10.0,
+  "best_metric": 0.9683195592286501,
+  "best_model_checkpoint": "swin-tiny-patch4-window7-224-finetuned-wsdmhar\\checkpoint-4187",
+  "epoch": 100.0,
   "eval_steps": 500,
-  "global_step": 530,
+  "global_step": 5300,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.18867924528301888,
-      "grad_norm": 5.205289840698242,
-      "learning_rate": 9.433962264150944e-06,
-      "loss": 1.7479,
+      "grad_norm": 9.041044235229492,
+      "learning_rate": 9.433962264150943e-07,
+      "loss": 1.4928,
       "step": 10
     },
     {
       "epoch": 0.37735849056603776,
-      "grad_norm": 4.392179012298584,
-      "learning_rate": 1.8867924528301888e-05,
-      "loss": 1.4506,
+      "grad_norm": 7.276942253112793,
+      "learning_rate": 1.8867924528301887e-06,
+      "loss": 1.4202,
       "step": 20
     },
     {
       "epoch": 0.5660377358490566,
-      "grad_norm": 6.868739604949951,
-      "learning_rate": 2.830188679245283e-05,
-      "loss": 1.0283,
+      "grad_norm": 8.077420234680176,
+      "learning_rate": 2.830188679245283e-06,
+      "loss": 1.3564,
       "step": 30
     },
     {
       "epoch": 0.7547169811320755,
-      "grad_norm": 9.729569435119629,
-      "learning_rate": 3.7735849056603776e-05,
-      "loss": 0.8064,
+      "grad_norm": 7.560600280761719,
+      "learning_rate": 3.7735849056603773e-06,
+      "loss": 1.1939,
       "step": 40
     },
     {
       "epoch": 0.9433962264150944,
-      "grad_norm": 12.286840438842773,
-      "learning_rate": 4.716981132075472e-05,
-      "loss": 0.7388,
+      "grad_norm": 5.84022855758667,
+      "learning_rate": 4.716981132075472e-06,
+      "loss": 1.0624,
       "step": 50
     },
     {
       "epoch": 1.0,
-      "eval_accuracy": 0.7117768595041323,
-      "eval_loss": 0.6308234930038452,
-      "eval_runtime": 24.0663,
-      "eval_samples_per_second": 120.667,
-      "eval_steps_per_second": 3.781,
+      "eval_accuracy": 0.6091597796143251,
+      "eval_loss": 0.8878794312477112,
+      "eval_runtime": 12.739,
+      "eval_samples_per_second": 227.961,
+      "eval_steps_per_second": 7.143,
       "step": 53
     },
     {
       "epoch": 1.1320754716981132,
-      "grad_norm": 9.05235481262207,
-      "learning_rate": 4.9266247379454926e-05,
-      "loss": 0.6616,
+      "grad_norm": 9.758731842041016,
+      "learning_rate": 5.660377358490566e-06,
+      "loss": 0.9299,
       "step": 60
     },
     {
       "epoch": 1.320754716981132,
-      "grad_norm": 10.109884262084961,
-      "learning_rate": 4.8218029350104823e-05,
-      "loss": 0.6363,
+      "grad_norm": 7.677206993103027,
+      "learning_rate": 6.60377358490566e-06,
+      "loss": 0.8724,
       "step": 70
     },
     {
       "epoch": 1.509433962264151,
-      "grad_norm": 10.793712615966797,
-      "learning_rate": 4.716981132075472e-05,
-      "loss": 0.554,
+      "grad_norm": 8.157505989074707,
+      "learning_rate": 7.547169811320755e-06,
+      "loss": 0.7971,
       "step": 80
     },
     {
       "epoch": 1.6981132075471699,
-      "grad_norm": 17.802772521972656,
-      "learning_rate": 4.612159329140461e-05,
-      "loss": 0.5516,
+      "grad_norm": 9.804584503173828,
+      "learning_rate": 8.49056603773585e-06,
+      "loss": 0.7717,
       "step": 90
     },
     {
       "epoch": 1.8867924528301887,
-      "grad_norm": 11.082262992858887,
-      "learning_rate": 4.5073375262054504e-05,
-      "loss": 0.5099,
+      "grad_norm": 8.317395210266113,
+      "learning_rate": 9.433962264150944e-06,
+      "loss": 0.6893,
       "step": 100
     },
     {
       "epoch": 2.0,
-      "eval_accuracy": 0.8484848484848485,
-      "eval_loss": 0.36690881848335266,
-      "eval_runtime": 13.3175,
-      "eval_samples_per_second": 218.059,
-      "eval_steps_per_second": 6.833,
+      "eval_accuracy": 0.7090220385674931,
+      "eval_loss": 0.6601111888885498,
+      "eval_runtime": 13.0303,
+      "eval_samples_per_second": 222.865,
+      "eval_steps_per_second": 6.984,
       "step": 106
     },
     {
       "epoch": 2.0754716981132075,
-      "grad_norm": 6.183718204498291,
-      "learning_rate": 4.402515723270441e-05,
-      "loss": 0.4778,
+      "grad_norm": 5.714374542236328,
+      "learning_rate": 1.0377358490566038e-05,
+      "loss": 0.6923,
       "step": 110
     },
     {
       "epoch": 2.2641509433962264,
-      "grad_norm": 13.359389305114746,
-      "learning_rate": 4.29769392033543e-05,
-      "loss": 0.4449,
+      "grad_norm": 8.416643142700195,
+      "learning_rate": 1.1320754716981132e-05,
+      "loss": 0.6475,
       "step": 120
     },
     {
       "epoch": 2.452830188679245,
-      "grad_norm": 6.166327953338623,
-      "learning_rate": 4.192872117400419e-05,
-      "loss": 0.4243,
+      "grad_norm": 8.871334075927734,
+      "learning_rate": 1.2264150943396227e-05,
+      "loss": 0.6156,
       "step": 130
     },
     {
       "epoch": 2.641509433962264,
-      "grad_norm": 6.538959980010986,
-      "learning_rate": 4.088050314465409e-05,
-      "loss": 0.4265,
+      "grad_norm": 10.934247970581055,
+      "learning_rate": 1.320754716981132e-05,
+      "loss": 0.597,
       "step": 140
     },
     {
       "epoch": 2.830188679245283,
-      "grad_norm": 11.39398193359375,
-      "learning_rate": 3.983228511530399e-05,
-      "loss": 0.4319,
+      "grad_norm": 12.311450004577637,
+      "learning_rate": 1.4150943396226415e-05,
+      "loss": 0.6152,
       "step": 150
     },
     {
       "epoch": 3.0,
-      "eval_accuracy": 0.8684573002754821,
-      "eval_loss": 0.33240532875061035,
-      "eval_runtime": 14.2442,
-      "eval_samples_per_second": 203.873,
-      "eval_steps_per_second": 6.389,
+      "eval_accuracy": 0.7854683195592287,
+      "eval_loss": 0.5114469528198242,
+      "eval_runtime": 13.0985,
+      "eval_samples_per_second": 221.706,
+      "eval_steps_per_second": 6.947,
       "step": 159
     },
     {
       "epoch": 3.018867924528302,
-      "grad_norm": 9.0410795211792,
-      "learning_rate": 3.878406708595388e-05,
-      "loss": 0.3985,
+      "grad_norm": 8.85444164276123,
+      "learning_rate": 1.509433962264151e-05,
+      "loss": 0.5785,
       "step": 160
     },
     {
       "epoch": 3.207547169811321,
-      "grad_norm": 6.786131381988525,
-      "learning_rate": 3.7735849056603776e-05,
-      "loss": 0.3912,
+      "grad_norm": 7.401244640350342,
+      "learning_rate": 1.6037735849056604e-05,
+      "loss": 0.5343,
       "step": 170
     },
     {
       "epoch": 3.3962264150943398,
-      "grad_norm": 8.235289573669434,
-      "learning_rate": 3.6687631027253674e-05,
-      "loss": 0.3496,
+      "grad_norm": 6.694042205810547,
+      "learning_rate": 1.69811320754717e-05,
+      "loss": 0.5322,
       "step": 180
     },
     {
       "epoch": 3.5849056603773586,
-      "grad_norm": 6.866382122039795,
-      "learning_rate": 3.5639412997903565e-05,
-      "loss": 0.3696,
+      "grad_norm": 7.124511241912842,
+      "learning_rate": 1.7924528301886792e-05,
+      "loss": 0.5533,
       "step": 190
     },
     {
       "epoch": 3.7735849056603774,
-      "grad_norm": 8.17050552368164,
-      "learning_rate": 3.4591194968553456e-05,
-      "loss": 0.3203,
+      "grad_norm": 11.249773025512695,
+      "learning_rate": 1.8867924528301888e-05,
+      "loss": 0.4865,
       "step": 200
     },
     {
       "epoch": 3.9622641509433962,
-      "grad_norm": 17.23760414123535,
-      "learning_rate": 3.354297693920336e-05,
-      "loss": 0.4002,
+      "grad_norm": 19.239667892456055,
+      "learning_rate": 1.9811320754716984e-05,
+      "loss": 0.5456,
       "step": 210
     },
     {
       "epoch": 4.0,
-      "eval_accuracy": 0.9028925619834711,
-      "eval_loss": 0.2758175730705261,
-      "eval_runtime": 14.4978,
-      "eval_samples_per_second": 200.306,
-      "eval_steps_per_second": 6.277,
+      "eval_accuracy": 0.8422865013774105,
+      "eval_loss": 0.3818637728691101,
+      "eval_runtime": 13.2694,
+      "eval_samples_per_second": 218.849,
+      "eval_steps_per_second": 6.858,
       "step": 212
     },
     {
       "epoch": 4.150943396226415,
-      "grad_norm": 17.225961685180664,
-      "learning_rate": 3.249475890985325e-05,
-      "loss": 0.3484,
+      "grad_norm": 11.330594062805176,
+      "learning_rate": 2.0754716981132076e-05,
+      "loss": 0.5114,
       "step": 220
     },
     {
       "epoch": 4.339622641509434,
-      "grad_norm": 12.645174980163574,
-      "learning_rate": 3.144654088050314e-05,
-      "loss": 0.353,
+      "grad_norm": 11.731520652770996,
+      "learning_rate": 2.1698113207547172e-05,
+      "loss": 0.4923,
       "step": 230
     },
     {
       "epoch": 4.528301886792453,
-      "grad_norm": 5.725615978240967,
-      "learning_rate": 3.0398322851153044e-05,
-      "loss": 0.2883,
+      "grad_norm": 6.535953521728516,
+      "learning_rate": 2.2641509433962265e-05,
+      "loss": 0.4251,
       "step": 240
     },
     {
       "epoch": 4.716981132075472,
-      "grad_norm": 5.914320468902588,
-      "learning_rate": 2.935010482180294e-05,
-      "loss": 0.3067,
+      "grad_norm": 5.095239639282227,
+      "learning_rate": 2.358490566037736e-05,
+      "loss": 0.4388,
       "step": 250
     },
     {
       "epoch": 4.90566037735849,
-      "grad_norm": 6.547543525695801,
-      "learning_rate": 2.830188679245283e-05,
-      "loss": 0.3589,
+      "grad_norm": 8.532243728637695,
+      "learning_rate": 2.4528301886792453e-05,
+      "loss": 0.4673,
       "step": 260
     },
     {
       "epoch": 5.0,
-      "eval_accuracy": 0.9132231404958677,
-      "eval_loss": 0.2503372132778168,
-      "eval_runtime": 14.406,
-      "eval_samples_per_second": 201.582,
-      "eval_steps_per_second": 6.317,
+      "eval_accuracy": 0.871900826446281,
+      "eval_loss": 0.32669946551322937,
+      "eval_runtime": 13.3969,
+      "eval_samples_per_second": 216.767,
+      "eval_steps_per_second": 6.793,
       "step": 265
     },
     {
       "epoch": 5.09433962264151,
-      "grad_norm": 8.123590469360352,
-      "learning_rate": 2.7253668763102725e-05,
-      "loss": 0.3272,
+      "grad_norm": 7.914596080780029,
+      "learning_rate": 2.547169811320755e-05,
+      "loss": 0.4331,
       "step": 270
     },
     {
       "epoch": 5.283018867924528,
-      "grad_norm": 5.585710048675537,
-      "learning_rate": 2.6205450733752623e-05,
-      "loss": 0.3535,
+      "grad_norm": 7.141151428222656,
+      "learning_rate": 2.641509433962264e-05,
+      "loss": 0.4553,
       "step": 280
     },
     {
       "epoch": 5.471698113207547,
-      "grad_norm": 8.529882431030273,
-      "learning_rate": 2.5157232704402517e-05,
-      "loss": 0.2797,
+      "grad_norm": 9.838526725769043,
+      "learning_rate": 2.7358490566037738e-05,
+      "loss": 0.3864,
       "step": 290
     },
     {
       "epoch": 5.660377358490566,
-      "grad_norm": 6.009942054748535,
-      "learning_rate": 2.4109014675052412e-05,
-      "loss": 0.3004,
+      "grad_norm": 6.91007137298584,
+      "learning_rate": 2.830188679245283e-05,
+      "loss": 0.4015,
       "step": 300
     },
     {
       "epoch": 5.849056603773585,
-      "grad_norm": 4.354741096496582,
-      "learning_rate": 2.3060796645702306e-05,
-      "loss": 0.3096,
+      "grad_norm": 6.716864585876465,
+      "learning_rate": 2.9245283018867926e-05,
+      "loss": 0.4166,
       "step": 310
     },
     {
       "epoch": 6.0,
-      "eval_accuracy": 0.9135674931129476,
-      "eval_loss": 0.24190402030944824,
-      "eval_runtime": 14.5262,
-      "eval_samples_per_second": 199.915,
-      "eval_steps_per_second": 6.265,
+      "eval_accuracy": 0.9039256198347108,
+      "eval_loss": 0.2804345488548279,
+      "eval_runtime": 13.5503,
+      "eval_samples_per_second": 214.312,
+      "eval_steps_per_second": 6.716,
       "step": 318
     },
     {
       "epoch": 6.037735849056604,
-      "grad_norm": 6.793937683105469,
-      "learning_rate": 2.2012578616352204e-05,
-      "loss": 0.3024,
+      "grad_norm": 19.735078811645508,
+      "learning_rate": 3.018867924528302e-05,
+      "loss": 0.4004,
       "step": 320
     },
     {
       "epoch": 6.226415094339623,
-      "grad_norm": 4.6357221603393555,
-      "learning_rate": 2.0964360587002095e-05,
-      "loss": 0.3013,
+      "grad_norm": 14.484492301940918,
+      "learning_rate": 3.113207547169811e-05,
+      "loss": 0.4114,
       "step": 330
     },
     {
       "epoch": 6.415094339622642,
-      "grad_norm": 5.533959865570068,
-      "learning_rate": 1.9916142557651993e-05,
-      "loss": 0.311,
+      "grad_norm": 8.806564331054688,
+      "learning_rate": 3.207547169811321e-05,
+      "loss": 0.398,
       "step": 340
     },
     {
       "epoch": 6.60377358490566,
-      "grad_norm": 5.0147600173950195,
-      "learning_rate": 1.8867924528301888e-05,
-      "loss": 0.2865,
+      "grad_norm": 11.519726753234863,
+      "learning_rate": 3.30188679245283e-05,
+      "loss": 0.413,
       "step": 350
     },
     {
       "epoch": 6.7924528301886795,
-      "grad_norm": 4.817899227142334,
-      "learning_rate": 1.7819706498951782e-05,
-      "loss": 0.2813,
+      "grad_norm": 4.552971363067627,
+      "learning_rate": 3.39622641509434e-05,
+      "loss": 0.3775,
       "step": 360
     },
     {
       "epoch": 6.981132075471698,
-      "grad_norm": 6.032566547393799,
-      "learning_rate": 1.677148846960168e-05,
-      "loss": 0.2708,
+      "grad_norm": 6.50352668762207,
+      "learning_rate": 3.490566037735849e-05,
+      "loss": 0.3757,
       "step": 370
     },
     {
       "epoch": 7.0,
-      "eval_accuracy": 0.9232093663911846,
-      "eval_loss": 0.22773109376430511,
-      "eval_runtime": 15.0479,
-      "eval_samples_per_second": 192.984,
-      "eval_steps_per_second": 6.047,
+      "eval_accuracy": 0.8994490358126722,
+      "eval_loss": 0.2881180942058563,
+      "eval_runtime": 13.6033,
+      "eval_samples_per_second": 213.478,
+      "eval_steps_per_second": 6.69,
       "step": 371
     },
     {
       "epoch": 7.169811320754717,
-      "grad_norm": 5.316783428192139,
-      "learning_rate": 1.572327044025157e-05,
-      "loss": 0.3009,
+      "grad_norm": 5.162815093994141,
+      "learning_rate": 3.5849056603773584e-05,
+      "loss": 0.3744,
       "step": 380
     },
     {
       "epoch": 7.3584905660377355,
-      "grad_norm": 5.2730607986450195,
-      "learning_rate": 1.467505241090147e-05,
-      "loss": 0.2422,
+      "grad_norm": 4.919179916381836,
+      "learning_rate": 3.679245283018868e-05,
+      "loss": 0.3084,
       "step": 390
     },
     {
       "epoch": 7.547169811320755,
-      "grad_norm": 5.739044189453125,
-      "learning_rate": 1.3626834381551362e-05,
-      "loss": 0.2667,
+      "grad_norm": 8.160808563232422,
+      "learning_rate": 3.7735849056603776e-05,
+      "loss": 0.3388,
       "step": 400
     },
     {
       "epoch": 7.735849056603773,
-      "grad_norm": 5.689853668212891,
-      "learning_rate": 1.2578616352201259e-05,
-      "loss": 0.2833,
+      "grad_norm": 7.450716972351074,
+      "learning_rate": 3.867924528301887e-05,
+      "loss": 0.3578,
       "step": 410
     },
     {
       "epoch": 7.9245283018867925,
-      "grad_norm": 6.43410062789917,
-      "learning_rate": 1.1530398322851153e-05,
-      "loss": 0.261,
+      "grad_norm": 7.976863861083984,
+      "learning_rate": 3.962264150943397e-05,
+      "loss": 0.3798,
       "step": 420
     },
     {
       "epoch": 8.0,
-      "eval_accuracy": 0.925275482093664,
-      "eval_loss": 0.21677668392658234,
-      "eval_runtime": 14.7132,
-      "eval_samples_per_second": 197.373,
-      "eval_steps_per_second": 6.185,
+      "eval_accuracy": 0.903236914600551,
+      "eval_loss": 0.26347047090530396,
+      "eval_runtime": 13.5658,
+      "eval_samples_per_second": 214.068,
+      "eval_steps_per_second": 6.708,
       "step": 424
     },
     {
       "epoch": 8.11320754716981,
-      "grad_norm": 5.5042643547058105,
-      "learning_rate": 1.0482180293501048e-05,
-      "loss": 0.2675,
+      "grad_norm": 6.626104354858398,
+      "learning_rate": 4.0566037735849064e-05,
+      "loss": 0.3583,
       "step": 430
     },
     {
       "epoch": 8.30188679245283,
-      "grad_norm": 4.880862236022949,
-      "learning_rate": 9.433962264150944e-06,
-      "loss": 0.2413,
+      "grad_norm": 6.100863933563232,
+      "learning_rate": 4.150943396226415e-05,
+      "loss": 0.3201,
       "step": 440
     },
     {
       "epoch": 8.49056603773585,
-      "grad_norm": 3.9594244956970215,
-      "learning_rate": 8.38574423480084e-06,
-      "loss": 0.2577,
+      "grad_norm": 5.2761969566345215,
+      "learning_rate": 4.245283018867925e-05,
+      "loss": 0.3452,
       "step": 450
     },
     {
       "epoch": 8.679245283018869,
-      "grad_norm": 6.740973949432373,
-      "learning_rate": 7.337526205450735e-06,
-      "loss": 0.2595,
+      "grad_norm": 4.848881244659424,
+      "learning_rate": 4.3396226415094345e-05,
+      "loss": 0.3602,
       "step": 460
     },
     {
       "epoch": 8.867924528301886,
-      "grad_norm": 7.188548564910889,
-      "learning_rate": 6.289308176100629e-06,
-      "loss": 0.2526,
+      "grad_norm": 10.011419296264648,
+      "learning_rate": 4.433962264150944e-05,
+      "loss": 0.3303,
       "step": 470
     },
     {
       "epoch": 9.0,
-      "eval_accuracy": 0.9245867768595041,
-      "eval_loss": 0.20991356670856476,
-      "eval_runtime": 14.793,
-      "eval_samples_per_second": 196.31,
-      "eval_steps_per_second": 6.152,
+      "eval_accuracy": 0.9073691460055097,
+      "eval_loss": 0.27034133672714233,
+      "eval_runtime": 13.7374,
+      "eval_samples_per_second": 211.393,
+      "eval_steps_per_second": 6.624,
       "step": 477
     },
     {
       "epoch": 9.056603773584905,
-      "grad_norm": 8.5458402633667,
-      "learning_rate": 5.241090146750524e-06,
-      "loss": 0.2465,
+      "grad_norm": 8.56981372833252,
+      "learning_rate": 4.528301886792453e-05,
+      "loss": 0.31,
       "step": 480
     },
     {
       "epoch": 9.245283018867925,
-      "grad_norm": 4.695093154907227,
-      "learning_rate": 4.19287211740042e-06,
-      "loss": 0.2172,
+      "grad_norm": 7.121793270111084,
+      "learning_rate": 4.6226415094339625e-05,
+      "loss": 0.3098,
       "step": 490
     },
     {
       "epoch": 9.433962264150944,
-      "grad_norm": 3.67899751663208,
-      "learning_rate": 3.1446540880503146e-06,
-      "loss": 0.2635,
+      "grad_norm": 4.458034515380859,
+      "learning_rate": 4.716981132075472e-05,
+      "loss": 0.3495,
       "step": 500
     },
     {
       "epoch": 9.622641509433961,
-      "grad_norm": 4.685281276702881,
-      "learning_rate": 2.09643605870021e-06,
-      "loss": 0.254,
+      "grad_norm": 6.7864484786987305,
+      "learning_rate": 4.811320754716982e-05,
+      "loss": 0.3383,
       "step": 510
     },
     {
       "epoch": 9.81132075471698,
-      "grad_norm": 5.518146991729736,
-      "learning_rate": 1.048218029350105e-06,
-      "loss": 0.2574,
+      "grad_norm": 6.478758811950684,
+      "learning_rate": 4.9056603773584906e-05,
+      "loss": 0.316,
       "step": 520
     },
     {
       "epoch": 10.0,
-      "grad_norm": 5.643453598022461,
-      "learning_rate": 0.0,
-      "loss": 0.2767,
+      "grad_norm": 6.849891185760498,
+      "learning_rate": 5e-05,
+      "loss": 0.3346,
       "step": 530
     },
     {
       "epoch": 10.0,
-      "eval_accuracy": 0.9297520661157025,
-      "eval_loss": 0.19896750152111053,
-      "eval_runtime": 15.5471,
-      "eval_samples_per_second": 186.787,
-      "eval_steps_per_second": 5.853,
+      "eval_accuracy": 0.9004820936639119,
+      "eval_loss": 0.25653818249702454,
+      "eval_runtime": 13.7209,
+      "eval_samples_per_second": 211.649,
+      "eval_steps_per_second": 6.632,
       "step": 530
     },
     {
-      "epoch": 10.0,
-      "step": 530,
-      "total_flos": 1.6836842977571635e+18,
-      "train_loss": 0.419760063009442,
-      "train_runtime": 917.7632,
-      "train_samples_per_second": 73.799,
-      "train_steps_per_second": 0.577
+      "epoch": 10.18867924528302,
+      "grad_norm": 5.824264049530029,
+      "learning_rate": 4.989517819706499e-05,
+      "loss": 0.3262,
+      "step": 540
+    },
+    {
+      "epoch": 10.377358490566039,
+      "grad_norm": 9.913765907287598,
+      "learning_rate": 4.979035639412998e-05,
+      "loss": 0.2917,
+      "step": 550
+    },
+    {
+      "epoch": 10.566037735849056,
+      "grad_norm": 8.706583976745605,
+      "learning_rate": 4.968553459119497e-05,
+      "loss": 0.325,
+      "step": 560
+    },
+    {
+      "epoch": 10.754716981132075,
+      "grad_norm": 4.388184070587158,
+      "learning_rate": 4.958071278825996e-05,
+      "loss": 0.2934,
+      "step": 570
+    },
+    {
+      "epoch": 10.943396226415095,
+      "grad_norm": 7.9404988288879395,
+      "learning_rate": 4.947589098532495e-05,
+      "loss": 0.2971,
+      "step": 580
+    },
+    {
+      "epoch": 11.0,
+      "eval_accuracy": 0.931129476584022,
+      "eval_loss": 0.21819160878658295,
+      "eval_runtime": 13.8344,
+      "eval_samples_per_second": 209.911,
+      "eval_steps_per_second": 6.578,
+      "step": 583
+    },
+    {
+      "epoch": 11.132075471698114,
+      "grad_norm": 4.344425678253174,
+      "learning_rate": 4.937106918238994e-05,
+      "loss": 0.25,
+      "step": 590
+    },
+    {
+      "epoch": 11.320754716981131,
+      "grad_norm": 7.321628570556641,
+      "learning_rate": 4.9266247379454926e-05,
+      "loss": 0.3084,
+      "step": 600
+    },
+    {
+      "epoch": 11.50943396226415,
+      "grad_norm": 12.875089645385742,
+      "learning_rate": 4.916142557651992e-05,
+      "loss": 0.3069,
+      "step": 610
+    },
+    {
+      "epoch": 11.69811320754717,
+      "grad_norm": 6.05894660949707,
+      "learning_rate": 4.9056603773584906e-05,
+      "loss": 0.2905,
+      "step": 620
+    },
+    {
+      "epoch": 11.88679245283019,
+      "grad_norm": 10.023378372192383,
+      "learning_rate": 4.8951781970649894e-05,
+      "loss": 0.2992,
+      "step": 630
+    },
+    {
+      "epoch": 12.0,
+      "eval_accuracy": 0.9256198347107438,
+      "eval_loss": 0.22397693991661072,
+      "eval_runtime": 13.8485,
+      "eval_samples_per_second": 209.698,
+      "eval_steps_per_second": 6.571,
+      "step": 636
+    },
+    {
+      "epoch": 12.075471698113208,
+      "grad_norm": 6.025568962097168,
+      "learning_rate": 4.884696016771489e-05,
+      "loss": 0.2936,
+      "step": 640
+    },
+    {
+      "epoch": 12.264150943396226,
+      "grad_norm": 5.965641021728516,
+      "learning_rate": 4.8742138364779875e-05,
+      "loss": 0.266,
+      "step": 650
+    },
+    {
+      "epoch": 12.452830188679245,
+      "grad_norm": 4.047191143035889,
+      "learning_rate": 4.863731656184486e-05,
+      "loss": 0.2696,
+      "step": 660
+    },
+    {
+      "epoch": 12.641509433962264,
+      "grad_norm": 4.262639999389648,
+      "learning_rate": 4.8532494758909855e-05,
+      "loss": 0.2642,
+      "step": 670
+    },
+    {
+      "epoch": 12.830188679245284,
+      "grad_norm": 4.320269584655762,
+      "learning_rate": 4.842767295597484e-05,
+      "loss": 0.2637,
+      "step": 680
+    },
+    {
+      "epoch": 13.0,
+      "eval_accuracy": 0.9238980716253443,
+      "eval_loss": 0.2130652368068695,
+      "eval_runtime": 13.8031,
+      "eval_samples_per_second": 210.388,
+      "eval_steps_per_second": 6.593,
+      "step": 689
+    },
+    {
+      "epoch": 13.018867924528301,
+      "grad_norm": 6.210031032562256,
+      "learning_rate": 4.8322851153039836e-05,
+      "loss": 0.2755,
+      "step": 690
+    },
+    {
+      "epoch": 13.20754716981132,
+      "grad_norm": 5.736047744750977,
+      "learning_rate": 4.8218029350104823e-05,
+      "loss": 0.2373,
+      "step": 700
+    },
+    {
+      "epoch": 13.39622641509434,
+      "grad_norm": 5.341845989227295,
+      "learning_rate": 4.811320754716982e-05,
+      "loss": 0.2462,
+      "step": 710
+    },
+    {
+      "epoch": 13.584905660377359,
+      "grad_norm": 4.025936126708984,
+      "learning_rate": 4.8008385744234804e-05,
+      "loss": 0.2491,
+      "step": 720
+    },
+    {
+      "epoch": 13.773584905660378,
+      "grad_norm": 4.850243091583252,
+      "learning_rate": 4.79035639412998e-05,
+      "loss": 0.2437,
+      "step": 730
+    },
+    {
+      "epoch": 13.962264150943396,
+      "grad_norm": 4.280818939208984,
+      "learning_rate": 4.7798742138364785e-05,
+      "loss": 0.2653,
+      "step": 740
+    },
+    {
+      "epoch": 14.0,
+      "eval_accuracy": 0.9397382920110193,
+      "eval_loss": 0.18005970120429993,
+      "eval_runtime": 13.9135,
+      "eval_samples_per_second": 208.718,
+      "eval_steps_per_second": 6.54,
+      "step": 742
+    },
+    {
+      "epoch": 14.150943396226415,
+      "grad_norm": 4.664927959442139,
+      "learning_rate": 4.769392033542977e-05,
+      "loss": 0.2444,
+      "step": 750
+    },
+    {
+      "epoch": 14.339622641509434,
+      "grad_norm": 4.106526851654053,
+      "learning_rate": 4.7589098532494766e-05,
+      "loss": 0.2468,
+      "step": 760
+    },
+    {
+      "epoch": 14.528301886792454,
+      "grad_norm": 5.82990837097168,
+      "learning_rate": 4.7484276729559753e-05,
+      "loss": 0.2534,
+      "step": 770
+    },
+    {
+      "epoch": 14.716981132075471,
+      "grad_norm": 5.992727279663086,
+      "learning_rate": 4.737945492662474e-05,
+      "loss": 0.2247,
+      "step": 780
+    },
+    {
+      "epoch": 14.90566037735849,
+      "grad_norm": 5.288634300231934,
+      "learning_rate": 4.7274633123689734e-05,
+      "loss": 0.2472,
+      "step": 790
+    },
+    {
+      "epoch": 15.0,
+      "eval_accuracy": 0.9376721763085399,
+      "eval_loss": 0.18071921169757843,
+      "eval_runtime": 13.77,
+      "eval_samples_per_second": 210.893,
+      "eval_steps_per_second": 6.609,
+      "step": 795
+    },
+    {
+      "epoch": 15.09433962264151,
+      "grad_norm": 5.116296768188477,
+      "learning_rate": 4.716981132075472e-05,
+      "loss": 0.2706,
+      "step": 800
+    },
+    {
+      "epoch": 15.283018867924529,
+      "grad_norm": 4.876150131225586,
+      "learning_rate": 4.706498951781971e-05,
+      "loss": 0.2174,
+      "step": 810
+    },
+    {
+      "epoch": 15.471698113207546,
+      "grad_norm": 4.544086456298828,
+      "learning_rate": 4.69601677148847e-05,
+      "loss": 0.1944,
+      "step": 820
+    },
+    {
+      "epoch": 15.660377358490566,
+      "grad_norm": 7.460907459259033,
+      "learning_rate": 4.685534591194969e-05,
+      "loss": 0.2432,
+      "step": 830
+    },
+    {
+      "epoch": 15.849056603773585,
+      "grad_norm": 5.503137111663818,
+      "learning_rate": 4.6750524109014677e-05,
+      "loss": 0.2263,
+      "step": 840
+    },
+    {
+      "epoch": 16.0,
+      "eval_accuracy": 0.946625344352617,
+      "eval_loss": 0.16115374863147736,
+      "eval_runtime": 13.8807,
+      "eval_samples_per_second": 209.211,
+      "eval_steps_per_second": 6.556,
+      "step": 848
+    },
+    {
+      "epoch": 16.037735849056602,
+      "grad_norm": 3.875622272491455,
+      "learning_rate": 4.664570230607967e-05,
+      "loss": 0.2365,
+      "step": 850
+    },
+    {
+      "epoch": 16.22641509433962,
+      "grad_norm": 4.338544845581055,
+      "learning_rate": 4.654088050314466e-05,
+      "loss": 0.1973,
+      "step": 860
+    },
+    {
+      "epoch": 16.41509433962264,
+      "grad_norm": 6.0123724937438965,
+      "learning_rate": 4.6436058700209645e-05,
+      "loss": 0.2341,
+      "step": 870
+    },
+    {
+      "epoch": 16.60377358490566,
+      "grad_norm": 3.352548837661743,
+      "learning_rate": 4.633123689727464e-05,
+      "loss": 0.195,
+      "step": 880
+    },
+    {
+      "epoch": 16.79245283018868,
+      "grad_norm": 3.6743788719177246,
+      "learning_rate": 4.6226415094339625e-05,
+      "loss": 0.2149,
+      "step": 890
+    },
+    {
+      "epoch": 16.9811320754717,
+      "grad_norm": 3.634098768234253,
+      "learning_rate": 4.612159329140461e-05,
+      "loss": 0.1786,
+      "step": 900
+    },
+    {
+      "epoch": 17.0,
+      "eval_accuracy": 0.9418044077134986,
+      "eval_loss": 0.17354166507720947,
+      "eval_runtime": 13.8691,
+      "eval_samples_per_second": 209.387,
+      "eval_steps_per_second": 6.561,
+      "step": 901
+    },
+    {
+      "epoch": 17.169811320754718,
+      "grad_norm": 4.430752277374268,
+      "learning_rate": 4.6016771488469606e-05,
+      "loss": 0.205,
+      "step": 910
+    },
+    {
+      "epoch": 17.358490566037737,
+      "grad_norm": 3.114032030105591,
+      "learning_rate": 4.5911949685534594e-05,
+      "loss": 0.2002,
+      "step": 920
+    },
+    {
+      "epoch": 17.547169811320753,
+      "grad_norm": 5.479780197143555,
+      "learning_rate": 4.580712788259958e-05,
+      "loss": 0.2224,
+      "step": 930
+    },
+    {
+      "epoch": 17.735849056603772,
+      "grad_norm": 5.793611526489258,
+      "learning_rate": 4.570230607966457e-05,
+      "loss": 0.1961,
+      "step": 940
+    },
+    {
+      "epoch": 17.92452830188679,
+      "grad_norm": 3.505262613296509,
+      "learning_rate": 4.559748427672956e-05,
+      "loss": 0.2103,
+      "step": 950
+    },
+    {
+      "epoch": 18.0,
+      "eval_accuracy": 0.9462809917355371,
+      "eval_loss": 0.1786232590675354,
+      "eval_runtime": 14.016,
+      "eval_samples_per_second": 207.192,
+      "eval_steps_per_second": 6.493,
+      "step": 954
+    },
+    {
+      "epoch": 18.11320754716981,
+      "grad_norm": 3.520742177963257,
+      "learning_rate": 4.549266247379455e-05,
+      "loss": 0.2158,
+      "step": 960
+    },
+    {
+      "epoch": 18.30188679245283,
+      "grad_norm": 5.575334072113037,
+      "learning_rate": 4.5387840670859536e-05,
+      "loss": 0.2004,
+      "step": 970
+    },
+    {
+      "epoch": 18.49056603773585,
+      "grad_norm": 4.227255344390869,
+      "learning_rate": 4.528301886792453e-05,
+      "loss": 0.173,
+      "step": 980
+    },
+    {
+      "epoch": 18.67924528301887,
+      "grad_norm": 4.518486022949219,
+      "learning_rate": 4.517819706498952e-05,
+      "loss": 0.1898,
+      "step": 990
+    },
+    {
+      "epoch": 18.867924528301888,
+      "grad_norm": 3.9365193843841553,
+      "learning_rate": 4.5073375262054504e-05,
+      "loss": 0.1725,
+      "step": 1000
+    },
+    {
+      "epoch": 19.0,
+      "eval_accuracy": 0.9473140495867769,
+      "eval_loss": 0.16307222843170166,
+      "eval_runtime": 14.0184,
+      "eval_samples_per_second": 207.157,
+      "eval_steps_per_second": 6.491,
+      "step": 1007
+    },
+    {
+      "epoch": 19.056603773584907,
+      "grad_norm": 4.218902587890625,
+      "learning_rate": 4.49685534591195e-05,
+      "loss": 0.1953,
+      "step": 1010
+    },
+    {
+      "epoch": 19.245283018867923,
+      "grad_norm": 5.483488082885742,
+      "learning_rate": 4.4863731656184485e-05,
+      "loss": 0.1648,
+      "step": 1020
+    },
+    {
+      "epoch": 19.433962264150942,
+      "grad_norm": 5.378849029541016,
+      "learning_rate": 4.475890985324948e-05,
+      "loss": 0.1807,
+      "step": 1030
+    },
+    {
+      "epoch": 19.62264150943396,
+      "grad_norm": 5.601091384887695,
+      "learning_rate": 4.4654088050314466e-05,
+      "loss": 0.2101,
+      "step": 1040
+    },
+    {
+      "epoch": 19.81132075471698,
+      "grad_norm": 2.9287290573120117,
+      "learning_rate": 4.454926624737946e-05,
+      "loss": 0.1628,
+      "step": 1050
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 6.378795146942139,
+      "learning_rate": 4.4444444444444447e-05,
+      "loss": 0.1787,
+      "step": 1060
+    },
+    {
+      "epoch": 20.0,
+      "eval_accuracy": 0.953168044077135,
+      "eval_loss": 0.14394132792949677,
+      "eval_runtime": 13.9963,
+      "eval_samples_per_second": 207.483,
+      "eval_steps_per_second": 6.502,
+      "step": 1060
+    },
+    {
+      "epoch": 20.18867924528302,
+      "grad_norm": 5.156091213226318,
+      "learning_rate": 4.433962264150944e-05,
+      "loss": 0.1518,
+      "step": 1070
+    },
+    {
+      "epoch": 20.37735849056604,
+      "grad_norm": 5.145476818084717,
+      "learning_rate": 4.423480083857443e-05,
+      "loss": 0.1779,
+      "step": 1080
+    },
+    {
+      "epoch": 20.566037735849058,
+      "grad_norm": 4.119344711303711,
+      "learning_rate": 4.4129979035639415e-05,
+      "loss": 0.2038,
+      "step": 1090
+    },
+    {
+      "epoch": 20.754716981132077,
+      "grad_norm": 4.134652137756348,
+      "learning_rate": 4.402515723270441e-05,
+      "loss": 0.1713,
+      "step": 1100
+    },
+    {
+      "epoch": 20.943396226415093,
+      "grad_norm": 5.130320072174072,
+      "learning_rate": 4.3920335429769396e-05,
+      "loss": 0.1924,
+      "step": 1110
+    },
+    {
+      "epoch": 21.0,
+      "eval_accuracy": 0.9504132231404959,
+      "eval_loss": 0.13879624009132385,
+      "eval_runtime": 13.8513,
+      "eval_samples_per_second": 209.655,
+      "eval_steps_per_second": 6.57,
+      "step": 1113
+    },
+    {
+      "epoch": 21.132075471698112,
+      "grad_norm": 5.733437538146973,
+      "learning_rate": 4.381551362683438e-05,
+      "loss": 0.164,
+      "step": 1120
+    },
+    {
+      "epoch": 21.32075471698113,
+      "grad_norm": 3.7996554374694824,
+      "learning_rate": 4.3710691823899376e-05,
+      "loss": 0.1479,
+      "step": 1130
+    },
+    {
+      "epoch": 21.50943396226415,
+      "grad_norm": 4.9020891189575195,
+      "learning_rate": 4.3605870020964364e-05,
+      "loss": 0.1657,
+      "step": 1140
+    },
+    {
+      "epoch": 21.69811320754717,
+      "grad_norm": 4.601658344268799,
+      "learning_rate": 4.350104821802935e-05,
+      "loss": 0.1644,
+      "step": 1150
+    },
+    {
+      "epoch": 21.88679245283019,
+      "grad_norm": 5.638570785522461,
+      "learning_rate": 4.3396226415094345e-05,
+      "loss": 0.1662,
+      "step": 1160
+    },
+    {
+      "epoch": 22.0,
+      "eval_accuracy": 0.9507575757575758,
+      "eval_loss": 0.1469692885875702,
+      "eval_runtime": 13.9412,
+      "eval_samples_per_second": 208.304,
+      "eval_steps_per_second": 6.527,
+      "step": 1166
+    },
+    {
+      "epoch": 22.07547169811321,
+      "grad_norm": 7.325222015380859,
+      "learning_rate": 4.329140461215933e-05,
+      "loss": 0.1926,
+      "step": 1170
+    },
+    {
+      "epoch": 22.264150943396228,
+      "grad_norm": 5.337314128875732,
+      "learning_rate": 4.318658280922432e-05,
+      "loss": 0.1908,
+      "step": 1180
+    },
+    {
+      "epoch": 22.452830188679247,
+      "grad_norm": 3.660973072052002,
+      "learning_rate": 4.308176100628931e-05,
+      "loss": 0.1663,
+      "step": 1190
+    },
+    {
+      "epoch": 22.641509433962263,
+      "grad_norm": 7.075264930725098,
+      "learning_rate": 4.29769392033543e-05,
+      "loss": 0.1778,
+      "step": 1200
+    },
+    {
+      "epoch": 22.830188679245282,
+      "grad_norm": 2.9561445713043213,
+      "learning_rate": 4.287211740041929e-05,
+      "loss": 0.1724,
+      "step": 1210
+    },
+    {
+      "epoch": 23.0,
+      "eval_accuracy": 0.949724517906336,
+      "eval_loss": 0.15376034379005432,
+      "eval_runtime": 13.9076,
+      "eval_samples_per_second": 208.807,
+      "eval_steps_per_second": 6.543,
+      "step": 1219
+    },
+    {
+      "epoch": 23.0188679245283,
+      "grad_norm": 6.19976806640625,
+      "learning_rate": 4.276729559748428e-05,
+      "loss": 0.1867,
+      "step": 1220
+    },
+    {
+      "epoch": 23.20754716981132,
+      "grad_norm": 5.15887451171875,
+      "learning_rate": 4.266247379454927e-05,
+      "loss": 0.1809,
+      "step": 1230
+    },
+    {
+      "epoch": 23.39622641509434,
+      "grad_norm": 5.114108562469482,
+      "learning_rate": 4.2557651991614255e-05,
+      "loss": 0.1636,
+      "step": 1240
+    },
+    {
+      "epoch": 23.58490566037736,
+      "grad_norm": 5.130180358886719,
+      "learning_rate": 4.245283018867925e-05,
+      "loss": 0.1835,
+      "step": 1250
+    },
+    {
+      "epoch": 23.77358490566038,
+      "grad_norm": 5.375781536102295,
+      "learning_rate": 4.2348008385744236e-05,
+      "loss": 0.1764,
+      "step": 1260
+    },
+    {
+      "epoch": 23.962264150943398,
+      "grad_norm": 3.6640334129333496,
+      "learning_rate": 4.224318658280922e-05,
+      "loss": 0.1633,
+      "step": 1270
+    },
+    {
+      "epoch": 24.0,
+      "eval_accuracy": 0.9383608815426997,
+      "eval_loss": 0.17309589684009552,
+      "eval_runtime": 22.6597,
+      "eval_samples_per_second": 128.157,
+      "eval_steps_per_second": 4.016,
+      "step": 1272
+    },
+    {
+      "epoch": 24.150943396226417,
+      "grad_norm": 4.2943196296691895,
+      "learning_rate": 4.213836477987422e-05,
+      "loss": 0.1458,
+      "step": 1280
+    },
+    {
+      "epoch": 24.339622641509433,
+      "grad_norm": 4.514562129974365,
+      "learning_rate": 4.2033542976939204e-05,
+      "loss": 0.159,
+      "step": 1290
+    },
+    {
+      "epoch": 24.528301886792452,
+      "grad_norm": 7.0057454109191895,
+      "learning_rate": 4.192872117400419e-05,
+      "loss": 0.1507,
+      "step": 1300
+    },
+    {
+      "epoch": 24.71698113207547,
+      "grad_norm": 5.325743198394775,
+      "learning_rate": 4.1823899371069185e-05,
+      "loss": 0.201,
+      "step": 1310
+    },
+    {
+      "epoch": 24.90566037735849,
+      "grad_norm": 4.204189300537109,
+      "learning_rate": 4.171907756813417e-05,
+      "loss": 0.174,
+      "step": 1320
+    },
+    {
+      "epoch": 25.0,
+      "eval_accuracy": 0.9538567493112947,
+      "eval_loss": 0.15548603236675262,
+      "eval_runtime": 20.4244,
+      "eval_samples_per_second": 142.183,
+      "eval_steps_per_second": 4.455,
+      "step": 1325
+    },
+    {
+      "epoch": 25.09433962264151,
+      "grad_norm": 5.918984413146973,
+      "learning_rate": 4.161425576519916e-05,
+      "loss": 0.1604,
+      "step": 1330
+    },
+    {
+      "epoch": 25.28301886792453,
+      "grad_norm": 3.3942575454711914,
+      "learning_rate": 4.150943396226415e-05,
+      "loss": 0.1297,
+      "step": 1340
+    },
+    {
+      "epoch": 25.471698113207548,
+      "grad_norm": 2.6150972843170166,
+      "learning_rate": 4.140461215932914e-05,
+      "loss": 0.1407,
+      "step": 1350
+    },
+    {
+      "epoch": 25.660377358490567,
+      "grad_norm": 4.893429756164551,
+      "learning_rate": 4.129979035639413e-05,
+      "loss": 0.1637,
+      "step": 1360
+    },
+    {
+      "epoch": 25.849056603773583,
+      "grad_norm": 5.928979396820068,
+      "learning_rate": 4.119496855345912e-05,
+      "loss": 0.1657,
+      "step": 1370
+    },
+    {
+      "epoch": 26.0,
+      "eval_accuracy": 0.9493801652892562,
+      "eval_loss": 0.15420162677764893,
+      "eval_runtime": 20.5055,
+      "eval_samples_per_second": 141.62,
+      "eval_steps_per_second": 4.438,
+      "step": 1378
+    },
+    {
+      "epoch": 26.037735849056602,
+      "grad_norm": 3.1796255111694336,
+      "learning_rate": 4.109014675052411e-05,
+      "loss": 0.1548,
+      "step": 1380
+    },
+    {
+      "epoch": 26.22641509433962,
+      "grad_norm": 3.422128438949585,
+      "learning_rate": 4.09853249475891e-05,
+      "loss": 0.1379,
+      "step": 1390
+    },
+    {
+      "epoch": 26.41509433962264,
+      "grad_norm": 4.308591365814209,
+      "learning_rate": 4.088050314465409e-05,
+      "loss": 0.1588,
+      "step": 1400
+    },
+    {
+      "epoch": 26.60377358490566,
+      "grad_norm": 5.928549766540527,
+      "learning_rate": 4.077568134171908e-05,
+      "loss": 0.1793,
+      "step": 1410
+    },
+    {
+      "epoch": 26.79245283018868,
+      "grad_norm": 4.973033905029297,
+      "learning_rate": 4.067085953878407e-05,
+      "loss": 0.114,
+      "step": 1420
+    },
+    {
+      "epoch": 26.9811320754717,
+      "grad_norm": 3.7284395694732666,
+      "learning_rate": 4.0566037735849064e-05,
+      "loss": 0.1513,
+      "step": 1430
+    },
+    {
+      "epoch": 27.0,
+      "eval_accuracy": 0.9507575757575758,
+      "eval_loss": 0.15260200202465057,
+      "eval_runtime": 20.5592,
+      "eval_samples_per_second": 141.251,
+      "eval_steps_per_second": 4.426,
+      "step": 1431
+    },
+    {
+      "epoch": 27.169811320754718,
+      "grad_norm": 4.743962287902832,
+      "learning_rate": 4.046121593291405e-05,
+      "loss": 0.1396,
+      "step": 1440
+    },
+    {
+      "epoch": 27.358490566037737,
+      "grad_norm": 5.549553871154785,
+      "learning_rate": 4.035639412997904e-05,
+      "loss": 0.1556,
+      "step": 1450
+    },
+    {
+      "epoch": 27.547169811320753,
+      "grad_norm": 4.103055953979492,
+      "learning_rate": 4.025157232704403e-05,
+      "loss": 0.1448,
+      "step": 1460
+    },
+    {
+      "epoch": 27.735849056603772,
+      "grad_norm": 3.5048537254333496,
+      "learning_rate": 4.014675052410902e-05,
+      "loss": 0.1536,
+      "step": 1470
+    },
+    {
+      "epoch": 27.92452830188679,
+      "grad_norm": 3.7681641578674316,
+      "learning_rate": 4.0041928721174006e-05,
+      "loss": 0.126,
+      "step": 1480
+    },
+    {
+      "epoch": 28.0,
+      "eval_accuracy": 0.9511019283746557,
+      "eval_loss": 0.15600712597370148,
+      "eval_runtime": 20.7331,
+      "eval_samples_per_second": 140.066,
+      "eval_steps_per_second": 4.389,
+      "step": 1484
+    },
+    {
+      "epoch": 28.11320754716981,
+      "grad_norm": 3.522587299346924,
+      "learning_rate": 3.9937106918239e-05,
+      "loss": 0.1438,
+      "step": 1490
+    },
+    {
+      "epoch": 28.30188679245283,
+      "grad_norm": 3.7148783206939697,
+      "learning_rate": 3.983228511530399e-05,
+      "loss": 0.154,
+      "step": 1500
+    },
+    {
+      "epoch": 28.49056603773585,
+      "grad_norm": 7.454770565032959,
+      "learning_rate": 3.9727463312368974e-05,
+      "loss": 0.1783,
+      "step": 1510
+    },
+    {
+      "epoch": 28.67924528301887,
+      "grad_norm": 3.996530532836914,
+      "learning_rate": 3.962264150943397e-05,
+      "loss": 0.1263,
+      "step": 1520
+    },
+    {
+      "epoch": 28.867924528301888,
+      "grad_norm": 5.041396141052246,
+      "learning_rate": 3.9517819706498955e-05,
+      "loss": 0.1508,
+      "step": 1530
+    },
+    {
+      "epoch": 29.0,
+      "eval_accuracy": 0.9480027548209367,
+      "eval_loss": 0.16071830689907074,
+      "eval_runtime": 21.0223,
+      "eval_samples_per_second": 138.139,
+      "eval_steps_per_second": 4.329,
+      "step": 1537
+    },
+    {
+      "epoch": 29.056603773584907,
+      "grad_norm": 3.525146007537842,
+      "learning_rate": 3.941299790356394e-05,
+      "loss": 0.1421,
+      "step": 1540
+    },
+    {
+      "epoch": 29.245283018867923,
+      "grad_norm": 5.157039642333984,
+      "learning_rate": 3.9308176100628936e-05,
+      "loss": 0.1276,
+      "step": 1550
+    },
+    {
+      "epoch": 29.433962264150942,
+      "grad_norm": 6.565103054046631,
+      "learning_rate": 3.920335429769392e-05,
+      "loss": 0.1662,
+      "step": 1560
+    },
+    {
+      "epoch": 29.62264150943396,
+      "grad_norm": 3.140432357788086,
+      "learning_rate": 3.909853249475891e-05,
+      "loss": 0.1446,
+      "step": 1570
+    },
+    {
+      "epoch": 29.81132075471698,
+      "grad_norm": 3.157646656036377,
+      "learning_rate": 3.8993710691823904e-05,
+      "loss": 0.1227,
+      "step": 1580
+    },
+    {
+      "epoch": 30.0,
+      "grad_norm": 3.9616310596466064,
+      "learning_rate": 3.888888888888889e-05,
+      "loss": 0.1368,
+      "step": 1590
+    },
+    {
+      "epoch": 30.0,
+      "eval_accuracy": 0.9435261707988981,
+      "eval_loss": 0.17288929224014282,
+      "eval_runtime": 20.8012,
+      "eval_samples_per_second": 139.608,
+      "eval_steps_per_second": 4.375,
+      "step": 1590
+    },
+    {
+      "epoch": 30.18867924528302,
+      "grad_norm": 2.845510482788086,
+      "learning_rate": 3.878406708595388e-05,
+      "loss": 0.1336,
+      "step": 1600
+    },
+    {
+      "epoch": 30.37735849056604,
+      "grad_norm": 3.7001242637634277,
+      "learning_rate": 3.867924528301887e-05,
+      "loss": 0.1202,
+      "step": 1610
+    },
+    {
+      "epoch": 30.566037735849058,
+      "grad_norm": 2.4213449954986572,
+      "learning_rate": 3.857442348008386e-05,
+      "loss": 0.1529,
+      "step": 1620
+    },
+    {
+      "epoch": 30.754716981132077,
+      "grad_norm": 3.6384825706481934,
+      "learning_rate": 3.8469601677148846e-05,
+      "loss": 0.1128,
+      "step": 1630
+    },
+    {
+      "epoch": 30.943396226415093,
+      "grad_norm": 9.870887756347656,
+      "learning_rate": 3.836477987421384e-05,
+      "loss": 0.1166,
+      "step": 1640
+    },
+    {
+      "epoch": 31.0,
+      "eval_accuracy": 0.953168044077135,
+      "eval_loss": 0.155534565448761,
+      "eval_runtime": 20.9232,
+      "eval_samples_per_second": 138.793,
+      "eval_steps_per_second": 4.349,
+      "step": 1643
+    },
+    {
+      "epoch": 31.132075471698112,
+      "grad_norm": 7.167893886566162,
+      "learning_rate": 3.825995807127883e-05,
+      "loss": 0.1234,
+      "step": 1650
+    },
+    {
+      "epoch": 31.32075471698113,
+      "grad_norm": 5.442743301391602,
+      "learning_rate": 3.8155136268343814e-05,
+      "loss": 0.1565,
+      "step": 1660
+    },
+    {
+      "epoch": 31.50943396226415,
+      "grad_norm": 4.499869346618652,
+      "learning_rate": 3.805031446540881e-05,
+      "loss": 0.1191,
+      "step": 1670
+    },
+    {
+      "epoch": 31.69811320754717,
+      "grad_norm": 4.722647666931152,
+      "learning_rate": 3.7945492662473795e-05,
+      "loss": 0.1239,
+      "step": 1680
+    },
+    {
+      "epoch": 31.88679245283019,
+      "grad_norm": 3.9234910011291504,
+      "learning_rate": 3.784067085953878e-05,
+      "loss": 0.1076,
+      "step": 1690
+    },
+    {
+      "epoch": 32.0,
+      "eval_accuracy": 0.9579889807162535,
+      "eval_loss": 0.14003022015094757,
+      "eval_runtime": 21.2564,
+      "eval_samples_per_second": 136.618,
+      "eval_steps_per_second": 4.281,
+      "step": 1696
+    },
+    {
+      "epoch": 32.075471698113205,
+      "grad_norm": 5.3656721115112305,
+      "learning_rate": 3.7735849056603776e-05,
+      "loss": 0.1086,
+      "step": 1700
+    },
+    {
+      "epoch": 32.264150943396224,
+      "grad_norm": 2.3153514862060547,
+      "learning_rate": 3.763102725366876e-05,
+      "loss": 0.1109,
+      "step": 1710
+    },
+    {
+      "epoch": 32.45283018867924,
+      "grad_norm": 6.487193584442139,
+      "learning_rate": 3.752620545073376e-05,
+      "loss": 0.1296,
+      "step": 1720
+    },
+    {
+      "epoch": 32.64150943396226,
+      "grad_norm": 4.362462997436523,
+      "learning_rate": 3.7421383647798744e-05,
+      "loss": 0.1078,
+      "step": 1730
+    },
+    {
+      "epoch": 32.83018867924528,
+      "grad_norm": 4.543455123901367,
+      "learning_rate": 3.731656184486374e-05,
+      "loss": 0.1189,
+      "step": 1740
+    },
+    {
+      "epoch": 33.0,
+      "eval_accuracy": 0.9590220385674931,
+      "eval_loss": 0.14192205667495728,
+      "eval_runtime": 21.093,
+      "eval_samples_per_second": 137.676,
+      "eval_steps_per_second": 4.314,
+      "step": 1749
+    },
+    {
+      "epoch": 33.0188679245283,
+      "grad_norm": 3.2465994358062744,
+      "learning_rate": 3.7211740041928725e-05,
+      "loss": 0.1553,
+      "step": 1750
+    },
+    {
+      "epoch": 33.20754716981132,
+      "grad_norm": 3.4110634326934814,
+      "learning_rate": 3.710691823899371e-05,
+      "loss": 0.1123,
+      "step": 1760
+    },
+    {
+      "epoch": 33.39622641509434,
+      "grad_norm": 6.8291802406311035,
+      "learning_rate": 3.7002096436058706e-05,
+      "loss": 0.1288,
+      "step": 1770
+    },
+    {
+      "epoch": 33.58490566037736,
+      "grad_norm": 5.650381565093994,
+      "learning_rate": 3.689727463312369e-05,
+      "loss": 0.1498,
+      "step": 1780
+    },
+    {
+      "epoch": 33.77358490566038,
+      "grad_norm": 4.716341018676758,
+      "learning_rate": 3.679245283018868e-05,
+      "loss": 0.137,
+      "step": 1790
+    },
+    {
+      "epoch": 33.9622641509434,
+      "grad_norm": 4.079151153564453,
+      "learning_rate": 3.6687631027253674e-05,
+      "loss": 0.1512,
+      "step": 1800
+    },
+    {
+      "epoch": 34.0,
+      "eval_accuracy": 0.9579889807162535,
+      "eval_loss": 0.13637615740299225,
+      "eval_runtime": 21.2019,
+      "eval_samples_per_second": 136.969,
+      "eval_steps_per_second": 4.292,
+      "step": 1802
+    },
+    {
+      "epoch": 34.15094339622642,
+      "grad_norm": 3.646527051925659,
+      "learning_rate": 3.658280922431866e-05,
+      "loss": 0.1018,
+      "step": 1810
+    },
+    {
+      "epoch": 34.339622641509436,
+      "grad_norm": 6.170238971710205,
+      "learning_rate": 3.647798742138365e-05,
+      "loss": 0.1085,
+      "step": 1820
+    },
+    {
+      "epoch": 34.528301886792455,
+      "grad_norm": 3.559018611907959,
+      "learning_rate": 3.637316561844864e-05,
+      "loss": 0.115,
+      "step": 1830
+    },
+    {
+      "epoch": 34.716981132075475,
+      "grad_norm": 5.245954990386963,
+      "learning_rate": 3.626834381551363e-05,
+      "loss": 0.1183,
+      "step": 1840
+    },
+    {
+      "epoch": 34.905660377358494,
+      "grad_norm": 4.156854629516602,
+      "learning_rate": 3.6163522012578616e-05,
+      "loss": 0.1323,
+      "step": 1850
+    },
+    {
+      "epoch": 35.0,
+      "eval_accuracy": 0.9538567493112947,
+      "eval_loss": 0.14969290792942047,
+      "eval_runtime": 21.2793,
+      "eval_samples_per_second": 136.471,
+      "eval_steps_per_second": 4.276,
+      "step": 1855
+    },
+    {
+      "epoch": 35.094339622641506,
+      "grad_norm": 4.309362888336182,
+      "learning_rate": 3.605870020964361e-05,
+      "loss": 0.1268,
+      "step": 1860
+    },
+    {
+      "epoch": 35.283018867924525,
+      "grad_norm": 3.494779348373413,
+      "learning_rate": 3.59538784067086e-05,
+      "loss": 0.1339,
+      "step": 1870
+    },
+    {
+      "epoch": 35.471698113207545,
+      "grad_norm": 4.567333221435547,
+      "learning_rate": 3.5849056603773584e-05,
+      "loss": 0.1003,
+      "step": 1880
+    },
+    {
+      "epoch": 35.660377358490564,
+      "grad_norm": 5.073373317718506,
+      "learning_rate": 3.574423480083858e-05,
+      "loss": 0.1413,
+      "step": 1890
+    },
+    {
+      "epoch": 35.84905660377358,
+      "grad_norm": 3.0305075645446777,
+      "learning_rate": 3.5639412997903565e-05,
+      "loss": 0.1031,
+      "step": 1900
+    },
+    {
+      "epoch": 36.0,
+      "eval_accuracy": 0.9579889807162535,
+      "eval_loss": 0.14369449019432068,
+      "eval_runtime": 21.3446,
+      "eval_samples_per_second": 136.053,
+      "eval_steps_per_second": 4.263,
+      "step": 1908
+    },
+    {
+      "epoch": 36.0377358490566,
+      "grad_norm": 2.8232624530792236,
+      "learning_rate": 3.553459119496855e-05,
+      "loss": 0.1213,
+      "step": 1910
+    },
+    {
+      "epoch": 36.22641509433962,
+      "grad_norm": 3.2020962238311768,
+      "learning_rate": 3.5429769392033546e-05,
+      "loss": 0.0914,
+      "step": 1920
+    },
+    {
+      "epoch": 36.41509433962264,
+      "grad_norm": 4.236616134643555,
+      "learning_rate": 3.532494758909853e-05,
+      "loss": 0.1012,
+      "step": 1930
+    },
+    {
+      "epoch": 36.60377358490566,
+      "grad_norm": 4.817173480987549,
+      "learning_rate": 3.522012578616352e-05,
+      "loss": 0.1082,
+      "step": 1940
+    },
+    {
+      "epoch": 36.79245283018868,
+      "grad_norm": 3.9018845558166504,
+      "learning_rate": 3.5115303983228514e-05,
+      "loss": 0.1074,
+      "step": 1950
+    },
+    {
+      "epoch": 36.9811320754717,
+      "grad_norm": 5.009905815124512,
+      "learning_rate": 3.50104821802935e-05,
+      "loss": 0.1215,
+      "step": 1960
+    },
+    {
+      "epoch": 37.0,
+      "eval_accuracy": 0.9559228650137741,
+      "eval_loss": 0.14596055448055267,
+      "eval_runtime": 21.2599,
+      "eval_samples_per_second": 136.595,
+      "eval_steps_per_second": 4.28,
+      "step": 1961
+    },
+    {
+      "epoch": 37.16981132075472,
+      "grad_norm": 2.7128424644470215,
+      "learning_rate": 3.490566037735849e-05,
+      "loss": 0.1129,
+      "step": 1970
+    },
+    {
+      "epoch": 37.35849056603774,
+      "grad_norm": 4.401316165924072,
+      "learning_rate": 3.480083857442348e-05,
+      "loss": 0.1205,
+      "step": 1980
+    },
+    {
+      "epoch": 37.54716981132076,
+      "grad_norm": 5.2666778564453125,
+      "learning_rate": 3.469601677148847e-05,
+      "loss": 0.127,
+      "step": 1990
+    },
+    {
+      "epoch": 37.735849056603776,
+      "grad_norm": 2.7217955589294434,
+      "learning_rate": 3.4591194968553456e-05,
+      "loss": 0.1068,
+      "step": 2000
+    },
+    {
+      "epoch": 37.924528301886795,
+      "grad_norm": 3.7162227630615234,
+      "learning_rate": 3.448637316561845e-05,
+      "loss": 0.1069,
+      "step": 2010
+    },
+    {
+      "epoch": 38.0,
+      "eval_accuracy": 0.9600550964187328,
+      "eval_loss": 0.13623014092445374,
+      "eval_runtime": 21.3213,
+      "eval_samples_per_second": 136.202,
+      "eval_steps_per_second": 4.268,
+      "step": 2014
+    },
+    {
+      "epoch": 38.113207547169814,
+      "grad_norm": 4.025697708129883,
+      "learning_rate": 3.438155136268344e-05,
+      "loss": 0.1095,
+      "step": 2020
+    },
+    {
+      "epoch": 38.301886792452834,
+      "grad_norm": 9.188610076904297,
+      "learning_rate": 3.4276729559748424e-05,
+      "loss": 0.1233,
+      "step": 2030
+    },
+    {
+      "epoch": 38.490566037735846,
+      "grad_norm": 4.473904609680176,
+      "learning_rate": 3.417190775681342e-05,
+      "loss": 0.1043,
+      "step": 2040
+    },
+    {
+      "epoch": 38.679245283018865,
+      "grad_norm": 3.2655704021453857,
+      "learning_rate": 3.4067085953878405e-05,
+      "loss": 0.0906,
+      "step": 2050
+    },
+    {
+      "epoch": 38.867924528301884,
+      "grad_norm": 5.039525985717773,
+      "learning_rate": 3.39622641509434e-05,
+      "loss": 0.129,
+      "step": 2060
+    },
+    {
+      "epoch": 39.0,
+      "eval_accuracy": 0.9590220385674931,
+      "eval_loss": 0.1490471512079239,
+      "eval_runtime": 21.334,
+      "eval_samples_per_second": 136.12,
+      "eval_steps_per_second": 4.265,
+      "step": 2067
+    },
+    {
+      "epoch": 39.056603773584904,
+      "grad_norm": 3.194096326828003,
+      "learning_rate": 3.3857442348008386e-05,
+      "loss": 0.0893,
+      "step": 2070
+    },
+    {
+      "epoch": 39.24528301886792,
+      "grad_norm": 1.9600547552108765,
+      "learning_rate": 3.375262054507338e-05,
+      "loss": 0.103,
+      "step": 2080
+    },
+    {
+      "epoch": 39.43396226415094,
+      "grad_norm": 3.6588046550750732,
+      "learning_rate": 3.364779874213837e-05,
+      "loss": 0.1004,
+      "step": 2090
+    },
+    {
+      "epoch": 39.62264150943396,
+      "grad_norm": 5.536744594573975,
+      "learning_rate": 3.354297693920336e-05,
+      "loss": 0.1163,
+      "step": 2100
+    },
+    {
+      "epoch": 39.81132075471698,
+      "grad_norm": 2.2924985885620117,
+      "learning_rate": 3.343815513626835e-05,
+      "loss": 0.0862,
+      "step": 2110
+    },
+    {
+      "epoch": 40.0,
+      "grad_norm": 4.559634685516357,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.1202,
+      "step": 2120
+    },
+    {
+      "epoch": 40.0,
+      "eval_accuracy": 0.9545454545454546,
+      "eval_loss": 0.16163212060928345,
+      "eval_runtime": 21.3342,
+      "eval_samples_per_second": 136.119,
+      "eval_steps_per_second": 4.265,
+      "step": 2120
+    },
+    {
+      "epoch": 40.18867924528302,
+      "grad_norm": 3.6690125465393066,
+      "learning_rate": 3.322851153039833e-05,
+      "loss": 0.111,
+      "step": 2130
+    },
+    {
+      "epoch": 40.37735849056604,
+      "grad_norm": 4.344234943389893,
+      "learning_rate": 3.3123689727463316e-05,
+      "loss": 0.1072,
+      "step": 2140
+    },
+    {
+      "epoch": 40.56603773584906,
+      "grad_norm": 3.5178353786468506,
+      "learning_rate": 3.30188679245283e-05,
+      "loss": 0.1036,
+      "step": 2150
+    },
+    {
+      "epoch": 40.75471698113208,
+      "grad_norm": 4.753892421722412,
+      "learning_rate": 3.29140461215933e-05,
+      "loss": 0.1173,
+      "step": 2160
+    },
+    {
+      "epoch": 40.943396226415096,
+      "grad_norm": 3.2936408519744873,
+      "learning_rate": 3.2809224318658284e-05,
+      "loss": 0.1011,
+      "step": 2170
+    },
+    {
+      "epoch": 41.0,
+      "eval_accuracy": 0.9569559228650137,
+      "eval_loss": 0.15179601311683655,
+      "eval_runtime": 21.3331,
+      "eval_samples_per_second": 136.127,
+      "eval_steps_per_second": 4.266,
+      "step": 2173
+    },
+    {
+      "epoch": 41.132075471698116,
+      "grad_norm": 2.5210537910461426,
+      "learning_rate": 3.270440251572327e-05,
+      "loss": 0.0855,
+      "step": 2180
+    },
+    {
+      "epoch": 41.320754716981135,
+      "grad_norm": 4.329328536987305,
+      "learning_rate": 3.2599580712788265e-05,
+      "loss": 0.1184,
+      "step": 2190
+    },
+    {
+      "epoch": 41.509433962264154,
+      "grad_norm": 7.0313801765441895,
+      "learning_rate": 3.249475890985325e-05,
+      "loss": 0.1234,
+      "step": 2200
+    },
+    {
+      "epoch": 41.698113207547166,
+      "grad_norm": 5.2681884765625,
+      "learning_rate": 3.238993710691824e-05,
+      "loss": 0.1292,
+      "step": 2210
+    },
+    {
+      "epoch": 41.886792452830186,
+      "grad_norm": 3.91233229637146,
+      "learning_rate": 3.228511530398323e-05,
+      "loss": 0.1092,
+      "step": 2220
+    },
+    {
+      "epoch": 42.0,
+      "eval_accuracy": 0.9617768595041323,
+      "eval_loss": 0.13080263137817383,
+      "eval_runtime": 21.5397,
+      "eval_samples_per_second": 134.821,
+      "eval_steps_per_second": 4.225,
+      "step": 2226
+    },
+    {
+      "epoch": 42.075471698113205,
+      "grad_norm": 3.9293324947357178,
+      "learning_rate": 3.218029350104822e-05,
+      "loss": 0.0893,
+      "step": 2230
+    },
+    {
+      "epoch": 42.264150943396224,
+      "grad_norm": 3.51786732673645,
+      "learning_rate": 3.207547169811321e-05,
+      "loss": 0.09,
+      "step": 2240
+    },
+    {
+      "epoch": 42.45283018867924,
+      "grad_norm": 3.4958715438842773,
+      "learning_rate": 3.19706498951782e-05,
+      "loss": 0.1196,
+      "step": 2250
+    },
+    {
+      "epoch": 42.64150943396226,
+      "grad_norm": 3.3699843883514404,
+      "learning_rate": 3.186582809224319e-05,
+      "loss": 0.1042,
+      "step": 2260
+    },
+    {
+      "epoch": 42.83018867924528,
+      "grad_norm": 3.706667423248291,
+      "learning_rate": 3.1761006289308175e-05,
+      "loss": 0.1163,
+      "step": 2270
+    },
+    {
+      "epoch": 43.0,
+      "eval_accuracy": 0.9590220385674931,
+      "eval_loss": 0.14582620561122894,
+      "eval_runtime": 21.7777,
+      "eval_samples_per_second": 133.347,
+      "eval_steps_per_second": 4.179,
+      "step": 2279
+    },
+    {
+      "epoch": 43.0188679245283,
+      "grad_norm": 3.969252347946167,
+      "learning_rate": 3.165618448637317e-05,
+      "loss": 0.0884,
+      "step": 2280
+    },
+    {
+      "epoch": 43.20754716981132,
+      "grad_norm": 5.795734882354736,
+      "learning_rate": 3.1551362683438156e-05,
+      "loss": 0.1031,
+      "step": 2290
+    },
+    {
+      "epoch": 43.39622641509434,
+      "grad_norm": 2.936450481414795,
+      "learning_rate": 3.144654088050314e-05,
+      "loss": 0.1086,
+      "step": 2300
+    },
+    {
+      "epoch": 43.58490566037736,
+      "grad_norm": 2.310685634613037,
+      "learning_rate": 3.134171907756814e-05,
+      "loss": 0.1055,
+      "step": 2310
+    },
+    {
+      "epoch": 43.77358490566038,
+      "grad_norm": 5.497471809387207,
+      "learning_rate": 3.1236897274633124e-05,
+      "loss": 0.0898,
+      "step": 2320
+    },
+    {
+      "epoch": 43.9622641509434,
+      "grad_norm": 3.115891456604004,
+      "learning_rate": 3.113207547169811e-05,
+      "loss": 0.1074,
+      "step": 2330
+    },
+    {
+      "epoch": 44.0,
+      "eval_accuracy": 0.9548898071625345,
+      "eval_loss": 0.14139670133590698,
+      "eval_runtime": 21.5889,
+      "eval_samples_per_second": 134.513,
+      "eval_steps_per_second": 4.215,
+      "step": 2332
+    },
+    {
+      "epoch": 44.15094339622642,
+      "grad_norm": 3.430103063583374,
+      "learning_rate": 3.1027253668763105e-05,
+      "loss": 0.0852,
+      "step": 2340
+    },
+    {
+      "epoch": 44.339622641509436,
+      "grad_norm": 3.0305774211883545,
+      "learning_rate": 3.092243186582809e-05,
+      "loss": 0.0818,
+      "step": 2350
+    },
+    {
+      "epoch": 44.528301886792455,
+      "grad_norm": 2.8462817668914795,
+      "learning_rate": 3.081761006289308e-05,
+      "loss": 0.0893,
+      "step": 2360
+    },
+    {
+      "epoch": 44.716981132075475,
+      "grad_norm": 2.0366451740264893,
+      "learning_rate": 3.071278825995807e-05,
+      "loss": 0.088,
+      "step": 2370
+    },
+    {
+      "epoch": 44.905660377358494,
+      "grad_norm": 2.5457682609558105,
+      "learning_rate": 3.060796645702306e-05,
+      "loss": 0.0814,
+      "step": 2380
+    },
+    {
+      "epoch": 45.0,
+      "eval_accuracy": 0.9579889807162535,
+      "eval_loss": 0.15091215074062347,
+      "eval_runtime": 21.5296,
+      "eval_samples_per_second": 134.884,
+      "eval_steps_per_second": 4.227,
+      "step": 2385
+    },
+    {
+      "epoch": 45.094339622641506,
+      "grad_norm": 2.5898990631103516,
+      "learning_rate": 3.050314465408805e-05,
+      "loss": 0.1118,
+      "step": 2390
+    },
+    {
+      "epoch": 45.283018867924525,
+      "grad_norm": 4.282632350921631,
+      "learning_rate": 3.0398322851153044e-05,
+      "loss": 0.0861,
+      "step": 2400
+    },
+    {
+      "epoch": 45.471698113207545,
+      "grad_norm": 3.0017223358154297,
+      "learning_rate": 3.029350104821803e-05,
+      "loss": 0.0882,
+      "step": 2410
+    },
+    {
+      "epoch": 45.660377358490564,
+      "grad_norm": 5.122268199920654,
+      "learning_rate": 3.018867924528302e-05,
+      "loss": 0.0847,
+      "step": 2420
+    },
+    {
+      "epoch": 45.84905660377358,
+      "grad_norm": 3.8469204902648926,
+      "learning_rate": 3.0083857442348012e-05,
+      "loss": 0.0985,
+      "step": 2430
+    },
+    {
+      "epoch": 46.0,
+      "eval_accuracy": 0.9628099173553719,
+      "eval_loss": 0.12866026163101196,
+      "eval_runtime": 21.7086,
+      "eval_samples_per_second": 133.772,
+      "eval_steps_per_second": 4.192,
+      "step": 2438
+    },
+    {
+      "epoch": 46.0377358490566,
+      "grad_norm": 3.621717691421509,
+      "learning_rate": 2.9979035639413e-05,
+      "loss": 0.0899,
+      "step": 2440
+    },
+    {
+      "epoch": 46.22641509433962,
+      "grad_norm": 5.262154579162598,
+      "learning_rate": 2.9874213836477987e-05,
+      "loss": 0.1056,
+      "step": 2450
+    },
+    {
+      "epoch": 46.41509433962264,
+      "grad_norm": 3.9734673500061035,
+      "learning_rate": 2.976939203354298e-05,
+      "loss": 0.0903,
+      "step": 2460
+    },
+    {
+      "epoch": 46.60377358490566,
+      "grad_norm": 5.937262535095215,
+      "learning_rate": 2.9664570230607968e-05,
+      "loss": 0.0728,
+      "step": 2470
+    },
+    {
+      "epoch": 46.79245283018868,
+      "grad_norm": 5.732816219329834,
+      "learning_rate": 2.9559748427672958e-05,
+      "loss": 0.1082,
+      "step": 2480
+    },
+    {
+      "epoch": 46.9811320754717,
+      "grad_norm": 4.877685070037842,
+      "learning_rate": 2.945492662473795e-05,
+      "loss": 0.0863,
+      "step": 2490
+    },
+    {
+      "epoch": 47.0,
+      "eval_accuracy": 0.962465564738292,
+      "eval_loss": 0.12769892811775208,
+      "eval_runtime": 21.6316,
+      "eval_samples_per_second": 134.248,
+      "eval_steps_per_second": 4.207,
+      "step": 2491
+    },
+    {
+      "epoch": 47.16981132075472,
+      "grad_norm": 4.740653991699219,
+      "learning_rate": 2.935010482180294e-05,
+      "loss": 0.0924,
+      "step": 2500
+    },
+    {
+      "epoch": 47.35849056603774,
+      "grad_norm": 3.5254249572753906,
+      "learning_rate": 2.9245283018867926e-05,
+      "loss": 0.0829,
+      "step": 2510
+    },
+    {
+      "epoch": 47.54716981132076,
+      "grad_norm": 2.8752875328063965,
+      "learning_rate": 2.9140461215932913e-05,
+      "loss": 0.0917,
+      "step": 2520
+    },
+    {
+      "epoch": 47.735849056603776,
+      "grad_norm": 3.466445207595825,
+      "learning_rate": 2.9035639412997907e-05,
+      "loss": 0.0683,
+      "step": 2530
+    },
+    {
+      "epoch": 47.924528301886795,
+      "grad_norm": 4.896220684051514,
+      "learning_rate": 2.8930817610062894e-05,
+      "loss": 0.0932,
+      "step": 2540
+    },
+    {
+      "epoch": 48.0,
+      "eval_accuracy": 0.9559228650137741,
+      "eval_loss": 0.14526014029979706,
+      "eval_runtime": 21.6957,
+      "eval_samples_per_second": 133.851,
+      "eval_steps_per_second": 4.194,
+      "step": 2544
+    },
+    {
+      "epoch": 48.113207547169814,
+      "grad_norm": 2.655581474304199,
+      "learning_rate": 2.882599580712788e-05,
+      "loss": 0.0709,
+      "step": 2550
+    },
+    {
+      "epoch": 48.301886792452834,
+      "grad_norm": 3.7453079223632812,
+      "learning_rate": 2.8721174004192875e-05,
+      "loss": 0.0856,
+      "step": 2560
+    },
+    {
+      "epoch": 48.490566037735846,
+      "grad_norm": 4.565659999847412,
+      "learning_rate": 2.8616352201257862e-05,
+      "loss": 0.0737,
+      "step": 2570
+    },
+    {
+      "epoch": 48.679245283018865,
+      "grad_norm": 2.7404532432556152,
+      "learning_rate": 2.851153039832285e-05,
+      "loss": 0.0843,
+      "step": 2580
+    },
+    {
+      "epoch": 48.867924528301884,
+      "grad_norm": 2.5788304805755615,
+      "learning_rate": 2.8406708595387843e-05,
+      "loss": 0.0863,
+      "step": 2590
+    },
+    {
+      "epoch": 49.0,
+      "eval_accuracy": 0.9566115702479339,
+      "eval_loss": 0.15200072526931763,
+      "eval_runtime": 21.6506,
+      "eval_samples_per_second": 134.13,
+      "eval_steps_per_second": 4.203,
+      "step": 2597
+    },
+    {
+      "epoch": 49.056603773584904,
+      "grad_norm": 2.0969185829162598,
+      "learning_rate": 2.830188679245283e-05,
+      "loss": 0.0779,
+      "step": 2600
+    },
+    {
+      "epoch": 49.24528301886792,
+      "grad_norm": 4.123626232147217,
+      "learning_rate": 2.8197064989517817e-05,
+      "loss": 0.1003,
+      "step": 2610
+    },
+    {
+      "epoch": 49.43396226415094,
+      "grad_norm": 3.9485299587249756,
+      "learning_rate": 2.809224318658281e-05,
+      "loss": 0.0862,
+      "step": 2620
+    },
+    {
+      "epoch": 49.62264150943396,
+      "grad_norm": 3.080941915512085,
+      "learning_rate": 2.7987421383647798e-05,
+      "loss": 0.072,
+      "step": 2630
+    },
+    {
+      "epoch": 49.81132075471698,
+      "grad_norm": 3.656919002532959,
+      "learning_rate": 2.788259958071279e-05,
+      "loss": 0.0945,
+      "step": 2640
+    },
+    {
+      "epoch": 50.0,
+      "grad_norm": 2.570844888687134,
+      "learning_rate": 2.777777777777778e-05,
+      "loss": 0.0887,
+      "step": 2650
+    },
+    {
+      "epoch": 50.0,
+      "eval_accuracy": 0.9655647382920111,
+      "eval_loss": 0.12789078056812286,
+      "eval_runtime": 21.6593,
+      "eval_samples_per_second": 134.076,
+      "eval_steps_per_second": 4.201,
+      "step": 2650
+    },
+    {
+      "epoch": 50.18867924528302,
+      "grad_norm": 2.957178831100464,
+      "learning_rate": 2.767295597484277e-05,
+      "loss": 0.0693,
+      "step": 2660
+    },
+    {
+      "epoch": 50.37735849056604,
+      "grad_norm": 4.276056289672852,
+      "learning_rate": 2.7568134171907757e-05,
+      "loss": 0.0892,
+      "step": 2670
+    },
+    {
+      "epoch": 50.56603773584906,
+      "grad_norm": 4.443928241729736,
+      "learning_rate": 2.746331236897275e-05,
+      "loss": 0.081,
+      "step": 2680
+    },
+    {
+      "epoch": 50.75471698113208,
+      "grad_norm": 2.7647054195404053,
+      "learning_rate": 2.7358490566037738e-05,
+      "loss": 0.0606,
+      "step": 2690
+    },
+    {
+      "epoch": 50.943396226415096,
+      "grad_norm": 5.863194942474365,
+      "learning_rate": 2.7253668763102725e-05,
+      "loss": 0.0744,
+      "step": 2700
+    },
+    {
+      "epoch": 51.0,
+      "eval_accuracy": 0.9566115702479339,
+      "eval_loss": 0.15517625212669373,
+      "eval_runtime": 21.6847,
+      "eval_samples_per_second": 133.919,
+      "eval_steps_per_second": 4.196,
+      "step": 2703
+    },
+    {
+      "epoch": 51.132075471698116,
+      "grad_norm": 3.68581485748291,
+      "learning_rate": 2.714884696016772e-05,
+      "loss": 0.0757,
+      "step": 2710
+    },
+    {
+      "epoch": 51.320754716981135,
+      "grad_norm": 6.279058933258057,
+      "learning_rate": 2.7044025157232706e-05,
+      "loss": 0.0839,
+      "step": 2720
+    },
+    {
+      "epoch": 51.509433962264154,
+      "grad_norm": 3.6847403049468994,
+      "learning_rate": 2.6939203354297693e-05,
+      "loss": 0.0754,
+      "step": 2730
+    },
+    {
+      "epoch": 51.698113207547166,
+      "grad_norm": 3.514678716659546,
+      "learning_rate": 2.6834381551362687e-05,
+      "loss": 0.0717,
+      "step": 2740
+    },
+    {
+      "epoch": 51.886792452830186,
+      "grad_norm": 5.7339768409729,
+      "learning_rate": 2.6729559748427674e-05,
+      "loss": 0.0928,
+      "step": 2750
+    },
+    {
+      "epoch": 52.0,
+      "eval_accuracy": 0.9621212121212122,
+      "eval_loss": 0.14646016061306,
+      "eval_runtime": 21.7451,
+      "eval_samples_per_second": 133.547,
+      "eval_steps_per_second": 4.185,
+      "step": 2756
+    },
+    {
+      "epoch": 52.075471698113205,
+      "grad_norm": 2.6166908740997314,
+      "learning_rate": 2.662473794549266e-05,
+      "loss": 0.0622,
+      "step": 2760
+    },
+    {
+      "epoch": 52.264150943396224,
+      "grad_norm": 4.80858850479126,
+      "learning_rate": 2.6519916142557655e-05,
+      "loss": 0.0869,
+      "step": 2770
+    },
+    {
+      "epoch": 52.45283018867924,
+      "grad_norm": 4.928915500640869,
+      "learning_rate": 2.641509433962264e-05,
+      "loss": 0.0796,
+      "step": 2780
+    },
+    {
+      "epoch": 52.64150943396226,
+      "grad_norm": 4.719991207122803,
+      "learning_rate": 2.631027253668763e-05,
+      "loss": 0.0789,
+      "step": 2790
+    },
+    {
+      "epoch": 52.83018867924528,
+      "grad_norm": 3.452692747116089,
+      "learning_rate": 2.6205450733752623e-05,
+      "loss": 0.0776,
+      "step": 2800
+    },
+    {
+      "epoch": 53.0,
+      "eval_accuracy": 0.9583333333333334,
+      "eval_loss": 0.15753231942653656,
+      "eval_runtime": 21.7056,
+      "eval_samples_per_second": 133.79,
+      "eval_steps_per_second": 4.192,
+      "step": 2809
+    },
+    {
+      "epoch": 53.0188679245283,
+      "grad_norm": 3.827479362487793,
+      "learning_rate": 2.610062893081761e-05,
+      "loss": 0.0614,
+      "step": 2810
+    },
+    {
+      "epoch": 53.20754716981132,
+      "grad_norm": 3.628530979156494,
+      "learning_rate": 2.59958071278826e-05,
+      "loss": 0.0836,
+      "step": 2820
+    },
+    {
+      "epoch": 53.39622641509434,
+      "grad_norm": 5.494470119476318,
+      "learning_rate": 2.589098532494759e-05,
+      "loss": 0.07,
+      "step": 2830
+    },
+    {
+      "epoch": 53.58490566037736,
+      "grad_norm": 2.920034646987915,
+      "learning_rate": 2.578616352201258e-05,
+      "loss": 0.0861,
+      "step": 2840
+    },
+    {
+      "epoch": 53.77358490566038,
+      "grad_norm": 2.367824077606201,
+      "learning_rate": 2.5681341719077568e-05,
+      "loss": 0.0824,
+      "step": 2850
+    },
+    {
+      "epoch": 53.9622641509434,
+      "grad_norm": 4.855032444000244,
+      "learning_rate": 2.5576519916142562e-05,
+      "loss": 0.088,
+      "step": 2860
+    },
+    {
+      "epoch": 54.0,
+      "eval_accuracy": 0.956267217630854,
+      "eval_loss": 0.16139821708202362,
+      "eval_runtime": 21.715,
+      "eval_samples_per_second": 133.733,
+      "eval_steps_per_second": 4.191,
+      "step": 2862
+    },
+    {
+      "epoch": 54.15094339622642,
+      "grad_norm": 4.724045753479004,
+      "learning_rate": 2.547169811320755e-05,
+      "loss": 0.076,
+      "step": 2870
+    },
+    {
+      "epoch": 54.339622641509436,
+      "grad_norm": 3.425045967102051,
+      "learning_rate": 2.5366876310272536e-05,
+      "loss": 0.0882,
+      "step": 2880
+    },
+    {
+      "epoch": 54.528301886792455,
+      "grad_norm": 2.366727590560913,
+      "learning_rate": 2.526205450733753e-05,
+      "loss": 0.0492,
+      "step": 2890
+    },
+    {
+      "epoch": 54.716981132075475,
+      "grad_norm": 3.3081116676330566,
+      "learning_rate": 2.5157232704402517e-05,
+      "loss": 0.0755,
+      "step": 2900
+    },
+    {
+      "epoch": 54.905660377358494,
+      "grad_norm": 4.172368049621582,
+      "learning_rate": 2.5052410901467504e-05,
+      "loss": 0.0909,
+      "step": 2910
+    },
+    {
+      "epoch": 55.0,
+      "eval_accuracy": 0.9638429752066116,
+      "eval_loss": 0.13122335076332092,
+      "eval_runtime": 22.1096,
+      "eval_samples_per_second": 131.346,
+      "eval_steps_per_second": 4.116,
+      "step": 2915
+    },
+    {
+      "epoch": 55.094339622641506,
+      "grad_norm": 4.725098609924316,
+      "learning_rate": 2.4947589098532495e-05,
+      "loss": 0.0612,
+      "step": 2920
+    },
+    {
+      "epoch": 55.283018867924525,
+      "grad_norm": 2.416428327560425,
+      "learning_rate": 2.4842767295597485e-05,
+      "loss": 0.067,
+      "step": 2930
+    },
+    {
+      "epoch": 55.471698113207545,
+      "grad_norm": 4.181457042694092,
+      "learning_rate": 2.4737945492662476e-05,
+      "loss": 0.0771,
+      "step": 2940
+    },
+    {
+      "epoch": 55.660377358490564,
+      "grad_norm": 5.096282005310059,
+      "learning_rate": 2.4633123689727463e-05,
+      "loss": 0.0747,
+      "step": 2950
+    },
+    {
+      "epoch": 55.84905660377358,
+      "grad_norm": 5.836996555328369,
+      "learning_rate": 2.4528301886792453e-05,
+      "loss": 0.089,
+      "step": 2960
+    },
+    {
+      "epoch": 56.0,
+      "eval_accuracy": 0.9652203856749312,
+      "eval_loss": 0.13570785522460938,
+      "eval_runtime": 21.9675,
+      "eval_samples_per_second": 132.195,
+      "eval_steps_per_second": 4.142,
+      "step": 2968
+    },
+    {
+      "epoch": 56.0377358490566,
+      "grad_norm": 3.5151453018188477,
+      "learning_rate": 2.4423480083857444e-05,
+      "loss": 0.0653,
+      "step": 2970
+    },
+    {
+      "epoch": 56.22641509433962,
+      "grad_norm": 5.056783676147461,
+      "learning_rate": 2.431865828092243e-05,
+      "loss": 0.0768,
+      "step": 2980
+    },
+    {
+      "epoch": 56.41509433962264,
+      "grad_norm": 2.0527374744415283,
+      "learning_rate": 2.421383647798742e-05,
+      "loss": 0.0694,
+      "step": 2990
+    },
+    {
+      "epoch": 56.60377358490566,
+      "grad_norm": 3.363852024078369,
+      "learning_rate": 2.4109014675052412e-05,
+      "loss": 0.0763,
+      "step": 3000
+    },
+    {
+      "epoch": 56.79245283018868,
+      "grad_norm": 2.2541282176971436,
+      "learning_rate": 2.4004192872117402e-05,
+      "loss": 0.0717,
+      "step": 3010
+    },
+    {
+      "epoch": 56.9811320754717,
+      "grad_norm": 1.5099104642868042,
+      "learning_rate": 2.3899371069182393e-05,
+      "loss": 0.0587,
+      "step": 3020
+    },
+    {
+      "epoch": 57.0,
+      "eval_accuracy": 0.9614325068870524,
+      "eval_loss": 0.15099208056926727,
+      "eval_runtime": 21.4231,
+      "eval_samples_per_second": 135.554,
+      "eval_steps_per_second": 4.248,
+      "step": 3021
+    },
+    {
+      "epoch": 57.16981132075472,
+      "grad_norm": 3.0756702423095703,
+      "learning_rate": 2.3794549266247383e-05,
+      "loss": 0.0931,
+      "step": 3030
+    },
+    {
+      "epoch": 57.35849056603774,
+      "grad_norm": 5.203828811645508,
+      "learning_rate": 2.368972746331237e-05,
+      "loss": 0.077,
+      "step": 3040
+    },
+    {
+      "epoch": 57.54716981132076,
+      "grad_norm": 3.4697303771972656,
+      "learning_rate": 2.358490566037736e-05,
+      "loss": 0.0773,
+      "step": 3050
+    },
+    {
+      "epoch": 57.735849056603776,
+      "grad_norm": 5.211211681365967,
+      "learning_rate": 2.348008385744235e-05,
+      "loss": 0.0829,
+      "step": 3060
+    },
+    {
+      "epoch": 57.924528301886795,
+      "grad_norm": 3.6697630882263184,
+      "learning_rate": 2.3375262054507338e-05,
+      "loss": 0.0931,
+      "step": 3070
+    },
+    {
+      "epoch": 58.0,
+      "eval_accuracy": 0.9579889807162535,
+      "eval_loss": 0.14657209813594818,
+      "eval_runtime": 22.1099,
+      "eval_samples_per_second": 131.344,
+      "eval_steps_per_second": 4.116,
+      "step": 3074
+    },
+    {
+      "epoch": 58.113207547169814,
+      "grad_norm": 5.555974960327148,
+      "learning_rate": 2.327044025157233e-05,
+      "loss": 0.0583,
+      "step": 3080
+    },
+    {
+      "epoch": 58.301886792452834,
+      "grad_norm": 4.070455551147461,
+      "learning_rate": 2.316561844863732e-05,
+      "loss": 0.0746,
+      "step": 3090
+    },
+    {
+      "epoch": 58.490566037735846,
+      "grad_norm": 2.739751100540161,
+      "learning_rate": 2.3060796645702306e-05,
+      "loss": 0.0698,
+      "step": 3100
+    },
+    {
+      "epoch": 58.679245283018865,
+      "grad_norm": 4.419424057006836,
+      "learning_rate": 2.2955974842767297e-05,
+      "loss": 0.0669,
+      "step": 3110
+    },
+    {
+      "epoch": 58.867924528301884,
+      "grad_norm": 2.998007297515869,
+      "learning_rate": 2.2851153039832284e-05,
+      "loss": 0.0878,
+      "step": 3120
+    },
+    {
+      "epoch": 59.0,
+      "eval_accuracy": 0.9590220385674931,
+      "eval_loss": 0.14993391931056976,
+      "eval_runtime": 21.3446,
+      "eval_samples_per_second": 136.053,
+      "eval_steps_per_second": 4.263,
+      "step": 3127
+    },
+    {
+      "epoch": 59.056603773584904,
+      "grad_norm": 3.094158411026001,
+      "learning_rate": 2.2746331236897274e-05,
+      "loss": 0.0679,
+      "step": 3130
+    },
+    {
+      "epoch": 59.24528301886792,
+      "grad_norm": 3.320544958114624,
+      "learning_rate": 2.2641509433962265e-05,
+      "loss": 0.066,
+      "step": 3140
+    },
+    {
+      "epoch": 59.43396226415094,
+      "grad_norm": 5.472837924957275,
+      "learning_rate": 2.2536687631027252e-05,
+      "loss": 0.071,
+      "step": 3150
+    },
+    {
+      "epoch": 59.62264150943396,
+      "grad_norm": 2.2281177043914795,
+      "learning_rate": 2.2431865828092242e-05,
+      "loss": 0.075,
+      "step": 3160
+    },
+    {
+      "epoch": 59.81132075471698,
+      "grad_norm": 4.7084503173828125,
+      "learning_rate": 2.2327044025157233e-05,
+      "loss": 0.0763,
+      "step": 3170
+    },
+    {
+      "epoch": 60.0,
+      "grad_norm": 4.050378322601318,
+      "learning_rate": 2.2222222222222223e-05,
+      "loss": 0.0725,
+      "step": 3180
+    },
+    {
+      "epoch": 60.0,
+      "eval_accuracy": 0.9597107438016529,
+      "eval_loss": 0.15237364172935486,
+      "eval_runtime": 21.4189,
+      "eval_samples_per_second": 135.581,
+      "eval_steps_per_second": 4.249,
+      "step": 3180
+    },
+    {
+      "epoch": 60.18867924528302,
+      "grad_norm": 5.035741329193115,
+      "learning_rate": 2.2117400419287214e-05,
+      "loss": 0.0754,
+      "step": 3190
+    },
+    {
+      "epoch": 60.37735849056604,
+      "grad_norm": 2.6131255626678467,
+      "learning_rate": 2.2012578616352204e-05,
+      "loss": 0.0727,
+      "step": 3200
+    },
+    {
+      "epoch": 60.56603773584906,
+      "grad_norm": 2.404343366622925,
+      "learning_rate": 2.190775681341719e-05,
+      "loss": 0.055,
+      "step": 3210
+    },
+    {
+      "epoch": 60.75471698113208,
+      "grad_norm": 3.9234628677368164,
+      "learning_rate": 2.1802935010482182e-05,
+      "loss": 0.0671,
+      "step": 3220
+    },
+    {
+      "epoch": 60.943396226415096,
+      "grad_norm": 2.465709686279297,
+      "learning_rate": 2.1698113207547172e-05,
+      "loss": 0.0543,
+      "step": 3230
+    },
+    {
+      "epoch": 61.0,
+      "eval_accuracy": 0.9583333333333334,
+      "eval_loss": 0.15427254140377045,
+      "eval_runtime": 21.2569,
+      "eval_samples_per_second": 136.615,
+      "eval_steps_per_second": 4.281,
+      "step": 3233
+    },
+    {
+      "epoch": 61.132075471698116,
+      "grad_norm": 4.535091876983643,
+      "learning_rate": 2.159329140461216e-05,
+      "loss": 0.0529,
+      "step": 3240
+    },
+    {
+      "epoch": 61.320754716981135,
+      "grad_norm": 2.738173484802246,
+      "learning_rate": 2.148846960167715e-05,
+      "loss": 0.0566,
+      "step": 3250
+    },
+    {
+      "epoch": 61.509433962264154,
+      "grad_norm": 3.950739860534668,
+      "learning_rate": 2.138364779874214e-05,
+      "loss": 0.0729,
+      "step": 3260
+    },
+    {
+      "epoch": 61.698113207547166,
+      "grad_norm": 2.284546375274658,
+      "learning_rate": 2.1278825995807127e-05,
+      "loss": 0.0555,
+      "step": 3270
+    },
+    {
+      "epoch": 61.886792452830186,
+      "grad_norm": 4.242305278778076,
+      "learning_rate": 2.1174004192872118e-05,
+      "loss": 0.0773,
+      "step": 3280
+    },
+    {
+      "epoch": 62.0,
+      "eval_accuracy": 0.9634986225895317,
+      "eval_loss": 0.15126247704029083,
+      "eval_runtime": 21.4863,
+      "eval_samples_per_second": 135.156,
+      "eval_steps_per_second": 4.235,
+      "step": 3286
+    },
+    {
+      "epoch": 62.075471698113205,
+      "grad_norm": 5.880760669708252,
+      "learning_rate": 2.106918238993711e-05,
+      "loss": 0.0663,
+      "step": 3290
+    },
+    {
+      "epoch": 62.264150943396224,
+      "grad_norm": 4.1386237144470215,
+      "learning_rate": 2.0964360587002095e-05,
+      "loss": 0.0643,
+      "step": 3300
+    },
+    {
+      "epoch": 62.45283018867924,
+      "grad_norm": 3.798180341720581,
+      "learning_rate": 2.0859538784067086e-05,
+      "loss": 0.0789,
+      "step": 3310
+    },
+    {
+      "epoch": 62.64150943396226,
+      "grad_norm": 2.6862704753875732,
+      "learning_rate": 2.0754716981132076e-05,
+      "loss": 0.0838,
+      "step": 3320
+    },
+    {
+      "epoch": 62.83018867924528,
+      "grad_norm": 4.914183616638184,
+      "learning_rate": 2.0649895178197063e-05,
+      "loss": 0.0626,
+      "step": 3330
+    },
+    {
+      "epoch": 63.0,
+      "eval_accuracy": 0.9600550964187328,
+      "eval_loss": 0.1511116325855255,
+      "eval_runtime": 21.3341,
+      "eval_samples_per_second": 136.12,
+      "eval_steps_per_second": 4.265,
+      "step": 3339
+    },
+    {
+      "epoch": 63.0188679245283,
+      "grad_norm": 3.25014591217041,
+      "learning_rate": 2.0545073375262054e-05,
+      "loss": 0.0534,
+      "step": 3340
+    },
+    {
+      "epoch": 63.20754716981132,
+      "grad_norm": 3.0906035900115967,
+      "learning_rate": 2.0440251572327044e-05,
+      "loss": 0.0598,
+      "step": 3350
+    },
+    {
+      "epoch": 63.39622641509434,
+      "grad_norm": 3.2928597927093506,
+      "learning_rate": 2.0335429769392035e-05,
+      "loss": 0.0704,
+      "step": 3360
+    },
+    {
+      "epoch": 63.58490566037736,
+      "grad_norm": 2.6541659832000732,
+      "learning_rate": 2.0230607966457025e-05,
+      "loss": 0.0541,
+      "step": 3370
+    },
+    {
+      "epoch": 63.77358490566038,
+      "grad_norm": 4.403809070587158,
+      "learning_rate": 2.0125786163522016e-05,
+      "loss": 0.0563,
+      "step": 3380
+    },
+    {
+      "epoch": 63.9622641509434,
+      "grad_norm": 4.669613361358643,
+      "learning_rate": 2.0020964360587003e-05,
+      "loss": 0.0649,
+      "step": 3390
+    },
+    {
+      "epoch": 64.0,
+      "eval_accuracy": 0.959366391184573,
+      "eval_loss": 0.1467009335756302,
+      "eval_runtime": 21.2656,
+      "eval_samples_per_second": 136.558,
+      "eval_steps_per_second": 4.279,
+      "step": 3392
+    },
+    {
+      "epoch": 64.15094339622641,
+      "grad_norm": 2.8066558837890625,
+      "learning_rate": 1.9916142557651993e-05,
+      "loss": 0.0746,
+      "step": 3400
+    },
+    {
+      "epoch": 64.33962264150944,
+      "grad_norm": 5.994083881378174,
+      "learning_rate": 1.9811320754716984e-05,
+      "loss": 0.0704,
+      "step": 3410
+    },
+    {
+      "epoch": 64.52830188679245,
+      "grad_norm": 4.9809441566467285,
+      "learning_rate": 1.970649895178197e-05,
+      "loss": 0.0773,
+      "step": 3420
+    },
+    {
+      "epoch": 64.71698113207547,
+      "grad_norm": 2.497436285018921,
+      "learning_rate": 1.960167714884696e-05,
+      "loss": 0.0781,
+      "step": 3430
+    },
+    {
+      "epoch": 64.90566037735849,
+      "grad_norm": 3.954669952392578,
+      "learning_rate": 1.9496855345911952e-05,
+      "loss": 0.0705,
+      "step": 3440
+    },
+    {
+      "epoch": 65.0,
+      "eval_accuracy": 0.9590220385674931,
+      "eval_loss": 0.1443195939064026,
+      "eval_runtime": 21.2667,
+      "eval_samples_per_second": 136.551,
+      "eval_steps_per_second": 4.279,
+      "step": 3445
+    },
+    {
+      "epoch": 65.09433962264151,
+      "grad_norm": 3.244377374649048,
+      "learning_rate": 1.939203354297694e-05,
+      "loss": 0.0515,
+      "step": 3450
+    },
+    {
+      "epoch": 65.28301886792453,
+      "grad_norm": 2.0447616577148438,
+      "learning_rate": 1.928721174004193e-05,
+      "loss": 0.0602,
+      "step": 3460
+    },
+    {
+      "epoch": 65.47169811320755,
+      "grad_norm": 9.920838356018066,
+      "learning_rate": 1.918238993710692e-05,
+      "loss": 0.072,
+      "step": 3470
+    },
+    {
+      "epoch": 65.66037735849056,
+      "grad_norm": 2.986557960510254,
+      "learning_rate": 1.9077568134171907e-05,
+      "loss": 0.0571,
+      "step": 3480
+    },
+    {
+      "epoch": 65.84905660377359,
+      "grad_norm": 3.244969606399536,
+      "learning_rate": 1.8972746331236897e-05,
+      "loss": 0.0737,
+      "step": 3490
+    },
+    {
+      "epoch": 66.0,
+      "eval_accuracy": 0.9607438016528925,
+      "eval_loss": 0.13612627983093262,
+      "eval_runtime": 21.1938,
+      "eval_samples_per_second": 137.021,
+      "eval_steps_per_second": 4.294,
+      "step": 3498
+    },
+    {
+      "epoch": 66.0377358490566,
+      "grad_norm": 3.0969536304473877,
+      "learning_rate": 1.8867924528301888e-05,
+      "loss": 0.0581,
+      "step": 3500
+    },
+    {
+      "epoch": 66.22641509433963,
+      "grad_norm": 5.167777061462402,
+      "learning_rate": 1.876310272536688e-05,
+      "loss": 0.0744,
+      "step": 3510
+    },
+    {
+      "epoch": 66.41509433962264,
+      "grad_norm": 3.00007700920105,
+      "learning_rate": 1.865828092243187e-05,
+      "loss": 0.0543,
+      "step": 3520
+    },
+    {
+      "epoch": 66.60377358490567,
+      "grad_norm": 2.6348941326141357,
+      "learning_rate": 1.8553459119496856e-05,
+      "loss": 0.0567,
+      "step": 3530
+    },
+    {
+      "epoch": 66.79245283018868,
+      "grad_norm": 5.631946563720703,
+      "learning_rate": 1.8448637316561846e-05,
+      "loss": 0.0779,
+      "step": 3540
+    },
+    {
+      "epoch": 66.98113207547169,
+      "grad_norm": 1.5974416732788086,
+      "learning_rate": 1.8343815513626837e-05,
+      "loss": 0.0518,
+      "step": 3550
+    },
+    {
+      "epoch": 67.0,
+      "eval_accuracy": 0.959366391184573,
+      "eval_loss": 0.14412498474121094,
+      "eval_runtime": 21.2408,
+      "eval_samples_per_second": 136.718,
+      "eval_steps_per_second": 4.284,
+      "step": 3551
+    },
+    {
+      "epoch": 67.16981132075472,
+      "grad_norm": 2.0217528343200684,
+      "learning_rate": 1.8238993710691824e-05,
+      "loss": 0.0645,
+      "step": 3560
+    },
+    {
+      "epoch": 67.35849056603773,
+      "grad_norm": 3.986748695373535,
+      "learning_rate": 1.8134171907756814e-05,
+      "loss": 0.0647,
+      "step": 3570
+    },
+    {
+      "epoch": 67.54716981132076,
+      "grad_norm": 4.671311855316162,
+      "learning_rate": 1.8029350104821805e-05,
+      "loss": 0.0666,
+      "step": 3580
+    },
+    {
+      "epoch": 67.73584905660377,
+      "grad_norm": 2.0372939109802246,
+      "learning_rate": 1.7924528301886792e-05,
+      "loss": 0.0554,
+      "step": 3590
+    },
+    {
+      "epoch": 67.9245283018868,
+      "grad_norm": 2.52120041847229,
+      "learning_rate": 1.7819706498951782e-05,
+      "loss": 0.0502,
+      "step": 3600
+    },
+    {
+      "epoch": 68.0,
+      "eval_accuracy": 0.9590220385674931,
+      "eval_loss": 0.1534823328256607,
+      "eval_runtime": 21.3789,
+      "eval_samples_per_second": 135.835,
+      "eval_steps_per_second": 4.257,
+      "step": 3604
+    },
+    {
+      "epoch": 68.11320754716981,
+      "grad_norm": 1.7292485237121582,
+      "learning_rate": 1.7714884696016773e-05,
+      "loss": 0.0696,
+      "step": 3610
+    },
+    {
+      "epoch": 68.30188679245283,
+      "grad_norm": 2.2730910778045654,
+      "learning_rate": 1.761006289308176e-05,
+      "loss": 0.0608,
+      "step": 3620
+    },
+    {
+      "epoch": 68.49056603773585,
+      "grad_norm": 3.564232110977173,
+      "learning_rate": 1.750524109014675e-05,
+      "loss": 0.0604,
+      "step": 3630
+    },
+    {
+      "epoch": 68.67924528301887,
+      "grad_norm": 2.5112924575805664,
+      "learning_rate": 1.740041928721174e-05,
+      "loss": 0.0415,
+      "step": 3640
+    },
+    {
+      "epoch": 68.86792452830188,
+      "grad_norm": 5.021323204040527,
+      "learning_rate": 1.7295597484276728e-05,
+      "loss": 0.0701,
+      "step": 3650
+    },
+    {
+      "epoch": 69.0,
+      "eval_accuracy": 0.9662534435261708,
+      "eval_loss": 0.1362384557723999,
+      "eval_runtime": 21.2026,
+      "eval_samples_per_second": 136.964,
+      "eval_steps_per_second": 4.292,
+      "step": 3657
+    },
+    {
+      "epoch": 69.05660377358491,
+      "grad_norm": 5.013925075531006,
+      "learning_rate": 1.719077568134172e-05,
+      "loss": 0.0744,
+      "step": 3660
+    },
+    {
+      "epoch": 69.24528301886792,
+      "grad_norm": 3.5540971755981445,
+      "learning_rate": 1.708595387840671e-05,
+      "loss": 0.0568,
+      "step": 3670
+    },
+    {
+      "epoch": 69.43396226415095,
+      "grad_norm": 3.4611597061157227,
+      "learning_rate": 1.69811320754717e-05,
+      "loss": 0.0513,
+      "step": 3680
+    },
+    {
+      "epoch": 69.62264150943396,
+      "grad_norm": 2.5300846099853516,
+      "learning_rate": 1.687631027253669e-05,
+      "loss": 0.0442,
+      "step": 3690
+    },
+    {
+      "epoch": 69.81132075471699,
+      "grad_norm": 2.6349620819091797,
+      "learning_rate": 1.677148846960168e-05,
+      "loss": 0.054,
+      "step": 3700
+    },
+    {
+      "epoch": 70.0,
+      "grad_norm": 3.122040033340454,
+      "learning_rate": 1.6666666666666667e-05,
+      "loss": 0.0826,
+      "step": 3710
+    },
+    {
+      "epoch": 70.0,
+      "eval_accuracy": 0.9610881542699724,
+      "eval_loss": 0.1492019146680832,
+      "eval_runtime": 21.2837,
+      "eval_samples_per_second": 136.442,
+      "eval_steps_per_second": 4.276,
+      "step": 3710
+    },
+    {
+      "epoch": 70.18867924528301,
+      "grad_norm": 3.5501227378845215,
+      "learning_rate": 1.6561844863731658e-05,
+      "loss": 0.0627,
+      "step": 3720
+    },
+    {
+      "epoch": 70.37735849056604,
+      "grad_norm": 2.6497879028320312,
+      "learning_rate": 1.645702306079665e-05,
+      "loss": 0.0461,
+      "step": 3730
+    },
+    {
+      "epoch": 70.56603773584905,
+      "grad_norm": 2.9843809604644775,
+      "learning_rate": 1.6352201257861635e-05,
+      "loss": 0.0484,
+      "step": 3740
+    },
+    {
+      "epoch": 70.75471698113208,
+      "grad_norm": 8.867347717285156,
+      "learning_rate": 1.6247379454926626e-05,
+      "loss": 0.0595,
+      "step": 3750
+    },
+    {
+      "epoch": 70.94339622641509,
+      "grad_norm": 4.957089900970459,
+      "learning_rate": 1.6142557651991616e-05,
+      "loss": 0.0715,
+      "step": 3760
+    },
+    {
+      "epoch": 71.0,
+      "eval_accuracy": 0.962465564738292,
+      "eval_loss": 0.16146376729011536,
+      "eval_runtime": 21.2899,
+      "eval_samples_per_second": 136.403,
+      "eval_steps_per_second": 4.274,
+      "step": 3763
+    },
+    {
+      "epoch": 71.13207547169812,
+      "grad_norm": 3.982633590698242,
+      "learning_rate": 1.6037735849056604e-05,
+      "loss": 0.0534,
+      "step": 3770
+    },
+    {
+      "epoch": 71.32075471698113,
+      "grad_norm": 4.253650188446045,
+      "learning_rate": 1.5932914046121594e-05,
+      "loss": 0.0677,
+      "step": 3780
+    },
+    {
+      "epoch": 71.50943396226415,
+      "grad_norm": 4.608425140380859,
+      "learning_rate": 1.5828092243186584e-05,
+      "loss": 0.0495,
+      "step": 3790
+    },
+    {
+      "epoch": 71.69811320754717,
+      "grad_norm": 5.12533712387085,
+      "learning_rate": 1.572327044025157e-05,
+      "loss": 0.0633,
+      "step": 3800
+    },
+    {
+      "epoch": 71.88679245283019,
+      "grad_norm": 4.220004558563232,
+      "learning_rate": 1.5618448637316562e-05,
+      "loss": 0.0635,
+      "step": 3810
+    },
+    {
+      "epoch": 72.0,
+      "eval_accuracy": 0.9641873278236914,
+      "eval_loss": 0.14879465103149414,
+      "eval_runtime": 21.2409,
+      "eval_samples_per_second": 136.717,
+      "eval_steps_per_second": 4.284,
+      "step": 3816
+    },
+    {
+      "epoch": 72.0754716981132,
+      "grad_norm": 1.841488242149353,
+      "learning_rate": 1.5513626834381552e-05,
+      "loss": 0.0608,
+      "step": 3820
+    },
+    {
+      "epoch": 72.26415094339623,
+      "grad_norm": 2.0446391105651855,
+      "learning_rate": 1.540880503144654e-05,
+      "loss": 0.0465,
+      "step": 3830
+    },
+    {
+      "epoch": 72.45283018867924,
+      "grad_norm": 3.4776628017425537,
+      "learning_rate": 1.530398322851153e-05,
+      "loss": 0.0575,
+      "step": 3840
+    },
+    {
+      "epoch": 72.64150943396227,
+      "grad_norm": 2.3915700912475586,
+      "learning_rate": 1.5199161425576522e-05,
+      "loss": 0.0564,
+      "step": 3850
+    },
+    {
+      "epoch": 72.83018867924528,
+      "grad_norm": 1.4522980451583862,
+      "learning_rate": 1.509433962264151e-05,
+      "loss": 0.0522,
+      "step": 3860
+    },
+    {
+      "epoch": 73.0,
+      "eval_accuracy": 0.9621212121212122,
+      "eval_loss": 0.14563634991645813,
+      "eval_runtime": 21.397,
+      "eval_samples_per_second": 135.72,
+      "eval_steps_per_second": 4.253,
+      "step": 3869
+    },
+    {
+      "epoch": 73.01886792452831,
+      "grad_norm": 2.048356056213379,
+      "learning_rate": 1.49895178197065e-05,
+      "loss": 0.0565,
+      "step": 3870
+    },
+    {
+      "epoch": 73.20754716981132,
+      "grad_norm": 2.7343058586120605,
+      "learning_rate": 1.488469601677149e-05,
+      "loss": 0.0403,
+      "step": 3880
+    },
+    {
+      "epoch": 73.39622641509433,
+      "grad_norm": 1.833511471748352,
+      "learning_rate": 1.4779874213836479e-05,
+      "loss": 0.0502,
+      "step": 3890
+    },
+    {
+      "epoch": 73.58490566037736,
+      "grad_norm": 2.4152145385742188,
+      "learning_rate": 1.467505241090147e-05,
+      "loss": 0.0539,
+      "step": 3900
+    },
+    {
+      "epoch": 73.77358490566037,
+      "grad_norm": 2.6949825286865234,
+      "learning_rate": 1.4570230607966457e-05,
+      "loss": 0.0499,
+      "step": 3910
+    },
+    {
+      "epoch": 73.9622641509434,
+      "grad_norm": 3.1027348041534424,
+      "learning_rate": 1.4465408805031447e-05,
+      "loss": 0.0485,
+      "step": 3920
+    },
+    {
+      "epoch": 74.0,
+      "eval_accuracy": 0.9645316804407713,
+      "eval_loss": 0.1386471837759018,
+      "eval_runtime": 21.3052,
+      "eval_samples_per_second": 136.304,
+      "eval_steps_per_second": 4.271,
+      "step": 3922
+    },
+    {
+      "epoch": 74.15094339622641,
+      "grad_norm": 1.7940607070922852,
+      "learning_rate": 1.4360587002096438e-05,
+      "loss": 0.0541,
+      "step": 3930
+    },
+    {
+      "epoch": 74.33962264150944,
+      "grad_norm": 2.581839084625244,
+      "learning_rate": 1.4255765199161425e-05,
+      "loss": 0.044,
+      "step": 3940
+    },
+    {
+      "epoch": 74.52830188679245,
+      "grad_norm": 1.7485500574111938,
+      "learning_rate": 1.4150943396226415e-05,
+      "loss": 0.043,
+      "step": 3950
+    },
+    {
+      "epoch": 74.71698113207547,
+      "grad_norm": 2.436922550201416,
+      "learning_rate": 1.4046121593291406e-05,
+      "loss": 0.0601,
+      "step": 3960
+    },
+    {
+      "epoch": 74.90566037735849,
+      "grad_norm": 2.460512638092041,
+      "learning_rate": 1.3941299790356394e-05,
+      "loss": 0.0629,
+      "step": 3970
+    },
+    {
+      "epoch": 75.0,
+      "eval_accuracy": 0.9631542699724518,
+      "eval_loss": 0.14631205797195435,
+      "eval_runtime": 21.2256,
+      "eval_samples_per_second": 136.816,
+      "eval_steps_per_second": 4.287,
+      "step": 3975
+    },
+    {
+      "epoch": 75.09433962264151,
+      "grad_norm": 2.925687551498413,
+      "learning_rate": 1.3836477987421385e-05,
+      "loss": 0.0535,
+      "step": 3980
+    },
+    {
+      "epoch": 75.28301886792453,
+      "grad_norm": 1.9862323999404907,
+      "learning_rate": 1.3731656184486375e-05,
+      "loss": 0.0586,
+      "step": 3990
+    },
+    {
+      "epoch": 75.47169811320755,
+      "grad_norm": 1.5170574188232422,
+      "learning_rate": 1.3626834381551362e-05,
+      "loss": 0.0471,
+      "step": 4000
+    },
+    {
+      "epoch": 75.66037735849056,
+      "grad_norm": 4.97707986831665,
+      "learning_rate": 1.3522012578616353e-05,
+      "loss": 0.0529,
+      "step": 4010
+    },
+    {
+      "epoch": 75.84905660377359,
+      "grad_norm": 1.9036014080047607,
+      "learning_rate": 1.3417190775681343e-05,
+      "loss": 0.0568,
+      "step": 4020
+    },
+    {
+      "epoch": 76.0,
+      "eval_accuracy": 0.9621212121212122,
+      "eval_loss": 0.14720916748046875,
+      "eval_runtime": 21.3087,
+      "eval_samples_per_second": 136.282,
+      "eval_steps_per_second": 4.271,
+      "step": 4028
+    },
+    {
+      "epoch": 76.0377358490566,
+      "grad_norm": 3.9243173599243164,
+      "learning_rate": 1.331236897274633e-05,
+      "loss": 0.0531,
+      "step": 4030
+    },
+    {
+      "epoch": 76.22641509433963,
+      "grad_norm": 3.3605711460113525,
+      "learning_rate": 1.320754716981132e-05,
+      "loss": 0.0625,
+      "step": 4040
+    },
+    {
+      "epoch": 76.41509433962264,
+      "grad_norm": 3.699404716491699,
+      "learning_rate": 1.3102725366876311e-05,
+      "loss": 0.0522,
+      "step": 4050
+    },
+    {
+      "epoch": 76.60377358490567,
+      "grad_norm": 4.595207691192627,
+      "learning_rate": 1.29979035639413e-05,
+      "loss": 0.0611,
+      "step": 4060
+    },
+    {
+      "epoch": 76.79245283018868,
+      "grad_norm": 6.970567226409912,
+      "learning_rate": 1.289308176100629e-05,
+      "loss": 0.0641,
+      "step": 4070
+    },
+    {
+      "epoch": 76.98113207547169,
+      "grad_norm": 5.223586559295654,
+      "learning_rate": 1.2788259958071281e-05,
+      "loss": 0.0556,
+      "step": 4080
+    },
+    {
+      "epoch": 77.0,
+      "eval_accuracy": 0.9659090909090909,
+      "eval_loss": 0.14402107894420624,
+      "eval_runtime": 21.2908,
+      "eval_samples_per_second": 136.397,
+      "eval_steps_per_second": 4.274,
+      "step": 4081
+    },
+    {
+      "epoch": 77.16981132075472,
+      "grad_norm": 1.7077960968017578,
+      "learning_rate": 1.2683438155136268e-05,
+      "loss": 0.0477,
+      "step": 4090
+    },
+    {
+      "epoch": 77.35849056603773,
+      "grad_norm": 3.9550302028656006,
+      "learning_rate": 1.2578616352201259e-05,
+      "loss": 0.0631,
+      "step": 4100
+    },
+    {
+      "epoch": 77.54716981132076,
+      "grad_norm": 3.574674129486084,
+      "learning_rate": 1.2473794549266247e-05,
+      "loss": 0.0357,
+      "step": 4110
+    },
+    {
+      "epoch": 77.73584905660377,
+      "grad_norm": 3.119210958480835,
+      "learning_rate": 1.2368972746331238e-05,
+      "loss": 0.0382,
+      "step": 4120
+    },
+    {
+      "epoch": 77.9245283018868,
+      "grad_norm": 1.9549647569656372,
+      "learning_rate": 1.2264150943396227e-05,
+      "loss": 0.0547,
+      "step": 4130
+    },
+    {
+      "epoch": 78.0,
+      "eval_accuracy": 0.9634986225895317,
+      "eval_loss": 0.14210809767246246,
+      "eval_runtime": 21.3339,
+      "eval_samples_per_second": 136.121,
+      "eval_steps_per_second": 4.266,
+      "step": 4134
+    },
+    {
+      "epoch": 78.11320754716981,
+      "grad_norm": 3.379382610321045,
+      "learning_rate": 1.2159329140461215e-05,
+      "loss": 0.0632,
+      "step": 4140
+    },
+    {
+      "epoch": 78.30188679245283,
+      "grad_norm": 5.461263656616211,
+      "learning_rate": 1.2054507337526206e-05,
+      "loss": 0.0689,
+      "step": 4150
+    },
+    {
+      "epoch": 78.49056603773585,
+      "grad_norm": 4.160185813903809,
+      "learning_rate": 1.1949685534591196e-05,
+      "loss": 0.054,
+      "step": 4160
+    },
+    {
+      "epoch": 78.67924528301887,
+      "grad_norm": 2.50945782661438,
+      "learning_rate": 1.1844863731656185e-05,
+      "loss": 0.0302,
+      "step": 4170
+    },
+    {
+      "epoch": 78.86792452830188,
+      "grad_norm": 3.265209197998047,
+      "learning_rate": 1.1740041928721176e-05,
+      "loss": 0.0527,
+      "step": 4180
+    },
+    {
+      "epoch": 79.0,
+      "eval_accuracy": 0.9683195592286501,
+      "eval_loss": 0.14441226422786713,
+      "eval_runtime": 21.2921,
+      "eval_samples_per_second": 136.389,
+      "eval_steps_per_second": 4.274,
+      "step": 4187
+    },
+    {
+      "epoch": 79.05660377358491,
+      "grad_norm": 1.1379858255386353,
+      "learning_rate": 1.1635220125786164e-05,
+      "loss": 0.0419,
+      "step": 4190
+    },
+    {
+      "epoch": 79.24528301886792,
+      "grad_norm": 5.1714043617248535,
+      "learning_rate": 1.1530398322851153e-05,
+      "loss": 0.053,
+      "step": 4200
+    },
+    {
+      "epoch": 79.43396226415095,
+      "grad_norm": 3.441499710083008,
+      "learning_rate": 1.1425576519916142e-05,
+      "loss": 0.0588,
+      "step": 4210
+    },
+    {
+      "epoch": 79.62264150943396,
+      "grad_norm": 1.575990915298462,
+      "learning_rate": 1.1320754716981132e-05,
+      "loss": 0.0333,
+      "step": 4220
+    },
+    {
+      "epoch": 79.81132075471699,
+      "grad_norm": 1.0692986249923706,
+      "learning_rate": 1.1215932914046121e-05,
+      "loss": 0.0375,
+      "step": 4230
+    },
+    {
+      "epoch": 80.0,
+      "grad_norm": 6.881436824798584,
+      "learning_rate": 1.1111111111111112e-05,
+      "loss": 0.054,
+      "step": 4240
+    },
+    {
+      "epoch": 80.0,
+      "eval_accuracy": 0.9628099173553719,
+      "eval_loss": 0.1463625133037567,
+      "eval_runtime": 21.276,
+      "eval_samples_per_second": 136.492,
+      "eval_steps_per_second": 4.277,
+      "step": 4240
+    },
+    {
+      "epoch": 80.18867924528301,
+      "grad_norm": 3.086602210998535,
+      "learning_rate": 1.1006289308176102e-05,
+      "loss": 0.0614,
+      "step": 4250
+    },
+    {
+      "epoch": 80.37735849056604,
+      "grad_norm": 1.642980933189392,
+      "learning_rate": 1.0901467505241091e-05,
+      "loss": 0.048,
+      "step": 4260
+    },
+    {
+      "epoch": 80.56603773584905,
+      "grad_norm": 5.067837715148926,
+      "learning_rate": 1.079664570230608e-05,
+      "loss": 0.0461,
+      "step": 4270
+    },
+    {
+      "epoch": 80.75471698113208,
+      "grad_norm": 3.674088478088379,
+      "learning_rate": 1.069182389937107e-05,
+      "loss": 0.0529,
+      "step": 4280
+    },
+    {
+      "epoch": 80.94339622641509,
+      "grad_norm": 3.663996934890747,
+      "learning_rate": 1.0587002096436059e-05,
+      "loss": 0.0641,
+      "step": 4290
+    },
+    {
+      "epoch": 81.0,
+      "eval_accuracy": 0.9634986225895317,
+      "eval_loss": 0.1491348147392273,
+      "eval_runtime": 21.3519,
+      "eval_samples_per_second": 136.007,
+      "eval_steps_per_second": 4.262,
+      "step": 4293
+    },
+    {
+      "epoch": 81.13207547169812,
+      "grad_norm": 3.4041941165924072,
+      "learning_rate": 1.0482180293501048e-05,
+      "loss": 0.0511,
+      "step": 4300
+    },
+    {
+      "epoch": 81.32075471698113,
+      "grad_norm": 5.157493591308594,
+      "learning_rate": 1.0377358490566038e-05,
+      "loss": 0.0471,
+      "step": 4310
+    },
+    {
+      "epoch": 81.50943396226415,
+      "grad_norm": 5.192855358123779,
+      "learning_rate": 1.0272536687631027e-05,
+      "loss": 0.0491,
+      "step": 4320
+    },
+    {
+      "epoch": 81.69811320754717,
+      "grad_norm": 2.236807346343994,
+      "learning_rate": 1.0167714884696017e-05,
+      "loss": 0.0358,
+      "step": 4330
+    },
+    {
+      "epoch": 81.88679245283019,
+      "grad_norm": 4.674067497253418,
+      "learning_rate": 1.0062893081761008e-05,
+      "loss": 0.0546,
+      "step": 4340
+    },
+    {
+      "epoch": 82.0,
+      "eval_accuracy": 0.9610881542699724,
+      "eval_loss": 0.15290114283561707,
+      "eval_runtime": 21.4172,
+      "eval_samples_per_second": 135.592,
+      "eval_steps_per_second": 4.249,
+      "step": 4346
+    },
+    {
+      "epoch": 82.0754716981132,
+      "grad_norm": 2.6720902919769287,
+      "learning_rate": 9.958071278825997e-06,
+      "loss": 0.0493,
+      "step": 4350
+    },
+    {
+      "epoch": 82.26415094339623,
+      "grad_norm": 3.191582441329956,
+      "learning_rate": 9.853249475890985e-06,
+      "loss": 0.0479,
+      "step": 4360
+    },
+    {
+      "epoch": 82.45283018867924,
+      "grad_norm": 6.1738481521606445,
+      "learning_rate": 9.748427672955976e-06,
+      "loss": 0.0517,
+      "step": 4370
+    },
+    {
+      "epoch": 82.64150943396227,
+      "grad_norm": 2.8287763595581055,
+      "learning_rate": 9.643605870020965e-06,
+      "loss": 0.0598,
+      "step": 4380
+    },
+    {
+      "epoch": 82.83018867924528,
+      "grad_norm": 2.7232823371887207,
+      "learning_rate": 9.538784067085953e-06,
+      "loss": 0.059,
+      "step": 4390
+    },
+    {
+      "epoch": 83.0,
+      "eval_accuracy": 0.9652203856749312,
+      "eval_loss": 0.14617380499839783,
+      "eval_runtime": 21.3846,
+      "eval_samples_per_second": 135.798,
+      "eval_steps_per_second": 4.255,
+      "step": 4399
+    },
+    {
+      "epoch": 83.01886792452831,
+      "grad_norm": 3.0003349781036377,
+      "learning_rate": 9.433962264150944e-06,
+      "loss": 0.061,
+      "step": 4400
+    },
+    {
+      "epoch": 83.20754716981132,
+      "grad_norm": 4.4709038734436035,
+      "learning_rate": 9.329140461215934e-06,
+      "loss": 0.0468,
+      "step": 4410
+    },
+    {
+      "epoch": 83.39622641509433,
+      "grad_norm": 3.809194564819336,
+      "learning_rate": 9.224318658280923e-06,
+      "loss": 0.048,
+      "step": 4420
+    },
+    {
+      "epoch": 83.58490566037736,
+      "grad_norm": 4.134964942932129,
+      "learning_rate": 9.119496855345912e-06,
+      "loss": 0.0593,
+      "step": 4430
+    },
+    {
+      "epoch": 83.77358490566037,
+      "grad_norm": 6.407557964324951,
+      "learning_rate": 9.014675052410902e-06,
+      "loss": 0.058,
+      "step": 4440
+    },
+    {
+      "epoch": 83.9622641509434,
+      "grad_norm": 2.055232048034668,
+      "learning_rate": 8.909853249475891e-06,
+      "loss": 0.0485,
+      "step": 4450
+    },
+    {
+      "epoch": 84.0,
+      "eval_accuracy": 0.9631542699724518,
+      "eval_loss": 0.15668189525604248,
+      "eval_runtime": 21.3505,
+      "eval_samples_per_second": 136.016,
+      "eval_steps_per_second": 4.262,
+      "step": 4452
+    },
+    {
+      "epoch": 84.15094339622641,
+      "grad_norm": 2.221158266067505,
+      "learning_rate": 8.80503144654088e-06,
+      "loss": 0.0555,
+      "step": 4460
+    },
+    {
+      "epoch": 84.33962264150944,
+      "grad_norm": 5.540987968444824,
+      "learning_rate": 8.70020964360587e-06,
+      "loss": 0.0507,
+      "step": 4470
+    },
+    {
+      "epoch": 84.52830188679245,
+      "grad_norm": 2.087411642074585,
+      "learning_rate": 8.59538784067086e-06,
+      "loss": 0.0485,
+      "step": 4480
+    },
+    {
+      "epoch": 84.71698113207547,
+      "grad_norm": 6.342270374298096,
+      "learning_rate": 8.49056603773585e-06,
+      "loss": 0.0533,
+      "step": 4490
+    },
+    {
+      "epoch": 84.90566037735849,
+      "grad_norm": 1.974360466003418,
+      "learning_rate": 8.38574423480084e-06,
+      "loss": 0.0388,
+      "step": 4500
+    },
+    {
+      "epoch": 85.0,
+      "eval_accuracy": 0.9621212121212122,
+      "eval_loss": 0.15479591488838196,
+      "eval_runtime": 21.3928,
+      "eval_samples_per_second": 135.747,
+      "eval_steps_per_second": 4.254,
+      "step": 4505
+    },
+    {
+      "epoch": 85.09433962264151,
+      "grad_norm": 5.822177410125732,
+      "learning_rate": 8.280922431865829e-06,
+      "loss": 0.0541,
+      "step": 4510
+    },
+    {
+      "epoch": 85.28301886792453,
+      "grad_norm": 6.9071149826049805,
+      "learning_rate": 8.176100628930818e-06,
+      "loss": 0.0382,
+      "step": 4520
+    },
+    {
+      "epoch": 85.47169811320755,
+      "grad_norm": 3.2203402519226074,
+      "learning_rate": 8.071278825995808e-06,
+      "loss": 0.0481,
+      "step": 4530
+    },
+    {
+      "epoch": 85.66037735849056,
+      "grad_norm": 2.526183843612671,
+      "learning_rate": 7.966457023060797e-06,
+      "loss": 0.0346,
+      "step": 4540
+    },
+    {
+      "epoch": 85.84905660377359,
+      "grad_norm": 1.8161990642547607,
+      "learning_rate": 7.861635220125786e-06,
+      "loss": 0.0421,
+      "step": 4550
+    },
+    {
+      "epoch": 86.0,
+      "eval_accuracy": 0.9621212121212122,
+      "eval_loss": 0.1483514904975891,
+      "eval_runtime": 21.3246,
+      "eval_samples_per_second": 136.181,
+      "eval_steps_per_second": 4.267,
+      "step": 4558
+    },
+    {
+      "epoch": 86.0377358490566,
+      "grad_norm": 4.167572021484375,
+      "learning_rate": 7.756813417190776e-06,
+      "loss": 0.0558,
+      "step": 4560
+    },
+    {
+      "epoch": 86.22641509433963,
+      "grad_norm": 7.080474853515625,
+      "learning_rate": 7.651991614255765e-06,
+      "loss": 0.0374,
+      "step": 4570
+    },
+    {
+      "epoch": 86.41509433962264,
+      "grad_norm": 3.921416759490967,
+      "learning_rate": 7.547169811320755e-06,
+      "loss": 0.0557,
+      "step": 4580
+    },
+    {
+      "epoch": 86.60377358490567,
+      "grad_norm": 2.932072639465332,
+      "learning_rate": 7.442348008385745e-06,
+      "loss": 0.0492,
+      "step": 4590
+    },
+    {
+      "epoch": 86.79245283018868,
+      "grad_norm": 3.091217041015625,
+      "learning_rate": 7.337526205450735e-06,
+      "loss": 0.0348,
+      "step": 4600
+    },
+    {
+      "epoch": 86.98113207547169,
+      "grad_norm": 3.1368813514709473,
+      "learning_rate": 7.2327044025157235e-06,
+      "loss": 0.0375,
+      "step": 4610
+    },
+    {
+      "epoch": 87.0,
+      "eval_accuracy": 0.9597107438016529,
+      "eval_loss": 0.1680881232023239,
+      "eval_runtime": 21.2539,
+      "eval_samples_per_second": 136.634,
+      "eval_steps_per_second": 4.282,
+      "step": 4611
+    },
+    {
+      "epoch": 87.16981132075472,
+      "grad_norm": 3.3242502212524414,
+      "learning_rate": 7.127882599580712e-06,
+      "loss": 0.0363,
+      "step": 4620
+    },
+    {
+      "epoch": 87.35849056603773,
+      "grad_norm": 5.762176990509033,
+      "learning_rate": 7.023060796645703e-06,
+      "loss": 0.0449,
+      "step": 4630
+    },
+    {
+      "epoch": 87.54716981132076,
+      "grad_norm": 2.077052593231201,
+      "learning_rate": 6.918238993710692e-06,
+      "loss": 0.0457,
+      "step": 4640
+    },
+    {
+      "epoch": 87.73584905660377,
+      "grad_norm": 2.2283737659454346,
+      "learning_rate": 6.813417190775681e-06,
+      "loss": 0.0397,
+      "step": 4650
+    },
+    {
+      "epoch": 87.9245283018868,
+      "grad_norm": 2.103003740310669,
+      "learning_rate": 6.708595387840672e-06,
+      "loss": 0.0376,
+      "step": 4660
+    },
+    {
+      "epoch": 88.0,
+      "eval_accuracy": 0.9631542699724518,
+      "eval_loss": 0.15125274658203125,
+      "eval_runtime": 21.3299,
+      "eval_samples_per_second": 136.147,
+      "eval_steps_per_second": 4.266,
+      "step": 4664
+    },
+    {
+      "epoch": 88.11320754716981,
+      "grad_norm": 9.418115615844727,
+      "learning_rate": 6.60377358490566e-06,
+      "loss": 0.0458,
+      "step": 4670
+    },
+    {
+      "epoch": 88.30188679245283,
+      "grad_norm": 2.476033926010132,
+      "learning_rate": 6.49895178197065e-06,
+      "loss": 0.0396,
+      "step": 4680
+    },
+    {
+      "epoch": 88.49056603773585,
+      "grad_norm": 2.3755970001220703,
+      "learning_rate": 6.3941299790356405e-06,
+      "loss": 0.0384,
+      "step": 4690
+    },
+    {
+      "epoch": 88.67924528301887,
+      "grad_norm": 2.3153653144836426,
+      "learning_rate": 6.289308176100629e-06,
+      "loss": 0.0357,
+      "step": 4700
+    },
+    {
+      "epoch": 88.86792452830188,
+      "grad_norm": 5.5892720222473145,
+      "learning_rate": 6.184486373165619e-06,
+      "loss": 0.0514,
+      "step": 4710
+    },
+    {
+      "epoch": 89.0,
+      "eval_accuracy": 0.9641873278236914,
+      "eval_loss": 0.148544043302536,
+      "eval_runtime": 21.3145,
+      "eval_samples_per_second": 136.245,
+      "eval_steps_per_second": 4.269,
+      "step": 4717
+    },
+    {
+      "epoch": 89.05660377358491,
+      "grad_norm": 3.5253398418426514,
+      "learning_rate": 6.079664570230608e-06,
+      "loss": 0.0286,
+      "step": 4720
+    },
+    {
+      "epoch": 89.24528301886792,
+      "grad_norm": 5.891650676727295,
+      "learning_rate": 5.974842767295598e-06,
+      "loss": 0.0607,
+      "step": 4730
+    },
+    {
+      "epoch": 89.43396226415095,
+      "grad_norm": 2.1808536052703857,
+      "learning_rate": 5.870020964360588e-06,
+      "loss": 0.0404,
+      "step": 4740
+    },
+    {
+      "epoch": 89.62264150943396,
+      "grad_norm": 6.388125896453857,
+      "learning_rate": 5.7651991614255766e-06,
+      "loss": 0.0469,
+      "step": 4750
+    },
+    {
+      "epoch": 89.81132075471699,
+      "grad_norm": 2.3668999671936035,
+      "learning_rate": 5.660377358490566e-06,
+      "loss": 0.0421,
+      "step": 4760
+    },
+    {
+      "epoch": 90.0,
+      "grad_norm": 3.1038739681243896,
+      "learning_rate": 5.555555555555556e-06,
+      "loss": 0.0598,
+      "step": 4770
+    },
+    {
+      "epoch": 90.0,
+      "eval_accuracy": 0.9638429752066116,
+      "eval_loss": 0.15414932370185852,
+      "eval_runtime": 21.2451,
+      "eval_samples_per_second": 136.69,
+      "eval_steps_per_second": 4.283,
+      "step": 4770
+    },
+    {
+      "epoch": 90.18867924528301,
+      "grad_norm": 3.297988176345825,
+      "learning_rate": 5.4507337526205454e-06,
+      "loss": 0.0397,
+      "step": 4780
+    },
+    {
+      "epoch": 90.37735849056604,
+      "grad_norm": 4.2701897621154785,
+      "learning_rate": 5.345911949685535e-06,
+      "loss": 0.0406,
+      "step": 4790
+    },
+    {
+      "epoch": 90.56603773584905,
+      "grad_norm": 3.7925121784210205,
+      "learning_rate": 5.241090146750524e-06,
+      "loss": 0.0458,
+      "step": 4800
+    },
+    {
+      "epoch": 90.75471698113208,
+      "grad_norm": 4.8097686767578125,
+      "learning_rate": 5.1362683438155135e-06,
+      "loss": 0.0555,
+      "step": 4810
+    },
+    {
+      "epoch": 90.94339622641509,
+      "grad_norm": 3.2213995456695557,
+      "learning_rate": 5.031446540880504e-06,
+      "loss": 0.0431,
+      "step": 4820
+    },
+    {
+      "epoch": 91.0,
+      "eval_accuracy": 0.9628099173553719,
+      "eval_loss": 0.14735093712806702,
+      "eval_runtime": 21.3261,
+      "eval_samples_per_second": 136.171,
+      "eval_steps_per_second": 4.267,
+      "step": 4823
+    },
+    {
+      "epoch": 91.13207547169812,
+      "grad_norm": 3.0545461177825928,
+      "learning_rate": 4.926624737945493e-06,
+      "loss": 0.0415,
+      "step": 4830
+    },
+    {
+      "epoch": 91.32075471698113,
+      "grad_norm": 3.0625436305999756,
+      "learning_rate": 4.821802935010482e-06,
+      "loss": 0.0454,
+      "step": 4840
+    },
+    {
+      "epoch": 91.50943396226415,
+      "grad_norm": 3.4635112285614014,
+      "learning_rate": 4.716981132075472e-06,
+      "loss": 0.0276,
+      "step": 4850
+    },
+    {
+      "epoch": 91.69811320754717,
+      "grad_norm": 2.14255428314209,
+      "learning_rate": 4.612159329140462e-06,
+      "loss": 0.0341,
+      "step": 4860
+    },
+    {
+      "epoch": 91.88679245283019,
+      "grad_norm": 5.496433258056641,
+      "learning_rate": 4.507337526205451e-06,
+      "loss": 0.0432,
+      "step": 4870
+    },
+    {
+      "epoch": 92.0,
+      "eval_accuracy": 0.9645316804407713,
+      "eval_loss": 0.14980562031269073,
+      "eval_runtime": 21.3199,
+      "eval_samples_per_second": 136.211,
+      "eval_steps_per_second": 4.268,
+      "step": 4876
+    },
+    {
+      "epoch": 92.0754716981132,
+      "grad_norm": 4.054795265197754,
+      "learning_rate": 4.40251572327044e-06,
+      "loss": 0.0529,
+      "step": 4880
+    },
+    {
+      "epoch": 92.26415094339623,
+      "grad_norm": 5.012380123138428,
+      "learning_rate": 4.29769392033543e-06,
+      "loss": 0.0492,
+      "step": 4890
+    },
+    {
+      "epoch": 92.45283018867924,
+      "grad_norm": 0.6688435673713684,
+      "learning_rate": 4.19287211740042e-06,
+      "loss": 0.052,
+      "step": 4900
+    },
+    {
+      "epoch": 92.64150943396227,
+      "grad_norm": 3.9398372173309326,
+      "learning_rate": 4.088050314465409e-06,
+      "loss": 0.0468,
+      "step": 4910
+    },
+    {
+      "epoch": 92.83018867924528,
+      "grad_norm": 1.439302682876587,
+      "learning_rate": 3.9832285115303985e-06,
+      "loss": 0.0391,
+      "step": 4920
+    },
+    {
+      "epoch": 93.0,
+      "eval_accuracy": 0.9645316804407713,
+      "eval_loss": 0.15064947307109833,
+      "eval_runtime": 21.2878,
+      "eval_samples_per_second": 136.416,
+      "eval_steps_per_second": 4.275,
+      "step": 4929
+    },
+    {
+      "epoch": 93.01886792452831,
+      "grad_norm": 2.215027093887329,
+      "learning_rate": 3.878406708595388e-06,
+      "loss": 0.0322,
+      "step": 4930
+    },
+    {
+      "epoch": 93.20754716981132,
+      "grad_norm": 4.157674312591553,
+      "learning_rate": 3.7735849056603773e-06,
+      "loss": 0.0481,
+      "step": 4940
+    },
+    {
+      "epoch": 93.39622641509433,
+      "grad_norm": 2.998509645462036,
+      "learning_rate": 3.6687631027253674e-06,
+      "loss": 0.0424,
+      "step": 4950
+    },
+    {
+      "epoch": 93.58490566037736,
+      "grad_norm": 2.701420783996582,
+      "learning_rate": 3.563941299790356e-06,
+      "loss": 0.0347,
+      "step": 4960
+    },
+    {
+      "epoch": 93.77358490566037,
+      "grad_norm": 2.793337821960449,
+      "learning_rate": 3.459119496855346e-06,
+      "loss": 0.0448,
+      "step": 4970
+    },
+    {
+      "epoch": 93.9622641509434,
+      "grad_norm": 2.3559460639953613,
+      "learning_rate": 3.354297693920336e-06,
+      "loss": 0.0408,
+      "step": 4980
+    },
+    {
+      "epoch": 94.0,
+      "eval_accuracy": 0.9641873278236914,
+      "eval_loss": 0.1462460607290268,
+      "eval_runtime": 21.3706,
+      "eval_samples_per_second": 135.888,
+      "eval_steps_per_second": 4.258,
+      "step": 4982
+    },
+    {
+      "epoch": 94.15094339622641,
+      "grad_norm": 1.9885700941085815,
+      "learning_rate": 3.249475890985325e-06,
+      "loss": 0.0373,
+      "step": 4990
+    },
+    {
+      "epoch": 94.33962264150944,
+      "grad_norm": 4.4361443519592285,
+      "learning_rate": 3.1446540880503146e-06,
+      "loss": 0.0598,
+      "step": 5000
+    },
+    {
+      "epoch": 94.52830188679245,
+      "grad_norm": 3.5327351093292236,
+      "learning_rate": 3.039832285115304e-06,
+      "loss": 0.0388,
+      "step": 5010
+    },
+    {
+      "epoch": 94.71698113207547,
+      "grad_norm": 0.833363950252533,
+      "learning_rate": 2.935010482180294e-06,
+      "loss": 0.0333,
+      "step": 5020
+    },
+    {
+      "epoch": 94.90566037735849,
+      "grad_norm": 3.4921486377716064,
+      "learning_rate": 2.830188679245283e-06,
+      "loss": 0.0335,
+      "step": 5030
+    },
+    {
+      "epoch": 95.0,
+      "eval_accuracy": 0.9652203856749312,
+      "eval_loss": 0.15087169408798218,
+      "eval_runtime": 21.4165,
+      "eval_samples_per_second": 135.597,
+      "eval_steps_per_second": 4.249,
+      "step": 5035
+    },
+    {
+      "epoch": 95.09433962264151,
+      "grad_norm": 4.499444484710693,
+      "learning_rate": 2.7253668763102727e-06,
+      "loss": 0.0394,
+      "step": 5040
+    },
+    {
+      "epoch": 95.28301886792453,
+      "grad_norm": 4.537922382354736,
+      "learning_rate": 2.620545073375262e-06,
+      "loss": 0.0428,
+      "step": 5050
+    },
+    {
+      "epoch": 95.47169811320755,
+      "grad_norm": 1.5033186674118042,
+      "learning_rate": 2.515723270440252e-06,
+      "loss": 0.0363,
+      "step": 5060
+    },
+    {
+      "epoch": 95.66037735849056,
+      "grad_norm": 0.8121969103813171,
+      "learning_rate": 2.410901467505241e-06,
+      "loss": 0.0298,
+      "step": 5070
+    },
+    {
+      "epoch": 95.84905660377359,
+      "grad_norm": 1.9728448390960693,
+      "learning_rate": 2.306079664570231e-06,
+      "loss": 0.0447,
+      "step": 5080
+    },
+    {
+      "epoch": 96.0,
+      "eval_accuracy": 0.9634986225895317,
+      "eval_loss": 0.15081696212291718,
+      "eval_runtime": 21.3522,
+      "eval_samples_per_second": 136.005,
+      "eval_steps_per_second": 4.262,
+      "step": 5088
+    },
+    {
+      "epoch": 96.0377358490566,
+      "grad_norm": 2.9549922943115234,
+      "learning_rate": 2.20125786163522e-06,
+      "loss": 0.0513,
+      "step": 5090
+    },
+    {
+      "epoch": 96.22641509433963,
+      "grad_norm": 4.062204837799072,
+      "learning_rate": 2.09643605870021e-06,
+      "loss": 0.0449,
+      "step": 5100
+    },
+    {
+      "epoch": 96.41509433962264,
+      "grad_norm": 2.071183443069458,
+      "learning_rate": 1.9916142557651992e-06,
+      "loss": 0.04,
+      "step": 5110
+    },
+    {
+      "epoch": 96.60377358490567,
+      "grad_norm": 0.9506546258926392,
+      "learning_rate": 1.8867924528301887e-06,
+      "loss": 0.0321,
+      "step": 5120
+    },
+    {
+      "epoch": 96.79245283018868,
+      "grad_norm": 2.33597469329834,
+      "learning_rate": 1.781970649895178e-06,
+      "loss": 0.024,
+      "step": 5130
+    },
+    {
+      "epoch": 96.98113207547169,
+      "grad_norm": 3.253387451171875,
+      "learning_rate": 1.677148846960168e-06,
+      "loss": 0.0477,
+      "step": 5140
+    },
+    {
+      "epoch": 97.0,
+      "eval_accuracy": 0.9634986225895317,
+      "eval_loss": 0.1510278433561325,
+      "eval_runtime": 21.3381,
+      "eval_samples_per_second": 136.095,
+      "eval_steps_per_second": 4.265,
+      "step": 5141
+    },
+    {
+      "epoch": 97.16981132075472,
+      "grad_norm": 2.858771324157715,
+      "learning_rate": 1.5723270440251573e-06,
+      "loss": 0.0386,
+      "step": 5150
+    },
+    {
+      "epoch": 97.35849056603773,
+      "grad_norm": 6.633811950683594,
+      "learning_rate": 1.467505241090147e-06,
+      "loss": 0.0419,
+      "step": 5160
+    },
+    {
+      "epoch": 97.54716981132076,
+      "grad_norm": 3.115361452102661,
+      "learning_rate": 1.3626834381551364e-06,
+      "loss": 0.0291,
+      "step": 5170
+    },
+    {
+      "epoch": 97.73584905660377,
+      "grad_norm": 3.6518330574035645,
+      "learning_rate": 1.257861635220126e-06,
+      "loss": 0.0372,
+      "step": 5180
+    },
+    {
+      "epoch": 97.9245283018868,
+      "grad_norm": 3.8466641902923584,
+      "learning_rate": 1.1530398322851154e-06,
+      "loss": 0.0504,
+      "step": 5190
+    },
+    {
+      "epoch": 98.0,
+      "eval_accuracy": 0.9641873278236914,
+      "eval_loss": 0.15101788938045502,
+      "eval_runtime": 21.5013,
+      "eval_samples_per_second": 135.061,
+      "eval_steps_per_second": 4.232,
+      "step": 5194
+    },
+    {
+      "epoch": 98.11320754716981,
+      "grad_norm": 4.916390895843506,
+      "learning_rate": 1.048218029350105e-06,
+      "loss": 0.0474,
+      "step": 5200
+    },
+    {
+      "epoch": 98.30188679245283,
+      "grad_norm": 3.228938102722168,
+      "learning_rate": 9.433962264150943e-07,
+      "loss": 0.0401,
+      "step": 5210
+    },
+    {
+      "epoch": 98.49056603773585,
+      "grad_norm": 1.5599745512008667,
+      "learning_rate": 8.38574423480084e-07,
+      "loss": 0.0439,
+      "step": 5220
+    },
+    {
+      "epoch": 98.67924528301887,
+      "grad_norm": 2.1779186725616455,
+      "learning_rate": 7.337526205450735e-07,
+      "loss": 0.0411,
+      "step": 5230
+    },
+    {
+      "epoch": 98.86792452830188,
+      "grad_norm": 3.2559657096862793,
+      "learning_rate": 6.28930817610063e-07,
+      "loss": 0.0406,
+      "step": 5240
+    },
+    {
+      "epoch": 99.0,
+      "eval_accuracy": 0.9648760330578512,
+      "eval_loss": 0.14794644713401794,
+      "eval_runtime": 21.5282,
+      "eval_samples_per_second": 134.893,
+      "eval_steps_per_second": 4.227,
+      "step": 5247
+    },
+    {
+      "epoch": 99.05660377358491,
+      "grad_norm": 2.303928852081299,
+      "learning_rate": 5.241090146750525e-07,
+      "loss": 0.0461,
+      "step": 5250
+    },
+    {
+      "epoch": 99.24528301886792,
+      "grad_norm": 5.052129745483398,
+      "learning_rate": 4.19287211740042e-07,
+      "loss": 0.0398,
+      "step": 5260
+    },
+    {
+      "epoch": 99.43396226415095,
+      "grad_norm": 5.838912010192871,
+      "learning_rate": 3.144654088050315e-07,
+      "loss": 0.0436,
+      "step": 5270
+    },
+    {
+      "epoch": 99.62264150943396,
+      "grad_norm": 1.3886405229568481,
+      "learning_rate": 2.09643605870021e-07,
+      "loss": 0.0281,
+      "step": 5280
+    },
+    {
+      "epoch": 99.81132075471699,
+      "grad_norm": 2.893517017364502,
+      "learning_rate": 1.048218029350105e-07,
+      "loss": 0.0495,
+      "step": 5290
+    },
+    {
+      "epoch": 100.0,
+      "grad_norm": 3.087257146835327,
+      "learning_rate": 0.0,
+      "loss": 0.0343,
+      "step": 5300
+    },
+    {
+      "epoch": 100.0,
+      "eval_accuracy": 0.9645316804407713,
+      "eval_loss": 0.1479804962873459,
+      "eval_runtime": 21.646,
+      "eval_samples_per_second": 134.159,
+      "eval_steps_per_second": 4.204,
+      "step": 5300
+    },
+    {
+      "epoch": 100.0,
+      "step": 5300,
+      "total_flos": 1.6838724615023002e+19,
+      "train_loss": 0.14425156983845638,
+      "train_runtime": 35971.5646,
+      "train_samples_per_second": 18.829,
+      "train_steps_per_second": 0.147
     }
   ],
   "logging_steps": 10,
-  "max_steps": 530,
+  "max_steps": 5300,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 10,
+  "num_train_epochs": 100,
   "save_steps": 500,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -496,7 +4645,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.6836842977571635e+18,
+  "total_flos": 1.6838724615023002e+19,
   "train_batch_size": 32,
   "trial_name": null,
   "trial_params": null