diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,1459 +1,6389 @@
 {
-  "best_metric": 0.5917085427135679,
-  "best_model_checkpoint": "swin-tiny-patch4-window7-224-finetuned-eurosat/checkpoint-224",
-  "epoch": 50.0,
+  "best_metric": 0.9816143497757848,
+  "best_model_checkpoint": "swin-tiny-patch4-window7-224-finetuned-eurosat/checkpoint-7800",
+  "epoch": 99.36305732484077,
   "eval_steps": 500,
-  "global_step": 1400,
+  "global_step": 7800,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.36,
-      "grad_norm": 2.619962215423584,
-      "learning_rate": 3.5714285714285714e-06,
-      "loss": 1.0415,
+      "epoch": 0.13,
+      "grad_norm": 2.7006075382232666,
+      "learning_rate": 6.41025641025641e-07,
+      "loss": 1.0769,
       "step": 10
     },
     {
-      "epoch": 0.71,
-      "grad_norm": 2.402837038040161,
-      "learning_rate": 7.142857142857143e-06,
-      "loss": 1.0339,
+      "epoch": 0.25,
+      "grad_norm": 2.1050777435302734,
+      "learning_rate": 1.282051282051282e-06,
+      "loss": 1.0669,
       "step": 20
     },
     {
-      "epoch": 1.0,
-      "eval_accuracy": 0.5640703517587939,
-      "eval_loss": 1.0541082620620728,
-      "eval_runtime": 5.8961,
-      "eval_samples_per_second": 270.009,
-      "eval_steps_per_second": 2.205,
-      "step": 28
-    },
-    {
-      "epoch": 1.07,
-      "grad_norm": 2.223407745361328,
-      "learning_rate": 1.0714285714285714e-05,
-      "loss": 1.025,
+      "epoch": 0.38,
+      "grad_norm": 3.808070182800293,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 1.0657,
       "step": 30
     },
     {
-      "epoch": 1.43,
-      "grad_norm": 2.5425541400909424,
-      "learning_rate": 1.4285714285714285e-05,
-      "loss": 1.0072,
+      "epoch": 0.51,
+      "grad_norm": 2.6294384002685547,
+      "learning_rate": 2.564102564102564e-06,
+      "loss": 1.0561,
       "step": 40
     },
     {
-      "epoch": 1.79,
-      "grad_norm": 3.219869375228882,
-      "learning_rate": 1.785714285714286e-05,
-      "loss": 1.0193,
+      "epoch": 0.64,
+      "grad_norm": 4.274094104766846,
+      "learning_rate": 3.205128205128205e-06,
+      "loss": 1.0414,
       "step": 50
     },
     {
-      "epoch": 2.0,
-      "eval_accuracy": 0.5621859296482412,
-      "eval_loss": 1.0463842153549194,
-      "eval_runtime": 5.8033,
-      "eval_samples_per_second": 274.326,
-      "eval_steps_per_second": 2.24,
-      "step": 56
-    },
-    {
-      "epoch": 2.14,
-      "grad_norm": 3.941281318664551,
-      "learning_rate": 2.1428571428571428e-05,
-      "loss": 1.0133,
+      "epoch": 0.76,
+      "grad_norm": 2.047131061553955,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 1.0187,
       "step": 60
     },
     {
-      "epoch": 2.5,
-      "grad_norm": 4.169382095336914,
-      "learning_rate": 2.5e-05,
-      "loss": 1.0182,
+      "epoch": 0.89,
+      "grad_norm": 2.6211767196655273,
+      "learning_rate": 4.487179487179488e-06,
+      "loss": 1.0105,
       "step": 70
     },
     {
-      "epoch": 2.86,
-      "grad_norm": 3.475947380065918,
-      "learning_rate": 2.857142857142857e-05,
-      "loss": 1.0348,
-      "step": 80
+      "epoch": 0.99,
+      "eval_accuracy": 0.5251121076233184,
+      "eval_loss": 0.9788457155227661,
+      "eval_runtime": 11.1593,
+      "eval_samples_per_second": 399.665,
+      "eval_steps_per_second": 3.136,
+      "step": 78
     },
     {
-      "epoch": 3.0,
-      "eval_accuracy": 0.5690954773869347,
-      "eval_loss": 1.0330955982208252,
-      "eval_runtime": 5.8594,
-      "eval_samples_per_second": 271.702,
-      "eval_steps_per_second": 2.219,
-      "step": 84
+      "epoch": 1.02,
+      "grad_norm": 2.5963022708892822,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 0.9866,
+      "step": 80
     },
     {
-      "epoch": 3.21,
-      "grad_norm": 3.1382639408111572,
-      "learning_rate": 3.2142857142857144e-05,
-      "loss": 1.0266,
+      "epoch": 1.15,
+      "grad_norm": 4.090935230255127,
+      "learning_rate": 5.76923076923077e-06,
+      "loss": 0.9689,
       "step": 90
     },
     {
-      "epoch": 3.57,
-      "grad_norm": 2.344207525253296,
-      "learning_rate": 3.571428571428572e-05,
-      "loss": 1.0075,
+      "epoch": 1.27,
+      "grad_norm": 3.1825904846191406,
+      "learning_rate": 6.41025641025641e-06,
+      "loss": 0.9478,
       "step": 100
     },
     {
-      "epoch": 3.93,
-      "grad_norm": 2.2122068405151367,
-      "learning_rate": 3.928571428571429e-05,
-      "loss": 1.0072,
+      "epoch": 1.4,
+      "grad_norm": 3.811648368835449,
+      "learning_rate": 7.051282051282052e-06,
+      "loss": 0.933,
       "step": 110
     },
     {
-      "epoch": 4.0,
-      "eval_accuracy": 0.5847989949748744,
-      "eval_loss": 1.0253527164459229,
-      "eval_runtime": 5.8726,
-      "eval_samples_per_second": 271.091,
-      "eval_steps_per_second": 2.214,
-      "step": 112
-    },
-    {
-      "epoch": 4.29,
-      "grad_norm": 2.9957728385925293,
-      "learning_rate": 4.2857142857142856e-05,
-      "loss": 1.0024,
+      "epoch": 1.53,
+      "grad_norm": 3.8240082263946533,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.9141,
       "step": 120
     },
     {
-      "epoch": 4.64,
-      "grad_norm": 2.9330532550811768,
-      "learning_rate": 4.642857142857143e-05,
-      "loss": 0.991,
+      "epoch": 1.66,
+      "grad_norm": 3.7604594230651855,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.8997,
       "step": 130
     },
     {
-      "epoch": 5.0,
-      "grad_norm": 2.355830430984497,
-      "learning_rate": 5e-05,
-      "loss": 0.9892,
-      "step": 140
-    },
-    {
-      "epoch": 5.0,
-      "eval_accuracy": 0.5753768844221105,
-      "eval_loss": 1.012104868888855,
-      "eval_runtime": 5.622,
-      "eval_samples_per_second": 283.175,
-      "eval_steps_per_second": 2.312,
+      "epoch": 1.78,
+      "grad_norm": 5.058235168457031,
+      "learning_rate": 8.974358974358976e-06,
+      "loss": 0.8839,
       "step": 140
     },
     {
-      "epoch": 5.36,
-      "grad_norm": 3.8320775032043457,
-      "learning_rate": 4.960317460317461e-05,
-      "loss": 0.9517,
+      "epoch": 1.91,
+      "grad_norm": 4.036779403686523,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.8749,
       "step": 150
     },
     {
-      "epoch": 5.71,
-      "grad_norm": 3.0432510375976562,
-      "learning_rate": 4.9206349206349204e-05,
-      "loss": 0.9379,
+      "epoch": 2.0,
+      "eval_accuracy": 0.6322869955156951,
+      "eval_loss": 0.8166685104370117,
+      "eval_runtime": 10.7913,
+      "eval_samples_per_second": 413.295,
+      "eval_steps_per_second": 3.243,
+      "step": 157
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 3.1002399921417236,
+      "learning_rate": 1.0256410256410256e-05,
+      "loss": 0.8482,
       "step": 160
     },
     {
-      "epoch": 6.0,
-      "eval_accuracy": 0.5810301507537688,
-      "eval_loss": 1.017525553703308,
-      "eval_runtime": 6.0814,
-      "eval_samples_per_second": 261.781,
-      "eval_steps_per_second": 2.138,
-      "step": 168
-    },
-    {
-      "epoch": 6.07,
-      "grad_norm": 3.3289706707000732,
-      "learning_rate": 4.880952380952381e-05,
-      "loss": 0.9475,
+      "epoch": 2.17,
+      "grad_norm": 4.336329460144043,
+      "learning_rate": 1.0897435897435898e-05,
+      "loss": 0.835,
       "step": 170
     },
     {
-      "epoch": 6.43,
-      "grad_norm": 3.335624933242798,
-      "learning_rate": 4.841269841269841e-05,
-      "loss": 0.9103,
+      "epoch": 2.29,
+      "grad_norm": 4.344544410705566,
+      "learning_rate": 1.153846153846154e-05,
+      "loss": 0.7967,
       "step": 180
     },
     {
-      "epoch": 6.79,
-      "grad_norm": 3.2209532260894775,
-      "learning_rate": 4.801587301587302e-05,
-      "loss": 0.9123,
+      "epoch": 2.42,
+      "grad_norm": 5.955235481262207,
+      "learning_rate": 1.217948717948718e-05,
+      "loss": 0.7853,
       "step": 190
     },
     {
-      "epoch": 7.0,
-      "eval_accuracy": 0.5866834170854272,
-      "eval_loss": 1.0119962692260742,
-      "eval_runtime": 5.8208,
-      "eval_samples_per_second": 273.502,
-      "eval_steps_per_second": 2.233,
-      "step": 196
-    },
-    {
-      "epoch": 7.14,
-      "grad_norm": 3.194115400314331,
-      "learning_rate": 4.761904761904762e-05,
-      "loss": 0.896,
+      "epoch": 2.55,
+      "grad_norm": 6.097214698791504,
+      "learning_rate": 1.282051282051282e-05,
+      "loss": 0.7587,
       "step": 200
     },
     {
-      "epoch": 7.5,
-      "grad_norm": 2.89758563041687,
-      "learning_rate": 4.722222222222222e-05,
-      "loss": 0.9013,
+      "epoch": 2.68,
+      "grad_norm": 6.144845485687256,
+      "learning_rate": 1.3461538461538462e-05,
+      "loss": 0.7538,
       "step": 210
     },
     {
-      "epoch": 7.86,
-      "grad_norm": 3.0895237922668457,
-      "learning_rate": 4.682539682539683e-05,
-      "loss": 0.8865,
+      "epoch": 2.8,
+      "grad_norm": 6.462852954864502,
+      "learning_rate": 1.4102564102564104e-05,
+      "loss": 0.7286,
       "step": 220
     },
     {
-      "epoch": 8.0,
-      "eval_accuracy": 0.5917085427135679,
-      "eval_loss": 1.0104140043258667,
-      "eval_runtime": 6.0762,
-      "eval_samples_per_second": 262.008,
-      "eval_steps_per_second": 2.14,
-      "step": 224
-    },
-    {
-      "epoch": 8.21,
-      "grad_norm": 2.715590715408325,
-      "learning_rate": 4.642857142857143e-05,
-      "loss": 0.8758,
+      "epoch": 2.93,
+      "grad_norm": 5.17026424407959,
+      "learning_rate": 1.4743589743589745e-05,
+      "loss": 0.7352,
       "step": 230
     },
     {
-      "epoch": 8.57,
-      "grad_norm": 3.639813184738159,
-      "learning_rate": 4.603174603174603e-05,
-      "loss": 0.8516,
-      "step": 240
+      "epoch": 2.99,
+      "eval_accuracy": 0.7179372197309417,
+      "eval_loss": 0.6497997641563416,
+      "eval_runtime": 10.8789,
+      "eval_samples_per_second": 409.968,
+      "eval_steps_per_second": 3.217,
+      "step": 235
     },
     {
-      "epoch": 8.93,
-      "grad_norm": 3.1830241680145264,
-      "learning_rate": 4.563492063492064e-05,
-      "loss": 0.8668,
-      "step": 250
+      "epoch": 3.06,
+      "grad_norm": 5.981796741485596,
+      "learning_rate": 1.5384615384615387e-05,
+      "loss": 0.6987,
+      "step": 240
     },
     {
-      "epoch": 9.0,
-      "eval_accuracy": 0.5873115577889447,
-      "eval_loss": 1.0236343145370483,
-      "eval_runtime": 6.0393,
-      "eval_samples_per_second": 263.608,
-      "eval_steps_per_second": 2.153,
-      "step": 252
+      "epoch": 3.18,
+      "grad_norm": 6.265073299407959,
+      "learning_rate": 1.602564102564103e-05,
+      "loss": 0.6902,
+      "step": 250
     },
     {
-      "epoch": 9.29,
-      "grad_norm": 3.1714346408843994,
-      "learning_rate": 4.523809523809524e-05,
-      "loss": 0.8352,
+      "epoch": 3.31,
+      "grad_norm": 3.7911741733551025,
+      "learning_rate": 1.6666666666666667e-05,
+      "loss": 0.6888,
       "step": 260
     },
     {
-      "epoch": 9.64,
-      "grad_norm": 3.9194183349609375,
-      "learning_rate": 4.4841269841269846e-05,
-      "loss": 0.8207,
+      "epoch": 3.44,
+      "grad_norm": 5.221859455108643,
+      "learning_rate": 1.730769230769231e-05,
+      "loss": 0.6771,
       "step": 270
     },
     {
-      "epoch": 10.0,
-      "grad_norm": 4.183494567871094,
-      "learning_rate": 4.4444444444444447e-05,
-      "loss": 0.8189,
-      "step": 280
-    },
-    {
-      "epoch": 10.0,
-      "eval_accuracy": 0.5829145728643216,
-      "eval_loss": 1.0359796285629272,
-      "eval_runtime": 6.1118,
-      "eval_samples_per_second": 260.481,
-      "eval_steps_per_second": 2.127,
+      "epoch": 3.57,
+      "grad_norm": 4.222895622253418,
+      "learning_rate": 1.794871794871795e-05,
+      "loss": 0.6579,
       "step": 280
     },
     {
-      "epoch": 10.36,
-      "grad_norm": 3.281873941421509,
-      "learning_rate": 4.404761904761905e-05,
-      "loss": 0.8051,
+      "epoch": 3.69,
+      "grad_norm": 4.25372314453125,
+      "learning_rate": 1.858974358974359e-05,
+      "loss": 0.6529,
       "step": 290
     },
     {
-      "epoch": 10.71,
-      "grad_norm": 2.9959723949432373,
-      "learning_rate": 4.3650793650793655e-05,
-      "loss": 0.7933,
+      "epoch": 3.82,
+      "grad_norm": 6.471427917480469,
+      "learning_rate": 1.923076923076923e-05,
+      "loss": 0.6488,
       "step": 300
     },
     {
-      "epoch": 11.0,
-      "eval_accuracy": 0.5835427135678392,
-      "eval_loss": 1.039525032043457,
-      "eval_runtime": 5.9823,
-      "eval_samples_per_second": 266.118,
-      "eval_steps_per_second": 2.173,
-      "step": 308
-    },
-    {
-      "epoch": 11.07,
-      "grad_norm": 3.470358371734619,
-      "learning_rate": 4.3253968253968256e-05,
-      "loss": 0.797,
+      "epoch": 3.95,
+      "grad_norm": 3.6297073364257812,
+      "learning_rate": 1.987179487179487e-05,
+      "loss": 0.647,
       "step": 310
     },
     {
-      "epoch": 11.43,
-      "grad_norm": 3.285288095474243,
-      "learning_rate": 4.2857142857142856e-05,
-      "loss": 0.7723,
+      "epoch": 4.0,
+      "eval_accuracy": 0.7630044843049327,
+      "eval_loss": 0.5618513226509094,
+      "eval_runtime": 10.9103,
+      "eval_samples_per_second": 408.789,
+      "eval_steps_per_second": 3.208,
+      "step": 314
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 3.593230724334717,
+      "learning_rate": 2.0512820512820512e-05,
+      "loss": 0.6343,
       "step": 320
     },
     {
-      "epoch": 11.79,
-      "grad_norm": 4.59849739074707,
-      "learning_rate": 4.2460317460317464e-05,
-      "loss": 0.7765,
+      "epoch": 4.2,
+      "grad_norm": 5.996413230895996,
+      "learning_rate": 2.1153846153846154e-05,
+      "loss": 0.6266,
       "step": 330
     },
     {
-      "epoch": 12.0,
-      "eval_accuracy": 0.5728643216080402,
-      "eval_loss": 1.0594085454940796,
-      "eval_runtime": 5.9856,
-      "eval_samples_per_second": 265.972,
-      "eval_steps_per_second": 2.172,
-      "step": 336
-    },
-    {
-      "epoch": 12.14,
-      "grad_norm": 4.7830305099487305,
-      "learning_rate": 4.2063492063492065e-05,
-      "loss": 0.7567,
+      "epoch": 4.33,
+      "grad_norm": 3.9877495765686035,
+      "learning_rate": 2.1794871794871795e-05,
+      "loss": 0.629,
       "step": 340
     },
     {
-      "epoch": 12.5,
-      "grad_norm": 3.7676315307617188,
-      "learning_rate": 4.166666666666667e-05,
-      "loss": 0.738,
+      "epoch": 4.46,
+      "grad_norm": 3.9979002475738525,
+      "learning_rate": 2.2435897435897437e-05,
+      "loss": 0.6199,
       "step": 350
     },
     {
-      "epoch": 12.86,
-      "grad_norm": 4.092833518981934,
-      "learning_rate": 4.126984126984127e-05,
-      "loss": 0.7538,
+      "epoch": 4.59,
+      "grad_norm": 5.1187639236450195,
+      "learning_rate": 2.307692307692308e-05,
+      "loss": 0.601,
       "step": 360
     },
     {
-      "epoch": 13.0,
-      "eval_accuracy": 0.5879396984924623,
-      "eval_loss": 1.0552036762237549,
-      "eval_runtime": 5.8071,
-      "eval_samples_per_second": 274.146,
-      "eval_steps_per_second": 2.239,
-      "step": 364
-    },
-    {
-      "epoch": 13.21,
-      "grad_norm": 4.582472801208496,
-      "learning_rate": 4.0873015873015874e-05,
-      "loss": 0.7175,
+      "epoch": 4.71,
+      "grad_norm": 3.832275629043579,
+      "learning_rate": 2.3717948717948718e-05,
+      "loss": 0.601,
       "step": 370
     },
     {
-      "epoch": 13.57,
-      "grad_norm": 4.479797840118408,
-      "learning_rate": 4.047619047619048e-05,
-      "loss": 0.7101,
+      "epoch": 4.84,
+      "grad_norm": 3.131157875061035,
+      "learning_rate": 2.435897435897436e-05,
+      "loss": 0.5921,
       "step": 380
     },
     {
-      "epoch": 13.93,
-      "grad_norm": 4.223387241363525,
-      "learning_rate": 4.007936507936508e-05,
-      "loss": 0.7146,
+      "epoch": 4.97,
+      "grad_norm": 5.246669769287109,
+      "learning_rate": 2.5e-05,
+      "loss": 0.5949,
       "step": 390
     },
     {
-      "epoch": 14.0,
-      "eval_accuracy": 0.5829145728643216,
-      "eval_loss": 1.0619994401931763,
-      "eval_runtime": 5.8703,
-      "eval_samples_per_second": 271.194,
-      "eval_steps_per_second": 2.215,
+      "epoch": 4.99,
+      "eval_accuracy": 0.7892376681614349,
+      "eval_loss": 0.5067703127861023,
+      "eval_runtime": 10.7897,
+      "eval_samples_per_second": 413.359,
+      "eval_steps_per_second": 3.244,
       "step": 392
     },
     {
-      "epoch": 14.29,
-      "grad_norm": 4.616257190704346,
-      "learning_rate": 3.968253968253968e-05,
-      "loss": 0.7031,
+      "epoch": 5.1,
+      "grad_norm": 3.668297052383423,
+      "learning_rate": 2.564102564102564e-05,
+      "loss": 0.5841,
       "step": 400
     },
     {
-      "epoch": 14.64,
-      "grad_norm": 3.6395843029022217,
-      "learning_rate": 3.928571428571429e-05,
-      "loss": 0.6901,
+      "epoch": 5.22,
+      "grad_norm": 3.633014440536499,
+      "learning_rate": 2.6282051282051285e-05,
+      "loss": 0.5847,
       "step": 410
     },
     {
-      "epoch": 15.0,
-      "grad_norm": 3.426710367202759,
-      "learning_rate": 3.888888888888889e-05,
-      "loss": 0.6885,
-      "step": 420
-    },
-    {
-      "epoch": 15.0,
-      "eval_accuracy": 0.5841708542713567,
-      "eval_loss": 1.0782768726348877,
-      "eval_runtime": 6.0171,
-      "eval_samples_per_second": 264.578,
-      "eval_steps_per_second": 2.16,
+      "epoch": 5.35,
+      "grad_norm": 4.339975833892822,
+      "learning_rate": 2.6923076923076923e-05,
+      "loss": 0.5921,
       "step": 420
     },
     {
-      "epoch": 15.36,
-      "grad_norm": 3.5050458908081055,
-      "learning_rate": 3.84920634920635e-05,
-      "loss": 0.6697,
+      "epoch": 5.48,
+      "grad_norm": 4.13716459274292,
+      "learning_rate": 2.756410256410257e-05,
+      "loss": 0.5648,
       "step": 430
     },
     {
-      "epoch": 15.71,
-      "grad_norm": 4.314873218536377,
-      "learning_rate": 3.809523809523809e-05,
-      "loss": 0.6556,
+      "epoch": 5.61,
+      "grad_norm": 5.4236063957214355,
+      "learning_rate": 2.8205128205128207e-05,
+      "loss": 0.5708,
       "step": 440
     },
     {
-      "epoch": 16.0,
-      "eval_accuracy": 0.5816582914572864,
-      "eval_loss": 1.1010228395462036,
-      "eval_runtime": 6.002,
-      "eval_samples_per_second": 265.244,
-      "eval_steps_per_second": 2.166,
-      "step": 448
-    },
-    {
-      "epoch": 16.07,
-      "grad_norm": 3.6204850673675537,
-      "learning_rate": 3.76984126984127e-05,
-      "loss": 0.6621,
+      "epoch": 5.73,
+      "grad_norm": 4.541626453399658,
+      "learning_rate": 2.8846153846153845e-05,
+      "loss": 0.5835,
       "step": 450
     },
     {
-      "epoch": 16.43,
-      "grad_norm": 5.87465238571167,
-      "learning_rate": 3.730158730158731e-05,
-      "loss": 0.6388,
+      "epoch": 5.86,
+      "grad_norm": 5.133627891540527,
+      "learning_rate": 2.948717948717949e-05,
+      "loss": 0.5705,
       "step": 460
     },
     {
-      "epoch": 16.79,
-      "grad_norm": 3.773968458175659,
-      "learning_rate": 3.690476190476191e-05,
-      "loss": 0.6453,
+      "epoch": 5.99,
+      "grad_norm": 4.043761253356934,
+      "learning_rate": 3.012820512820513e-05,
+      "loss": 0.5773,
       "step": 470
     },
     {
-      "epoch": 17.0,
-      "eval_accuracy": 0.5734924623115578,
-      "eval_loss": 1.1130640506744385,
-      "eval_runtime": 5.9355,
-      "eval_samples_per_second": 268.215,
-      "eval_steps_per_second": 2.19,
-      "step": 476
-    },
-    {
-      "epoch": 17.14,
-      "grad_norm": 3.7962214946746826,
-      "learning_rate": 3.650793650793651e-05,
-      "loss": 0.6504,
+      "epoch": 6.0,
+      "eval_accuracy": 0.805829596412556,
+      "eval_loss": 0.4673232436180115,
+      "eval_runtime": 10.582,
+      "eval_samples_per_second": 421.47,
+      "eval_steps_per_second": 3.307,
+      "step": 471
+    },
+    {
+      "epoch": 6.11,
+      "grad_norm": 4.192721843719482,
+      "learning_rate": 3.0769230769230774e-05,
+      "loss": 0.5543,
       "step": 480
     },
     {
-      "epoch": 17.5,
-      "grad_norm": 4.2159013748168945,
-      "learning_rate": 3.611111111111111e-05,
-      "loss": 0.6058,
+      "epoch": 6.24,
+      "grad_norm": 5.365765571594238,
+      "learning_rate": 3.141025641025641e-05,
+      "loss": 0.5417,
       "step": 490
     },
     {
-      "epoch": 17.86,
-      "grad_norm": 4.258510112762451,
-      "learning_rate": 3.571428571428572e-05,
-      "loss": 0.6175,
+      "epoch": 6.37,
+      "grad_norm": 5.424711227416992,
+      "learning_rate": 3.205128205128206e-05,
+      "loss": 0.5687,
       "step": 500
     },
     {
-      "epoch": 18.0,
-      "eval_accuracy": 0.5891959798994975,
-      "eval_loss": 1.1074261665344238,
-      "eval_runtime": 5.4208,
-      "eval_samples_per_second": 293.686,
-      "eval_steps_per_second": 2.398,
-      "step": 504
-    },
-    {
-      "epoch": 18.21,
-      "grad_norm": 4.8816914558410645,
-      "learning_rate": 3.5317460317460324e-05,
-      "loss": 0.5942,
+      "epoch": 6.5,
+      "grad_norm": 4.212834358215332,
+      "learning_rate": 3.269230769230769e-05,
+      "loss": 0.5573,
       "step": 510
     },
     {
-      "epoch": 18.57,
-      "grad_norm": 4.609433650970459,
-      "learning_rate": 3.492063492063492e-05,
-      "loss": 0.5943,
+      "epoch": 6.62,
+      "grad_norm": 4.826545238494873,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.5387,
       "step": 520
     },
     {
-      "epoch": 18.93,
-      "grad_norm": 7.7561211585998535,
-      "learning_rate": 3.4523809523809526e-05,
-      "loss": 0.5993,
+      "epoch": 6.75,
+      "grad_norm": 4.972387313842773,
+      "learning_rate": 3.397435897435898e-05,
+      "loss": 0.5361,
       "step": 530
     },
     {
-      "epoch": 19.0,
-      "eval_accuracy": 0.5741206030150754,
-      "eval_loss": 1.1327540874481201,
-      "eval_runtime": 5.7963,
-      "eval_samples_per_second": 274.658,
-      "eval_steps_per_second": 2.243,
-      "step": 532
-    },
-    {
-      "epoch": 19.29,
-      "grad_norm": 4.787009239196777,
-      "learning_rate": 3.412698412698413e-05,
-      "loss": 0.5741,
+      "epoch": 6.88,
+      "grad_norm": 4.788640022277832,
+      "learning_rate": 3.461538461538462e-05,
+      "loss": 0.5401,
       "step": 540
     },
     {
-      "epoch": 19.64,
-      "grad_norm": 7.622641086578369,
-      "learning_rate": 3.3730158730158734e-05,
-      "loss": 0.5803,
-      "step": 550
+      "epoch": 6.99,
+      "eval_accuracy": 0.8210762331838565,
+      "eval_loss": 0.4277366101741791,
+      "eval_runtime": 10.3618,
+      "eval_samples_per_second": 430.427,
+      "eval_steps_per_second": 3.378,
+      "step": 549
     },
     {
-      "epoch": 20.0,
-      "grad_norm": 6.530401706695557,
-      "learning_rate": 3.3333333333333335e-05,
-      "loss": 0.5683,
-      "step": 560
+      "epoch": 7.01,
+      "grad_norm": 5.113460540771484,
+      "learning_rate": 3.525641025641026e-05,
+      "loss": 0.5512,
+      "step": 550
     },
     {
-      "epoch": 20.0,
-      "eval_accuracy": 0.5791457286432161,
-      "eval_loss": 1.1423227787017822,
-      "eval_runtime": 5.7458,
-      "eval_samples_per_second": 277.073,
-      "eval_steps_per_second": 2.263,
+      "epoch": 7.13,
+      "grad_norm": 4.552058696746826,
+      "learning_rate": 3.58974358974359e-05,
+      "loss": 0.5286,
       "step": 560
     },
     {
-      "epoch": 20.36,
-      "grad_norm": 4.021312236785889,
-      "learning_rate": 3.2936507936507936e-05,
-      "loss": 0.5457,
+      "epoch": 7.26,
+      "grad_norm": 3.3090872764587402,
+      "learning_rate": 3.653846153846154e-05,
+      "loss": 0.5329,
       "step": 570
     },
     {
-      "epoch": 20.71,
-      "grad_norm": 5.022521018981934,
-      "learning_rate": 3.253968253968254e-05,
-      "loss": 0.5524,
+      "epoch": 7.39,
+      "grad_norm": 4.843341827392578,
+      "learning_rate": 3.717948717948718e-05,
+      "loss": 0.5134,
       "step": 580
     },
     {
-      "epoch": 21.0,
-      "eval_accuracy": 0.5873115577889447,
-      "eval_loss": 1.1516565084457397,
-      "eval_runtime": 6.4553,
-      "eval_samples_per_second": 246.619,
-      "eval_steps_per_second": 2.014,
-      "step": 588
-    },
-    {
-      "epoch": 21.07,
-      "grad_norm": 4.585721969604492,
-      "learning_rate": 3.2142857142857144e-05,
-      "loss": 0.5493,
+      "epoch": 7.52,
+      "grad_norm": 4.348685264587402,
+      "learning_rate": 3.782051282051282e-05,
+      "loss": 0.525,
       "step": 590
     },
     {
-      "epoch": 21.43,
-      "grad_norm": 3.6502797603607178,
-      "learning_rate": 3.1746031746031745e-05,
-      "loss": 0.5393,
+      "epoch": 7.64,
+      "grad_norm": 5.437953472137451,
+      "learning_rate": 3.846153846153846e-05,
+      "loss": 0.5024,
       "step": 600
     },
     {
-      "epoch": 21.79,
-      "grad_norm": 4.494349002838135,
-      "learning_rate": 3.134920634920635e-05,
-      "loss": 0.5151,
+      "epoch": 7.77,
+      "grad_norm": 4.2977986335754395,
+      "learning_rate": 3.9102564102564105e-05,
+      "loss": 0.538,
       "step": 610
     },
     {
-      "epoch": 22.0,
-      "eval_accuracy": 0.5766331658291457,
-      "eval_loss": 1.1673014163970947,
-      "eval_runtime": 6.0498,
-      "eval_samples_per_second": 263.15,
-      "eval_steps_per_second": 2.149,
-      "step": 616
-    },
-    {
-      "epoch": 22.14,
-      "grad_norm": 4.292243957519531,
-      "learning_rate": 3.095238095238095e-05,
-      "loss": 0.5241,
+      "epoch": 7.9,
+      "grad_norm": 3.921816110610962,
+      "learning_rate": 3.974358974358974e-05,
+      "loss": 0.5116,
       "step": 620
     },
     {
-      "epoch": 22.5,
-      "grad_norm": 4.669267177581787,
-      "learning_rate": 3.055555555555556e-05,
-      "loss": 0.5095,
+      "epoch": 8.0,
+      "eval_accuracy": 0.831390134529148,
+      "eval_loss": 0.40886422991752625,
+      "eval_runtime": 10.3582,
+      "eval_samples_per_second": 430.575,
+      "eval_steps_per_second": 3.379,
+      "step": 628
+    },
+    {
+      "epoch": 8.03,
+      "grad_norm": 3.6436734199523926,
+      "learning_rate": 4.038461538461539e-05,
+      "loss": 0.4756,
       "step": 630
     },
     {
-      "epoch": 22.86,
-      "grad_norm": 4.2560882568359375,
-      "learning_rate": 3.0158730158730158e-05,
-      "loss": 0.5096,
+      "epoch": 8.15,
+      "grad_norm": 7.276066303253174,
+      "learning_rate": 4.1025641025641023e-05,
+      "loss": 0.5389,
       "step": 640
     },
     {
-      "epoch": 23.0,
-      "eval_accuracy": 0.5797738693467337,
-      "eval_loss": 1.17599618434906,
-      "eval_runtime": 5.8279,
-      "eval_samples_per_second": 273.169,
-      "eval_steps_per_second": 2.231,
-      "step": 644
-    },
-    {
-      "epoch": 23.21,
-      "grad_norm": 4.331119537353516,
-      "learning_rate": 2.9761904761904762e-05,
-      "loss": 0.5092,
+      "epoch": 8.28,
+      "grad_norm": 5.59036922454834,
+      "learning_rate": 4.166666666666667e-05,
+      "loss": 0.5104,
       "step": 650
     },
     {
-      "epoch": 23.57,
-      "grad_norm": 5.246380805969238,
-      "learning_rate": 2.9365079365079366e-05,
-      "loss": 0.5016,
+      "epoch": 8.41,
+      "grad_norm": 5.391291618347168,
+      "learning_rate": 4.230769230769231e-05,
+      "loss": 0.4793,
       "step": 660
     },
     {
-      "epoch": 23.93,
-      "grad_norm": 4.598259449005127,
-      "learning_rate": 2.8968253968253974e-05,
-      "loss": 0.4937,
+      "epoch": 8.54,
+      "grad_norm": 3.3373210430145264,
+      "learning_rate": 4.294871794871795e-05,
+      "loss": 0.4997,
       "step": 670
     },
     {
-      "epoch": 24.0,
-      "eval_accuracy": 0.5816582914572864,
-      "eval_loss": 1.193081259727478,
-      "eval_runtime": 5.7093,
-      "eval_samples_per_second": 278.844,
-      "eval_steps_per_second": 2.277,
-      "step": 672
-    },
-    {
-      "epoch": 24.29,
-      "grad_norm": 4.1697893142700195,
-      "learning_rate": 2.857142857142857e-05,
-      "loss": 0.4713,
+      "epoch": 8.66,
+      "grad_norm": 3.6154773235321045,
+      "learning_rate": 4.358974358974359e-05,
+      "loss": 0.4856,
       "step": 680
     },
     {
-      "epoch": 24.64,
-      "grad_norm": 3.761561870574951,
-      "learning_rate": 2.8174603174603175e-05,
-      "loss": 0.469,
+      "epoch": 8.79,
+      "grad_norm": 4.880829811096191,
+      "learning_rate": 4.423076923076923e-05,
+      "loss": 0.4944,
       "step": 690
     },
     {
-      "epoch": 25.0,
-      "grad_norm": 3.7730281352996826,
-      "learning_rate": 2.777777777777778e-05,
-      "loss": 0.487,
+      "epoch": 8.92,
+      "grad_norm": 4.5230584144592285,
+      "learning_rate": 4.4871794871794874e-05,
+      "loss": 0.4751,
       "step": 700
     },
     {
-      "epoch": 25.0,
-      "eval_accuracy": 0.5734924623115578,
-      "eval_loss": 1.2083638906478882,
-      "eval_runtime": 6.0189,
-      "eval_samples_per_second": 264.498,
-      "eval_steps_per_second": 2.16,
-      "step": 700
+      "epoch": 8.99,
+      "eval_accuracy": 0.8399103139013453,
+      "eval_loss": 0.3890719711780548,
+      "eval_runtime": 10.3075,
+      "eval_samples_per_second": 432.694,
+      "eval_steps_per_second": 3.396,
+      "step": 706
     },
     {
-      "epoch": 25.36,
-      "grad_norm": 4.0658674240112305,
-      "learning_rate": 2.7380952380952383e-05,
-      "loss": 0.4587,
+      "epoch": 9.04,
+      "grad_norm": 3.3694958686828613,
+      "learning_rate": 4.5512820512820516e-05,
+      "loss": 0.4895,
       "step": 710
     },
     {
-      "epoch": 25.71,
-      "grad_norm": 3.6383256912231445,
-      "learning_rate": 2.6984126984126984e-05,
-      "loss": 0.4597,
+      "epoch": 9.17,
+      "grad_norm": 4.626150608062744,
+      "learning_rate": 4.615384615384616e-05,
+      "loss": 0.4721,
       "step": 720
     },
     {
-      "epoch": 26.0,
-      "eval_accuracy": 0.571608040201005,
-      "eval_loss": 1.2270249128341675,
-      "eval_runtime": 6.2134,
-      "eval_samples_per_second": 256.222,
-      "eval_steps_per_second": 2.092,
-      "step": 728
-    },
-    {
-      "epoch": 26.07,
-      "grad_norm": 4.403080940246582,
-      "learning_rate": 2.6587301587301588e-05,
-      "loss": 0.4461,
+      "epoch": 9.3,
+      "grad_norm": 3.687791109085083,
+      "learning_rate": 4.67948717948718e-05,
+      "loss": 0.4672,
       "step": 730
     },
     {
-      "epoch": 26.43,
-      "grad_norm": 4.59806489944458,
-      "learning_rate": 2.6190476190476192e-05,
-      "loss": 0.4487,
+      "epoch": 9.43,
+      "grad_norm": 3.2490782737731934,
+      "learning_rate": 4.7435897435897435e-05,
+      "loss": 0.4665,
       "step": 740
     },
     {
-      "epoch": 26.79,
-      "grad_norm": 3.9561307430267334,
-      "learning_rate": 2.5793650793650796e-05,
-      "loss": 0.4482,
+      "epoch": 9.55,
+      "grad_norm": 4.15757417678833,
+      "learning_rate": 4.8076923076923084e-05,
+      "loss": 0.4646,
       "step": 750
     },
     {
-      "epoch": 27.0,
-      "eval_accuracy": 0.5829145728643216,
-      "eval_loss": 1.2389401197433472,
-      "eval_runtime": 5.923,
-      "eval_samples_per_second": 268.783,
-      "eval_steps_per_second": 2.195,
-      "step": 756
-    },
-    {
-      "epoch": 27.14,
-      "grad_norm": 4.864688396453857,
-      "learning_rate": 2.5396825396825397e-05,
-      "loss": 0.4678,
+      "epoch": 9.68,
+      "grad_norm": 6.2055864334106445,
+      "learning_rate": 4.871794871794872e-05,
+      "loss": 0.4853,
       "step": 760
     },
     {
-      "epoch": 27.5,
-      "grad_norm": 11.65251636505127,
-      "learning_rate": 2.5e-05,
-      "loss": 0.4347,
+      "epoch": 9.81,
+      "grad_norm": 3.2382359504699707,
+      "learning_rate": 4.935897435897436e-05,
+      "loss": 0.4837,
       "step": 770
     },
     {
-      "epoch": 27.86,
-      "grad_norm": 4.065871715545654,
-      "learning_rate": 2.4603174603174602e-05,
-      "loss": 0.4183,
+      "epoch": 9.94,
+      "grad_norm": 4.219330310821533,
+      "learning_rate": 5e-05,
+      "loss": 0.4659,
       "step": 780
     },
     {
-      "epoch": 28.0,
-      "eval_accuracy": 0.5772613065326633,
-      "eval_loss": 1.2430446147918701,
-      "eval_runtime": 5.8698,
-      "eval_samples_per_second": 271.218,
-      "eval_steps_per_second": 2.215,
-      "step": 784
-    },
-    {
-      "epoch": 28.21,
-      "grad_norm": 3.8666417598724365,
-      "learning_rate": 2.4206349206349206e-05,
-      "loss": 0.42,
+      "epoch": 10.0,
+      "eval_accuracy": 0.8573991031390135,
+      "eval_loss": 0.3598979413509369,
+      "eval_runtime": 10.2377,
+      "eval_samples_per_second": 435.643,
+      "eval_steps_per_second": 3.419,
+      "step": 785
+    },
+    {
+      "epoch": 10.06,
+      "grad_norm": 3.231353759765625,
+      "learning_rate": 4.992877492877493e-05,
+      "loss": 0.4682,
       "step": 790
     },
     {
-      "epoch": 28.57,
-      "grad_norm": 4.105574131011963,
-      "learning_rate": 2.380952380952381e-05,
-      "loss": 0.4359,
+      "epoch": 10.19,
+      "grad_norm": 4.601670742034912,
+      "learning_rate": 4.985754985754986e-05,
+      "loss": 0.4553,
       "step": 800
     },
     {
-      "epoch": 28.93,
-      "grad_norm": 4.677915573120117,
-      "learning_rate": 2.3412698412698414e-05,
-      "loss": 0.4228,
+      "epoch": 10.32,
+      "grad_norm": 4.098852634429932,
+      "learning_rate": 4.978632478632479e-05,
+      "loss": 0.4374,
       "step": 810
     },
     {
-      "epoch": 29.0,
-      "eval_accuracy": 0.5741206030150754,
-      "eval_loss": 1.2637208700180054,
-      "eval_runtime": 5.5801,
-      "eval_samples_per_second": 285.302,
-      "eval_steps_per_second": 2.33,
-      "step": 812
-    },
-    {
-      "epoch": 29.29,
-      "grad_norm": 4.680498123168945,
-      "learning_rate": 2.3015873015873015e-05,
-      "loss": 0.4268,
+      "epoch": 10.45,
+      "grad_norm": 3.5977771282196045,
+      "learning_rate": 4.971509971509972e-05,
+      "loss": 0.4478,
       "step": 820
     },
     {
-      "epoch": 29.64,
-      "grad_norm": 3.9155895709991455,
-      "learning_rate": 2.261904761904762e-05,
-      "loss": 0.3908,
+      "epoch": 10.57,
+      "grad_norm": 4.338653564453125,
+      "learning_rate": 4.9643874643874646e-05,
+      "loss": 0.4441,
       "step": 830
     },
     {
-      "epoch": 30.0,
-      "grad_norm": 4.745033264160156,
-      "learning_rate": 2.2222222222222223e-05,
-      "loss": 0.4116,
-      "step": 840
-    },
-    {
-      "epoch": 30.0,
-      "eval_accuracy": 0.5778894472361809,
-      "eval_loss": 1.2687839269638062,
-      "eval_runtime": 5.7938,
-      "eval_samples_per_second": 274.776,
-      "eval_steps_per_second": 2.244,
+      "epoch": 10.7,
+      "grad_norm": 4.484642505645752,
+      "learning_rate": 4.9572649572649575e-05,
+      "loss": 0.4597,
       "step": 840
     },
     {
-      "epoch": 30.36,
-      "grad_norm": 4.2474236488342285,
-      "learning_rate": 2.1825396825396827e-05,
-      "loss": 0.3889,
+      "epoch": 10.83,
+      "grad_norm": 4.2885966300964355,
+      "learning_rate": 4.95014245014245e-05,
+      "loss": 0.4315,
       "step": 850
     },
     {
-      "epoch": 30.71,
-      "grad_norm": 4.243285655975342,
-      "learning_rate": 2.1428571428571428e-05,
-      "loss": 0.3942,
+      "epoch": 10.96,
+      "grad_norm": 5.804717540740967,
+      "learning_rate": 4.943019943019943e-05,
+      "loss": 0.4457,
       "step": 860
     },
     {
-      "epoch": 31.0,
-      "eval_accuracy": 0.5879396984924623,
-      "eval_loss": 1.2986161708831787,
-      "eval_runtime": 5.7249,
-      "eval_samples_per_second": 278.081,
-      "eval_steps_per_second": 2.271,
-      "step": 868
+      "epoch": 10.99,
+      "eval_accuracy": 0.8591928251121076,
+      "eval_loss": 0.34933042526245117,
+      "eval_runtime": 10.3712,
+      "eval_samples_per_second": 430.038,
+      "eval_steps_per_second": 3.375,
+      "step": 863
     },
     {
-      "epoch": 31.07,
-      "grad_norm": 4.418118000030518,
-      "learning_rate": 2.1031746031746032e-05,
-      "loss": 0.3868,
+      "epoch": 11.08,
+      "grad_norm": 3.4534051418304443,
+      "learning_rate": 4.935897435897436e-05,
+      "loss": 0.431,
       "step": 870
     },
     {
-      "epoch": 31.43,
-      "grad_norm": 4.63828182220459,
-      "learning_rate": 2.0634920634920636e-05,
-      "loss": 0.3935,
+      "epoch": 11.21,
+      "grad_norm": 4.114124774932861,
+      "learning_rate": 4.928774928774929e-05,
+      "loss": 0.4171,
       "step": 880
     },
     {
-      "epoch": 31.79,
-      "grad_norm": 4.932127952575684,
-      "learning_rate": 2.023809523809524e-05,
-      "loss": 0.3815,
+      "epoch": 11.34,
+      "grad_norm": 12.598810195922852,
+      "learning_rate": 4.921652421652422e-05,
+      "loss": 0.445,
       "step": 890
     },
     {
-      "epoch": 32.0,
-      "eval_accuracy": 0.5766331658291457,
-      "eval_loss": 1.2911109924316406,
-      "eval_runtime": 5.5086,
-      "eval_samples_per_second": 289.004,
-      "eval_steps_per_second": 2.36,
-      "step": 896
-    },
-    {
-      "epoch": 32.14,
-      "grad_norm": 5.120588779449463,
-      "learning_rate": 1.984126984126984e-05,
-      "loss": 0.3814,
+      "epoch": 11.46,
+      "grad_norm": 3.874229907989502,
+      "learning_rate": 4.9145299145299147e-05,
+      "loss": 0.397,
       "step": 900
     },
     {
-      "epoch": 32.5,
-      "grad_norm": 4.348486423492432,
-      "learning_rate": 1.9444444444444445e-05,
-      "loss": 0.3891,
+      "epoch": 11.59,
+      "grad_norm": 4.273011684417725,
+      "learning_rate": 4.9074074074074075e-05,
+      "loss": 0.413,
       "step": 910
     },
     {
-      "epoch": 32.86,
-      "grad_norm": 5.042387008666992,
-      "learning_rate": 1.9047619047619046e-05,
-      "loss": 0.3828,
+      "epoch": 11.72,
+      "grad_norm": 3.6375491619110107,
+      "learning_rate": 4.9002849002849004e-05,
+      "loss": 0.4079,
       "step": 920
     },
     {
-      "epoch": 33.0,
-      "eval_accuracy": 0.5772613065326633,
-      "eval_loss": 1.311328411102295,
-      "eval_runtime": 5.8619,
-      "eval_samples_per_second": 271.584,
-      "eval_steps_per_second": 2.218,
-      "step": 924
-    },
-    {
-      "epoch": 33.21,
-      "grad_norm": 4.366072177886963,
-      "learning_rate": 1.8650793650793654e-05,
-      "loss": 0.3692,
+      "epoch": 11.85,
+      "grad_norm": 4.725768089294434,
+      "learning_rate": 4.893162393162393e-05,
+      "loss": 0.437,
       "step": 930
     },
     {
-      "epoch": 33.57,
-      "grad_norm": 3.8073909282684326,
-      "learning_rate": 1.8253968253968254e-05,
-      "loss": 0.3617,
+      "epoch": 11.97,
+      "grad_norm": 5.251430988311768,
+      "learning_rate": 4.886039886039887e-05,
+      "loss": 0.4021,
       "step": 940
     },
     {
-      "epoch": 33.93,
-      "grad_norm": 4.39243221282959,
-      "learning_rate": 1.785714285714286e-05,
+      "epoch": 12.0,
+      "eval_accuracy": 0.8751121076233184,
+      "eval_loss": 0.317982017993927,
+      "eval_runtime": 10.1336,
+      "eval_samples_per_second": 440.118,
+      "eval_steps_per_second": 3.454,
+      "step": 942
+    },
+    {
+      "epoch": 12.1,
+      "grad_norm": 3.173100471496582,
+      "learning_rate": 4.878917378917379e-05,
       "loss": 0.3791,
       "step": 950
     },
     {
-      "epoch": 34.0,
-      "eval_accuracy": 0.5766331658291457,
-      "eval_loss": 1.3316831588745117,
-      "eval_runtime": 5.7518,
-      "eval_samples_per_second": 276.782,
-      "eval_steps_per_second": 2.26,
-      "step": 952
-    },
-    {
-      "epoch": 34.29,
-      "grad_norm": 4.339027404785156,
-      "learning_rate": 1.746031746031746e-05,
-      "loss": 0.3716,
+      "epoch": 12.23,
+      "grad_norm": 4.874154090881348,
+      "learning_rate": 4.871794871794872e-05,
+      "loss": 0.4091,
       "step": 960
     },
     {
-      "epoch": 34.64,
-      "grad_norm": 4.211394786834717,
-      "learning_rate": 1.7063492063492063e-05,
-      "loss": 0.3644,
+      "epoch": 12.36,
+      "grad_norm": 3.875936985015869,
+      "learning_rate": 4.864672364672365e-05,
+      "loss": 0.4073,
       "step": 970
     },
     {
-      "epoch": 35.0,
-      "grad_norm": 4.398725509643555,
-      "learning_rate": 1.6666666666666667e-05,
-      "loss": 0.3701,
-      "step": 980
-    },
-    {
-      "epoch": 35.0,
-      "eval_accuracy": 0.5772613065326633,
-      "eval_loss": 1.3383643627166748,
-      "eval_runtime": 5.7692,
-      "eval_samples_per_second": 275.948,
-      "eval_steps_per_second": 2.253,
+      "epoch": 12.48,
+      "grad_norm": 4.053490161895752,
+      "learning_rate": 4.8575498575498576e-05,
+      "loss": 0.3908,
       "step": 980
     },
     {
-      "epoch": 35.36,
-      "grad_norm": 5.39536714553833,
-      "learning_rate": 1.626984126984127e-05,
-      "loss": 0.3625,
+      "epoch": 12.61,
+      "grad_norm": 4.173293113708496,
+      "learning_rate": 4.8504273504273505e-05,
+      "loss": 0.3962,
       "step": 990
     },
     {
-      "epoch": 35.71,
-      "grad_norm": 4.096094131469727,
-      "learning_rate": 1.5873015873015872e-05,
-      "loss": 0.3566,
+      "epoch": 12.74,
+      "grad_norm": 2.876476764678955,
+      "learning_rate": 4.8433048433048433e-05,
+      "loss": 0.409,
       "step": 1000
     },
     {
-      "epoch": 36.0,
-      "eval_accuracy": 0.5753768844221105,
-      "eval_loss": 1.3406319618225098,
-      "eval_runtime": 5.7402,
-      "eval_samples_per_second": 277.342,
-      "eval_steps_per_second": 2.265,
-      "step": 1008
-    },
-    {
-      "epoch": 36.07,
-      "grad_norm": 4.44684362411499,
-      "learning_rate": 1.5476190476190476e-05,
-      "loss": 0.3621,
+      "epoch": 12.87,
+      "grad_norm": 4.63971471786499,
+      "learning_rate": 4.836182336182337e-05,
+      "loss": 0.3814,
       "step": 1010
     },
     {
-      "epoch": 36.43,
-      "grad_norm": 4.413670063018799,
-      "learning_rate": 1.5079365079365079e-05,
-      "loss": 0.3435,
+      "epoch": 12.99,
+      "grad_norm": 4.614744663238525,
+      "learning_rate": 4.829059829059829e-05,
+      "loss": 0.3925,
       "step": 1020
     },
     {
-      "epoch": 36.79,
-      "grad_norm": 4.409304618835449,
-      "learning_rate": 1.4682539682539683e-05,
-      "loss": 0.3551,
-      "step": 1030
+      "epoch": 12.99,
+      "eval_accuracy": 0.886322869955157,
+      "eval_loss": 0.2823837697505951,
+      "eval_runtime": 10.629,
+      "eval_samples_per_second": 419.606,
+      "eval_steps_per_second": 3.293,
+      "step": 1020
     },
     {
-      "epoch": 37.0,
-      "eval_accuracy": 0.5766331658291457,
-      "eval_loss": 1.34104323387146,
-      "eval_runtime": 5.5746,
-      "eval_samples_per_second": 285.58,
-      "eval_steps_per_second": 2.332,
-      "step": 1036
+      "epoch": 13.12,
+      "grad_norm": 3.8458189964294434,
+      "learning_rate": 4.821937321937322e-05,
+      "loss": 0.3691,
+      "step": 1030
     },
     {
-      "epoch": 37.14,
-      "grad_norm": 4.187081813812256,
-      "learning_rate": 1.4285714285714285e-05,
-      "loss": 0.3558,
+      "epoch": 13.25,
+      "grad_norm": 5.067654609680176,
+      "learning_rate": 4.814814814814815e-05,
+      "loss": 0.3906,
       "step": 1040
     },
     {
-      "epoch": 37.5,
-      "grad_norm": 4.235673427581787,
-      "learning_rate": 1.388888888888889e-05,
-      "loss": 0.3392,
+      "epoch": 13.38,
+      "grad_norm": 3.770333766937256,
+      "learning_rate": 4.8076923076923084e-05,
+      "loss": 0.3871,
       "step": 1050
     },
     {
-      "epoch": 37.86,
-      "grad_norm": 4.1030497550964355,
-      "learning_rate": 1.3492063492063492e-05,
-      "loss": 0.3487,
+      "epoch": 13.5,
+      "grad_norm": 3.551605463027954,
+      "learning_rate": 4.8005698005698006e-05,
+      "loss": 0.3768,
       "step": 1060
     },
     {
-      "epoch": 38.0,
-      "eval_accuracy": 0.5866834170854272,
-      "eval_loss": 1.3364226818084717,
-      "eval_runtime": 5.6497,
-      "eval_samples_per_second": 281.783,
-      "eval_steps_per_second": 2.301,
-      "step": 1064
-    },
-    {
-      "epoch": 38.21,
-      "grad_norm": 5.5322394371032715,
-      "learning_rate": 1.3095238095238096e-05,
-      "loss": 0.3359,
+      "epoch": 13.63,
+      "grad_norm": 4.692648887634277,
+      "learning_rate": 4.7934472934472934e-05,
+      "loss": 0.3758,
       "step": 1070
     },
     {
-      "epoch": 38.57,
-      "grad_norm": 4.478665828704834,
-      "learning_rate": 1.2698412698412699e-05,
-      "loss": 0.3445,
+      "epoch": 13.76,
+      "grad_norm": 4.032926082611084,
+      "learning_rate": 4.786324786324787e-05,
+      "loss": 0.3616,
       "step": 1080
     },
     {
-      "epoch": 38.93,
-      "grad_norm": 4.051063060760498,
-      "learning_rate": 1.2301587301587301e-05,
-      "loss": 0.3463,
+      "epoch": 13.89,
+      "grad_norm": 3.6287736892700195,
+      "learning_rate": 4.779202279202279e-05,
+      "loss": 0.3737,
       "step": 1090
     },
     {
-      "epoch": 39.0,
-      "eval_accuracy": 0.5810301507537688,
-      "eval_loss": 1.3495649099349976,
-      "eval_runtime": 5.6668,
-      "eval_samples_per_second": 280.934,
-      "eval_steps_per_second": 2.294,
-      "step": 1092
-    },
-    {
-      "epoch": 39.29,
-      "grad_norm": 3.2469401359558105,
-      "learning_rate": 1.1904761904761905e-05,
-      "loss": 0.3348,
+      "epoch": 14.0,
+      "eval_accuracy": 0.8959641255605382,
+      "eval_loss": 0.2685752809047699,
+      "eval_runtime": 10.3242,
+      "eval_samples_per_second": 431.993,
+      "eval_steps_per_second": 3.39,
+      "step": 1099
+    },
+    {
+      "epoch": 14.01,
+      "grad_norm": 4.15897274017334,
+      "learning_rate": 4.772079772079772e-05,
+      "loss": 0.3678,
       "step": 1100
     },
     {
-      "epoch": 39.64,
-      "grad_norm": 3.4843826293945312,
-      "learning_rate": 1.1507936507936508e-05,
-      "loss": 0.3162,
+      "epoch": 14.14,
+      "grad_norm": 5.19406795501709,
+      "learning_rate": 4.764957264957265e-05,
+      "loss": 0.3729,
       "step": 1110
     },
     {
-      "epoch": 40.0,
-      "grad_norm": 4.144495010375977,
-      "learning_rate": 1.1111111111111112e-05,
-      "loss": 0.3242,
-      "step": 1120
-    },
-    {
-      "epoch": 40.0,
-      "eval_accuracy": 0.574748743718593,
-      "eval_loss": 1.3639838695526123,
-      "eval_runtime": 5.8819,
-      "eval_samples_per_second": 270.661,
-      "eval_steps_per_second": 2.21,
+      "epoch": 14.27,
+      "grad_norm": 3.7079544067382812,
+      "learning_rate": 4.7578347578347584e-05,
+      "loss": 0.3597,
       "step": 1120
     },
     {
-      "epoch": 40.36,
-      "grad_norm": 4.064784526824951,
-      "learning_rate": 1.0714285714285714e-05,
-      "loss": 0.3376,
+      "epoch": 14.39,
+      "grad_norm": 5.892209053039551,
+      "learning_rate": 4.7507122507122506e-05,
+      "loss": 0.358,
       "step": 1130
     },
     {
-      "epoch": 40.71,
-      "grad_norm": 3.94600248336792,
-      "learning_rate": 1.0317460317460318e-05,
-      "loss": 0.3308,
+      "epoch": 14.52,
+      "grad_norm": 3.5323147773742676,
+      "learning_rate": 4.7435897435897435e-05,
+      "loss": 0.3534,
       "step": 1140
     },
     {
-      "epoch": 41.0,
-      "eval_accuracy": 0.571608040201005,
-      "eval_loss": 1.3626537322998047,
-      "eval_runtime": 5.5788,
-      "eval_samples_per_second": 285.366,
-      "eval_steps_per_second": 2.33,
-      "step": 1148
-    },
-    {
-      "epoch": 41.07,
-      "grad_norm": 3.8364109992980957,
-      "learning_rate": 9.92063492063492e-06,
-      "loss": 0.3196,
+      "epoch": 14.65,
+      "grad_norm": 5.116568565368652,
+      "learning_rate": 4.736467236467237e-05,
+      "loss": 0.3418,
       "step": 1150
     },
     {
-      "epoch": 41.43,
-      "grad_norm": 3.929502010345459,
-      "learning_rate": 9.523809523809523e-06,
-      "loss": 0.3224,
+      "epoch": 14.78,
+      "grad_norm": 2.994664430618286,
+      "learning_rate": 4.72934472934473e-05,
+      "loss": 0.3647,
       "step": 1160
     },
     {
-      "epoch": 41.79,
-      "grad_norm": 4.359261989593506,
-      "learning_rate": 9.126984126984127e-06,
-      "loss": 0.3255,
+      "epoch": 14.9,
+      "grad_norm": 5.119503974914551,
+      "learning_rate": 4.722222222222222e-05,
+      "loss": 0.3573,
       "step": 1170
     },
     {
-      "epoch": 42.0,
-      "eval_accuracy": 0.5804020100502513,
-      "eval_loss": 1.379508376121521,
-      "eval_runtime": 5.7679,
-      "eval_samples_per_second": 276.011,
-      "eval_steps_per_second": 2.254,
-      "step": 1176
-    },
-    {
-      "epoch": 42.14,
-      "grad_norm": 4.006887912750244,
-      "learning_rate": 8.73015873015873e-06,
-      "loss": 0.3268,
+      "epoch": 14.99,
+      "eval_accuracy": 0.8977578475336323,
+      "eval_loss": 0.2508743107318878,
+      "eval_runtime": 10.38,
+      "eval_samples_per_second": 429.674,
+      "eval_steps_per_second": 3.372,
+      "step": 1177
+    },
+    {
+      "epoch": 15.03,
+      "grad_norm": 3.1542904376983643,
+      "learning_rate": 4.7150997150997157e-05,
+      "loss": 0.3428,
       "step": 1180
     },
     {
-      "epoch": 42.5,
-      "grad_norm": 3.5152969360351562,
-      "learning_rate": 8.333333333333334e-06,
-      "loss": 0.3166,
+      "epoch": 15.16,
+      "grad_norm": 3.757366418838501,
+      "learning_rate": 4.7079772079772085e-05,
+      "loss": 0.3365,
       "step": 1190
     },
     {
-      "epoch": 42.86,
-      "grad_norm": 3.868173122406006,
-      "learning_rate": 7.936507936507936e-06,
-      "loss": 0.3295,
+      "epoch": 15.29,
+      "grad_norm": 3.979665517807007,
+      "learning_rate": 4.700854700854701e-05,
+      "loss": 0.3345,
       "step": 1200
     },
     {
-      "epoch": 43.0,
-      "eval_accuracy": 0.5797738693467337,
-      "eval_loss": 1.374683141708374,
-      "eval_runtime": 5.5853,
-      "eval_samples_per_second": 285.032,
-      "eval_steps_per_second": 2.328,
-      "step": 1204
-    },
-    {
-      "epoch": 43.21,
-      "grad_norm": 3.3808319568634033,
-      "learning_rate": 7.5396825396825394e-06,
-      "loss": 0.3201,
+      "epoch": 15.41,
+      "grad_norm": 4.517212867736816,
+      "learning_rate": 4.6937321937321936e-05,
+      "loss": 0.3251,
       "step": 1210
     },
     {
-      "epoch": 43.57,
-      "grad_norm": 4.094415187835693,
-      "learning_rate": 7.142857142857143e-06,
-      "loss": 0.3223,
+      "epoch": 15.54,
+      "grad_norm": 3.9786815643310547,
+      "learning_rate": 4.686609686609687e-05,
+      "loss": 0.338,
       "step": 1220
     },
     {
-      "epoch": 43.93,
-      "grad_norm": 3.389286518096924,
-      "learning_rate": 6.746031746031746e-06,
-      "loss": 0.3147,
+      "epoch": 15.67,
+      "grad_norm": 6.16834831237793,
+      "learning_rate": 4.67948717948718e-05,
+      "loss": 0.3425,
       "step": 1230
     },
     {
-      "epoch": 44.0,
-      "eval_accuracy": 0.5860552763819096,
-      "eval_loss": 1.3746650218963623,
-      "eval_runtime": 6.0707,
-      "eval_samples_per_second": 262.244,
-      "eval_steps_per_second": 2.141,
-      "step": 1232
-    },
-    {
-      "epoch": 44.29,
-      "grad_norm": 3.764704942703247,
-      "learning_rate": 6.349206349206349e-06,
-      "loss": 0.3107,
+      "epoch": 15.8,
+      "grad_norm": 3.120864152908325,
+      "learning_rate": 4.672364672364672e-05,
+      "loss": 0.3374,
       "step": 1240
     },
     {
-      "epoch": 44.64,
-      "grad_norm": 4.65610933303833,
-      "learning_rate": 5.9523809523809525e-06,
-      "loss": 0.3058,
+      "epoch": 15.92,
+      "grad_norm": 5.18550443649292,
+      "learning_rate": 4.665242165242166e-05,
+      "loss": 0.3463,
       "step": 1250
     },
     {
-      "epoch": 45.0,
-      "grad_norm": 4.179033279418945,
-      "learning_rate": 5.555555555555556e-06,
-      "loss": 0.3125,
-      "step": 1260
-    },
-    {
-      "epoch": 45.0,
-      "eval_accuracy": 0.5816582914572864,
-      "eval_loss": 1.3839383125305176,
-      "eval_runtime": 5.8047,
-      "eval_samples_per_second": 274.262,
-      "eval_steps_per_second": 2.24,
+      "epoch": 16.0,
+      "eval_accuracy": 0.9100896860986547,
+      "eval_loss": 0.22138647735118866,
+      "eval_runtime": 10.4458,
+      "eval_samples_per_second": 426.965,
+      "eval_steps_per_second": 3.351,
+      "step": 1256
+    },
+    {
+      "epoch": 16.05,
+      "grad_norm": 3.5026066303253174,
+      "learning_rate": 4.6581196581196586e-05,
+      "loss": 0.3289,
       "step": 1260
     },
     {
-      "epoch": 45.36,
-      "grad_norm": 7.927102088928223,
-      "learning_rate": 5.158730158730159e-06,
-      "loss": 0.3077,
+      "epoch": 16.18,
+      "grad_norm": 3.7427220344543457,
+      "learning_rate": 4.6509971509971515e-05,
+      "loss": 0.3376,
       "step": 1270
     },
     {
-      "epoch": 45.71,
-      "grad_norm": 3.8612356185913086,
-      "learning_rate": 4.7619047619047615e-06,
-      "loss": 0.3276,
+      "epoch": 16.31,
+      "grad_norm": 4.764074325561523,
+      "learning_rate": 4.643874643874644e-05,
+      "loss": 0.3082,
       "step": 1280
     },
     {
-      "epoch": 46.0,
-      "eval_accuracy": 0.5841708542713567,
-      "eval_loss": 1.3805787563323975,
-      "eval_runtime": 5.5264,
-      "eval_samples_per_second": 288.072,
-      "eval_steps_per_second": 2.352,
-      "step": 1288
-    },
-    {
-      "epoch": 46.07,
-      "grad_norm": 4.099473476409912,
-      "learning_rate": 4.365079365079365e-06,
-      "loss": 0.3076,
+      "epoch": 16.43,
+      "grad_norm": 5.684610366821289,
+      "learning_rate": 4.636752136752137e-05,
+      "loss": 0.307,
       "step": 1290
     },
     {
-      "epoch": 46.43,
-      "grad_norm": 3.8270063400268555,
-      "learning_rate": 3.968253968253968e-06,
-      "loss": 0.3097,
+      "epoch": 16.56,
+      "grad_norm": 3.970574140548706,
+      "learning_rate": 4.62962962962963e-05,
+      "loss": 0.3215,
       "step": 1300
     },
     {
-      "epoch": 46.79,
-      "grad_norm": 3.907658576965332,
-      "learning_rate": 3.5714285714285714e-06,
-      "loss": 0.2989,
+      "epoch": 16.69,
+      "grad_norm": 3.9064781665802,
+      "learning_rate": 4.622507122507122e-05,
+      "loss": 0.3219,
       "step": 1310
     },
     {
-      "epoch": 47.0,
-      "eval_accuracy": 0.5885678391959799,
-      "eval_loss": 1.3905527591705322,
-      "eval_runtime": 5.6646,
-      "eval_samples_per_second": 281.046,
-      "eval_steps_per_second": 2.295,
-      "step": 1316
-    },
-    {
-      "epoch": 47.14,
-      "grad_norm": 4.716944694519043,
-      "learning_rate": 3.1746031746031746e-06,
-      "loss": 0.3173,
+      "epoch": 16.82,
+      "grad_norm": 5.803124904632568,
+      "learning_rate": 4.615384615384616e-05,
+      "loss": 0.316,
       "step": 1320
     },
     {
-      "epoch": 47.5,
-      "grad_norm": 3.6047604084014893,
-      "learning_rate": 2.777777777777778e-06,
-      "loss": 0.3172,
+      "epoch": 16.94,
+      "grad_norm": 5.820592880249023,
+      "learning_rate": 4.608262108262109e-05,
+      "loss": 0.3174,
       "step": 1330
     },
     {
-      "epoch": 47.86,
-      "grad_norm": 4.362003803253174,
-      "learning_rate": 2.3809523809523808e-06,
-      "loss": 0.2941,
+      "epoch": 16.99,
+      "eval_accuracy": 0.9143497757847534,
+      "eval_loss": 0.22333355247974396,
+      "eval_runtime": 10.5076,
+      "eval_samples_per_second": 424.456,
+      "eval_steps_per_second": 3.331,
+      "step": 1334
+    },
+    {
+      "epoch": 17.07,
+      "grad_norm": 3.744340419769287,
+      "learning_rate": 4.6011396011396016e-05,
+      "loss": 0.3105,
       "step": 1340
     },
     {
-      "epoch": 48.0,
-      "eval_accuracy": 0.5866834170854272,
-      "eval_loss": 1.3876359462738037,
-      "eval_runtime": 5.5384,
-      "eval_samples_per_second": 287.447,
-      "eval_steps_per_second": 2.347,
-      "step": 1344
-    },
-    {
-      "epoch": 48.21,
-      "grad_norm": 4.251241207122803,
-      "learning_rate": 1.984126984126984e-06,
-      "loss": 0.3138,
+      "epoch": 17.2,
+      "grad_norm": 4.38726806640625,
+      "learning_rate": 4.594017094017094e-05,
+      "loss": 0.3017,
       "step": 1350
     },
     {
-      "epoch": 48.57,
-      "grad_norm": 3.9117441177368164,
-      "learning_rate": 1.5873015873015873e-06,
-      "loss": 0.3042,
+      "epoch": 17.32,
+      "grad_norm": 3.789621353149414,
+      "learning_rate": 4.586894586894587e-05,
+      "loss": 0.3139,
       "step": 1360
     },
     {
-      "epoch": 48.93,
-      "grad_norm": 3.703327178955078,
-      "learning_rate": 1.1904761904761904e-06,
-      "loss": 0.3131,
+      "epoch": 17.45,
+      "grad_norm": 4.819761753082275,
+      "learning_rate": 4.57977207977208e-05,
+      "loss": 0.2945,
       "step": 1370
     },
     {
-      "epoch": 49.0,
-      "eval_accuracy": 0.582286432160804,
-      "eval_loss": 1.3895643949508667,
-      "eval_runtime": 5.5003,
-      "eval_samples_per_second": 289.44,
-      "eval_steps_per_second": 2.364,
-      "step": 1372
-    },
-    {
-      "epoch": 49.29,
-      "grad_norm": 6.085123538970947,
-      "learning_rate": 7.936507936507937e-07,
-      "loss": 0.2984,
+      "epoch": 17.58,
+      "grad_norm": 4.197057723999023,
+      "learning_rate": 4.572649572649573e-05,
+      "loss": 0.3118,
       "step": 1380
     },
     {
-      "epoch": 49.64,
-      "grad_norm": 3.707207679748535,
-      "learning_rate": 3.9682539682539683e-07,
-      "loss": 0.3092,
+      "epoch": 17.71,
+      "grad_norm": 4.414229393005371,
+      "learning_rate": 4.565527065527066e-05,
+      "loss": 0.3006,
       "step": 1390
     },
     {
-      "epoch": 50.0,
-      "grad_norm": 4.588694095611572,
-      "learning_rate": 0.0,
-      "loss": 0.2975,
+      "epoch": 17.83,
+      "grad_norm": 6.089674472808838,
+      "learning_rate": 4.558404558404559e-05,
+      "loss": 0.3064,
       "step": 1400
     },
     {
-      "epoch": 50.0,
-      "eval_accuracy": 0.5835427135678392,
-      "eval_loss": 1.3905625343322754,
-      "eval_runtime": 5.7976,
-      "eval_samples_per_second": 274.595,
-      "eval_steps_per_second": 2.242,
-      "step": 1400
+      "epoch": 17.96,
+      "grad_norm": 3.0857927799224854,
+      "learning_rate": 4.5512820512820516e-05,
+      "loss": 0.3029,
+      "step": 1410
     },
     {
-      "epoch": 50.0,
-      "step": 1400,
-      "total_flos": 1.7807825640923136e+19,
-      "train_loss": 0.5565686808313642,
-      "train_runtime": 3248.0342,
-      "train_samples_per_second": 220.564,
-      "train_steps_per_second": 0.431
+      "epoch": 18.0,
+      "eval_accuracy": 0.9221973094170404,
+      "eval_loss": 0.20649677515029907,
+      "eval_runtime": 10.5315,
+      "eval_samples_per_second": 423.491,
+      "eval_steps_per_second": 3.323,
+      "step": 1413
+    },
+    {
+      "epoch": 18.09,
+      "grad_norm": 3.7757556438446045,
+      "learning_rate": 4.544159544159544e-05,
+      "loss": 0.3047,
+      "step": 1420
+    },
+    {
+      "epoch": 18.22,
+      "grad_norm": 3.210659980773926,
+      "learning_rate": 4.5370370370370374e-05,
+      "loss": 0.2969,
+      "step": 1430
+    },
+    {
+      "epoch": 18.34,
+      "grad_norm": 3.7553064823150635,
+      "learning_rate": 4.52991452991453e-05,
+      "loss": 0.2778,
+      "step": 1440
+    },
+    {
+      "epoch": 18.47,
+      "grad_norm": 4.472036361694336,
+      "learning_rate": 4.522792022792023e-05,
+      "loss": 0.2919,
+      "step": 1450
+    },
+    {
+      "epoch": 18.6,
+      "grad_norm": 3.068850517272949,
+      "learning_rate": 4.515669515669516e-05,
+      "loss": 0.3074,
+      "step": 1460
+    },
+    {
+      "epoch": 18.73,
+      "grad_norm": 4.442654609680176,
+      "learning_rate": 4.508547008547009e-05,
+      "loss": 0.3014,
+      "step": 1470
+    },
+    {
+      "epoch": 18.85,
+      "grad_norm": 3.905217409133911,
+      "learning_rate": 4.501424501424502e-05,
+      "loss": 0.2956,
+      "step": 1480
+    },
+    {
+      "epoch": 18.98,
+      "grad_norm": 4.504239559173584,
+      "learning_rate": 4.4943019943019946e-05,
+      "loss": 0.27,
+      "step": 1490
+    },
+    {
+      "epoch": 18.99,
+      "eval_accuracy": 0.9286995515695067,
+      "eval_loss": 0.18995393812656403,
+      "eval_runtime": 10.258,
+      "eval_samples_per_second": 434.783,
+      "eval_steps_per_second": 3.412,
+      "step": 1491
+    },
+    {
+      "epoch": 19.11,
+      "grad_norm": 4.248488426208496,
+      "learning_rate": 4.4871794871794874e-05,
+      "loss": 0.2797,
+      "step": 1500
+    },
+    {
+      "epoch": 19.24,
+      "grad_norm": 3.5203890800476074,
+      "learning_rate": 4.48005698005698e-05,
+      "loss": 0.2737,
+      "step": 1510
+    },
+    {
+      "epoch": 19.36,
+      "grad_norm": 3.1374619007110596,
+      "learning_rate": 4.472934472934473e-05,
+      "loss": 0.2821,
+      "step": 1520
+    },
+    {
+      "epoch": 19.49,
+      "grad_norm": 3.23164439201355,
+      "learning_rate": 4.465811965811966e-05,
+      "loss": 0.2728,
+      "step": 1530
+    },
+    {
+      "epoch": 19.62,
+      "grad_norm": 3.956149101257324,
+      "learning_rate": 4.458689458689459e-05,
+      "loss": 0.2831,
+      "step": 1540
+    },
+    {
+      "epoch": 19.75,
+      "grad_norm": 2.9227089881896973,
+      "learning_rate": 4.451566951566952e-05,
+      "loss": 0.2742,
+      "step": 1550
+    },
+    {
+      "epoch": 19.87,
+      "grad_norm": 4.3883490562438965,
+      "learning_rate": 4.4444444444444447e-05,
+      "loss": 0.2637,
+      "step": 1560
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 3.7743494510650635,
+      "learning_rate": 4.4373219373219375e-05,
+      "loss": 0.2796,
+      "step": 1570
+    },
+    {
+      "epoch": 20.0,
+      "eval_accuracy": 0.929372197309417,
+      "eval_loss": 0.18432486057281494,
+      "eval_runtime": 10.4019,
+      "eval_samples_per_second": 428.767,
+      "eval_steps_per_second": 3.365,
+      "step": 1570
+    },
+    {
+      "epoch": 20.13,
+      "grad_norm": 3.4726107120513916,
+      "learning_rate": 4.4301994301994304e-05,
+      "loss": 0.2705,
+      "step": 1580
+    },
+    {
+      "epoch": 20.25,
+      "grad_norm": 3.691387891769409,
+      "learning_rate": 4.423076923076923e-05,
+      "loss": 0.2622,
+      "step": 1590
+    },
+    {
+      "epoch": 20.38,
+      "grad_norm": 3.4888954162597656,
+      "learning_rate": 4.415954415954416e-05,
+      "loss": 0.2666,
+      "step": 1600
+    },
+    {
+      "epoch": 20.51,
+      "grad_norm": 3.5790562629699707,
+      "learning_rate": 4.408831908831909e-05,
+      "loss": 0.26,
+      "step": 1610
+    },
+    {
+      "epoch": 20.64,
+      "grad_norm": 3.888726234436035,
+      "learning_rate": 4.401709401709402e-05,
+      "loss": 0.2599,
+      "step": 1620
+    },
+    {
+      "epoch": 20.76,
+      "grad_norm": 2.829801082611084,
+      "learning_rate": 4.394586894586895e-05,
+      "loss": 0.2547,
+      "step": 1630
+    },
+    {
+      "epoch": 20.89,
+      "grad_norm": 4.912312030792236,
+      "learning_rate": 4.3874643874643876e-05,
+      "loss": 0.2649,
+      "step": 1640
+    },
+    {
+      "epoch": 20.99,
+      "eval_accuracy": 0.9291479820627803,
+      "eval_loss": 0.1811482161283493,
+      "eval_runtime": 10.3055,
+      "eval_samples_per_second": 432.78,
+      "eval_steps_per_second": 3.396,
+      "step": 1648
+    },
+    {
+      "epoch": 21.02,
+      "grad_norm": 2.7831737995147705,
+      "learning_rate": 4.3803418803418805e-05,
+      "loss": 0.2606,
+      "step": 1650
+    },
+    {
+      "epoch": 21.15,
+      "grad_norm": 4.410768508911133,
+      "learning_rate": 4.3732193732193733e-05,
+      "loss": 0.2461,
+      "step": 1660
+    },
+    {
+      "epoch": 21.27,
+      "grad_norm": 4.2035675048828125,
+      "learning_rate": 4.366096866096866e-05,
+      "loss": 0.2497,
+      "step": 1670
+    },
+    {
+      "epoch": 21.4,
+      "grad_norm": 4.2231011390686035,
+      "learning_rate": 4.358974358974359e-05,
+      "loss": 0.2647,
+      "step": 1680
+    },
+    {
+      "epoch": 21.53,
+      "grad_norm": 3.350215435028076,
+      "learning_rate": 4.351851851851852e-05,
+      "loss": 0.2486,
+      "step": 1690
+    },
+    {
+      "epoch": 21.66,
+      "grad_norm": 4.015892505645752,
+      "learning_rate": 4.344729344729345e-05,
+      "loss": 0.2756,
+      "step": 1700
+    },
+    {
+      "epoch": 21.78,
+      "grad_norm": 2.7854413986206055,
+      "learning_rate": 4.337606837606838e-05,
+      "loss": 0.2442,
+      "step": 1710
+    },
+    {
+      "epoch": 21.91,
+      "grad_norm": 4.043717861175537,
+      "learning_rate": 4.3304843304843306e-05,
+      "loss": 0.2561,
+      "step": 1720
+    },
+    {
+      "epoch": 22.0,
+      "eval_accuracy": 0.9345291479820628,
+      "eval_loss": 0.16786783933639526,
+      "eval_runtime": 10.4139,
+      "eval_samples_per_second": 428.274,
+      "eval_steps_per_second": 3.361,
+      "step": 1727
+    },
+    {
+      "epoch": 22.04,
+      "grad_norm": 2.745814085006714,
+      "learning_rate": 4.3233618233618234e-05,
+      "loss": 0.2697,
+      "step": 1730
+    },
+    {
+      "epoch": 22.17,
+      "grad_norm": 3.69480562210083,
+      "learning_rate": 4.316239316239317e-05,
+      "loss": 0.2559,
+      "step": 1740
+    },
+    {
+      "epoch": 22.29,
+      "grad_norm": 3.843909502029419,
+      "learning_rate": 4.309116809116809e-05,
+      "loss": 0.2476,
+      "step": 1750
+    },
+    {
+      "epoch": 22.42,
+      "grad_norm": 3.822366237640381,
+      "learning_rate": 4.301994301994302e-05,
+      "loss": 0.2268,
+      "step": 1760
+    },
+    {
+      "epoch": 22.55,
+      "grad_norm": 3.71618390083313,
+      "learning_rate": 4.294871794871795e-05,
+      "loss": 0.2584,
+      "step": 1770
+    },
+    {
+      "epoch": 22.68,
+      "grad_norm": 4.780421257019043,
+      "learning_rate": 4.287749287749288e-05,
+      "loss": 0.2486,
+      "step": 1780
+    },
+    {
+      "epoch": 22.8,
+      "grad_norm": 4.906135082244873,
+      "learning_rate": 4.2806267806267806e-05,
+      "loss": 0.2454,
+      "step": 1790
+    },
+    {
+      "epoch": 22.93,
+      "grad_norm": 3.733837366104126,
+      "learning_rate": 4.2735042735042735e-05,
+      "loss": 0.2384,
+      "step": 1800
+    },
+    {
+      "epoch": 22.99,
+      "eval_accuracy": 0.940134529147982,
+      "eval_loss": 0.15794554352760315,
+      "eval_runtime": 10.3863,
+      "eval_samples_per_second": 429.413,
+      "eval_steps_per_second": 3.37,
+      "step": 1805
+    },
+    {
+      "epoch": 23.06,
+      "grad_norm": 3.1447577476501465,
+      "learning_rate": 4.266381766381767e-05,
+      "loss": 0.24,
+      "step": 1810
+    },
+    {
+      "epoch": 23.18,
+      "grad_norm": 4.22783899307251,
+      "learning_rate": 4.259259259259259e-05,
+      "loss": 0.2304,
+      "step": 1820
+    },
+    {
+      "epoch": 23.31,
+      "grad_norm": 3.896044969558716,
+      "learning_rate": 4.252136752136752e-05,
+      "loss": 0.2209,
+      "step": 1830
+    },
+    {
+      "epoch": 23.44,
+      "grad_norm": 3.943885326385498,
+      "learning_rate": 4.2450142450142457e-05,
+      "loss": 0.2286,
+      "step": 1840
+    },
+    {
+      "epoch": 23.57,
+      "grad_norm": 3.7429778575897217,
+      "learning_rate": 4.2378917378917385e-05,
+      "loss": 0.2209,
+      "step": 1850
+    },
+    {
+      "epoch": 23.69,
+      "grad_norm": 3.867816209793091,
+      "learning_rate": 4.230769230769231e-05,
+      "loss": 0.25,
+      "step": 1860
+    },
+    {
+      "epoch": 23.82,
+      "grad_norm": 3.7919929027557373,
+      "learning_rate": 4.2236467236467236e-05,
+      "loss": 0.255,
+      "step": 1870
+    },
+    {
+      "epoch": 23.95,
+      "grad_norm": 3.726590156555176,
+      "learning_rate": 4.216524216524217e-05,
+      "loss": 0.2415,
+      "step": 1880
+    },
+    {
+      "epoch": 24.0,
+      "eval_accuracy": 0.9461883408071748,
+      "eval_loss": 0.1414378136396408,
+      "eval_runtime": 10.1308,
+      "eval_samples_per_second": 440.241,
+      "eval_steps_per_second": 3.455,
+      "step": 1884
+    },
+    {
+      "epoch": 24.08,
+      "grad_norm": 4.867549419403076,
+      "learning_rate": 4.209401709401709e-05,
+      "loss": 0.2319,
+      "step": 1890
+    },
+    {
+      "epoch": 24.2,
+      "grad_norm": 3.615717649459839,
+      "learning_rate": 4.202279202279202e-05,
+      "loss": 0.2315,
+      "step": 1900
+    },
+    {
+      "epoch": 24.33,
+      "grad_norm": 3.8111448287963867,
+      "learning_rate": 4.195156695156696e-05,
+      "loss": 0.2395,
+      "step": 1910
+    },
+    {
+      "epoch": 24.46,
+      "grad_norm": 3.761992931365967,
+      "learning_rate": 4.1880341880341886e-05,
+      "loss": 0.2211,
+      "step": 1920
+    },
+    {
+      "epoch": 24.59,
+      "grad_norm": 4.226995468139648,
+      "learning_rate": 4.180911680911681e-05,
+      "loss": 0.2236,
+      "step": 1930
+    },
+    {
+      "epoch": 24.71,
+      "grad_norm": 3.9813332557678223,
+      "learning_rate": 4.1737891737891737e-05,
+      "loss": 0.2384,
+      "step": 1940
+    },
+    {
+      "epoch": 24.84,
+      "grad_norm": 3.4749677181243896,
+      "learning_rate": 4.166666666666667e-05,
+      "loss": 0.2181,
+      "step": 1950
+    },
+    {
+      "epoch": 24.97,
+      "grad_norm": 3.7109665870666504,
+      "learning_rate": 4.15954415954416e-05,
+      "loss": 0.2282,
+      "step": 1960
+    },
+    {
+      "epoch": 24.99,
+      "eval_accuracy": 0.9439461883408071,
+      "eval_loss": 0.147759348154068,
+      "eval_runtime": 10.416,
+      "eval_samples_per_second": 428.189,
+      "eval_steps_per_second": 3.36,
+      "step": 1962
+    },
+    {
+      "epoch": 25.1,
+      "grad_norm": 5.970840930938721,
+      "learning_rate": 4.152421652421652e-05,
+      "loss": 0.2231,
+      "step": 1970
+    },
+    {
+      "epoch": 25.22,
+      "grad_norm": 4.683918476104736,
+      "learning_rate": 4.145299145299146e-05,
+      "loss": 0.2314,
+      "step": 1980
+    },
+    {
+      "epoch": 25.35,
+      "grad_norm": 4.046987533569336,
+      "learning_rate": 4.138176638176639e-05,
+      "loss": 0.2415,
+      "step": 1990
+    },
+    {
+      "epoch": 25.48,
+      "grad_norm": 4.222659111022949,
+      "learning_rate": 4.131054131054131e-05,
+      "loss": 0.2305,
+      "step": 2000
+    },
+    {
+      "epoch": 25.61,
+      "grad_norm": 3.906933069229126,
+      "learning_rate": 4.123931623931624e-05,
+      "loss": 0.2053,
+      "step": 2010
+    },
+    {
+      "epoch": 25.73,
+      "grad_norm": 3.3867435455322266,
+      "learning_rate": 4.116809116809117e-05,
+      "loss": 0.2116,
+      "step": 2020
+    },
+    {
+      "epoch": 25.86,
+      "grad_norm": 3.771847724914551,
+      "learning_rate": 4.10968660968661e-05,
+      "loss": 0.2448,
+      "step": 2030
+    },
+    {
+      "epoch": 25.99,
+      "grad_norm": 2.8068020343780518,
+      "learning_rate": 4.1025641025641023e-05,
+      "loss": 0.2345,
+      "step": 2040
+    },
+    {
+      "epoch": 26.0,
+      "eval_accuracy": 0.9419282511210763,
+      "eval_loss": 0.15225176513195038,
+      "eval_runtime": 11.029,
+      "eval_samples_per_second": 404.39,
+      "eval_steps_per_second": 3.173,
+      "step": 2041
+    },
+    {
+      "epoch": 26.11,
+      "grad_norm": 4.059516906738281,
+      "learning_rate": 4.095441595441596e-05,
+      "loss": 0.2241,
+      "step": 2050
+    },
+    {
+      "epoch": 26.24,
+      "grad_norm": 4.394323825836182,
+      "learning_rate": 4.088319088319089e-05,
+      "loss": 0.2269,
+      "step": 2060
+    },
+    {
+      "epoch": 26.37,
+      "grad_norm": 4.457634925842285,
+      "learning_rate": 4.0811965811965816e-05,
+      "loss": 0.2171,
+      "step": 2070
+    },
+    {
+      "epoch": 26.5,
+      "grad_norm": 4.27789306640625,
+      "learning_rate": 4.074074074074074e-05,
+      "loss": 0.2229,
+      "step": 2080
+    },
+    {
+      "epoch": 26.62,
+      "grad_norm": 4.969176292419434,
+      "learning_rate": 4.0669515669515674e-05,
+      "loss": 0.2116,
+      "step": 2090
+    },
+    {
+      "epoch": 26.75,
+      "grad_norm": 3.5934197902679443,
+      "learning_rate": 4.05982905982906e-05,
+      "loss": 0.2135,
+      "step": 2100
+    },
+    {
+      "epoch": 26.88,
+      "grad_norm": 6.578746795654297,
+      "learning_rate": 4.0527065527065524e-05,
+      "loss": 0.2196,
+      "step": 2110
+    },
+    {
+      "epoch": 26.99,
+      "eval_accuracy": 0.952914798206278,
+      "eval_loss": 0.12402135878801346,
+      "eval_runtime": 10.1196,
+      "eval_samples_per_second": 440.729,
+      "eval_steps_per_second": 3.459,
+      "step": 2119
+    },
+    {
+      "epoch": 27.01,
+      "grad_norm": 4.893234729766846,
+      "learning_rate": 4.045584045584046e-05,
+      "loss": 0.2126,
+      "step": 2120
+    },
+    {
+      "epoch": 27.13,
+      "grad_norm": 4.168965816497803,
+      "learning_rate": 4.038461538461539e-05,
+      "loss": 0.2142,
+      "step": 2130
+    },
+    {
+      "epoch": 27.26,
+      "grad_norm": 3.5886390209198,
+      "learning_rate": 4.031339031339032e-05,
+      "loss": 0.2014,
+      "step": 2140
+    },
+    {
+      "epoch": 27.39,
+      "grad_norm": 3.832911252975464,
+      "learning_rate": 4.024216524216524e-05,
+      "loss": 0.207,
+      "step": 2150
+    },
+    {
+      "epoch": 27.52,
+      "grad_norm": 5.326879978179932,
+      "learning_rate": 4.0170940170940174e-05,
+      "loss": 0.2033,
+      "step": 2160
+    },
+    {
+      "epoch": 27.64,
+      "grad_norm": 3.8248493671417236,
+      "learning_rate": 4.00997150997151e-05,
+      "loss": 0.2112,
+      "step": 2170
+    },
+    {
+      "epoch": 27.77,
+      "grad_norm": 3.867006540298462,
+      "learning_rate": 4.002849002849003e-05,
+      "loss": 0.2135,
+      "step": 2180
+    },
+    {
+      "epoch": 27.9,
+      "grad_norm": 3.587290048599243,
+      "learning_rate": 3.995726495726496e-05,
+      "loss": 0.2118,
+      "step": 2190
+    },
+    {
+      "epoch": 28.0,
+      "eval_accuracy": 0.9502242152466368,
+      "eval_loss": 0.12278711795806885,
+      "eval_runtime": 10.6512,
+      "eval_samples_per_second": 418.732,
+      "eval_steps_per_second": 3.286,
+      "step": 2198
+    },
+    {
+      "epoch": 28.03,
+      "grad_norm": 5.586561679840088,
+      "learning_rate": 3.988603988603989e-05,
+      "loss": 0.213,
+      "step": 2200
+    },
+    {
+      "epoch": 28.15,
+      "grad_norm": 3.0234127044677734,
+      "learning_rate": 3.981481481481482e-05,
+      "loss": 0.206,
+      "step": 2210
+    },
+    {
+      "epoch": 28.28,
+      "grad_norm": 4.138052463531494,
+      "learning_rate": 3.974358974358974e-05,
+      "loss": 0.2212,
+      "step": 2220
+    },
+    {
+      "epoch": 28.41,
+      "grad_norm": 3.2922720909118652,
+      "learning_rate": 3.9672364672364675e-05,
+      "loss": 0.1937,
+      "step": 2230
+    },
+    {
+      "epoch": 28.54,
+      "grad_norm": 3.6713385581970215,
+      "learning_rate": 3.9601139601139604e-05,
+      "loss": 0.2051,
+      "step": 2240
+    },
+    {
+      "epoch": 28.66,
+      "grad_norm": 3.211871385574341,
+      "learning_rate": 3.952991452991453e-05,
+      "loss": 0.2117,
+      "step": 2250
+    },
+    {
+      "epoch": 28.79,
+      "grad_norm": 3.307561159133911,
+      "learning_rate": 3.945868945868946e-05,
+      "loss": 0.2161,
+      "step": 2260
+    },
+    {
+      "epoch": 28.92,
+      "grad_norm": 3.385369300842285,
+      "learning_rate": 3.938746438746439e-05,
+      "loss": 0.1977,
+      "step": 2270
+    },
+    {
+      "epoch": 28.99,
+      "eval_accuracy": 0.9565022421524664,
+      "eval_loss": 0.11919865012168884,
+      "eval_runtime": 10.3622,
+      "eval_samples_per_second": 430.411,
+      "eval_steps_per_second": 3.378,
+      "step": 2276
+    },
+    {
+      "epoch": 29.04,
+      "grad_norm": 5.586241245269775,
+      "learning_rate": 3.931623931623932e-05,
+      "loss": 0.21,
+      "step": 2280
+    },
+    {
+      "epoch": 29.17,
+      "grad_norm": 3.34932017326355,
+      "learning_rate": 3.924501424501425e-05,
+      "loss": 0.2062,
+      "step": 2290
+    },
+    {
+      "epoch": 29.3,
+      "grad_norm": 3.922372579574585,
+      "learning_rate": 3.9173789173789176e-05,
+      "loss": 0.2166,
+      "step": 2300
+    },
+    {
+      "epoch": 29.43,
+      "grad_norm": 3.6068267822265625,
+      "learning_rate": 3.9102564102564105e-05,
+      "loss": 0.2046,
+      "step": 2310
+    },
+    {
+      "epoch": 29.55,
+      "grad_norm": 4.987060070037842,
+      "learning_rate": 3.903133903133903e-05,
+      "loss": 0.1985,
+      "step": 2320
+    },
+    {
+      "epoch": 29.68,
+      "grad_norm": 3.772573232650757,
+      "learning_rate": 3.896011396011396e-05,
+      "loss": 0.2037,
+      "step": 2330
+    },
+    {
+      "epoch": 29.81,
+      "grad_norm": 4.787131309509277,
+      "learning_rate": 3.888888888888889e-05,
+      "loss": 0.2023,
+      "step": 2340
+    },
+    {
+      "epoch": 29.94,
+      "grad_norm": 3.7376861572265625,
+      "learning_rate": 3.881766381766382e-05,
+      "loss": 0.1994,
+      "step": 2350
+    },
+    {
+      "epoch": 30.0,
+      "eval_accuracy": 0.952914798206278,
+      "eval_loss": 0.12083474546670914,
+      "eval_runtime": 10.1108,
+      "eval_samples_per_second": 441.111,
+      "eval_steps_per_second": 3.462,
+      "step": 2355
+    },
+    {
+      "epoch": 30.06,
+      "grad_norm": 2.6949622631073,
+      "learning_rate": 3.874643874643875e-05,
+      "loss": 0.199,
+      "step": 2360
+    },
+    {
+      "epoch": 30.19,
+      "grad_norm": 4.504613876342773,
+      "learning_rate": 3.867521367521368e-05,
+      "loss": 0.1856,
+      "step": 2370
+    },
+    {
+      "epoch": 30.32,
+      "grad_norm": 3.8520376682281494,
+      "learning_rate": 3.8603988603988605e-05,
+      "loss": 0.1982,
+      "step": 2380
+    },
+    {
+      "epoch": 30.45,
+      "grad_norm": 4.474611282348633,
+      "learning_rate": 3.8532763532763534e-05,
+      "loss": 0.199,
+      "step": 2390
+    },
+    {
+      "epoch": 30.57,
+      "grad_norm": 3.9707252979278564,
+      "learning_rate": 3.846153846153846e-05,
+      "loss": 0.1993,
+      "step": 2400
+    },
+    {
+      "epoch": 30.7,
+      "grad_norm": 4.545299530029297,
+      "learning_rate": 3.839031339031339e-05,
+      "loss": 0.2002,
+      "step": 2410
+    },
+    {
+      "epoch": 30.83,
+      "grad_norm": 4.200654029846191,
+      "learning_rate": 3.831908831908832e-05,
+      "loss": 0.1955,
+      "step": 2420
+    },
+    {
+      "epoch": 30.96,
+      "grad_norm": 3.393235206604004,
+      "learning_rate": 3.824786324786325e-05,
+      "loss": 0.1959,
+      "step": 2430
+    },
+    {
+      "epoch": 30.99,
+      "eval_accuracy": 0.9562780269058296,
+      "eval_loss": 0.11892925947904587,
+      "eval_runtime": 10.326,
+      "eval_samples_per_second": 431.92,
+      "eval_steps_per_second": 3.39,
+      "step": 2433
+    },
+    {
+      "epoch": 31.08,
+      "grad_norm": 5.270153045654297,
+      "learning_rate": 3.817663817663818e-05,
+      "loss": 0.2021,
+      "step": 2440
+    },
+    {
+      "epoch": 31.21,
+      "grad_norm": 2.6245462894439697,
+      "learning_rate": 3.8105413105413106e-05,
+      "loss": 0.1928,
+      "step": 2450
+    },
+    {
+      "epoch": 31.34,
+      "grad_norm": 3.6266441345214844,
+      "learning_rate": 3.8034188034188035e-05,
+      "loss": 0.2017,
+      "step": 2460
+    },
+    {
+      "epoch": 31.46,
+      "grad_norm": 4.613752365112305,
+      "learning_rate": 3.7962962962962964e-05,
+      "loss": 0.1922,
+      "step": 2470
+    },
+    {
+      "epoch": 31.59,
+      "grad_norm": 3.5917015075683594,
+      "learning_rate": 3.789173789173789e-05,
+      "loss": 0.1888,
+      "step": 2480
+    },
+    {
+      "epoch": 31.72,
+      "grad_norm": 4.044875621795654,
+      "learning_rate": 3.782051282051282e-05,
+      "loss": 0.195,
+      "step": 2490
+    },
+    {
+      "epoch": 31.85,
+      "grad_norm": 3.4889109134674072,
+      "learning_rate": 3.774928774928775e-05,
+      "loss": 0.1872,
+      "step": 2500
+    },
+    {
+      "epoch": 31.97,
+      "grad_norm": 3.380016565322876,
+      "learning_rate": 3.767806267806268e-05,
+      "loss": 0.1946,
+      "step": 2510
+    },
+    {
+      "epoch": 32.0,
+      "eval_accuracy": 0.9616591928251121,
+      "eval_loss": 0.11063708364963531,
+      "eval_runtime": 10.3813,
+      "eval_samples_per_second": 429.619,
+      "eval_steps_per_second": 3.371,
+      "step": 2512
+    },
+    {
+      "epoch": 32.1,
+      "grad_norm": 2.8205275535583496,
+      "learning_rate": 3.760683760683761e-05,
+      "loss": 0.181,
+      "step": 2520
+    },
+    {
+      "epoch": 32.23,
+      "grad_norm": 3.5672075748443604,
+      "learning_rate": 3.7535612535612536e-05,
+      "loss": 0.1799,
+      "step": 2530
+    },
+    {
+      "epoch": 32.36,
+      "grad_norm": 2.782327651977539,
+      "learning_rate": 3.746438746438747e-05,
+      "loss": 0.1867,
+      "step": 2540
+    },
+    {
+      "epoch": 32.48,
+      "grad_norm": 4.034256935119629,
+      "learning_rate": 3.739316239316239e-05,
+      "loss": 0.1787,
+      "step": 2550
+    },
+    {
+      "epoch": 32.61,
+      "grad_norm": 4.922691345214844,
+      "learning_rate": 3.732193732193732e-05,
+      "loss": 0.1917,
+      "step": 2560
+    },
+    {
+      "epoch": 32.74,
+      "grad_norm": 3.8344011306762695,
+      "learning_rate": 3.725071225071226e-05,
+      "loss": 0.1926,
+      "step": 2570
+    },
+    {
+      "epoch": 32.87,
+      "grad_norm": 3.1199018955230713,
+      "learning_rate": 3.717948717948718e-05,
+      "loss": 0.1879,
+      "step": 2580
+    },
+    {
+      "epoch": 32.99,
+      "grad_norm": 4.506677150726318,
+      "learning_rate": 3.710826210826211e-05,
+      "loss": 0.1777,
+      "step": 2590
+    },
+    {
+      "epoch": 32.99,
+      "eval_accuracy": 0.9587443946188341,
+      "eval_loss": 0.10475672781467438,
+      "eval_runtime": 10.5123,
+      "eval_samples_per_second": 424.265,
+      "eval_steps_per_second": 3.329,
+      "step": 2590
+    },
+    {
+      "epoch": 33.12,
+      "grad_norm": 3.60703444480896,
+      "learning_rate": 3.7037037037037037e-05,
+      "loss": 0.1859,
+      "step": 2600
+    },
+    {
+      "epoch": 33.25,
+      "grad_norm": 2.695091485977173,
+      "learning_rate": 3.696581196581197e-05,
+      "loss": 0.1898,
+      "step": 2610
+    },
+    {
+      "epoch": 33.38,
+      "grad_norm": 5.429329872131348,
+      "learning_rate": 3.6894586894586894e-05,
+      "loss": 0.203,
+      "step": 2620
+    },
+    {
+      "epoch": 33.5,
+      "grad_norm": 4.286501407623291,
+      "learning_rate": 3.682336182336182e-05,
+      "loss": 0.189,
+      "step": 2630
+    },
+    {
+      "epoch": 33.63,
+      "grad_norm": 3.461913585662842,
+      "learning_rate": 3.675213675213676e-05,
+      "loss": 0.1844,
+      "step": 2640
+    },
+    {
+      "epoch": 33.76,
+      "grad_norm": 3.4661142826080322,
+      "learning_rate": 3.668091168091169e-05,
+      "loss": 0.1765,
+      "step": 2650
+    },
+    {
+      "epoch": 33.89,
+      "grad_norm": 3.5034396648406982,
+      "learning_rate": 3.660968660968661e-05,
+      "loss": 0.1895,
+      "step": 2660
+    },
+    {
+      "epoch": 34.0,
+      "eval_accuracy": 0.9643497757847533,
+      "eval_loss": 0.09648270159959793,
+      "eval_runtime": 10.4535,
+      "eval_samples_per_second": 426.651,
+      "eval_steps_per_second": 3.348,
+      "step": 2669
+    },
+    {
+      "epoch": 34.01,
+      "grad_norm": 3.3810720443725586,
+      "learning_rate": 3.653846153846154e-05,
+      "loss": 0.1745,
+      "step": 2670
+    },
+    {
+      "epoch": 34.14,
+      "grad_norm": 3.2430992126464844,
+      "learning_rate": 3.646723646723647e-05,
+      "loss": 0.1728,
+      "step": 2680
+    },
+    {
+      "epoch": 34.27,
+      "grad_norm": 3.9114370346069336,
+      "learning_rate": 3.6396011396011395e-05,
+      "loss": 0.1795,
+      "step": 2690
+    },
+    {
+      "epoch": 34.39,
+      "grad_norm": 5.329428195953369,
+      "learning_rate": 3.6324786324786323e-05,
+      "loss": 0.1911,
+      "step": 2700
+    },
+    {
+      "epoch": 34.52,
+      "grad_norm": 3.0880038738250732,
+      "learning_rate": 3.625356125356126e-05,
+      "loss": 0.1864,
+      "step": 2710
+    },
+    {
+      "epoch": 34.65,
+      "grad_norm": 3.305945634841919,
+      "learning_rate": 3.618233618233619e-05,
+      "loss": 0.1756,
+      "step": 2720
+    },
+    {
+      "epoch": 34.78,
+      "grad_norm": 3.8540592193603516,
+      "learning_rate": 3.611111111111111e-05,
+      "loss": 0.1808,
+      "step": 2730
+    },
+    {
+      "epoch": 34.9,
+      "grad_norm": 3.3947036266326904,
+      "learning_rate": 3.603988603988604e-05,
+      "loss": 0.1779,
+      "step": 2740
+    },
+    {
+      "epoch": 34.99,
+      "eval_accuracy": 0.9623318385650225,
+      "eval_loss": 0.10805962979793549,
+      "eval_runtime": 10.2969,
+      "eval_samples_per_second": 433.139,
+      "eval_steps_per_second": 3.399,
+      "step": 2747
+    },
+    {
+      "epoch": 35.03,
+      "grad_norm": 3.559788942337036,
+      "learning_rate": 3.5968660968660974e-05,
+      "loss": 0.1743,
+      "step": 2750
+    },
+    {
+      "epoch": 35.16,
+      "grad_norm": 4.594892978668213,
+      "learning_rate": 3.58974358974359e-05,
+      "loss": 0.175,
+      "step": 2760
+    },
+    {
+      "epoch": 35.29,
+      "grad_norm": 5.519185543060303,
+      "learning_rate": 3.5826210826210824e-05,
+      "loss": 0.1728,
+      "step": 2770
+    },
+    {
+      "epoch": 35.41,
+      "grad_norm": 3.953115463256836,
+      "learning_rate": 3.575498575498576e-05,
+      "loss": 0.1758,
+      "step": 2780
+    },
+    {
+      "epoch": 35.54,
+      "grad_norm": 4.416403770446777,
+      "learning_rate": 3.568376068376069e-05,
+      "loss": 0.175,
+      "step": 2790
+    },
+    {
+      "epoch": 35.67,
+      "grad_norm": 3.546534776687622,
+      "learning_rate": 3.561253561253561e-05,
+      "loss": 0.186,
+      "step": 2800
+    },
+    {
+      "epoch": 35.8,
+      "grad_norm": 2.469801664352417,
+      "learning_rate": 3.554131054131054e-05,
+      "loss": 0.1721,
+      "step": 2810
+    },
+    {
+      "epoch": 35.92,
+      "grad_norm": 2.8391408920288086,
+      "learning_rate": 3.5470085470085474e-05,
+      "loss": 0.1717,
+      "step": 2820
+    },
+    {
+      "epoch": 36.0,
+      "eval_accuracy": 0.9589686098654708,
+      "eval_loss": 0.11545290797948837,
+      "eval_runtime": 10.546,
+      "eval_samples_per_second": 422.908,
+      "eval_steps_per_second": 3.319,
+      "step": 2826
+    },
+    {
+      "epoch": 36.05,
+      "grad_norm": 2.674395799636841,
+      "learning_rate": 3.53988603988604e-05,
+      "loss": 0.1903,
+      "step": 2830
+    },
+    {
+      "epoch": 36.18,
+      "grad_norm": 3.417855739593506,
+      "learning_rate": 3.5327635327635325e-05,
+      "loss": 0.179,
+      "step": 2840
+    },
+    {
+      "epoch": 36.31,
+      "grad_norm": 3.361017942428589,
+      "learning_rate": 3.525641025641026e-05,
+      "loss": 0.1597,
+      "step": 2850
+    },
+    {
+      "epoch": 36.43,
+      "grad_norm": 3.462763786315918,
+      "learning_rate": 3.518518518518519e-05,
+      "loss": 0.1754,
+      "step": 2860
+    },
+    {
+      "epoch": 36.56,
+      "grad_norm": 4.020880699157715,
+      "learning_rate": 3.511396011396012e-05,
+      "loss": 0.1784,
+      "step": 2870
+    },
+    {
+      "epoch": 36.69,
+      "grad_norm": 4.944223403930664,
+      "learning_rate": 3.504273504273504e-05,
+      "loss": 0.168,
+      "step": 2880
+    },
+    {
+      "epoch": 36.82,
+      "grad_norm": 3.0020694732666016,
+      "learning_rate": 3.4971509971509975e-05,
+      "loss": 0.1765,
+      "step": 2890
+    },
+    {
+      "epoch": 36.94,
+      "grad_norm": 3.5330681800842285,
+      "learning_rate": 3.4900284900284904e-05,
+      "loss": 0.1715,
+      "step": 2900
+    },
+    {
+      "epoch": 36.99,
+      "eval_accuracy": 0.9630044843049327,
+      "eval_loss": 0.1006338968873024,
+      "eval_runtime": 10.331,
+      "eval_samples_per_second": 431.709,
+      "eval_steps_per_second": 3.388,
+      "step": 2904
+    },
+    {
+      "epoch": 37.07,
+      "grad_norm": 3.5816547870635986,
+      "learning_rate": 3.4829059829059826e-05,
+      "loss": 0.163,
+      "step": 2910
+    },
+    {
+      "epoch": 37.2,
+      "grad_norm": 3.187361240386963,
+      "learning_rate": 3.475783475783476e-05,
+      "loss": 0.1637,
+      "step": 2920
+    },
+    {
+      "epoch": 37.32,
+      "grad_norm": 4.5370612144470215,
+      "learning_rate": 3.468660968660969e-05,
+      "loss": 0.1698,
+      "step": 2930
+    },
+    {
+      "epoch": 37.45,
+      "grad_norm": 4.927834510803223,
+      "learning_rate": 3.461538461538462e-05,
+      "loss": 0.1664,
+      "step": 2940
+    },
+    {
+      "epoch": 37.58,
+      "grad_norm": 4.095933437347412,
+      "learning_rate": 3.454415954415954e-05,
+      "loss": 0.1554,
+      "step": 2950
+    },
+    {
+      "epoch": 37.71,
+      "grad_norm": 3.43340802192688,
+      "learning_rate": 3.4472934472934476e-05,
+      "loss": 0.1735,
+      "step": 2960
+    },
+    {
+      "epoch": 37.83,
+      "grad_norm": 4.661231994628906,
+      "learning_rate": 3.4401709401709405e-05,
+      "loss": 0.1866,
+      "step": 2970
+    },
+    {
+      "epoch": 37.96,
+      "grad_norm": 3.5270962715148926,
+      "learning_rate": 3.433048433048433e-05,
+      "loss": 0.1748,
+      "step": 2980
+    },
+    {
+      "epoch": 38.0,
+      "eval_accuracy": 0.9695067264573991,
+      "eval_loss": 0.0894242599606514,
+      "eval_runtime": 10.4549,
+      "eval_samples_per_second": 426.594,
+      "eval_steps_per_second": 3.348,
+      "step": 2983
+    },
+    {
+      "epoch": 38.09,
+      "grad_norm": 3.2506449222564697,
+      "learning_rate": 3.425925925925926e-05,
+      "loss": 0.1792,
+      "step": 2990
+    },
+    {
+      "epoch": 38.22,
+      "grad_norm": 3.383273124694824,
+      "learning_rate": 3.418803418803419e-05,
+      "loss": 0.1672,
+      "step": 3000
+    },
+    {
+      "epoch": 38.34,
+      "grad_norm": 4.260499000549316,
+      "learning_rate": 3.411680911680912e-05,
+      "loss": 0.1728,
+      "step": 3010
+    },
+    {
+      "epoch": 38.47,
+      "grad_norm": 3.8010692596435547,
+      "learning_rate": 3.404558404558404e-05,
+      "loss": 0.1649,
+      "step": 3020
+    },
+    {
+      "epoch": 38.6,
+      "grad_norm": 3.387493371963501,
+      "learning_rate": 3.397435897435898e-05,
+      "loss": 0.1653,
+      "step": 3030
+    },
+    {
+      "epoch": 38.73,
+      "grad_norm": 2.461514711380005,
+      "learning_rate": 3.3903133903133905e-05,
+      "loss": 0.1598,
+      "step": 3040
+    },
+    {
+      "epoch": 38.85,
+      "grad_norm": 2.754783868789673,
+      "learning_rate": 3.3831908831908834e-05,
+      "loss": 0.1613,
+      "step": 3050
+    },
+    {
+      "epoch": 38.98,
+      "grad_norm": 3.63181209564209,
+      "learning_rate": 3.376068376068376e-05,
+      "loss": 0.1677,
+      "step": 3060
+    },
+    {
+      "epoch": 38.99,
+      "eval_accuracy": 0.9639013452914799,
+      "eval_loss": 0.10370239615440369,
+      "eval_runtime": 10.8803,
+      "eval_samples_per_second": 409.915,
+      "eval_steps_per_second": 3.217,
+      "step": 3061
+    },
+    {
+      "epoch": 39.11,
+      "grad_norm": 3.2453739643096924,
+      "learning_rate": 3.368945868945869e-05,
+      "loss": 0.1749,
+      "step": 3070
+    },
+    {
+      "epoch": 39.24,
+      "grad_norm": 3.9649202823638916,
+      "learning_rate": 3.361823361823362e-05,
+      "loss": 0.1663,
+      "step": 3080
+    },
+    {
+      "epoch": 39.36,
+      "grad_norm": 4.201056957244873,
+      "learning_rate": 3.354700854700855e-05,
+      "loss": 0.1669,
+      "step": 3090
+    },
+    {
+      "epoch": 39.49,
+      "grad_norm": 4.199280738830566,
+      "learning_rate": 3.347578347578348e-05,
+      "loss": 0.1699,
+      "step": 3100
+    },
+    {
+      "epoch": 39.62,
+      "grad_norm": 2.895860433578491,
+      "learning_rate": 3.3404558404558406e-05,
+      "loss": 0.1686,
+      "step": 3110
+    },
+    {
+      "epoch": 39.75,
+      "grad_norm": 2.9467713832855225,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.1662,
+      "step": 3120
+    },
+    {
+      "epoch": 39.87,
+      "grad_norm": 3.6263599395751953,
+      "learning_rate": 3.3262108262108264e-05,
+      "loss": 0.1587,
+      "step": 3130
+    },
+    {
+      "epoch": 40.0,
+      "grad_norm": 3.581874370574951,
+      "learning_rate": 3.319088319088319e-05,
+      "loss": 0.1723,
+      "step": 3140
+    },
+    {
+      "epoch": 40.0,
+      "eval_accuracy": 0.963677130044843,
+      "eval_loss": 0.10090423375368118,
+      "eval_runtime": 10.487,
+      "eval_samples_per_second": 425.289,
+      "eval_steps_per_second": 3.337,
+      "step": 3140
+    },
+    {
+      "epoch": 40.13,
+      "grad_norm": 3.810126543045044,
+      "learning_rate": 3.311965811965812e-05,
+      "loss": 0.1535,
+      "step": 3150
+    },
+    {
+      "epoch": 40.25,
+      "grad_norm": 3.254972219467163,
+      "learning_rate": 3.304843304843305e-05,
+      "loss": 0.165,
+      "step": 3160
+    },
+    {
+      "epoch": 40.38,
+      "grad_norm": 3.7977054119110107,
+      "learning_rate": 3.297720797720798e-05,
+      "loss": 0.161,
+      "step": 3170
+    },
+    {
+      "epoch": 40.51,
+      "grad_norm": 2.7377121448516846,
+      "learning_rate": 3.290598290598291e-05,
+      "loss": 0.1622,
+      "step": 3180
+    },
+    {
+      "epoch": 40.64,
+      "grad_norm": 3.7134432792663574,
+      "learning_rate": 3.2834757834757836e-05,
+      "loss": 0.1672,
+      "step": 3190
+    },
+    {
+      "epoch": 40.76,
+      "grad_norm": 4.389490127563477,
+      "learning_rate": 3.2763532763532764e-05,
+      "loss": 0.1653,
+      "step": 3200
+    },
+    {
+      "epoch": 40.89,
+      "grad_norm": 2.8473117351531982,
+      "learning_rate": 3.269230769230769e-05,
+      "loss": 0.1552,
+      "step": 3210
+    },
+    {
+      "epoch": 40.99,
+      "eval_accuracy": 0.9697309417040358,
+      "eval_loss": 0.08673026412725449,
+      "eval_runtime": 10.5751,
+      "eval_samples_per_second": 421.744,
+      "eval_steps_per_second": 3.31,
+      "step": 3218
+    },
+    {
+      "epoch": 41.02,
+      "grad_norm": 3.270151138305664,
+      "learning_rate": 3.262108262108262e-05,
+      "loss": 0.1578,
+      "step": 3220
+    },
+    {
+      "epoch": 41.15,
+      "grad_norm": 2.5095419883728027,
+      "learning_rate": 3.254985754985755e-05,
+      "loss": 0.1576,
+      "step": 3230
+    },
+    {
+      "epoch": 41.27,
+      "grad_norm": 4.268132209777832,
+      "learning_rate": 3.247863247863248e-05,
+      "loss": 0.1619,
+      "step": 3240
+    },
+    {
+      "epoch": 41.4,
+      "grad_norm": 3.5522594451904297,
+      "learning_rate": 3.240740740740741e-05,
+      "loss": 0.1545,
+      "step": 3250
+    },
+    {
+      "epoch": 41.53,
+      "grad_norm": 4.531596660614014,
+      "learning_rate": 3.2336182336182337e-05,
+      "loss": 0.1627,
+      "step": 3260
+    },
+    {
+      "epoch": 41.66,
+      "grad_norm": 2.89469575881958,
+      "learning_rate": 3.2264957264957265e-05,
+      "loss": 0.1576,
+      "step": 3270
+    },
+    {
+      "epoch": 41.78,
+      "grad_norm": 3.506288766860962,
+      "learning_rate": 3.2193732193732194e-05,
+      "loss": 0.1675,
+      "step": 3280
+    },
+    {
+      "epoch": 41.91,
+      "grad_norm": 2.518845796585083,
+      "learning_rate": 3.212250712250712e-05,
+      "loss": 0.1658,
+      "step": 3290
+    },
+    {
+      "epoch": 42.0,
+      "eval_accuracy": 0.9701793721973094,
+      "eval_loss": 0.0859610065817833,
+      "eval_runtime": 10.0539,
+      "eval_samples_per_second": 443.611,
+      "eval_steps_per_second": 3.481,
+      "step": 3297
+    },
+    {
+      "epoch": 42.04,
+      "grad_norm": 3.802457332611084,
+      "learning_rate": 3.205128205128206e-05,
+      "loss": 0.1662,
+      "step": 3300
+    },
+    {
+      "epoch": 42.17,
+      "grad_norm": 3.1138010025024414,
+      "learning_rate": 3.198005698005698e-05,
+      "loss": 0.1531,
+      "step": 3310
+    },
+    {
+      "epoch": 42.29,
+      "grad_norm": 6.375518321990967,
+      "learning_rate": 3.190883190883191e-05,
+      "loss": 0.1617,
+      "step": 3320
+    },
+    {
+      "epoch": 42.42,
+      "grad_norm": 5.619466304779053,
+      "learning_rate": 3.183760683760684e-05,
+      "loss": 0.1585,
+      "step": 3330
+    },
+    {
+      "epoch": 42.55,
+      "grad_norm": 2.600783586502075,
+      "learning_rate": 3.176638176638177e-05,
+      "loss": 0.1607,
+      "step": 3340
+    },
+    {
+      "epoch": 42.68,
+      "grad_norm": 4.039291858673096,
+      "learning_rate": 3.1695156695156695e-05,
+      "loss": 0.1515,
+      "step": 3350
+    },
+    {
+      "epoch": 42.8,
+      "grad_norm": 4.614203929901123,
+      "learning_rate": 3.162393162393162e-05,
+      "loss": 0.1368,
+      "step": 3360
+    },
+    {
+      "epoch": 42.93,
+      "grad_norm": 2.795916795730591,
+      "learning_rate": 3.155270655270656e-05,
+      "loss": 0.1531,
+      "step": 3370
+    },
+    {
+      "epoch": 42.99,
+      "eval_accuracy": 0.9672645739910314,
+      "eval_loss": 0.09214626997709274,
+      "eval_runtime": 10.26,
+      "eval_samples_per_second": 434.7,
+      "eval_steps_per_second": 3.411,
+      "step": 3375
+    },
+    {
+      "epoch": 43.06,
+      "grad_norm": 3.106384038925171,
+      "learning_rate": 3.148148148148148e-05,
+      "loss": 0.1597,
+      "step": 3380
+    },
+    {
+      "epoch": 43.18,
+      "grad_norm": 3.675632953643799,
+      "learning_rate": 3.141025641025641e-05,
+      "loss": 0.1531,
+      "step": 3390
+    },
+    {
+      "epoch": 43.31,
+      "grad_norm": 3.027937173843384,
+      "learning_rate": 3.133903133903134e-05,
+      "loss": 0.161,
+      "step": 3400
+    },
+    {
+      "epoch": 43.44,
+      "grad_norm": 3.678048849105835,
+      "learning_rate": 3.1267806267806274e-05,
+      "loss": 0.162,
+      "step": 3410
+    },
+    {
+      "epoch": 43.57,
+      "grad_norm": 3.193264961242676,
+      "learning_rate": 3.1196581196581195e-05,
+      "loss": 0.1574,
+      "step": 3420
+    },
+    {
+      "epoch": 43.69,
+      "grad_norm": 4.626323699951172,
+      "learning_rate": 3.1125356125356124e-05,
+      "loss": 0.1633,
+      "step": 3430
+    },
+    {
+      "epoch": 43.82,
+      "grad_norm": 4.2914018630981445,
+      "learning_rate": 3.105413105413106e-05,
+      "loss": 0.1563,
+      "step": 3440
+    },
+    {
+      "epoch": 43.95,
+      "grad_norm": 3.208162784576416,
+      "learning_rate": 3.098290598290599e-05,
+      "loss": 0.1593,
+      "step": 3450
+    },
+    {
+      "epoch": 44.0,
+      "eval_accuracy": 0.9645739910313902,
+      "eval_loss": 0.09411517530679703,
+      "eval_runtime": 10.3819,
+      "eval_samples_per_second": 429.592,
+      "eval_steps_per_second": 3.371,
+      "step": 3454
+    },
+    {
+      "epoch": 44.08,
+      "grad_norm": 4.032459735870361,
+      "learning_rate": 3.091168091168091e-05,
+      "loss": 0.1518,
+      "step": 3460
+    },
+    {
+      "epoch": 44.2,
+      "grad_norm": 3.239495038986206,
+      "learning_rate": 3.084045584045584e-05,
+      "loss": 0.1519,
+      "step": 3470
+    },
+    {
+      "epoch": 44.33,
+      "grad_norm": 2.621792793273926,
+      "learning_rate": 3.0769230769230774e-05,
+      "loss": 0.1385,
+      "step": 3480
+    },
+    {
+      "epoch": 44.46,
+      "grad_norm": 4.725817680358887,
+      "learning_rate": 3.0698005698005696e-05,
+      "loss": 0.142,
+      "step": 3490
+    },
+    {
+      "epoch": 44.59,
+      "grad_norm": 4.686051368713379,
+      "learning_rate": 3.0626780626780625e-05,
+      "loss": 0.1542,
+      "step": 3500
+    },
+    {
+      "epoch": 44.71,
+      "grad_norm": 4.348143577575684,
+      "learning_rate": 3.055555555555556e-05,
+      "loss": 0.1529,
+      "step": 3510
+    },
+    {
+      "epoch": 44.84,
+      "grad_norm": 2.644388437271118,
+      "learning_rate": 3.0484330484330486e-05,
+      "loss": 0.1524,
+      "step": 3520
+    },
+    {
+      "epoch": 44.97,
+      "grad_norm": 3.8867974281311035,
+      "learning_rate": 3.0413105413105414e-05,
+      "loss": 0.1568,
+      "step": 3530
+    },
+    {
+      "epoch": 44.99,
+      "eval_accuracy": 0.968609865470852,
+      "eval_loss": 0.08786384016275406,
+      "eval_runtime": 10.3766,
+      "eval_samples_per_second": 429.814,
+      "eval_steps_per_second": 3.373,
+      "step": 3532
+    },
+    {
+      "epoch": 45.1,
+      "grad_norm": 3.1211063861846924,
+      "learning_rate": 3.034188034188034e-05,
+      "loss": 0.1518,
+      "step": 3540
+    },
+    {
+      "epoch": 45.22,
+      "grad_norm": 4.649716854095459,
+      "learning_rate": 3.0270655270655275e-05,
+      "loss": 0.1519,
+      "step": 3550
+    },
+    {
+      "epoch": 45.35,
+      "grad_norm": 2.5936279296875,
+      "learning_rate": 3.01994301994302e-05,
+      "loss": 0.1462,
+      "step": 3560
+    },
+    {
+      "epoch": 45.48,
+      "grad_norm": 3.2736833095550537,
+      "learning_rate": 3.012820512820513e-05,
+      "loss": 0.1516,
+      "step": 3570
+    },
+    {
+      "epoch": 45.61,
+      "grad_norm": 3.4682116508483887,
+      "learning_rate": 3.005698005698006e-05,
+      "loss": 0.1482,
+      "step": 3580
+    },
+    {
+      "epoch": 45.73,
+      "grad_norm": 4.235909938812256,
+      "learning_rate": 2.9985754985754986e-05,
+      "loss": 0.147,
+      "step": 3590
+    },
+    {
+      "epoch": 45.86,
+      "grad_norm": 2.4908273220062256,
+      "learning_rate": 2.9914529914529915e-05,
+      "loss": 0.1431,
+      "step": 3600
+    },
+    {
+      "epoch": 45.99,
+      "grad_norm": 4.006827354431152,
+      "learning_rate": 2.9843304843304844e-05,
+      "loss": 0.1621,
+      "step": 3610
+    },
+    {
+      "epoch": 46.0,
+      "eval_accuracy": 0.9692825112107624,
+      "eval_loss": 0.09294471889734268,
+      "eval_runtime": 10.4845,
+      "eval_samples_per_second": 425.39,
+      "eval_steps_per_second": 3.338,
+      "step": 3611
+    },
+    {
+      "epoch": 46.11,
+      "grad_norm": 3.718818187713623,
+      "learning_rate": 2.9772079772079776e-05,
+      "loss": 0.1594,
+      "step": 3620
+    },
+    {
+      "epoch": 46.24,
+      "grad_norm": 3.664454936981201,
+      "learning_rate": 2.97008547008547e-05,
+      "loss": 0.1466,
+      "step": 3630
+    },
+    {
+      "epoch": 46.37,
+      "grad_norm": 3.0173661708831787,
+      "learning_rate": 2.962962962962963e-05,
+      "loss": 0.1544,
+      "step": 3640
+    },
+    {
+      "epoch": 46.5,
+      "grad_norm": 3.583787679672241,
+      "learning_rate": 2.9558404558404562e-05,
+      "loss": 0.1559,
+      "step": 3650
+    },
+    {
+      "epoch": 46.62,
+      "grad_norm": 3.0325767993927,
+      "learning_rate": 2.948717948717949e-05,
+      "loss": 0.1347,
+      "step": 3660
+    },
+    {
+      "epoch": 46.75,
+      "grad_norm": 3.1503076553344727,
+      "learning_rate": 2.9415954415954416e-05,
+      "loss": 0.1531,
+      "step": 3670
+    },
+    {
+      "epoch": 46.88,
+      "grad_norm": 3.430950164794922,
+      "learning_rate": 2.9344729344729345e-05,
+      "loss": 0.1602,
+      "step": 3680
+    },
+    {
+      "epoch": 46.99,
+      "eval_accuracy": 0.974439461883408,
+      "eval_loss": 0.0777370035648346,
+      "eval_runtime": 10.1977,
+      "eval_samples_per_second": 437.355,
+      "eval_steps_per_second": 3.432,
+      "step": 3689
+    },
+    {
+      "epoch": 47.01,
+      "grad_norm": 3.286393165588379,
+      "learning_rate": 2.9273504273504277e-05,
+      "loss": 0.1461,
+      "step": 3690
+    },
+    {
+      "epoch": 47.13,
+      "grad_norm": 4.135302543640137,
+      "learning_rate": 2.9202279202279202e-05,
+      "loss": 0.1468,
+      "step": 3700
+    },
+    {
+      "epoch": 47.26,
+      "grad_norm": 3.4055960178375244,
+      "learning_rate": 2.913105413105413e-05,
+      "loss": 0.1402,
+      "step": 3710
+    },
+    {
+      "epoch": 47.39,
+      "grad_norm": 3.821648359298706,
+      "learning_rate": 2.9059829059829063e-05,
+      "loss": 0.1449,
+      "step": 3720
+    },
+    {
+      "epoch": 47.52,
+      "grad_norm": 3.148038148880005,
+      "learning_rate": 2.898860398860399e-05,
+      "loss": 0.1386,
+      "step": 3730
+    },
+    {
+      "epoch": 47.64,
+      "grad_norm": 2.2282421588897705,
+      "learning_rate": 2.8917378917378917e-05,
+      "loss": 0.1423,
+      "step": 3740
+    },
+    {
+      "epoch": 47.77,
+      "grad_norm": 3.5172359943389893,
+      "learning_rate": 2.8846153846153845e-05,
+      "loss": 0.1531,
+      "step": 3750
+    },
+    {
+      "epoch": 47.9,
+      "grad_norm": 3.5688581466674805,
+      "learning_rate": 2.8774928774928778e-05,
+      "loss": 0.1397,
+      "step": 3760
+    },
+    {
+      "epoch": 48.0,
+      "eval_accuracy": 0.9733183856502242,
+      "eval_loss": 0.08208809792995453,
+      "eval_runtime": 10.4302,
+      "eval_samples_per_second": 427.603,
+      "eval_steps_per_second": 3.356,
+      "step": 3768
+    },
+    {
+      "epoch": 48.03,
+      "grad_norm": 2.843202829360962,
+      "learning_rate": 2.8703703703703706e-05,
+      "loss": 0.1411,
+      "step": 3770
+    },
+    {
+      "epoch": 48.15,
+      "grad_norm": 3.353113889694214,
+      "learning_rate": 2.863247863247863e-05,
+      "loss": 0.145,
+      "step": 3780
+    },
+    {
+      "epoch": 48.28,
+      "grad_norm": 3.2403688430786133,
+      "learning_rate": 2.8561253561253564e-05,
+      "loss": 0.1338,
+      "step": 3790
+    },
+    {
+      "epoch": 48.41,
+      "grad_norm": 2.347370147705078,
+      "learning_rate": 2.8490028490028492e-05,
+      "loss": 0.1355,
+      "step": 3800
+    },
+    {
+      "epoch": 48.54,
+      "grad_norm": 3.2337448596954346,
+      "learning_rate": 2.8418803418803418e-05,
+      "loss": 0.1424,
+      "step": 3810
+    },
+    {
+      "epoch": 48.66,
+      "grad_norm": 4.670347213745117,
+      "learning_rate": 2.8347578347578346e-05,
+      "loss": 0.1485,
+      "step": 3820
+    },
+    {
+      "epoch": 48.79,
+      "grad_norm": 5.514860153198242,
+      "learning_rate": 2.827635327635328e-05,
+      "loss": 0.1551,
+      "step": 3830
+    },
+    {
+      "epoch": 48.92,
+      "grad_norm": 4.811940670013428,
+      "learning_rate": 2.8205128205128207e-05,
+      "loss": 0.1475,
+      "step": 3840
+    },
+    {
+      "epoch": 48.99,
+      "eval_accuracy": 0.9710762331838565,
+      "eval_loss": 0.08098708093166351,
+      "eval_runtime": 10.2161,
+      "eval_samples_per_second": 436.568,
+      "eval_steps_per_second": 3.426,
+      "step": 3846
+    },
+    {
+      "epoch": 49.04,
+      "grad_norm": 2.1994664669036865,
+      "learning_rate": 2.8133903133903132e-05,
+      "loss": 0.1395,
+      "step": 3850
+    },
+    {
+      "epoch": 49.17,
+      "grad_norm": 3.298173666000366,
+      "learning_rate": 2.8062678062678068e-05,
+      "loss": 0.1446,
+      "step": 3860
+    },
+    {
+      "epoch": 49.3,
+      "grad_norm": 3.0983335971832275,
+      "learning_rate": 2.7991452991452993e-05,
+      "loss": 0.1445,
+      "step": 3870
+    },
+    {
+      "epoch": 49.43,
+      "grad_norm": 2.741196393966675,
+      "learning_rate": 2.7920227920227922e-05,
+      "loss": 0.1398,
+      "step": 3880
+    },
+    {
+      "epoch": 49.55,
+      "grad_norm": 4.882288932800293,
+      "learning_rate": 2.7849002849002847e-05,
+      "loss": 0.1393,
+      "step": 3890
+    },
+    {
+      "epoch": 49.68,
+      "grad_norm": 3.602112054824829,
+      "learning_rate": 2.777777777777778e-05,
+      "loss": 0.1358,
+      "step": 3900
+    },
+    {
+      "epoch": 49.81,
+      "grad_norm": 3.6309425830841064,
+      "learning_rate": 2.7706552706552708e-05,
+      "loss": 0.1431,
+      "step": 3910
+    },
+    {
+      "epoch": 49.94,
+      "grad_norm": 4.7522101402282715,
+      "learning_rate": 2.7635327635327633e-05,
+      "loss": 0.1452,
+      "step": 3920
+    },
+    {
+      "epoch": 50.0,
+      "eval_accuracy": 0.9697309417040358,
+      "eval_loss": 0.0942649319767952,
+      "eval_runtime": 10.3942,
+      "eval_samples_per_second": 429.086,
+      "eval_steps_per_second": 3.367,
+      "step": 3925
+    },
+    {
+      "epoch": 50.06,
+      "grad_norm": 2.828382730484009,
+      "learning_rate": 2.756410256410257e-05,
+      "loss": 0.1416,
+      "step": 3930
+    },
+    {
+      "epoch": 50.19,
+      "grad_norm": 2.9188289642333984,
+      "learning_rate": 2.7492877492877494e-05,
+      "loss": 0.1428,
+      "step": 3940
+    },
+    {
+      "epoch": 50.32,
+      "grad_norm": 3.1731650829315186,
+      "learning_rate": 2.7421652421652423e-05,
+      "loss": 0.1275,
+      "step": 3950
+    },
+    {
+      "epoch": 50.45,
+      "grad_norm": 3.445089340209961,
+      "learning_rate": 2.7350427350427355e-05,
+      "loss": 0.138,
+      "step": 3960
+    },
+    {
+      "epoch": 50.57,
+      "grad_norm": 3.163792133331299,
+      "learning_rate": 2.7279202279202283e-05,
+      "loss": 0.1387,
+      "step": 3970
+    },
+    {
+      "epoch": 50.7,
+      "grad_norm": 2.461280107498169,
+      "learning_rate": 2.720797720797721e-05,
+      "loss": 0.1439,
+      "step": 3980
+    },
+    {
+      "epoch": 50.83,
+      "grad_norm": 3.221646785736084,
+      "learning_rate": 2.7136752136752137e-05,
+      "loss": 0.125,
+      "step": 3990
+    },
+    {
+      "epoch": 50.96,
+      "grad_norm": 4.313431262969971,
+      "learning_rate": 2.706552706552707e-05,
+      "loss": 0.148,
+      "step": 4000
+    },
+    {
+      "epoch": 50.99,
+      "eval_accuracy": 0.9704035874439462,
+      "eval_loss": 0.0943058580160141,
+      "eval_runtime": 10.7055,
+      "eval_samples_per_second": 416.608,
+      "eval_steps_per_second": 3.269,
+      "step": 4003
+    },
+    {
+      "epoch": 51.08,
+      "grad_norm": 3.418119192123413,
+      "learning_rate": 2.6994301994301995e-05,
+      "loss": 0.1299,
+      "step": 4010
+    },
+    {
+      "epoch": 51.21,
+      "grad_norm": 3.145573854446411,
+      "learning_rate": 2.6923076923076923e-05,
+      "loss": 0.1406,
+      "step": 4020
+    },
+    {
+      "epoch": 51.34,
+      "grad_norm": 3.141972541809082,
+      "learning_rate": 2.6851851851851855e-05,
+      "loss": 0.1463,
+      "step": 4030
+    },
+    {
+      "epoch": 51.46,
+      "grad_norm": 2.996499538421631,
+      "learning_rate": 2.6780626780626784e-05,
+      "loss": 0.1453,
+      "step": 4040
+    },
+    {
+      "epoch": 51.59,
+      "grad_norm": 8.0499849319458,
+      "learning_rate": 2.670940170940171e-05,
+      "loss": 0.1423,
+      "step": 4050
+    },
+    {
+      "epoch": 51.72,
+      "grad_norm": 2.710381507873535,
+      "learning_rate": 2.6638176638176638e-05,
+      "loss": 0.145,
+      "step": 4060
+    },
+    {
+      "epoch": 51.85,
+      "grad_norm": 2.674241304397583,
+      "learning_rate": 2.656695156695157e-05,
+      "loss": 0.1447,
+      "step": 4070
+    },
+    {
+      "epoch": 51.97,
+      "grad_norm": 4.219669342041016,
+      "learning_rate": 2.64957264957265e-05,
+      "loss": 0.1392,
+      "step": 4080
+    },
+    {
+      "epoch": 52.0,
+      "eval_accuracy": 0.9695067264573991,
+      "eval_loss": 0.08689579367637634,
+      "eval_runtime": 10.4402,
+      "eval_samples_per_second": 427.196,
+      "eval_steps_per_second": 3.352,
+      "step": 4082
+    },
+    {
+      "epoch": 52.1,
+      "grad_norm": 3.582681179046631,
+      "learning_rate": 2.6424501424501424e-05,
+      "loss": 0.1486,
+      "step": 4090
+    },
+    {
+      "epoch": 52.23,
+      "grad_norm": 5.496790409088135,
+      "learning_rate": 2.6353276353276356e-05,
+      "loss": 0.1415,
+      "step": 4100
+    },
+    {
+      "epoch": 52.36,
+      "grad_norm": 3.7762444019317627,
+      "learning_rate": 2.6282051282051285e-05,
+      "loss": 0.1414,
+      "step": 4110
+    },
+    {
+      "epoch": 52.48,
+      "grad_norm": 3.185096502304077,
+      "learning_rate": 2.621082621082621e-05,
+      "loss": 0.1394,
+      "step": 4120
+    },
+    {
+      "epoch": 52.61,
+      "grad_norm": 3.969970941543579,
+      "learning_rate": 2.613960113960114e-05,
+      "loss": 0.1471,
+      "step": 4130
+    },
+    {
+      "epoch": 52.74,
+      "grad_norm": 3.0576138496398926,
+      "learning_rate": 2.606837606837607e-05,
+      "loss": 0.1425,
+      "step": 4140
+    },
+    {
+      "epoch": 52.87,
+      "grad_norm": 3.9330620765686035,
+      "learning_rate": 2.5997150997151e-05,
+      "loss": 0.1474,
+      "step": 4150
+    },
+    {
+      "epoch": 52.99,
+      "grad_norm": 4.062915325164795,
+      "learning_rate": 2.5925925925925925e-05,
+      "loss": 0.1389,
+      "step": 4160
+    },
+    {
+      "epoch": 52.99,
+      "eval_accuracy": 0.9737668161434978,
+      "eval_loss": 0.08755916357040405,
+      "eval_runtime": 10.2788,
+      "eval_samples_per_second": 433.903,
+      "eval_steps_per_second": 3.405,
+      "step": 4160
+    },
+    {
+      "epoch": 53.12,
+      "grad_norm": 5.062565803527832,
+      "learning_rate": 2.5854700854700857e-05,
+      "loss": 0.1362,
+      "step": 4170
+    },
+    {
+      "epoch": 53.25,
+      "grad_norm": 3.4343032836914062,
+      "learning_rate": 2.5783475783475786e-05,
+      "loss": 0.1416,
+      "step": 4180
+    },
+    {
+      "epoch": 53.38,
+      "grad_norm": 2.700774669647217,
+      "learning_rate": 2.5712250712250714e-05,
+      "loss": 0.1326,
+      "step": 4190
+    },
+    {
+      "epoch": 53.5,
+      "grad_norm": 3.6101787090301514,
+      "learning_rate": 2.564102564102564e-05,
+      "loss": 0.1425,
+      "step": 4200
+    },
+    {
+      "epoch": 53.63,
+      "grad_norm": 3.596529960632324,
+      "learning_rate": 2.5569800569800572e-05,
+      "loss": 0.1282,
+      "step": 4210
+    },
+    {
+      "epoch": 53.76,
+      "grad_norm": 3.493985891342163,
+      "learning_rate": 2.54985754985755e-05,
+      "loss": 0.1426,
+      "step": 4220
+    },
+    {
+      "epoch": 53.89,
+      "grad_norm": 4.897692680358887,
+      "learning_rate": 2.5427350427350426e-05,
+      "loss": 0.1295,
+      "step": 4230
+    },
+    {
+      "epoch": 54.0,
+      "eval_accuracy": 0.9748878923766816,
+      "eval_loss": 0.07158580422401428,
+      "eval_runtime": 10.2959,
+      "eval_samples_per_second": 433.183,
+      "eval_steps_per_second": 3.399,
+      "step": 4239
+    },
+    {
+      "epoch": 54.01,
+      "grad_norm": 2.9666993618011475,
+      "learning_rate": 2.535612535612536e-05,
+      "loss": 0.1345,
+      "step": 4240
+    },
+    {
+      "epoch": 54.14,
+      "grad_norm": 3.897339105606079,
+      "learning_rate": 2.5284900284900286e-05,
+      "loss": 0.1323,
+      "step": 4250
+    },
+    {
+      "epoch": 54.27,
+      "grad_norm": 2.5065248012542725,
+      "learning_rate": 2.5213675213675215e-05,
+      "loss": 0.1279,
+      "step": 4260
+    },
+    {
+      "epoch": 54.39,
+      "grad_norm": 5.935027122497559,
+      "learning_rate": 2.514245014245014e-05,
+      "loss": 0.1365,
+      "step": 4270
+    },
+    {
+      "epoch": 54.52,
+      "grad_norm": 3.2909152507781982,
+      "learning_rate": 2.5071225071225073e-05,
+      "loss": 0.1465,
+      "step": 4280
+    },
+    {
+      "epoch": 54.65,
+      "grad_norm": 3.701512336730957,
+      "learning_rate": 2.5e-05,
+      "loss": 0.1278,
+      "step": 4290
+    },
+    {
+      "epoch": 54.78,
+      "grad_norm": 3.227336883544922,
+      "learning_rate": 2.492877492877493e-05,
+      "loss": 0.1398,
+      "step": 4300
+    },
+    {
+      "epoch": 54.9,
+      "grad_norm": 4.817599773406982,
+      "learning_rate": 2.485754985754986e-05,
+      "loss": 0.1394,
+      "step": 4310
+    },
+    {
+      "epoch": 54.99,
+      "eval_accuracy": 0.9751121076233183,
+      "eval_loss": 0.07604929804801941,
+      "eval_runtime": 10.3795,
+      "eval_samples_per_second": 429.695,
+      "eval_steps_per_second": 3.372,
+      "step": 4317
+    },
+    {
+      "epoch": 55.03,
+      "grad_norm": 3.4743528366088867,
+      "learning_rate": 2.4786324786324787e-05,
+      "loss": 0.139,
+      "step": 4320
+    },
+    {
+      "epoch": 55.16,
+      "grad_norm": 3.180518865585327,
+      "learning_rate": 2.4715099715099716e-05,
+      "loss": 0.1309,
+      "step": 4330
+    },
+    {
+      "epoch": 55.29,
+      "grad_norm": 4.160658359527588,
+      "learning_rate": 2.4643874643874645e-05,
+      "loss": 0.1303,
+      "step": 4340
+    },
+    {
+      "epoch": 55.41,
+      "grad_norm": 4.241569519042969,
+      "learning_rate": 2.4572649572649573e-05,
+      "loss": 0.1269,
+      "step": 4350
+    },
+    {
+      "epoch": 55.54,
+      "grad_norm": 3.5420875549316406,
+      "learning_rate": 2.4501424501424502e-05,
+      "loss": 0.1363,
+      "step": 4360
+    },
+    {
+      "epoch": 55.67,
+      "grad_norm": 3.2201666831970215,
+      "learning_rate": 2.4430199430199434e-05,
+      "loss": 0.1339,
+      "step": 4370
+    },
+    {
+      "epoch": 55.8,
+      "grad_norm": 3.6691396236419678,
+      "learning_rate": 2.435897435897436e-05,
+      "loss": 0.1212,
+      "step": 4380
+    },
+    {
+      "epoch": 55.92,
+      "grad_norm": 3.5013792514801025,
+      "learning_rate": 2.4287749287749288e-05,
+      "loss": 0.1354,
+      "step": 4390
+    },
+    {
+      "epoch": 56.0,
+      "eval_accuracy": 0.9726457399103139,
+      "eval_loss": 0.07711587846279144,
+      "eval_runtime": 10.631,
+      "eval_samples_per_second": 419.529,
+      "eval_steps_per_second": 3.292,
+      "step": 4396
+    },
+    {
+      "epoch": 56.05,
+      "grad_norm": 3.276965618133545,
+      "learning_rate": 2.4216524216524217e-05,
+      "loss": 0.1331,
+      "step": 4400
+    },
+    {
+      "epoch": 56.18,
+      "grad_norm": 4.311830997467041,
+      "learning_rate": 2.4145299145299145e-05,
+      "loss": 0.1376,
+      "step": 4410
+    },
+    {
+      "epoch": 56.31,
+      "grad_norm": 3.6508076190948486,
+      "learning_rate": 2.4074074074074074e-05,
+      "loss": 0.1415,
+      "step": 4420
+    },
+    {
+      "epoch": 56.43,
+      "grad_norm": 4.303562641143799,
+      "learning_rate": 2.4002849002849003e-05,
+      "loss": 0.1298,
+      "step": 4430
+    },
+    {
+      "epoch": 56.56,
+      "grad_norm": 3.281374216079712,
+      "learning_rate": 2.3931623931623935e-05,
+      "loss": 0.1273,
+      "step": 4440
+    },
+    {
+      "epoch": 56.69,
+      "grad_norm": 4.081586837768555,
+      "learning_rate": 2.386039886039886e-05,
+      "loss": 0.1287,
+      "step": 4450
+    },
+    {
+      "epoch": 56.82,
+      "grad_norm": 5.350743293762207,
+      "learning_rate": 2.3789173789173792e-05,
+      "loss": 0.1336,
+      "step": 4460
+    },
+    {
+      "epoch": 56.94,
+      "grad_norm": 3.5683207511901855,
+      "learning_rate": 2.3717948717948718e-05,
+      "loss": 0.1303,
+      "step": 4470
+    },
+    {
+      "epoch": 56.99,
+      "eval_accuracy": 0.9737668161434978,
+      "eval_loss": 0.07378443330526352,
+      "eval_runtime": 10.3059,
+      "eval_samples_per_second": 432.762,
+      "eval_steps_per_second": 3.396,
+      "step": 4474
+    },
+    {
+      "epoch": 57.07,
+      "grad_norm": 4.208488941192627,
+      "learning_rate": 2.364672364672365e-05,
+      "loss": 0.1318,
+      "step": 4480
+    },
+    {
+      "epoch": 57.2,
+      "grad_norm": 3.7957112789154053,
+      "learning_rate": 2.3575498575498578e-05,
+      "loss": 0.1296,
+      "step": 4490
+    },
+    {
+      "epoch": 57.32,
+      "grad_norm": 3.4968860149383545,
+      "learning_rate": 2.3504273504273504e-05,
+      "loss": 0.1208,
+      "step": 4500
+    },
+    {
+      "epoch": 57.45,
+      "grad_norm": 3.205312728881836,
+      "learning_rate": 2.3433048433048436e-05,
+      "loss": 0.129,
+      "step": 4510
+    },
+    {
+      "epoch": 57.58,
+      "grad_norm": 3.5443649291992188,
+      "learning_rate": 2.336182336182336e-05,
+      "loss": 0.1249,
+      "step": 4520
+    },
+    {
+      "epoch": 57.71,
+      "grad_norm": 4.887803554534912,
+      "learning_rate": 2.3290598290598293e-05,
+      "loss": 0.1384,
+      "step": 4530
+    },
+    {
+      "epoch": 57.83,
+      "grad_norm": 3.6573920249938965,
+      "learning_rate": 2.321937321937322e-05,
+      "loss": 0.1215,
+      "step": 4540
+    },
+    {
+      "epoch": 57.96,
+      "grad_norm": 2.767714500427246,
+      "learning_rate": 2.314814814814815e-05,
+      "loss": 0.1274,
+      "step": 4550
+    },
+    {
+      "epoch": 58.0,
+      "eval_accuracy": 0.9724215246636772,
+      "eval_loss": 0.08340190351009369,
+      "eval_runtime": 10.2608,
+      "eval_samples_per_second": 434.665,
+      "eval_steps_per_second": 3.411,
+      "step": 4553
+    },
+    {
+      "epoch": 58.09,
+      "grad_norm": 2.663431406021118,
+      "learning_rate": 2.307692307692308e-05,
+      "loss": 0.1336,
+      "step": 4560
+    },
+    {
+      "epoch": 58.22,
+      "grad_norm": 3.646914482116699,
+      "learning_rate": 2.3005698005698008e-05,
+      "loss": 0.1225,
+      "step": 4570
+    },
+    {
+      "epoch": 58.34,
+      "grad_norm": 4.5192461013793945,
+      "learning_rate": 2.2934472934472936e-05,
+      "loss": 0.1305,
+      "step": 4580
+    },
+    {
+      "epoch": 58.47,
+      "grad_norm": 3.2621636390686035,
+      "learning_rate": 2.2863247863247865e-05,
+      "loss": 0.1223,
+      "step": 4590
+    },
+    {
+      "epoch": 58.6,
+      "grad_norm": 4.150683879852295,
+      "learning_rate": 2.2792022792022794e-05,
+      "loss": 0.1307,
+      "step": 4600
+    },
+    {
+      "epoch": 58.73,
+      "grad_norm": 2.834435224533081,
+      "learning_rate": 2.272079772079772e-05,
+      "loss": 0.1312,
+      "step": 4610
+    },
+    {
+      "epoch": 58.85,
+      "grad_norm": 4.989063739776611,
+      "learning_rate": 2.264957264957265e-05,
+      "loss": 0.1275,
+      "step": 4620
+    },
+    {
+      "epoch": 58.98,
+      "grad_norm": 2.5661489963531494,
+      "learning_rate": 2.257834757834758e-05,
+      "loss": 0.1276,
+      "step": 4630
+    },
+    {
+      "epoch": 58.99,
+      "eval_accuracy": 0.9746636771300449,
+      "eval_loss": 0.08008856326341629,
+      "eval_runtime": 11.2045,
+      "eval_samples_per_second": 398.054,
+      "eval_steps_per_second": 3.124,
+      "step": 4631
+    },
+    {
+      "epoch": 59.11,
+      "grad_norm": 5.07080602645874,
+      "learning_rate": 2.250712250712251e-05,
+      "loss": 0.1212,
+      "step": 4640
+    },
+    {
+      "epoch": 59.24,
+      "grad_norm": 3.80442476272583,
+      "learning_rate": 2.2435897435897437e-05,
+      "loss": 0.1254,
+      "step": 4650
+    },
+    {
+      "epoch": 59.36,
+      "grad_norm": 2.6264405250549316,
+      "learning_rate": 2.2364672364672366e-05,
+      "loss": 0.1206,
+      "step": 4660
+    },
+    {
+      "epoch": 59.49,
+      "grad_norm": 2.7287278175354004,
+      "learning_rate": 2.2293447293447295e-05,
+      "loss": 0.1206,
+      "step": 4670
+    },
+    {
+      "epoch": 59.62,
+      "grad_norm": 4.177188873291016,
+      "learning_rate": 2.2222222222222223e-05,
+      "loss": 0.1283,
+      "step": 4680
+    },
+    {
+      "epoch": 59.75,
+      "grad_norm": 3.6389882564544678,
+      "learning_rate": 2.2150997150997152e-05,
+      "loss": 0.1373,
+      "step": 4690
+    },
+    {
+      "epoch": 59.87,
+      "grad_norm": 3.7176778316497803,
+      "learning_rate": 2.207977207977208e-05,
+      "loss": 0.1225,
+      "step": 4700
+    },
+    {
+      "epoch": 60.0,
+      "grad_norm": 3.90850567817688,
+      "learning_rate": 2.200854700854701e-05,
+      "loss": 0.1372,
+      "step": 4710
+    },
+    {
+      "epoch": 60.0,
+      "eval_accuracy": 0.9762331838565023,
+      "eval_loss": 0.06674081087112427,
+      "eval_runtime": 10.292,
+      "eval_samples_per_second": 433.347,
+      "eval_steps_per_second": 3.401,
+      "step": 4710
+    },
+    {
+      "epoch": 60.13,
+      "grad_norm": 2.7248566150665283,
+      "learning_rate": 2.1937321937321938e-05,
+      "loss": 0.1263,
+      "step": 4720
+    },
+    {
+      "epoch": 60.25,
+      "grad_norm": 4.5060319900512695,
+      "learning_rate": 2.1866096866096867e-05,
+      "loss": 0.1239,
+      "step": 4730
+    },
+    {
+      "epoch": 60.38,
+      "grad_norm": 3.740729331970215,
+      "learning_rate": 2.1794871794871795e-05,
+      "loss": 0.1179,
+      "step": 4740
+    },
+    {
+      "epoch": 60.51,
+      "grad_norm": 3.870370388031006,
+      "learning_rate": 2.1723646723646724e-05,
+      "loss": 0.1281,
+      "step": 4750
+    },
+    {
+      "epoch": 60.64,
+      "grad_norm": 3.314084768295288,
+      "learning_rate": 2.1652421652421653e-05,
+      "loss": 0.1097,
+      "step": 4760
+    },
+    {
+      "epoch": 60.76,
+      "grad_norm": 3.503286838531494,
+      "learning_rate": 2.1581196581196585e-05,
+      "loss": 0.1349,
+      "step": 4770
+    },
+    {
+      "epoch": 60.89,
+      "grad_norm": 4.666501045227051,
+      "learning_rate": 2.150997150997151e-05,
+      "loss": 0.1417,
+      "step": 4780
+    },
+    {
+      "epoch": 60.99,
+      "eval_accuracy": 0.9737668161434978,
+      "eval_loss": 0.07472814619541168,
+      "eval_runtime": 10.7094,
+      "eval_samples_per_second": 416.456,
+      "eval_steps_per_second": 3.268,
+      "step": 4788
+    },
+    {
+      "epoch": 61.02,
+      "grad_norm": 3.3184502124786377,
+      "learning_rate": 2.143874643874644e-05,
+      "loss": 0.1315,
+      "step": 4790
+    },
+    {
+      "epoch": 61.15,
+      "grad_norm": 2.5868165493011475,
+      "learning_rate": 2.1367521367521368e-05,
+      "loss": 0.125,
+      "step": 4800
+    },
+    {
+      "epoch": 61.27,
+      "grad_norm": 2.161036729812622,
+      "learning_rate": 2.1296296296296296e-05,
+      "loss": 0.1302,
+      "step": 4810
+    },
+    {
+      "epoch": 61.4,
+      "grad_norm": 4.309581279754639,
+      "learning_rate": 2.1225071225071228e-05,
+      "loss": 0.1263,
+      "step": 4820
+    },
+    {
+      "epoch": 61.53,
+      "grad_norm": 4.2532148361206055,
+      "learning_rate": 2.1153846153846154e-05,
+      "loss": 0.1235,
+      "step": 4830
+    },
+    {
+      "epoch": 61.66,
+      "grad_norm": 3.6365723609924316,
+      "learning_rate": 2.1082621082621086e-05,
+      "loss": 0.1237,
+      "step": 4840
+    },
+    {
+      "epoch": 61.78,
+      "grad_norm": 2.715269088745117,
+      "learning_rate": 2.101139601139601e-05,
+      "loss": 0.1224,
+      "step": 4850
+    },
+    {
+      "epoch": 61.91,
+      "grad_norm": 1.8564375638961792,
+      "learning_rate": 2.0940170940170943e-05,
+      "loss": 0.1249,
+      "step": 4860
+    },
+    {
+      "epoch": 62.0,
+      "eval_accuracy": 0.974439461883408,
+      "eval_loss": 0.07346644997596741,
+      "eval_runtime": 11.0856,
+      "eval_samples_per_second": 402.324,
+      "eval_steps_per_second": 3.157,
+      "step": 4867
+    },
+    {
+      "epoch": 62.04,
+      "grad_norm": 6.367058753967285,
+      "learning_rate": 2.0868945868945868e-05,
+      "loss": 0.1208,
+      "step": 4870
+    },
+    {
+      "epoch": 62.17,
+      "grad_norm": 2.9588770866394043,
+      "learning_rate": 2.07977207977208e-05,
+      "loss": 0.1199,
+      "step": 4880
+    },
+    {
+      "epoch": 62.29,
+      "grad_norm": 2.998056650161743,
+      "learning_rate": 2.072649572649573e-05,
+      "loss": 0.1325,
+      "step": 4890
+    },
+    {
+      "epoch": 62.42,
+      "grad_norm": 3.0851778984069824,
+      "learning_rate": 2.0655270655270654e-05,
+      "loss": 0.1289,
+      "step": 4900
+    },
+    {
+      "epoch": 62.55,
+      "grad_norm": 3.708634376525879,
+      "learning_rate": 2.0584045584045586e-05,
+      "loss": 0.1262,
+      "step": 4910
+    },
+    {
+      "epoch": 62.68,
+      "grad_norm": 3.1951818466186523,
+      "learning_rate": 2.0512820512820512e-05,
+      "loss": 0.1289,
+      "step": 4920
+    },
+    {
+      "epoch": 62.8,
+      "grad_norm": 3.3787269592285156,
+      "learning_rate": 2.0441595441595444e-05,
+      "loss": 0.12,
+      "step": 4930
+    },
+    {
+      "epoch": 62.93,
+      "grad_norm": 3.423536539077759,
+      "learning_rate": 2.037037037037037e-05,
+      "loss": 0.1212,
+      "step": 4940
+    },
+    {
+      "epoch": 62.99,
+      "eval_accuracy": 0.9780269058295964,
+      "eval_loss": 0.06646457314491272,
+      "eval_runtime": 11.0112,
+      "eval_samples_per_second": 405.041,
+      "eval_steps_per_second": 3.179,
+      "step": 4945
+    },
+    {
+      "epoch": 63.06,
+      "grad_norm": 2.744992733001709,
+      "learning_rate": 2.02991452991453e-05,
+      "loss": 0.1274,
+      "step": 4950
+    },
+    {
+      "epoch": 63.18,
+      "grad_norm": 2.38899564743042,
+      "learning_rate": 2.022792022792023e-05,
+      "loss": 0.1175,
+      "step": 4960
+    },
+    {
+      "epoch": 63.31,
+      "grad_norm": 3.2532944679260254,
+      "learning_rate": 2.015669515669516e-05,
+      "loss": 0.127,
+      "step": 4970
+    },
+    {
+      "epoch": 63.44,
+      "grad_norm": 3.285688877105713,
+      "learning_rate": 2.0085470085470087e-05,
+      "loss": 0.1229,
+      "step": 4980
+    },
+    {
+      "epoch": 63.57,
+      "grad_norm": 3.6270945072174072,
+      "learning_rate": 2.0014245014245016e-05,
+      "loss": 0.1299,
+      "step": 4990
+    },
+    {
+      "epoch": 63.69,
+      "grad_norm": 3.3796772956848145,
+      "learning_rate": 1.9943019943019945e-05,
+      "loss": 0.1328,
+      "step": 5000
+    },
+    {
+      "epoch": 63.82,
+      "grad_norm": 2.5864744186401367,
+      "learning_rate": 1.987179487179487e-05,
+      "loss": 0.1214,
+      "step": 5010
+    },
+    {
+      "epoch": 63.95,
+      "grad_norm": 3.8097081184387207,
+      "learning_rate": 1.9800569800569802e-05,
+      "loss": 0.1218,
+      "step": 5020
+    },
+    {
+      "epoch": 64.0,
+      "eval_accuracy": 0.9771300448430493,
+      "eval_loss": 0.06802598387002945,
+      "eval_runtime": 10.3686,
+      "eval_samples_per_second": 430.144,
+      "eval_steps_per_second": 3.376,
+      "step": 5024
+    },
+    {
+      "epoch": 64.08,
+      "grad_norm": 4.177213668823242,
+      "learning_rate": 1.972934472934473e-05,
+      "loss": 0.1228,
+      "step": 5030
+    },
+    {
+      "epoch": 64.2,
+      "grad_norm": 2.647430419921875,
+      "learning_rate": 1.965811965811966e-05,
+      "loss": 0.1301,
+      "step": 5040
+    },
+    {
+      "epoch": 64.33,
+      "grad_norm": 2.9064316749572754,
+      "learning_rate": 1.9586894586894588e-05,
+      "loss": 0.1324,
+      "step": 5050
+    },
+    {
+      "epoch": 64.46,
+      "grad_norm": 5.072793006896973,
+      "learning_rate": 1.9515669515669517e-05,
+      "loss": 0.1187,
+      "step": 5060
+    },
+    {
+      "epoch": 64.59,
+      "grad_norm": 3.9644272327423096,
+      "learning_rate": 1.9444444444444445e-05,
+      "loss": 0.1193,
+      "step": 5070
+    },
+    {
+      "epoch": 64.71,
+      "grad_norm": 3.9974265098571777,
+      "learning_rate": 1.9373219373219374e-05,
+      "loss": 0.1102,
+      "step": 5080
+    },
+    {
+      "epoch": 64.84,
+      "grad_norm": 4.116329669952393,
+      "learning_rate": 1.9301994301994303e-05,
+      "loss": 0.1259,
+      "step": 5090
+    },
+    {
+      "epoch": 64.97,
+      "grad_norm": 3.9444773197174072,
+      "learning_rate": 1.923076923076923e-05,
+      "loss": 0.1193,
+      "step": 5100
+    },
+    {
+      "epoch": 64.99,
+      "eval_accuracy": 0.9760089686098655,
+      "eval_loss": 0.06792016327381134,
+      "eval_runtime": 10.2369,
+      "eval_samples_per_second": 435.679,
+      "eval_steps_per_second": 3.419,
+      "step": 5102
+    },
+    {
+      "epoch": 65.1,
+      "grad_norm": 3.6853489875793457,
+      "learning_rate": 1.915954415954416e-05,
+      "loss": 0.1193,
+      "step": 5110
+    },
+    {
+      "epoch": 65.22,
+      "grad_norm": 3.1782078742980957,
+      "learning_rate": 1.908831908831909e-05,
+      "loss": 0.1237,
+      "step": 5120
+    },
+    {
+      "epoch": 65.35,
+      "grad_norm": 2.1836137771606445,
+      "learning_rate": 1.9017094017094017e-05,
+      "loss": 0.1157,
+      "step": 5130
+    },
+    {
+      "epoch": 65.48,
+      "grad_norm": 3.0715341567993164,
+      "learning_rate": 1.8945868945868946e-05,
+      "loss": 0.1252,
+      "step": 5140
+    },
+    {
+      "epoch": 65.61,
+      "grad_norm": 4.32092809677124,
+      "learning_rate": 1.8874643874643875e-05,
+      "loss": 0.1203,
+      "step": 5150
+    },
+    {
+      "epoch": 65.73,
+      "grad_norm": 2.6298437118530273,
+      "learning_rate": 1.8803418803418804e-05,
+      "loss": 0.1175,
+      "step": 5160
+    },
+    {
+      "epoch": 65.86,
+      "grad_norm": 3.1503069400787354,
+      "learning_rate": 1.8732193732193736e-05,
+      "loss": 0.1277,
+      "step": 5170
+    },
+    {
+      "epoch": 65.99,
+      "grad_norm": 2.627164363861084,
+      "learning_rate": 1.866096866096866e-05,
+      "loss": 0.1148,
+      "step": 5180
+    },
+    {
+      "epoch": 66.0,
+      "eval_accuracy": 0.9773542600896861,
+      "eval_loss": 0.0685017853975296,
+      "eval_runtime": 10.4375,
+      "eval_samples_per_second": 427.307,
+      "eval_steps_per_second": 3.353,
+      "step": 5181
+    },
+    {
+      "epoch": 66.11,
+      "grad_norm": 3.094813346862793,
+      "learning_rate": 1.858974358974359e-05,
+      "loss": 0.1221,
+      "step": 5190
+    },
+    {
+      "epoch": 66.24,
+      "grad_norm": 2.4847347736358643,
+      "learning_rate": 1.8518518518518518e-05,
+      "loss": 0.1395,
+      "step": 5200
+    },
+    {
+      "epoch": 66.37,
+      "grad_norm": 2.964954137802124,
+      "learning_rate": 1.8447293447293447e-05,
+      "loss": 0.1105,
+      "step": 5210
+    },
+    {
+      "epoch": 66.5,
+      "grad_norm": 4.3417439460754395,
+      "learning_rate": 1.837606837606838e-05,
+      "loss": 0.1168,
+      "step": 5220
+    },
+    {
+      "epoch": 66.62,
+      "grad_norm": 3.419844388961792,
+      "learning_rate": 1.8304843304843304e-05,
+      "loss": 0.1181,
+      "step": 5230
+    },
+    {
+      "epoch": 66.75,
+      "grad_norm": 3.0345187187194824,
+      "learning_rate": 1.8233618233618236e-05,
+      "loss": 0.1174,
+      "step": 5240
+    },
+    {
+      "epoch": 66.88,
+      "grad_norm": 2.317073345184326,
+      "learning_rate": 1.8162393162393162e-05,
+      "loss": 0.1242,
+      "step": 5250
+    },
+    {
+      "epoch": 66.99,
+      "eval_accuracy": 0.9800448430493274,
+      "eval_loss": 0.06472171097993851,
+      "eval_runtime": 11.3805,
+      "eval_samples_per_second": 391.9,
+      "eval_steps_per_second": 3.075,
+      "step": 5259
+    },
+    {
+      "epoch": 67.01,
+      "grad_norm": 2.867168664932251,
+      "learning_rate": 1.8091168091168094e-05,
+      "loss": 0.1135,
+      "step": 5260
+    },
+    {
+      "epoch": 67.13,
+      "grad_norm": 3.222736358642578,
+      "learning_rate": 1.801994301994302e-05,
+      "loss": 0.1148,
+      "step": 5270
+    },
+    {
+      "epoch": 67.26,
+      "grad_norm": 2.9340291023254395,
+      "learning_rate": 1.794871794871795e-05,
+      "loss": 0.1194,
+      "step": 5280
+    },
+    {
+      "epoch": 67.39,
+      "grad_norm": 3.0920584201812744,
+      "learning_rate": 1.787749287749288e-05,
+      "loss": 0.1256,
+      "step": 5290
+    },
+    {
+      "epoch": 67.52,
+      "grad_norm": 3.2834219932556152,
+      "learning_rate": 1.7806267806267805e-05,
+      "loss": 0.1177,
+      "step": 5300
+    },
+    {
+      "epoch": 67.64,
+      "grad_norm": 2.8663060665130615,
+      "learning_rate": 1.7735042735042737e-05,
+      "loss": 0.1135,
+      "step": 5310
+    },
+    {
+      "epoch": 67.77,
+      "grad_norm": 2.839386224746704,
+      "learning_rate": 1.7663817663817662e-05,
+      "loss": 0.1273,
+      "step": 5320
+    },
+    {
+      "epoch": 67.9,
+      "grad_norm": 3.285187005996704,
+      "learning_rate": 1.7592592592592595e-05,
+      "loss": 0.1167,
+      "step": 5330
+    },
+    {
+      "epoch": 68.0,
+      "eval_accuracy": 0.9782511210762331,
+      "eval_loss": 0.06463192403316498,
+      "eval_runtime": 11.1553,
+      "eval_samples_per_second": 399.809,
+      "eval_steps_per_second": 3.138,
+      "step": 5338
+    },
+    {
+      "epoch": 68.03,
+      "grad_norm": 3.3999927043914795,
+      "learning_rate": 1.752136752136752e-05,
+      "loss": 0.1185,
+      "step": 5340
+    },
+    {
+      "epoch": 68.15,
+      "grad_norm": 3.005258798599243,
+      "learning_rate": 1.7450142450142452e-05,
+      "loss": 0.1158,
+      "step": 5350
+    },
+    {
+      "epoch": 68.28,
+      "grad_norm": 2.6332740783691406,
+      "learning_rate": 1.737891737891738e-05,
+      "loss": 0.1251,
+      "step": 5360
+    },
+    {
+      "epoch": 68.41,
+      "grad_norm": 3.3177764415740967,
+      "learning_rate": 1.730769230769231e-05,
+      "loss": 0.1149,
+      "step": 5370
+    },
+    {
+      "epoch": 68.54,
+      "grad_norm": 2.7641139030456543,
+      "learning_rate": 1.7236467236467238e-05,
+      "loss": 0.1051,
+      "step": 5380
+    },
+    {
+      "epoch": 68.66,
+      "grad_norm": 2.804523468017578,
+      "learning_rate": 1.7165242165242167e-05,
+      "loss": 0.1142,
+      "step": 5390
+    },
+    {
+      "epoch": 68.79,
+      "grad_norm": 2.723928928375244,
+      "learning_rate": 1.7094017094017095e-05,
+      "loss": 0.1225,
+      "step": 5400
+    },
+    {
+      "epoch": 68.92,
+      "grad_norm": 3.367866039276123,
+      "learning_rate": 1.702279202279202e-05,
+      "loss": 0.117,
+      "step": 5410
+    },
+    {
+      "epoch": 68.99,
+      "eval_accuracy": 0.976457399103139,
+      "eval_loss": 0.07633925974369049,
+      "eval_runtime": 10.4599,
+      "eval_samples_per_second": 426.391,
+      "eval_steps_per_second": 3.346,
+      "step": 5416
+    },
+    {
+      "epoch": 69.04,
+      "grad_norm": 3.8155672550201416,
+      "learning_rate": 1.6951566951566953e-05,
+      "loss": 0.1192,
+      "step": 5420
+    },
+    {
+      "epoch": 69.17,
+      "grad_norm": 3.2433383464813232,
+      "learning_rate": 1.688034188034188e-05,
+      "loss": 0.124,
+      "step": 5430
+    },
+    {
+      "epoch": 69.3,
+      "grad_norm": 3.6944828033447266,
+      "learning_rate": 1.680911680911681e-05,
+      "loss": 0.1164,
+      "step": 5440
+    },
+    {
+      "epoch": 69.43,
+      "grad_norm": 3.125169515609741,
+      "learning_rate": 1.673789173789174e-05,
+      "loss": 0.1271,
+      "step": 5450
+    },
+    {
+      "epoch": 69.55,
+      "grad_norm": 3.082037925720215,
+      "learning_rate": 1.6666666666666667e-05,
+      "loss": 0.1121,
+      "step": 5460
+    },
+    {
+      "epoch": 69.68,
+      "grad_norm": 2.831787586212158,
+      "learning_rate": 1.6595441595441596e-05,
+      "loss": 0.113,
+      "step": 5470
+    },
+    {
+      "epoch": 69.81,
+      "grad_norm": 4.720918655395508,
+      "learning_rate": 1.6524216524216525e-05,
+      "loss": 0.1211,
+      "step": 5480
+    },
+    {
+      "epoch": 69.94,
+      "grad_norm": 3.05661940574646,
+      "learning_rate": 1.6452991452991454e-05,
+      "loss": 0.1153,
+      "step": 5490
+    },
+    {
+      "epoch": 70.0,
+      "eval_accuracy": 0.9753363228699552,
+      "eval_loss": 0.07196246832609177,
+      "eval_runtime": 10.3439,
+      "eval_samples_per_second": 431.173,
+      "eval_steps_per_second": 3.384,
+      "step": 5495
+    },
+    {
+      "epoch": 70.06,
+      "grad_norm": 2.578516960144043,
+      "learning_rate": 1.6381766381766382e-05,
+      "loss": 0.1139,
+      "step": 5500
+    },
+    {
+      "epoch": 70.19,
+      "grad_norm": 3.541548013687134,
+      "learning_rate": 1.631054131054131e-05,
+      "loss": 0.1125,
+      "step": 5510
+    },
+    {
+      "epoch": 70.32,
+      "grad_norm": 2.709437847137451,
+      "learning_rate": 1.623931623931624e-05,
+      "loss": 0.1264,
+      "step": 5520
+    },
+    {
+      "epoch": 70.45,
+      "grad_norm": 1.947713851928711,
+      "learning_rate": 1.6168091168091168e-05,
+      "loss": 0.1165,
+      "step": 5530
+    },
+    {
+      "epoch": 70.57,
+      "grad_norm": 2.832228899002075,
+      "learning_rate": 1.6096866096866097e-05,
+      "loss": 0.1242,
+      "step": 5540
+    },
+    {
+      "epoch": 70.7,
+      "grad_norm": 1.871283769607544,
+      "learning_rate": 1.602564102564103e-05,
+      "loss": 0.1175,
+      "step": 5550
+    },
+    {
+      "epoch": 70.83,
+      "grad_norm": 4.204667568206787,
+      "learning_rate": 1.5954415954415954e-05,
+      "loss": 0.1097,
+      "step": 5560
+    },
+    {
+      "epoch": 70.96,
+      "grad_norm": 4.079117298126221,
+      "learning_rate": 1.5883190883190886e-05,
+      "loss": 0.12,
+      "step": 5570
+    },
+    {
+      "epoch": 70.99,
+      "eval_accuracy": 0.9771300448430493,
+      "eval_loss": 0.07174661755561829,
+      "eval_runtime": 10.2885,
+      "eval_samples_per_second": 433.494,
+      "eval_steps_per_second": 3.402,
+      "step": 5573
+    },
+    {
+      "epoch": 71.08,
+      "grad_norm": 2.7620456218719482,
+      "learning_rate": 1.581196581196581e-05,
+      "loss": 0.1112,
+      "step": 5580
+    },
+    {
+      "epoch": 71.21,
+      "grad_norm": 2.8281893730163574,
+      "learning_rate": 1.574074074074074e-05,
+      "loss": 0.1207,
+      "step": 5590
+    },
+    {
+      "epoch": 71.34,
+      "grad_norm": 3.6722238063812256,
+      "learning_rate": 1.566951566951567e-05,
+      "loss": 0.1306,
+      "step": 5600
+    },
+    {
+      "epoch": 71.46,
+      "grad_norm": 3.395185708999634,
+      "learning_rate": 1.5598290598290598e-05,
+      "loss": 0.1093,
+      "step": 5610
+    },
+    {
+      "epoch": 71.59,
+      "grad_norm": 1.971070647239685,
+      "learning_rate": 1.552706552706553e-05,
+      "loss": 0.1136,
+      "step": 5620
+    },
+    {
+      "epoch": 71.72,
+      "grad_norm": 2.9130399227142334,
+      "learning_rate": 1.5455840455840455e-05,
+      "loss": 0.1003,
+      "step": 5630
+    },
+    {
+      "epoch": 71.85,
+      "grad_norm": 2.5622756481170654,
+      "learning_rate": 1.5384615384615387e-05,
+      "loss": 0.1079,
+      "step": 5640
+    },
+    {
+      "epoch": 71.97,
+      "grad_norm": 3.1971945762634277,
+      "learning_rate": 1.5313390313390312e-05,
+      "loss": 0.1054,
+      "step": 5650
+    },
+    {
+      "epoch": 72.0,
+      "eval_accuracy": 0.9766816143497757,
+      "eval_loss": 0.067739337682724,
+      "eval_runtime": 10.4674,
+      "eval_samples_per_second": 426.084,
+      "eval_steps_per_second": 3.344,
+      "step": 5652
+    },
+    {
+      "epoch": 72.1,
+      "grad_norm": 2.922377109527588,
+      "learning_rate": 1.5242165242165243e-05,
+      "loss": 0.1139,
+      "step": 5660
+    },
+    {
+      "epoch": 72.23,
+      "grad_norm": 2.4021689891815186,
+      "learning_rate": 1.517094017094017e-05,
+      "loss": 0.108,
+      "step": 5670
+    },
+    {
+      "epoch": 72.36,
+      "grad_norm": 2.4300858974456787,
+      "learning_rate": 1.50997150997151e-05,
+      "loss": 0.0996,
+      "step": 5680
+    },
+    {
+      "epoch": 72.48,
+      "grad_norm": 2.820833206176758,
+      "learning_rate": 1.502849002849003e-05,
+      "loss": 0.1135,
+      "step": 5690
+    },
+    {
+      "epoch": 72.61,
+      "grad_norm": 3.1016552448272705,
+      "learning_rate": 1.4957264957264958e-05,
+      "loss": 0.1242,
+      "step": 5700
+    },
+    {
+      "epoch": 72.74,
+      "grad_norm": 3.2477970123291016,
+      "learning_rate": 1.4886039886039888e-05,
+      "loss": 0.1238,
+      "step": 5710
+    },
+    {
+      "epoch": 72.87,
+      "grad_norm": 3.47007155418396,
+      "learning_rate": 1.4814814814814815e-05,
+      "loss": 0.1118,
+      "step": 5720
+    },
+    {
+      "epoch": 72.99,
+      "grad_norm": 4.1029534339904785,
+      "learning_rate": 1.4743589743589745e-05,
+      "loss": 0.1183,
+      "step": 5730
+    },
+    {
+      "epoch": 72.99,
+      "eval_accuracy": 0.9755605381165919,
+      "eval_loss": 0.07409787178039551,
+      "eval_runtime": 10.2462,
+      "eval_samples_per_second": 435.283,
+      "eval_steps_per_second": 3.416,
+      "step": 5730
+    },
+    {
+      "epoch": 73.12,
+      "grad_norm": 2.9067156314849854,
+      "learning_rate": 1.4672364672364672e-05,
+      "loss": 0.1128,
+      "step": 5740
+    },
+    {
+      "epoch": 73.25,
+      "grad_norm": 2.93924880027771,
+      "learning_rate": 1.4601139601139601e-05,
+      "loss": 0.1098,
+      "step": 5750
+    },
+    {
+      "epoch": 73.38,
+      "grad_norm": 2.739440679550171,
+      "learning_rate": 1.4529914529914531e-05,
+      "loss": 0.1261,
+      "step": 5760
+    },
+    {
+      "epoch": 73.5,
+      "grad_norm": 2.8033335208892822,
+      "learning_rate": 1.4458689458689458e-05,
+      "loss": 0.1106,
+      "step": 5770
+    },
+    {
+      "epoch": 73.63,
+      "grad_norm": 2.91092586517334,
+      "learning_rate": 1.4387464387464389e-05,
+      "loss": 0.1042,
+      "step": 5780
+    },
+    {
+      "epoch": 73.76,
+      "grad_norm": 3.8227293491363525,
+      "learning_rate": 1.4316239316239316e-05,
+      "loss": 0.1074,
+      "step": 5790
+    },
+    {
+      "epoch": 73.89,
+      "grad_norm": 2.5820155143737793,
+      "learning_rate": 1.4245014245014246e-05,
+      "loss": 0.1082,
+      "step": 5800
+    },
+    {
+      "epoch": 74.0,
+      "eval_accuracy": 0.9786995515695067,
+      "eval_loss": 0.06757389008998871,
+      "eval_runtime": 11.2378,
+      "eval_samples_per_second": 396.874,
+      "eval_steps_per_second": 3.114,
+      "step": 5809
+    },
+    {
+      "epoch": 74.01,
+      "grad_norm": 3.325789451599121,
+      "learning_rate": 1.4173789173789173e-05,
+      "loss": 0.1094,
+      "step": 5810
+    },
+    {
+      "epoch": 74.14,
+      "grad_norm": 4.414618968963623,
+      "learning_rate": 1.4102564102564104e-05,
+      "loss": 0.1132,
+      "step": 5820
+    },
+    {
+      "epoch": 74.27,
+      "grad_norm": 3.79183292388916,
+      "learning_rate": 1.4031339031339034e-05,
+      "loss": 0.1049,
+      "step": 5830
+    },
+    {
+      "epoch": 74.39,
+      "grad_norm": 3.2433505058288574,
+      "learning_rate": 1.3960113960113961e-05,
+      "loss": 0.106,
+      "step": 5840
+    },
+    {
+      "epoch": 74.52,
+      "grad_norm": 2.762228488922119,
+      "learning_rate": 1.388888888888889e-05,
+      "loss": 0.1164,
+      "step": 5850
+    },
+    {
+      "epoch": 74.65,
+      "grad_norm": 3.4645419120788574,
+      "learning_rate": 1.3817663817663817e-05,
+      "loss": 0.1207,
+      "step": 5860
+    },
+    {
+      "epoch": 74.78,
+      "grad_norm": 2.7387356758117676,
+      "learning_rate": 1.3746438746438747e-05,
+      "loss": 0.117,
+      "step": 5870
+    },
+    {
+      "epoch": 74.9,
+      "grad_norm": 3.4120523929595947,
+      "learning_rate": 1.3675213675213677e-05,
+      "loss": 0.1088,
+      "step": 5880
+    },
+    {
+      "epoch": 74.99,
+      "eval_accuracy": 0.9751121076233183,
+      "eval_loss": 0.06995895504951477,
+      "eval_runtime": 11.4177,
+      "eval_samples_per_second": 390.621,
+      "eval_steps_per_second": 3.065,
+      "step": 5887
+    },
+    {
+      "epoch": 75.03,
+      "grad_norm": 2.275909662246704,
+      "learning_rate": 1.3603988603988604e-05,
+      "loss": 0.1245,
+      "step": 5890
+    },
+    {
+      "epoch": 75.16,
+      "grad_norm": 2.343656539916992,
+      "learning_rate": 1.3532763532763535e-05,
+      "loss": 0.1044,
+      "step": 5900
+    },
+    {
+      "epoch": 75.29,
+      "grad_norm": 4.431203842163086,
+      "learning_rate": 1.3461538461538462e-05,
+      "loss": 0.1126,
+      "step": 5910
+    },
+    {
+      "epoch": 75.41,
+      "grad_norm": 2.1456689834594727,
+      "learning_rate": 1.3390313390313392e-05,
+      "loss": 0.1094,
+      "step": 5920
+    },
+    {
+      "epoch": 75.54,
+      "grad_norm": 2.2616610527038574,
+      "learning_rate": 1.3319088319088319e-05,
+      "loss": 0.1117,
+      "step": 5930
+    },
+    {
+      "epoch": 75.67,
+      "grad_norm": 3.3451130390167236,
+      "learning_rate": 1.324786324786325e-05,
+      "loss": 0.1229,
+      "step": 5940
+    },
+    {
+      "epoch": 75.8,
+      "grad_norm": 3.578305721282959,
+      "learning_rate": 1.3176638176638178e-05,
+      "loss": 0.1133,
+      "step": 5950
+    },
+    {
+      "epoch": 75.92,
+      "grad_norm": 2.3085949420928955,
+      "learning_rate": 1.3105413105413105e-05,
+      "loss": 0.1125,
+      "step": 5960
+    },
+    {
+      "epoch": 76.0,
+      "eval_accuracy": 0.97847533632287,
+      "eval_loss": 0.0663037970662117,
+      "eval_runtime": 10.5423,
+      "eval_samples_per_second": 423.057,
+      "eval_steps_per_second": 3.32,
+      "step": 5966
+    },
+    {
+      "epoch": 76.05,
+      "grad_norm": 3.370089530944824,
+      "learning_rate": 1.3034188034188035e-05,
+      "loss": 0.1103,
+      "step": 5970
+    },
+    {
+      "epoch": 76.18,
+      "grad_norm": 3.2268216609954834,
+      "learning_rate": 1.2962962962962962e-05,
+      "loss": 0.1133,
+      "step": 5980
+    },
+    {
+      "epoch": 76.31,
+      "grad_norm": 2.5671768188476562,
+      "learning_rate": 1.2891737891737893e-05,
+      "loss": 0.1083,
+      "step": 5990
+    },
+    {
+      "epoch": 76.43,
+      "grad_norm": 3.591700553894043,
+      "learning_rate": 1.282051282051282e-05,
+      "loss": 0.1078,
+      "step": 6000
+    },
+    {
+      "epoch": 76.56,
+      "grad_norm": 3.825608968734741,
+      "learning_rate": 1.274928774928775e-05,
+      "loss": 0.1116,
+      "step": 6010
+    },
+    {
+      "epoch": 76.69,
+      "grad_norm": 3.0779523849487305,
+      "learning_rate": 1.267806267806268e-05,
+      "loss": 0.1174,
+      "step": 6020
+    },
+    {
+      "epoch": 76.82,
+      "grad_norm": 2.5936954021453857,
+      "learning_rate": 1.2606837606837608e-05,
+      "loss": 0.1062,
+      "step": 6030
+    },
+    {
+      "epoch": 76.94,
+      "grad_norm": 2.4793176651000977,
+      "learning_rate": 1.2535612535612536e-05,
+      "loss": 0.1099,
+      "step": 6040
+    },
+    {
+      "epoch": 76.99,
+      "eval_accuracy": 0.9789237668161435,
+      "eval_loss": 0.0621563158929348,
+      "eval_runtime": 10.8775,
+      "eval_samples_per_second": 410.02,
+      "eval_steps_per_second": 3.218,
+      "step": 6044
+    },
+    {
+      "epoch": 77.07,
+      "grad_norm": 2.338437557220459,
+      "learning_rate": 1.2464387464387465e-05,
+      "loss": 0.1113,
+      "step": 6050
+    },
+    {
+      "epoch": 77.2,
+      "grad_norm": 3.1221530437469482,
+      "learning_rate": 1.2393162393162394e-05,
+      "loss": 0.1088,
+      "step": 6060
+    },
+    {
+      "epoch": 77.32,
+      "grad_norm": 4.532721519470215,
+      "learning_rate": 1.2321937321937322e-05,
+      "loss": 0.1117,
+      "step": 6070
+    },
+    {
+      "epoch": 77.45,
+      "grad_norm": 4.488461494445801,
+      "learning_rate": 1.2250712250712251e-05,
+      "loss": 0.1079,
+      "step": 6080
+    },
+    {
+      "epoch": 77.58,
+      "grad_norm": 2.4823977947235107,
+      "learning_rate": 1.217948717948718e-05,
+      "loss": 0.1161,
+      "step": 6090
+    },
+    {
+      "epoch": 77.71,
+      "grad_norm": 3.6858134269714355,
+      "learning_rate": 1.2108262108262108e-05,
+      "loss": 0.1035,
+      "step": 6100
+    },
+    {
+      "epoch": 77.83,
+      "grad_norm": 2.647006034851074,
+      "learning_rate": 1.2037037037037037e-05,
+      "loss": 0.1158,
+      "step": 6110
+    },
+    {
+      "epoch": 77.96,
+      "grad_norm": 2.6735477447509766,
+      "learning_rate": 1.1965811965811967e-05,
+      "loss": 0.1128,
+      "step": 6120
+    },
+    {
+      "epoch": 78.0,
+      "eval_accuracy": 0.979372197309417,
+      "eval_loss": 0.06598477810621262,
+      "eval_runtime": 10.8917,
+      "eval_samples_per_second": 409.486,
+      "eval_steps_per_second": 3.213,
+      "step": 6123
+    },
+    {
+      "epoch": 78.09,
+      "grad_norm": 2.4398152828216553,
+      "learning_rate": 1.1894586894586896e-05,
+      "loss": 0.1175,
+      "step": 6130
+    },
+    {
+      "epoch": 78.22,
+      "grad_norm": 2.345200538635254,
+      "learning_rate": 1.1823361823361825e-05,
+      "loss": 0.1119,
+      "step": 6140
+    },
+    {
+      "epoch": 78.34,
+      "grad_norm": 2.4092376232147217,
+      "learning_rate": 1.1752136752136752e-05,
+      "loss": 0.1166,
+      "step": 6150
+    },
+    {
+      "epoch": 78.47,
+      "grad_norm": 2.720093011856079,
+      "learning_rate": 1.168091168091168e-05,
+      "loss": 0.1072,
+      "step": 6160
+    },
+    {
+      "epoch": 78.6,
+      "grad_norm": 3.7120041847229004,
+      "learning_rate": 1.160968660968661e-05,
+      "loss": 0.1047,
+      "step": 6170
+    },
+    {
+      "epoch": 78.73,
+      "grad_norm": 2.8657994270324707,
+      "learning_rate": 1.153846153846154e-05,
+      "loss": 0.11,
+      "step": 6180
+    },
+    {
+      "epoch": 78.85,
+      "grad_norm": 4.595070838928223,
+      "learning_rate": 1.1467236467236468e-05,
+      "loss": 0.106,
+      "step": 6190
+    },
+    {
+      "epoch": 78.98,
+      "grad_norm": 3.3609421253204346,
+      "learning_rate": 1.1396011396011397e-05,
+      "loss": 0.1178,
+      "step": 6200
+    },
+    {
+      "epoch": 78.99,
+      "eval_accuracy": 0.9780269058295964,
+      "eval_loss": 0.06985889375209808,
+      "eval_runtime": 10.464,
+      "eval_samples_per_second": 426.224,
+      "eval_steps_per_second": 3.345,
+      "step": 6201
+    },
+    {
+      "epoch": 79.11,
+      "grad_norm": 3.0199007987976074,
+      "learning_rate": 1.1324786324786326e-05,
+      "loss": 0.1126,
+      "step": 6210
+    },
+    {
+      "epoch": 79.24,
+      "grad_norm": 2.6402392387390137,
+      "learning_rate": 1.1253561253561254e-05,
+      "loss": 0.1042,
+      "step": 6220
+    },
+    {
+      "epoch": 79.36,
+      "grad_norm": 3.8217875957489014,
+      "learning_rate": 1.1182336182336183e-05,
+      "loss": 0.1007,
+      "step": 6230
+    },
+    {
+      "epoch": 79.49,
+      "grad_norm": 3.7943570613861084,
+      "learning_rate": 1.1111111111111112e-05,
+      "loss": 0.1115,
+      "step": 6240
+    },
+    {
+      "epoch": 79.62,
+      "grad_norm": 3.9829814434051514,
+      "learning_rate": 1.103988603988604e-05,
+      "loss": 0.0956,
+      "step": 6250
+    },
+    {
+      "epoch": 79.75,
+      "grad_norm": 3.254031181335449,
+      "learning_rate": 1.0968660968660969e-05,
+      "loss": 0.1088,
+      "step": 6260
+    },
+    {
+      "epoch": 79.87,
+      "grad_norm": 3.34173583984375,
+      "learning_rate": 1.0897435897435898e-05,
+      "loss": 0.1117,
+      "step": 6270
+    },
+    {
+      "epoch": 80.0,
+      "grad_norm": 3.977674961090088,
+      "learning_rate": 1.0826210826210826e-05,
+      "loss": 0.1129,
+      "step": 6280
+    },
+    {
+      "epoch": 80.0,
+      "eval_accuracy": 0.979372197309417,
+      "eval_loss": 0.0604555681347847,
+      "eval_runtime": 10.4032,
+      "eval_samples_per_second": 428.716,
+      "eval_steps_per_second": 3.364,
+      "step": 6280
+    },
+    {
+      "epoch": 80.13,
+      "grad_norm": 6.222870349884033,
+      "learning_rate": 1.0754985754985755e-05,
+      "loss": 0.0968,
+      "step": 6290
+    },
+    {
+      "epoch": 80.25,
+      "grad_norm": 3.2638728618621826,
+      "learning_rate": 1.0683760683760684e-05,
+      "loss": 0.1026,
+      "step": 6300
+    },
+    {
+      "epoch": 80.38,
+      "grad_norm": 4.064828872680664,
+      "learning_rate": 1.0612535612535614e-05,
+      "loss": 0.1073,
+      "step": 6310
+    },
+    {
+      "epoch": 80.51,
+      "grad_norm": 3.2518603801727295,
+      "learning_rate": 1.0541310541310543e-05,
+      "loss": 0.1039,
+      "step": 6320
+    },
+    {
+      "epoch": 80.64,
+      "grad_norm": 4.303738117218018,
+      "learning_rate": 1.0470085470085471e-05,
+      "loss": 0.1145,
+      "step": 6330
+    },
+    {
+      "epoch": 80.76,
+      "grad_norm": 3.027240514755249,
+      "learning_rate": 1.03988603988604e-05,
+      "loss": 0.1142,
+      "step": 6340
+    },
+    {
+      "epoch": 80.89,
+      "grad_norm": 6.462448596954346,
+      "learning_rate": 1.0327635327635327e-05,
+      "loss": 0.1013,
+      "step": 6350
+    },
+    {
+      "epoch": 80.99,
+      "eval_accuracy": 0.9778026905829597,
+      "eval_loss": 0.06936540454626083,
+      "eval_runtime": 10.4243,
+      "eval_samples_per_second": 427.848,
+      "eval_steps_per_second": 3.358,
+      "step": 6358
+    },
+    {
+      "epoch": 81.02,
+      "grad_norm": 3.1497411727905273,
+      "learning_rate": 1.0256410256410256e-05,
+      "loss": 0.0973,
+      "step": 6360
+    },
+    {
+      "epoch": 81.15,
+      "grad_norm": 2.9892778396606445,
+      "learning_rate": 1.0185185185185185e-05,
+      "loss": 0.1011,
+      "step": 6370
+    },
+    {
+      "epoch": 81.27,
+      "grad_norm": 3.1823902130126953,
+      "learning_rate": 1.0113960113960115e-05,
+      "loss": 0.1039,
+      "step": 6380
+    },
+    {
+      "epoch": 81.4,
+      "grad_norm": 2.418894052505493,
+      "learning_rate": 1.0042735042735044e-05,
+      "loss": 0.1163,
+      "step": 6390
+    },
+    {
+      "epoch": 81.53,
+      "grad_norm": 2.359311580657959,
+      "learning_rate": 9.971509971509972e-06,
+      "loss": 0.106,
+      "step": 6400
+    },
+    {
+      "epoch": 81.66,
+      "grad_norm": 4.2471604347229,
+      "learning_rate": 9.900284900284901e-06,
+      "loss": 0.1007,
+      "step": 6410
+    },
+    {
+      "epoch": 81.78,
+      "grad_norm": 4.758132457733154,
+      "learning_rate": 9.82905982905983e-06,
+      "loss": 0.1024,
+      "step": 6420
+    },
+    {
+      "epoch": 81.91,
+      "grad_norm": 2.4870598316192627,
+      "learning_rate": 9.757834757834758e-06,
+      "loss": 0.1078,
+      "step": 6430
+    },
+    {
+      "epoch": 82.0,
+      "eval_accuracy": 0.9782511210762331,
+      "eval_loss": 0.06523539870977402,
+      "eval_runtime": 10.4302,
+      "eval_samples_per_second": 427.604,
+      "eval_steps_per_second": 3.356,
+      "step": 6437
+    },
+    {
+      "epoch": 82.04,
+      "grad_norm": 2.7343454360961914,
+      "learning_rate": 9.686609686609687e-06,
+      "loss": 0.1012,
+      "step": 6440
+    },
+    {
+      "epoch": 82.17,
+      "grad_norm": 3.698735475540161,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.1075,
+      "step": 6450
+    },
+    {
+      "epoch": 82.29,
+      "grad_norm": 3.699899673461914,
+      "learning_rate": 9.544159544159544e-06,
+      "loss": 0.1085,
+      "step": 6460
+    },
+    {
+      "epoch": 82.42,
+      "grad_norm": 2.9340295791625977,
+      "learning_rate": 9.472934472934473e-06,
+      "loss": 0.1013,
+      "step": 6470
+    },
+    {
+      "epoch": 82.55,
+      "grad_norm": 2.8767290115356445,
+      "learning_rate": 9.401709401709402e-06,
+      "loss": 0.0997,
+      "step": 6480
+    },
+    {
+      "epoch": 82.68,
+      "grad_norm": 4.756886959075928,
+      "learning_rate": 9.33048433048433e-06,
+      "loss": 0.104,
+      "step": 6490
+    },
+    {
+      "epoch": 82.8,
+      "grad_norm": 5.544289588928223,
+      "learning_rate": 9.259259259259259e-06,
+      "loss": 0.1067,
+      "step": 6500
+    },
+    {
+      "epoch": 82.93,
+      "grad_norm": 2.162623882293701,
+      "learning_rate": 9.18803418803419e-06,
+      "loss": 0.0994,
+      "step": 6510
+    },
+    {
+      "epoch": 82.99,
+      "eval_accuracy": 0.9811659192825112,
+      "eval_loss": 0.060371555387973785,
+      "eval_runtime": 10.7881,
+      "eval_samples_per_second": 413.42,
+      "eval_steps_per_second": 3.244,
+      "step": 6515
+    },
+    {
+      "epoch": 83.06,
+      "grad_norm": 3.7604353427886963,
+      "learning_rate": 9.116809116809118e-06,
+      "loss": 0.1025,
+      "step": 6520
+    },
+    {
+      "epoch": 83.18,
+      "grad_norm": 2.88057279586792,
+      "learning_rate": 9.045584045584047e-06,
+      "loss": 0.1064,
+      "step": 6530
+    },
+    {
+      "epoch": 83.31,
+      "grad_norm": 2.340789556503296,
+      "learning_rate": 8.974358974358976e-06,
+      "loss": 0.1113,
+      "step": 6540
+    },
+    {
+      "epoch": 83.44,
+      "grad_norm": 1.9274300336837769,
+      "learning_rate": 8.903133903133903e-06,
+      "loss": 0.0986,
+      "step": 6550
+    },
+    {
+      "epoch": 83.57,
+      "grad_norm": 3.982344627380371,
+      "learning_rate": 8.831908831908831e-06,
+      "loss": 0.1172,
+      "step": 6560
+    },
+    {
+      "epoch": 83.69,
+      "grad_norm": 3.080048084259033,
+      "learning_rate": 8.76068376068376e-06,
+      "loss": 0.1034,
+      "step": 6570
+    },
+    {
+      "epoch": 83.82,
+      "grad_norm": 2.989431619644165,
+      "learning_rate": 8.68945868945869e-06,
+      "loss": 0.1117,
+      "step": 6580
+    },
+    {
+      "epoch": 83.95,
+      "grad_norm": 2.0674595832824707,
+      "learning_rate": 8.618233618233619e-06,
+      "loss": 0.1093,
+      "step": 6590
+    },
+    {
+      "epoch": 84.0,
+      "eval_accuracy": 0.9804932735426009,
+      "eval_loss": 0.06000491976737976,
+      "eval_runtime": 10.3525,
+      "eval_samples_per_second": 430.812,
+      "eval_steps_per_second": 3.381,
+      "step": 6594
+    },
+    {
+      "epoch": 84.08,
+      "grad_norm": 2.6291697025299072,
+      "learning_rate": 8.547008547008548e-06,
+      "loss": 0.1044,
+      "step": 6600
+    },
+    {
+      "epoch": 84.2,
+      "grad_norm": 2.8807177543640137,
+      "learning_rate": 8.475783475783476e-06,
+      "loss": 0.1062,
+      "step": 6610
+    },
+    {
+      "epoch": 84.33,
+      "grad_norm": 2.1789872646331787,
+      "learning_rate": 8.404558404558405e-06,
+      "loss": 0.106,
+      "step": 6620
+    },
+    {
+      "epoch": 84.46,
+      "grad_norm": 3.824889898300171,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.1086,
+      "step": 6630
+    },
+    {
+      "epoch": 84.59,
+      "grad_norm": 3.5731046199798584,
+      "learning_rate": 8.262108262108262e-06,
+      "loss": 0.1015,
+      "step": 6640
+    },
+    {
+      "epoch": 84.71,
+      "grad_norm": 2.0508885383605957,
+      "learning_rate": 8.190883190883191e-06,
+      "loss": 0.1016,
+      "step": 6650
+    },
+    {
+      "epoch": 84.84,
+      "grad_norm": 3.047819137573242,
+      "learning_rate": 8.11965811965812e-06,
+      "loss": 0.1017,
+      "step": 6660
+    },
+    {
+      "epoch": 84.97,
+      "grad_norm": 3.115485191345215,
+      "learning_rate": 8.048433048433048e-06,
+      "loss": 0.1039,
+      "step": 6670
+    },
+    {
+      "epoch": 84.99,
+      "eval_accuracy": 0.9786995515695067,
+      "eval_loss": 0.06455076485872269,
+      "eval_runtime": 10.2681,
+      "eval_samples_per_second": 434.357,
+      "eval_steps_per_second": 3.409,
+      "step": 6672
+    },
+    {
+      "epoch": 85.1,
+      "grad_norm": 2.3923819065093994,
+      "learning_rate": 7.977207977207977e-06,
+      "loss": 0.1023,
+      "step": 6680
+    },
+    {
+      "epoch": 85.22,
+      "grad_norm": 4.972830772399902,
+      "learning_rate": 7.905982905982906e-06,
+      "loss": 0.1017,
+      "step": 6690
+    },
+    {
+      "epoch": 85.35,
+      "grad_norm": 3.3479678630828857,
+      "learning_rate": 7.834757834757835e-06,
+      "loss": 0.1057,
+      "step": 6700
+    },
+    {
+      "epoch": 85.48,
+      "grad_norm": 4.136251449584961,
+      "learning_rate": 7.763532763532765e-06,
+      "loss": 0.1159,
+      "step": 6710
+    },
+    {
+      "epoch": 85.61,
+      "grad_norm": 3.715238094329834,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.1068,
+      "step": 6720
+    },
+    {
+      "epoch": 85.73,
+      "grad_norm": 2.57712459564209,
+      "learning_rate": 7.6210826210826214e-06,
+      "loss": 0.1127,
+      "step": 6730
+    },
+    {
+      "epoch": 85.86,
+      "grad_norm": 2.153548240661621,
+      "learning_rate": 7.54985754985755e-06,
+      "loss": 0.108,
+      "step": 6740
+    },
+    {
+      "epoch": 85.99,
+      "grad_norm": 2.7407655715942383,
+      "learning_rate": 7.478632478632479e-06,
+      "loss": 0.0963,
+      "step": 6750
+    },
+    {
+      "epoch": 86.0,
+      "eval_accuracy": 0.9789237668161435,
+      "eval_loss": 0.06539863348007202,
+      "eval_runtime": 10.5875,
+      "eval_samples_per_second": 421.253,
+      "eval_steps_per_second": 3.306,
+      "step": 6751
+    },
+    {
+      "epoch": 86.11,
+      "grad_norm": 2.8693339824676514,
+      "learning_rate": 7.4074074074074075e-06,
+      "loss": 0.0998,
+      "step": 6760
+    },
+    {
+      "epoch": 86.24,
+      "grad_norm": 5.288963317871094,
+      "learning_rate": 7.336182336182336e-06,
+      "loss": 0.1129,
+      "step": 6770
+    },
+    {
+      "epoch": 86.37,
+      "grad_norm": 3.063286066055298,
+      "learning_rate": 7.264957264957266e-06,
+      "loss": 0.1013,
+      "step": 6780
+    },
+    {
+      "epoch": 86.5,
+      "grad_norm": 3.49326229095459,
+      "learning_rate": 7.193732193732194e-06,
+      "loss": 0.1135,
+      "step": 6790
+    },
+    {
+      "epoch": 86.62,
+      "grad_norm": 3.090437412261963,
+      "learning_rate": 7.122507122507123e-06,
+      "loss": 0.1067,
+      "step": 6800
+    },
+    {
+      "epoch": 86.75,
+      "grad_norm": 2.2138142585754395,
+      "learning_rate": 7.051282051282052e-06,
+      "loss": 0.0982,
+      "step": 6810
+    },
+    {
+      "epoch": 86.88,
+      "grad_norm": 2.791416883468628,
+      "learning_rate": 6.9800569800569804e-06,
+      "loss": 0.1053,
+      "step": 6820
+    },
+    {
+      "epoch": 86.99,
+      "eval_accuracy": 0.9802690582959641,
+      "eval_loss": 0.06266650557518005,
+      "eval_runtime": 10.4188,
+      "eval_samples_per_second": 428.073,
+      "eval_steps_per_second": 3.359,
+      "step": 6829
+    },
+    {
+      "epoch": 87.01,
+      "grad_norm": 2.712346076965332,
+      "learning_rate": 6.908831908831908e-06,
+      "loss": 0.1096,
+      "step": 6830
+    },
+    {
+      "epoch": 87.13,
+      "grad_norm": 3.091013193130493,
+      "learning_rate": 6.837606837606839e-06,
+      "loss": 0.1135,
+      "step": 6840
+    },
+    {
+      "epoch": 87.26,
+      "grad_norm": 2.5728962421417236,
+      "learning_rate": 6.766381766381767e-06,
+      "loss": 0.0967,
+      "step": 6850
+    },
+    {
+      "epoch": 87.39,
+      "grad_norm": 2.6085259914398193,
+      "learning_rate": 6.695156695156696e-06,
+      "loss": 0.103,
+      "step": 6860
+    },
+    {
+      "epoch": 87.52,
+      "grad_norm": 2.8301172256469727,
+      "learning_rate": 6.623931623931625e-06,
+      "loss": 0.1051,
+      "step": 6870
+    },
+    {
+      "epoch": 87.64,
+      "grad_norm": 3.9067742824554443,
+      "learning_rate": 6.5527065527065525e-06,
+      "loss": 0.1025,
+      "step": 6880
+    },
+    {
+      "epoch": 87.77,
+      "grad_norm": 2.3788182735443115,
+      "learning_rate": 6.481481481481481e-06,
+      "loss": 0.1077,
+      "step": 6890
+    },
+    {
+      "epoch": 87.9,
+      "grad_norm": 2.764779806137085,
+      "learning_rate": 6.41025641025641e-06,
+      "loss": 0.0982,
+      "step": 6900
+    },
+    {
+      "epoch": 88.0,
+      "eval_accuracy": 0.9800448430493274,
+      "eval_loss": 0.061907876282930374,
+      "eval_runtime": 10.4602,
+      "eval_samples_per_second": 426.379,
+      "eval_steps_per_second": 3.346,
+      "step": 6908
+    },
+    {
+      "epoch": 88.03,
+      "grad_norm": 3.2199273109436035,
+      "learning_rate": 6.33903133903134e-06,
+      "loss": 0.1019,
+      "step": 6910
+    },
+    {
+      "epoch": 88.15,
+      "grad_norm": 2.0769879817962646,
+      "learning_rate": 6.267806267806268e-06,
+      "loss": 0.1063,
+      "step": 6920
+    },
+    {
+      "epoch": 88.28,
+      "grad_norm": 2.188361406326294,
+      "learning_rate": 6.196581196581197e-06,
+      "loss": 0.0914,
+      "step": 6930
+    },
+    {
+      "epoch": 88.41,
+      "grad_norm": 2.7577333450317383,
+      "learning_rate": 6.1253561253561255e-06,
+      "loss": 0.103,
+      "step": 6940
+    },
+    {
+      "epoch": 88.54,
+      "grad_norm": 4.330448627471924,
+      "learning_rate": 6.054131054131054e-06,
+      "loss": 0.1075,
+      "step": 6950
+    },
+    {
+      "epoch": 88.66,
+      "grad_norm": 2.961698055267334,
+      "learning_rate": 5.982905982905984e-06,
+      "loss": 0.1003,
+      "step": 6960
+    },
+    {
+      "epoch": 88.79,
+      "grad_norm": 3.129741668701172,
+      "learning_rate": 5.911680911680912e-06,
+      "loss": 0.101,
+      "step": 6970
+    },
+    {
+      "epoch": 88.92,
+      "grad_norm": 4.230170726776123,
+      "learning_rate": 5.84045584045584e-06,
+      "loss": 0.0944,
+      "step": 6980
+    },
+    {
+      "epoch": 88.99,
+      "eval_accuracy": 0.9795964125560538,
+      "eval_loss": 0.060706257820129395,
+      "eval_runtime": 11.2714,
+      "eval_samples_per_second": 395.693,
+      "eval_steps_per_second": 3.105,
+      "step": 6986
+    },
+    {
+      "epoch": 89.04,
+      "grad_norm": 3.656252145767212,
+      "learning_rate": 5.76923076923077e-06,
+      "loss": 0.1078,
+      "step": 6990
+    },
+    {
+      "epoch": 89.17,
+      "grad_norm": 3.191498041152954,
+      "learning_rate": 5.6980056980056985e-06,
+      "loss": 0.0929,
+      "step": 7000
+    },
+    {
+      "epoch": 89.3,
+      "grad_norm": 2.7151176929473877,
+      "learning_rate": 5.626780626780627e-06,
+      "loss": 0.1012,
+      "step": 7010
+    },
+    {
+      "epoch": 89.43,
+      "grad_norm": 2.8162662982940674,
+      "learning_rate": 5.555555555555556e-06,
+      "loss": 0.1071,
+      "step": 7020
+    },
+    {
+      "epoch": 89.55,
+      "grad_norm": 3.192384719848633,
+      "learning_rate": 5.4843304843304845e-06,
+      "loss": 0.095,
+      "step": 7030
+    },
+    {
+      "epoch": 89.68,
+      "grad_norm": 2.0794003009796143,
+      "learning_rate": 5.413105413105413e-06,
+      "loss": 0.1047,
+      "step": 7040
+    },
+    {
+      "epoch": 89.81,
+      "grad_norm": 5.687547206878662,
+      "learning_rate": 5.341880341880342e-06,
+      "loss": 0.1027,
+      "step": 7050
+    },
+    {
+      "epoch": 89.94,
+      "grad_norm": 2.7075958251953125,
+      "learning_rate": 5.270655270655271e-06,
+      "loss": 0.0959,
+      "step": 7060
+    },
+    {
+      "epoch": 90.0,
+      "eval_accuracy": 0.9800448430493274,
+      "eval_loss": 0.06611481308937073,
+      "eval_runtime": 10.5465,
+      "eval_samples_per_second": 422.889,
+      "eval_steps_per_second": 3.319,
+      "step": 7065
+    },
+    {
+      "epoch": 90.06,
+      "grad_norm": 2.5442237854003906,
+      "learning_rate": 5.1994301994302e-06,
+      "loss": 0.1023,
+      "step": 7070
+    },
+    {
+      "epoch": 90.19,
+      "grad_norm": 2.451735019683838,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 0.0936,
+      "step": 7080
+    },
+    {
+      "epoch": 90.32,
+      "grad_norm": 3.2897391319274902,
+      "learning_rate": 5.0569800569800575e-06,
+      "loss": 0.1067,
+      "step": 7090
+    },
+    {
+      "epoch": 90.45,
+      "grad_norm": 2.906238079071045,
+      "learning_rate": 4.985754985754986e-06,
+      "loss": 0.0985,
+      "step": 7100
+    },
+    {
+      "epoch": 90.57,
+      "grad_norm": 3.004779100418091,
+      "learning_rate": 4.914529914529915e-06,
+      "loss": 0.0874,
+      "step": 7110
+    },
+    {
+      "epoch": 90.7,
+      "grad_norm": 2.9214158058166504,
+      "learning_rate": 4.8433048433048435e-06,
+      "loss": 0.1001,
+      "step": 7120
+    },
+    {
+      "epoch": 90.83,
+      "grad_norm": 5.096872329711914,
+      "learning_rate": 4.772079772079772e-06,
+      "loss": 0.0938,
+      "step": 7130
+    },
+    {
+      "epoch": 90.96,
+      "grad_norm": 2.0840651988983154,
+      "learning_rate": 4.700854700854701e-06,
+      "loss": 0.101,
+      "step": 7140
+    },
+    {
+      "epoch": 90.99,
+      "eval_accuracy": 0.9809417040358744,
+      "eval_loss": 0.0642046332359314,
+      "eval_runtime": 10.3719,
+      "eval_samples_per_second": 430.007,
+      "eval_steps_per_second": 3.374,
+      "step": 7143
+    },
+    {
+      "epoch": 91.08,
+      "grad_norm": 3.488114356994629,
+      "learning_rate": 4.6296296296296296e-06,
+      "loss": 0.0997,
+      "step": 7150
+    },
+    {
+      "epoch": 91.21,
+      "grad_norm": 2.822373390197754,
+      "learning_rate": 4.558404558404559e-06,
+      "loss": 0.1048,
+      "step": 7160
+    },
+    {
+      "epoch": 91.34,
+      "grad_norm": 4.034416198730469,
+      "learning_rate": 4.487179487179488e-06,
+      "loss": 0.1008,
+      "step": 7170
+    },
+    {
+      "epoch": 91.46,
+      "grad_norm": 2.5757200717926025,
+      "learning_rate": 4.415954415954416e-06,
+      "loss": 0.1077,
+      "step": 7180
+    },
+    {
+      "epoch": 91.59,
+      "grad_norm": 4.5958662033081055,
+      "learning_rate": 4.344729344729345e-06,
+      "loss": 0.1003,
+      "step": 7190
+    },
+    {
+      "epoch": 91.72,
+      "grad_norm": 4.227506637573242,
+      "learning_rate": 4.273504273504274e-06,
+      "loss": 0.1032,
+      "step": 7200
+    },
+    {
+      "epoch": 91.85,
+      "grad_norm": 3.0407326221466064,
+      "learning_rate": 4.2022792022792025e-06,
+      "loss": 0.1012,
+      "step": 7210
+    },
+    {
+      "epoch": 91.97,
+      "grad_norm": 4.446778774261475,
+      "learning_rate": 4.131054131054131e-06,
+      "loss": 0.1095,
+      "step": 7220
+    },
+    {
+      "epoch": 92.0,
+      "eval_accuracy": 0.9807174887892377,
+      "eval_loss": 0.060695547610521317,
+      "eval_runtime": 10.4866,
+      "eval_samples_per_second": 425.304,
+      "eval_steps_per_second": 3.338,
+      "step": 7222
+    },
+    {
+      "epoch": 92.1,
+      "grad_norm": 2.901665449142456,
+      "learning_rate": 4.05982905982906e-06,
+      "loss": 0.0995,
+      "step": 7230
+    },
+    {
+      "epoch": 92.23,
+      "grad_norm": 2.991088390350342,
+      "learning_rate": 3.988603988603989e-06,
+      "loss": 0.098,
+      "step": 7240
+    },
+    {
+      "epoch": 92.36,
+      "grad_norm": 2.8773093223571777,
+      "learning_rate": 3.917378917378917e-06,
+      "loss": 0.1009,
+      "step": 7250
+    },
+    {
+      "epoch": 92.48,
+      "grad_norm": 2.93807053565979,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 0.0943,
+      "step": 7260
+    },
+    {
+      "epoch": 92.61,
+      "grad_norm": 2.7378273010253906,
+      "learning_rate": 3.774928774928775e-06,
+      "loss": 0.0974,
+      "step": 7270
+    },
+    {
+      "epoch": 92.74,
+      "grad_norm": 2.933560609817505,
+      "learning_rate": 3.7037037037037037e-06,
+      "loss": 0.1009,
+      "step": 7280
+    },
+    {
+      "epoch": 92.87,
+      "grad_norm": 3.682366132736206,
+      "learning_rate": 3.632478632478633e-06,
+      "loss": 0.0935,
+      "step": 7290
+    },
+    {
+      "epoch": 92.99,
+      "grad_norm": 3.3825466632843018,
+      "learning_rate": 3.5612535612535615e-06,
+      "loss": 0.1079,
+      "step": 7300
+    },
+    {
+      "epoch": 92.99,
+      "eval_accuracy": 0.9802690582959641,
+      "eval_loss": 0.06095505878329277,
+      "eval_runtime": 10.2605,
+      "eval_samples_per_second": 434.675,
+      "eval_steps_per_second": 3.411,
+      "step": 7300
+    },
+    {
+      "epoch": 93.12,
+      "grad_norm": 2.7189486026763916,
+      "learning_rate": 3.4900284900284902e-06,
+      "loss": 0.0916,
+      "step": 7310
+    },
+    {
+      "epoch": 93.25,
+      "grad_norm": 2.3126866817474365,
+      "learning_rate": 3.4188034188034193e-06,
+      "loss": 0.099,
+      "step": 7320
+    },
+    {
+      "epoch": 93.38,
+      "grad_norm": 2.1135940551757812,
+      "learning_rate": 3.347578347578348e-06,
+      "loss": 0.089,
+      "step": 7330
+    },
+    {
+      "epoch": 93.5,
+      "grad_norm": 4.08342981338501,
+      "learning_rate": 3.2763532763532763e-06,
+      "loss": 0.1045,
+      "step": 7340
+    },
+    {
+      "epoch": 93.63,
+      "grad_norm": 3.6746959686279297,
+      "learning_rate": 3.205128205128205e-06,
+      "loss": 0.1017,
+      "step": 7350
+    },
+    {
+      "epoch": 93.76,
+      "grad_norm": 2.6264023780822754,
+      "learning_rate": 3.133903133903134e-06,
+      "loss": 0.1037,
+      "step": 7360
+    },
+    {
+      "epoch": 93.89,
+      "grad_norm": 3.9072980880737305,
+      "learning_rate": 3.0626780626780627e-06,
+      "loss": 0.1153,
+      "step": 7370
+    },
+    {
+      "epoch": 94.0,
+      "eval_accuracy": 0.9798206278026906,
+      "eval_loss": 0.06323741376399994,
+      "eval_runtime": 10.4676,
+      "eval_samples_per_second": 426.078,
+      "eval_steps_per_second": 3.344,
+      "step": 7379
+    },
+    {
+      "epoch": 94.01,
+      "grad_norm": 2.2694411277770996,
+      "learning_rate": 2.991452991452992e-06,
+      "loss": 0.0948,
+      "step": 7380
+    },
+    {
+      "epoch": 94.14,
+      "grad_norm": 2.713989019393921,
+      "learning_rate": 2.92022792022792e-06,
+      "loss": 0.0993,
+      "step": 7390
+    },
+    {
+      "epoch": 94.27,
+      "grad_norm": 2.7938032150268555,
+      "learning_rate": 2.8490028490028492e-06,
+      "loss": 0.1007,
+      "step": 7400
+    },
+    {
+      "epoch": 94.39,
+      "grad_norm": 3.5172252655029297,
+      "learning_rate": 2.777777777777778e-06,
+      "loss": 0.0922,
+      "step": 7410
+    },
+    {
+      "epoch": 94.52,
+      "grad_norm": 2.6634156703948975,
+      "learning_rate": 2.7065527065527066e-06,
+      "loss": 0.0965,
+      "step": 7420
+    },
+    {
+      "epoch": 94.65,
+      "grad_norm": 3.6004881858825684,
+      "learning_rate": 2.6353276353276357e-06,
+      "loss": 0.1091,
+      "step": 7430
+    },
+    {
+      "epoch": 94.78,
+      "grad_norm": 2.2654991149902344,
+      "learning_rate": 2.564102564102564e-06,
+      "loss": 0.0934,
+      "step": 7440
+    },
+    {
+      "epoch": 94.9,
+      "grad_norm": 2.590730905532837,
+      "learning_rate": 2.492877492877493e-06,
+      "loss": 0.1022,
+      "step": 7450
+    },
+    {
+      "epoch": 94.99,
+      "eval_accuracy": 0.9811659192825112,
+      "eval_loss": 0.061776064336299896,
+      "eval_runtime": 10.6691,
+      "eval_samples_per_second": 418.031,
+      "eval_steps_per_second": 3.281,
+      "step": 7457
+    },
+    {
+      "epoch": 95.03,
+      "grad_norm": 2.9376516342163086,
+      "learning_rate": 2.4216524216524218e-06,
+      "loss": 0.0883,
+      "step": 7460
+    },
+    {
+      "epoch": 95.16,
+      "grad_norm": 3.090703248977661,
+      "learning_rate": 2.3504273504273504e-06,
+      "loss": 0.1053,
+      "step": 7470
+    },
+    {
+      "epoch": 95.29,
+      "grad_norm": 2.322587728500366,
+      "learning_rate": 2.2792022792022796e-06,
+      "loss": 0.0911,
+      "step": 7480
+    },
+    {
+      "epoch": 95.41,
+      "grad_norm": 3.8820295333862305,
+      "learning_rate": 2.207977207977208e-06,
+      "loss": 0.0998,
+      "step": 7490
+    },
+    {
+      "epoch": 95.54,
+      "grad_norm": 2.962207078933716,
+      "learning_rate": 2.136752136752137e-06,
+      "loss": 0.0934,
+      "step": 7500
+    },
+    {
+      "epoch": 95.67,
+      "grad_norm": 3.2972400188446045,
+      "learning_rate": 2.0655270655270656e-06,
+      "loss": 0.0996,
+      "step": 7510
+    },
+    {
+      "epoch": 95.8,
+      "grad_norm": 3.117957830429077,
+      "learning_rate": 1.9943019943019943e-06,
+      "loss": 0.1057,
+      "step": 7520
+    },
+    {
+      "epoch": 95.92,
+      "grad_norm": 3.654308557510376,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 0.1079,
+      "step": 7530
+    },
+    {
+      "epoch": 96.0,
+      "eval_accuracy": 0.9809417040358744,
+      "eval_loss": 0.06057490035891533,
+      "eval_runtime": 10.8603,
+      "eval_samples_per_second": 410.671,
+      "eval_steps_per_second": 3.223,
+      "step": 7536
+    },
+    {
+      "epoch": 96.05,
+      "grad_norm": 3.377328395843506,
+      "learning_rate": 1.8518518518518519e-06,
+      "loss": 0.0884,
+      "step": 7540
+    },
+    {
+      "epoch": 96.18,
+      "grad_norm": 3.9910852909088135,
+      "learning_rate": 1.7806267806267808e-06,
+      "loss": 0.0989,
+      "step": 7550
+    },
+    {
+      "epoch": 96.31,
+      "grad_norm": 3.2044425010681152,
+      "learning_rate": 1.7094017094017097e-06,
+      "loss": 0.099,
+      "step": 7560
+    },
+    {
+      "epoch": 96.43,
+      "grad_norm": 2.561246156692505,
+      "learning_rate": 1.6381766381766381e-06,
+      "loss": 0.0914,
+      "step": 7570
+    },
+    {
+      "epoch": 96.56,
+      "grad_norm": 3.3193604946136475,
+      "learning_rate": 1.566951566951567e-06,
+      "loss": 0.0996,
+      "step": 7580
+    },
+    {
+      "epoch": 96.69,
+      "grad_norm": 2.501915216445923,
+      "learning_rate": 1.495726495726496e-06,
+      "loss": 0.1051,
+      "step": 7590
+    },
+    {
+      "epoch": 96.82,
+      "grad_norm": 3.687192916870117,
+      "learning_rate": 1.4245014245014246e-06,
+      "loss": 0.1013,
+      "step": 7600
+    },
+    {
+      "epoch": 96.94,
+      "grad_norm": 2.6535141468048096,
+      "learning_rate": 1.3532763532763533e-06,
+      "loss": 0.0942,
+      "step": 7610
+    },
+    {
+      "epoch": 96.99,
+      "eval_accuracy": 0.9800448430493274,
+      "eval_loss": 0.0612373948097229,
+      "eval_runtime": 10.5685,
+      "eval_samples_per_second": 422.011,
+      "eval_steps_per_second": 3.312,
+      "step": 7614
+    },
+    {
+      "epoch": 97.07,
+      "grad_norm": 4.711557865142822,
+      "learning_rate": 1.282051282051282e-06,
+      "loss": 0.103,
+      "step": 7620
+    },
+    {
+      "epoch": 97.2,
+      "grad_norm": 4.340113162994385,
+      "learning_rate": 1.2108262108262109e-06,
+      "loss": 0.0955,
+      "step": 7630
+    },
+    {
+      "epoch": 97.32,
+      "grad_norm": 3.0971832275390625,
+      "learning_rate": 1.1396011396011398e-06,
+      "loss": 0.098,
+      "step": 7640
+    },
+    {
+      "epoch": 97.45,
+      "grad_norm": 2.318181276321411,
+      "learning_rate": 1.0683760683760685e-06,
+      "loss": 0.0952,
+      "step": 7650
+    },
+    {
+      "epoch": 97.58,
+      "grad_norm": 2.7054574489593506,
+      "learning_rate": 9.971509971509971e-07,
+      "loss": 0.102,
+      "step": 7660
+    },
+    {
+      "epoch": 97.71,
+      "grad_norm": 2.653917074203491,
+      "learning_rate": 9.259259259259259e-07,
+      "loss": 0.0943,
+      "step": 7670
+    },
+    {
+      "epoch": 97.83,
+      "grad_norm": 1.9929085969924927,
+      "learning_rate": 8.547008547008548e-07,
+      "loss": 0.1015,
+      "step": 7680
+    },
+    {
+      "epoch": 97.96,
+      "grad_norm": 2.189223289489746,
+      "learning_rate": 7.834757834757835e-07,
+      "loss": 0.0927,
+      "step": 7690
+    },
+    {
+      "epoch": 98.0,
+      "eval_accuracy": 0.9809417040358744,
+      "eval_loss": 0.05984415113925934,
+      "eval_runtime": 10.4501,
+      "eval_samples_per_second": 426.792,
+      "eval_steps_per_second": 3.349,
+      "step": 7693
+    },
+    {
+      "epoch": 98.09,
+      "grad_norm": 2.4412922859191895,
+      "learning_rate": 7.122507122507123e-07,
+      "loss": 0.1048,
+      "step": 7700
+    },
+    {
+      "epoch": 98.22,
+      "grad_norm": 2.839982748031616,
+      "learning_rate": 6.41025641025641e-07,
+      "loss": 0.1044,
+      "step": 7710
+    },
+    {
+      "epoch": 98.34,
+      "grad_norm": 1.9306455850601196,
+      "learning_rate": 5.698005698005699e-07,
+      "loss": 0.0911,
+      "step": 7720
+    },
+    {
+      "epoch": 98.47,
+      "grad_norm": 2.3731818199157715,
+      "learning_rate": 4.985754985754986e-07,
+      "loss": 0.0971,
+      "step": 7730
+    },
+    {
+      "epoch": 98.6,
+      "grad_norm": 3.8493146896362305,
+      "learning_rate": 4.273504273504274e-07,
+      "loss": 0.0941,
+      "step": 7740
+    },
+    {
+      "epoch": 98.73,
+      "grad_norm": 2.2244791984558105,
+      "learning_rate": 3.5612535612535615e-07,
+      "loss": 0.0998,
+      "step": 7750
+    },
+    {
+      "epoch": 98.85,
+      "grad_norm": 2.7025341987609863,
+      "learning_rate": 2.8490028490028494e-07,
+      "loss": 0.1015,
+      "step": 7760
+    },
+    {
+      "epoch": 98.98,
+      "grad_norm": 2.6602606773376465,
+      "learning_rate": 2.136752136752137e-07,
+      "loss": 0.1032,
+      "step": 7770
+    },
+    {
+      "epoch": 98.99,
+      "eval_accuracy": 0.981390134529148,
+      "eval_loss": 0.06043354794383049,
+      "eval_runtime": 10.296,
+      "eval_samples_per_second": 433.177,
+      "eval_steps_per_second": 3.399,
+      "step": 7771
+    },
+    {
+      "epoch": 99.11,
+      "grad_norm": 2.8605153560638428,
+      "learning_rate": 1.4245014245014247e-07,
+      "loss": 0.1033,
+      "step": 7780
+    },
+    {
+      "epoch": 99.24,
+      "grad_norm": 1.583977460861206,
+      "learning_rate": 7.122507122507124e-08,
+      "loss": 0.0991,
+      "step": 7790
+    },
+    {
+      "epoch": 99.36,
+      "grad_norm": 2.7007992267608643,
+      "learning_rate": 0.0,
+      "loss": 0.0925,
+      "step": 7800
+    },
+    {
+      "epoch": 99.36,
+      "eval_accuracy": 0.9816143497757848,
+      "eval_loss": 0.06034744158387184,
+      "eval_runtime": 10.4984,
+      "eval_samples_per_second": 424.827,
+      "eval_steps_per_second": 3.334,
+      "step": 7800
+    },
+    {
+      "epoch": 99.36,
+      "step": 7800,
+      "total_flos": 9.913968910687958e+19,
+      "train_loss": 0.21531945743621925,
+      "train_runtime": 13739.5225,
+      "train_samples_per_second": 292.15,
+      "train_steps_per_second": 0.568
     }
   ],
   "logging_steps": 10,
-  "max_steps": 1400,
+  "max_steps": 7800,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 50,
+  "num_train_epochs": 100,
   "save_steps": 500,
-  "total_flos": 1.7807825640923136e+19,
+  "total_flos": 9.913968910687958e+19,
   "train_batch_size": 128,
   "trial_name": null,
   "trial_params": null