diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,901 +1,4623 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 86.66666666666667,
+  "epoch": 148.8,
   "eval_steps": 500,
-  "global_step": 130,
+  "global_step": 4650,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 1.0,
-      "eval_accuracy": 0.0,
-      "eval_loss": 2.030040740966797,
-      "eval_runtime": 0.2932,
-      "eval_samples_per_second": 27.282,
-      "eval_steps_per_second": 3.41,
-      "step": 1
+      "epoch": 0.32,
+      "grad_norm": 3.6344285011291504,
+      "learning_rate": 1.0752688172043011e-06,
+      "loss": 3.6013,
+      "step": 10
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 3.322525978088379,
+      "learning_rate": 2.1505376344086023e-06,
+      "loss": 3.6033,
+      "step": 20
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 3.9304728507995605,
+      "learning_rate": 3.225806451612903e-06,
+      "loss": 3.5919,
+      "step": 30
+    },
+    {
+      "epoch": 0.992,
+      "eval_accuracy": 0.03153153153153153,
+      "eval_loss": 3.5735630989074707,
+      "eval_runtime": 10.6737,
+      "eval_samples_per_second": 41.598,
+      "eval_steps_per_second": 1.312,
+      "step": 31
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 4.976346492767334,
+      "learning_rate": 4.3010752688172045e-06,
+      "loss": 3.5714,
+      "step": 40
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 6.2756218910217285,
+      "learning_rate": 5.376344086021506e-06,
+      "loss": 3.5632,
+      "step": 50
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 6.759893894195557,
+      "learning_rate": 6.451612903225806e-06,
+      "loss": 3.5392,
+      "step": 60
+    },
+    {
+      "epoch": 1.984,
+      "eval_accuracy": 0.0472972972972973,
+      "eval_loss": 3.4989891052246094,
+      "eval_runtime": 9.3519,
+      "eval_samples_per_second": 47.477,
+      "eval_steps_per_second": 1.497,
+      "step": 62
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 7.38181209564209,
+      "learning_rate": 7.526881720430108e-06,
+      "loss": 3.5077,
+      "step": 70
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 8.292854309082031,
+      "learning_rate": 8.602150537634409e-06,
+      "loss": 3.4781,
+      "step": 80
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 10.34473705291748,
+      "learning_rate": 9.67741935483871e-06,
+      "loss": 3.4591,
+      "step": 90
+    },
+    {
+      "epoch": 2.976,
+      "eval_accuracy": 0.11036036036036036,
+      "eval_loss": 3.3992254734039307,
+      "eval_runtime": 9.265,
+      "eval_samples_per_second": 47.922,
+      "eval_steps_per_second": 1.511,
+      "step": 93
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 14.919889450073242,
+      "learning_rate": 1.0752688172043012e-05,
+      "loss": 3.425,
+      "step": 100
     },
     {
-      "epoch": 2.0,
-      "eval_accuracy": 0.0,
-      "eval_loss": 2.020792007446289,
-      "eval_runtime": 0.2487,
-      "eval_samples_per_second": 32.169,
-      "eval_steps_per_second": 4.021,
-      "step": 3
+      "epoch": 3.52,
+      "grad_norm": 13.331893920898438,
+      "learning_rate": 1.1827956989247313e-05,
+      "loss": 3.3924,
+      "step": 110
     },
     {
-      "epoch": 3.0,
-      "eval_accuracy": 0.0,
-      "eval_loss": 1.996970534324646,
-      "eval_runtime": 0.2271,
-      "eval_samples_per_second": 35.23,
-      "eval_steps_per_second": 4.404,
-      "step": 5
+      "epoch": 3.84,
+      "grad_norm": 10.675400733947754,
+      "learning_rate": 1.2903225806451613e-05,
+      "loss": 3.3361,
+      "step": 120
     },
     {
       "epoch": 4.0,
-      "eval_accuracy": 0.125,
-      "eval_loss": 1.9852917194366455,
-      "eval_runtime": 0.2294,
-      "eval_samples_per_second": 34.875,
-      "eval_steps_per_second": 4.359,
-      "step": 6
-    },
-    {
-      "epoch": 5.0,
-      "eval_accuracy": 0.125,
-      "eval_loss": 1.9666370153427124,
-      "eval_runtime": 0.23,
-      "eval_samples_per_second": 34.788,
-      "eval_steps_per_second": 4.348,
-      "step": 7
-    },
-    {
-      "epoch": 6.0,
-      "eval_accuracy": 0.25,
-      "eval_loss": 1.9215028285980225,
-      "eval_runtime": 0.2245,
-      "eval_samples_per_second": 35.632,
-      "eval_steps_per_second": 4.454,
-      "step": 9
-    },
-    {
-      "epoch": 6.666666666666667,
-      "grad_norm": 14.784972190856934,
-      "learning_rate": 3.846153846153846e-05,
-      "loss": 1.024,
-      "step": 10
+      "eval_accuracy": 0.25225225225225223,
+      "eval_loss": 3.2525722980499268,
+      "eval_runtime": 9.2756,
+      "eval_samples_per_second": 47.868,
+      "eval_steps_per_second": 1.509,
+      "step": 125
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 14.011612892150879,
+      "learning_rate": 1.3978494623655914e-05,
+      "loss": 3.2768,
+      "step": 130
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 12.37881851196289,
+      "learning_rate": 1.5053763440860215e-05,
+      "loss": 3.2526,
+      "step": 140
+    },
+    {
+      "epoch": 4.8,
+      "grad_norm": 11.936776161193848,
+      "learning_rate": 1.6129032258064517e-05,
+      "loss": 3.2066,
+      "step": 150
+    },
+    {
+      "epoch": 4.992,
+      "eval_accuracy": 0.3761261261261261,
+      "eval_loss": 3.085122585296631,
+      "eval_runtime": 9.2654,
+      "eval_samples_per_second": 47.92,
+      "eval_steps_per_second": 1.511,
+      "step": 156
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 15.466531753540039,
+      "learning_rate": 1.7204301075268818e-05,
+      "loss": 3.1437,
+      "step": 160
+    },
+    {
+      "epoch": 5.44,
+      "grad_norm": 12.19348430633545,
+      "learning_rate": 1.827956989247312e-05,
+      "loss": 3.0506,
+      "step": 170
+    },
+    {
+      "epoch": 5.76,
+      "grad_norm": 18.181127548217773,
+      "learning_rate": 1.935483870967742e-05,
+      "loss": 3.0024,
+      "step": 180
+    },
+    {
+      "epoch": 5.984,
+      "eval_accuracy": 0.4436936936936937,
+      "eval_loss": 2.837559938430786,
+      "eval_runtime": 9.2717,
+      "eval_samples_per_second": 47.888,
+      "eval_steps_per_second": 1.51,
+      "step": 187
+    },
+    {
+      "epoch": 6.08,
+      "grad_norm": 11.255278587341309,
+      "learning_rate": 2.0430107526881722e-05,
+      "loss": 2.926,
+      "step": 190
+    },
+    {
+      "epoch": 6.4,
+      "grad_norm": 27.72000503540039,
+      "learning_rate": 2.1505376344086024e-05,
+      "loss": 2.8533,
+      "step": 200
+    },
+    {
+      "epoch": 6.72,
+      "grad_norm": 29.05733299255371,
+      "learning_rate": 2.258064516129032e-05,
+      "loss": 2.8094,
+      "step": 210
+    },
+    {
+      "epoch": 6.976,
+      "eval_accuracy": 0.49099099099099097,
+      "eval_loss": 2.631969928741455,
+      "eval_runtime": 9.3148,
+      "eval_samples_per_second": 47.666,
+      "eval_steps_per_second": 1.503,
+      "step": 218
+    },
+    {
+      "epoch": 7.04,
+      "grad_norm": 14.886007308959961,
+      "learning_rate": 2.3655913978494626e-05,
+      "loss": 2.7366,
+      "step": 220
+    },
+    {
+      "epoch": 7.36,
+      "grad_norm": 16.759021759033203,
+      "learning_rate": 2.4731182795698928e-05,
+      "loss": 2.6044,
+      "step": 230
+    },
+    {
+      "epoch": 7.68,
+      "grad_norm": 83.9926528930664,
+      "learning_rate": 2.5806451612903226e-05,
+      "loss": 2.5626,
+      "step": 240
     },
     {
-      "epoch": 7.0,
-      "eval_accuracy": 0.125,
-      "eval_loss": 1.8757238388061523,
-      "eval_runtime": 0.2259,
-      "eval_samples_per_second": 35.411,
-      "eval_steps_per_second": 4.426,
-      "step": 11
+      "epoch": 8.0,
+      "grad_norm": 101.79939270019531,
+      "learning_rate": 2.6881720430107527e-05,
+      "loss": 2.509,
+      "step": 250
     },
     {
       "epoch": 8.0,
-      "eval_accuracy": 0.125,
-      "eval_loss": 1.8579578399658203,
-      "eval_runtime": 0.2276,
-      "eval_samples_per_second": 35.149,
-      "eval_steps_per_second": 4.394,
-      "step": 12
+      "eval_accuracy": 0.536036036036036,
+      "eval_loss": 2.376469850540161,
+      "eval_runtime": 9.414,
+      "eval_samples_per_second": 47.164,
+      "eval_steps_per_second": 1.487,
+      "step": 250
+    },
+    {
+      "epoch": 8.32,
+      "grad_norm": 28.594270706176758,
+      "learning_rate": 2.7956989247311828e-05,
+      "loss": 2.4039,
+      "step": 260
+    },
+    {
+      "epoch": 8.64,
+      "grad_norm": 34.03124237060547,
+      "learning_rate": 2.9032258064516133e-05,
+      "loss": 2.2813,
+      "step": 270
+    },
+    {
+      "epoch": 8.96,
+      "grad_norm": 13.96226978302002,
+      "learning_rate": 3.010752688172043e-05,
+      "loss": 2.2526,
+      "step": 280
+    },
+    {
+      "epoch": 8.992,
+      "eval_accuracy": 0.5923423423423423,
+      "eval_loss": 2.040027141571045,
+      "eval_runtime": 9.1841,
+      "eval_samples_per_second": 48.344,
+      "eval_steps_per_second": 1.524,
+      "step": 281
+    },
+    {
+      "epoch": 9.28,
+      "grad_norm": 29.609779357910156,
+      "learning_rate": 3.118279569892473e-05,
+      "loss": 2.0909,
+      "step": 290
+    },
+    {
+      "epoch": 9.6,
+      "grad_norm": 29.265300750732422,
+      "learning_rate": 3.2258064516129034e-05,
+      "loss": 2.0485,
+      "step": 300
+    },
+    {
+      "epoch": 9.92,
+      "grad_norm": 32.22190856933594,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 1.9442,
+      "step": 310
+    },
+    {
+      "epoch": 9.984,
+      "eval_accuracy": 0.6396396396396397,
+      "eval_loss": 1.7939640283584595,
+      "eval_runtime": 9.3505,
+      "eval_samples_per_second": 47.484,
+      "eval_steps_per_second": 1.497,
+      "step": 312
+    },
+    {
+      "epoch": 10.24,
+      "grad_norm": 21.0910701751709,
+      "learning_rate": 3.4408602150537636e-05,
+      "loss": 1.9181,
+      "step": 320
+    },
+    {
+      "epoch": 10.56,
+      "grad_norm": 45.26454544067383,
+      "learning_rate": 3.548387096774194e-05,
+      "loss": 1.7878,
+      "step": 330
+    },
+    {
+      "epoch": 10.88,
+      "grad_norm": 35.63391876220703,
+      "learning_rate": 3.655913978494624e-05,
+      "loss": 1.7672,
+      "step": 340
+    },
+    {
+      "epoch": 10.975999999999999,
+      "eval_accuracy": 0.6824324324324325,
+      "eval_loss": 1.5892232656478882,
+      "eval_runtime": 9.2159,
+      "eval_samples_per_second": 48.177,
+      "eval_steps_per_second": 1.519,
+      "step": 343
     },
     {
-      "epoch": 9.0,
-      "eval_accuracy": 0.125,
-      "eval_loss": 1.8413233757019043,
-      "eval_runtime": 0.232,
-      "eval_samples_per_second": 34.49,
-      "eval_steps_per_second": 4.311,
-      "step": 13
+      "epoch": 11.2,
+      "grad_norm": 14.121517181396484,
+      "learning_rate": 3.763440860215054e-05,
+      "loss": 1.6401,
+      "step": 350
     },
     {
-      "epoch": 10.0,
-      "eval_accuracy": 0.375,
-      "eval_loss": 1.7954258918762207,
-      "eval_runtime": 0.2267,
-      "eval_samples_per_second": 35.29,
-      "eval_steps_per_second": 4.411,
-      "step": 15
+      "epoch": 11.52,
+      "grad_norm": 17.638029098510742,
+      "learning_rate": 3.870967741935484e-05,
+      "loss": 1.6047,
+      "step": 360
     },
     {
-      "epoch": 11.0,
-      "eval_accuracy": 0.5,
-      "eval_loss": 1.7509543895721436,
-      "eval_runtime": 0.2337,
-      "eval_samples_per_second": 34.23,
-      "eval_steps_per_second": 4.279,
-      "step": 17
+      "epoch": 11.84,
+      "grad_norm": 28.33017921447754,
+      "learning_rate": 3.978494623655914e-05,
+      "loss": 1.5273,
+      "step": 370
     },
     {
       "epoch": 12.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.7309346199035645,
-      "eval_runtime": 0.2261,
-      "eval_samples_per_second": 35.381,
-      "eval_steps_per_second": 4.423,
-      "step": 18
-    },
-    {
-      "epoch": 13.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.7131527662277222,
-      "eval_runtime": 0.2297,
-      "eval_samples_per_second": 34.829,
-      "eval_steps_per_second": 4.354,
-      "step": 19
-    },
-    {
-      "epoch": 13.333333333333334,
-      "grad_norm": 9.231935501098633,
-      "learning_rate": 4.700854700854701e-05,
-      "loss": 0.8487,
-      "step": 20
+      "eval_accuracy": 0.7184684684684685,
+      "eval_loss": 1.3499852418899536,
+      "eval_runtime": 9.247,
+      "eval_samples_per_second": 48.016,
+      "eval_steps_per_second": 1.514,
+      "step": 375
+    },
+    {
+      "epoch": 12.16,
+      "grad_norm": 24.126209259033203,
+      "learning_rate": 4.0860215053763444e-05,
+      "loss": 1.4728,
+      "step": 380
+    },
+    {
+      "epoch": 12.48,
+      "grad_norm": 27.138376235961914,
+      "learning_rate": 4.1935483870967746e-05,
+      "loss": 1.4202,
+      "step": 390
+    },
+    {
+      "epoch": 12.8,
+      "grad_norm": 37.29874038696289,
+      "learning_rate": 4.301075268817205e-05,
+      "loss": 1.3854,
+      "step": 400
+    },
+    {
+      "epoch": 12.992,
+      "eval_accuracy": 0.7162162162162162,
+      "eval_loss": 1.224295735359192,
+      "eval_runtime": 9.3269,
+      "eval_samples_per_second": 47.604,
+      "eval_steps_per_second": 1.501,
+      "step": 406
+    },
+    {
+      "epoch": 13.12,
+      "grad_norm": 23.704998016357422,
+      "learning_rate": 4.408602150537635e-05,
+      "loss": 1.3024,
+      "step": 410
+    },
+    {
+      "epoch": 13.44,
+      "grad_norm": 31.681236267089844,
+      "learning_rate": 4.516129032258064e-05,
+      "loss": 1.2981,
+      "step": 420
+    },
+    {
+      "epoch": 13.76,
+      "grad_norm": 26.51857566833496,
+      "learning_rate": 4.6236559139784944e-05,
+      "loss": 1.197,
+      "step": 430
     },
     {
-      "epoch": 14.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.6768202781677246,
-      "eval_runtime": 0.2263,
-      "eval_samples_per_second": 35.354,
-      "eval_steps_per_second": 4.419,
-      "step": 21
+      "epoch": 13.984,
+      "eval_accuracy": 0.7387387387387387,
+      "eval_loss": 1.1022223234176636,
+      "eval_runtime": 9.3715,
+      "eval_samples_per_second": 47.378,
+      "eval_steps_per_second": 1.494,
+      "step": 437
     },
     {
-      "epoch": 15.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.64021635055542,
-      "eval_runtime": 0.2376,
-      "eval_samples_per_second": 33.673,
-      "eval_steps_per_second": 4.209,
-      "step": 23
+      "epoch": 14.08,
+      "grad_norm": 19.252714157104492,
+      "learning_rate": 4.731182795698925e-05,
+      "loss": 1.148,
+      "step": 440
+    },
+    {
+      "epoch": 14.4,
+      "grad_norm": 39.495262145996094,
+      "learning_rate": 4.8387096774193554e-05,
+      "loss": 1.1297,
+      "step": 450
+    },
+    {
+      "epoch": 14.72,
+      "grad_norm": 35.67964172363281,
+      "learning_rate": 4.9462365591397855e-05,
+      "loss": 1.1114,
+      "step": 460
+    },
+    {
+      "epoch": 14.975999999999999,
+      "eval_accuracy": 0.7612612612612613,
+      "eval_loss": 1.0137522220611572,
+      "eval_runtime": 9.3185,
+      "eval_samples_per_second": 47.647,
+      "eval_steps_per_second": 1.502,
+      "step": 468
+    },
+    {
+      "epoch": 15.04,
+      "grad_norm": 13.771783828735352,
+      "learning_rate": 4.994026284348865e-05,
+      "loss": 1.0861,
+      "step": 470
+    },
+    {
+      "epoch": 15.36,
+      "grad_norm": 17.045921325683594,
+      "learning_rate": 4.982078853046595e-05,
+      "loss": 1.0129,
+      "step": 480
+    },
+    {
+      "epoch": 15.68,
+      "grad_norm": 30.740581512451172,
+      "learning_rate": 4.9701314217443254e-05,
+      "loss": 1.0051,
+      "step": 490
     },
     {
       "epoch": 16.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.6197324991226196,
-      "eval_runtime": 0.2274,
-      "eval_samples_per_second": 35.183,
-      "eval_steps_per_second": 4.398,
-      "step": 24
+      "grad_norm": 29.636211395263672,
+      "learning_rate": 4.9581839904420555e-05,
+      "loss": 0.9364,
+      "step": 500
     },
     {
-      "epoch": 17.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.5951614379882812,
-      "eval_runtime": 0.2246,
-      "eval_samples_per_second": 35.616,
-      "eval_steps_per_second": 4.452,
-      "step": 25
+      "epoch": 16.0,
+      "eval_accuracy": 0.7747747747747747,
+      "eval_loss": 0.9164464473724365,
+      "eval_runtime": 9.2849,
+      "eval_samples_per_second": 47.819,
+      "eval_steps_per_second": 1.508,
+      "step": 500
     },
     {
-      "epoch": 18.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.525948166847229,
-      "eval_runtime": 0.229,
-      "eval_samples_per_second": 34.929,
-      "eval_steps_per_second": 4.366,
-      "step": 27
+      "epoch": 16.32,
+      "grad_norm": 15.4671049118042,
+      "learning_rate": 4.9462365591397855e-05,
+      "loss": 0.8783,
+      "step": 510
     },
     {
-      "epoch": 19.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4598863124847412,
-      "eval_runtime": 0.228,
-      "eval_samples_per_second": 35.095,
-      "eval_steps_per_second": 4.387,
-      "step": 29
+      "epoch": 16.64,
+      "grad_norm": 17.964916229248047,
+      "learning_rate": 4.9342891278375156e-05,
+      "loss": 0.8561,
+      "step": 520
     },
     {
-      "epoch": 20.0,
-      "grad_norm": 15.795075416564941,
-      "learning_rate": 4.2735042735042735e-05,
-      "loss": 0.6549,
-      "step": 30
+      "epoch": 16.96,
+      "grad_norm": 24.931062698364258,
+      "learning_rate": 4.9223416965352456e-05,
+      "loss": 0.8755,
+      "step": 530
+    },
+    {
+      "epoch": 16.992,
+      "eval_accuracy": 0.7522522522522522,
+      "eval_loss": 0.9057530164718628,
+      "eval_runtime": 9.2393,
+      "eval_samples_per_second": 48.055,
+      "eval_steps_per_second": 1.515,
+      "step": 531
+    },
+    {
+      "epoch": 17.28,
+      "grad_norm": 24.417903900146484,
+      "learning_rate": 4.910394265232976e-05,
+      "loss": 0.8487,
+      "step": 540
+    },
+    {
+      "epoch": 17.6,
+      "grad_norm": 25.21699333190918,
+      "learning_rate": 4.898446833930705e-05,
+      "loss": 0.8157,
+      "step": 550
+    },
+    {
+      "epoch": 17.92,
+      "grad_norm": 17.92466926574707,
+      "learning_rate": 4.886499402628435e-05,
+      "loss": 0.7473,
+      "step": 560
+    },
+    {
+      "epoch": 17.984,
+      "eval_accuracy": 0.7927927927927928,
+      "eval_loss": 0.804523766040802,
+      "eval_runtime": 9.2619,
+      "eval_samples_per_second": 47.938,
+      "eval_steps_per_second": 1.512,
+      "step": 562
+    },
+    {
+      "epoch": 18.24,
+      "grad_norm": 25.38798713684082,
+      "learning_rate": 4.874551971326165e-05,
+      "loss": 0.7434,
+      "step": 570
+    },
+    {
+      "epoch": 18.56,
+      "grad_norm": 22.557018280029297,
+      "learning_rate": 4.862604540023895e-05,
+      "loss": 0.7647,
+      "step": 580
+    },
+    {
+      "epoch": 18.88,
+      "grad_norm": 15.224944114685059,
+      "learning_rate": 4.850657108721625e-05,
+      "loss": 0.7189,
+      "step": 590
+    },
+    {
+      "epoch": 18.976,
+      "eval_accuracy": 0.7882882882882883,
+      "eval_loss": 0.77347731590271,
+      "eval_runtime": 9.2397,
+      "eval_samples_per_second": 48.053,
+      "eval_steps_per_second": 1.515,
+      "step": 593
+    },
+    {
+      "epoch": 19.2,
+      "grad_norm": 36.273311614990234,
+      "learning_rate": 4.8387096774193554e-05,
+      "loss": 0.6816,
+      "step": 600
+    },
+    {
+      "epoch": 19.52,
+      "grad_norm": 19.850025177001953,
+      "learning_rate": 4.8267622461170854e-05,
+      "loss": 0.6405,
+      "step": 610
+    },
+    {
+      "epoch": 19.84,
+      "grad_norm": 27.864734649658203,
+      "learning_rate": 4.814814814814815e-05,
+      "loss": 0.6461,
+      "step": 620
     },
     {
       "epoch": 20.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4526441097259521,
-      "eval_runtime": 0.2272,
-      "eval_samples_per_second": 35.213,
-      "eval_steps_per_second": 4.402,
-      "step": 30
+      "eval_accuracy": 0.8198198198198198,
+      "eval_loss": 0.6875730156898499,
+      "eval_runtime": 9.2749,
+      "eval_samples_per_second": 47.871,
+      "eval_steps_per_second": 1.509,
+      "step": 625
     },
     {
-      "epoch": 21.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4458982944488525,
-      "eval_runtime": 0.227,
-      "eval_samples_per_second": 35.238,
-      "eval_steps_per_second": 4.405,
-      "step": 31
+      "epoch": 20.16,
+      "grad_norm": 19.48863410949707,
+      "learning_rate": 4.802867383512545e-05,
+      "loss": 0.6304,
+      "step": 630
+    },
+    {
+      "epoch": 20.48,
+      "grad_norm": 20.945470809936523,
+      "learning_rate": 4.790919952210275e-05,
+      "loss": 0.535,
+      "step": 640
+    },
+    {
+      "epoch": 20.8,
+      "grad_norm": 24.0327205657959,
+      "learning_rate": 4.778972520908005e-05,
+      "loss": 0.6041,
+      "step": 650
+    },
+    {
+      "epoch": 20.992,
+      "eval_accuracy": 0.7972972972972973,
+      "eval_loss": 0.7211897373199463,
+      "eval_runtime": 9.4517,
+      "eval_samples_per_second": 46.976,
+      "eval_steps_per_second": 1.481,
+      "step": 656
     },
     {
-      "epoch": 22.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4222255945205688,
-      "eval_runtime": 0.2296,
-      "eval_samples_per_second": 34.847,
-      "eval_steps_per_second": 4.356,
-      "step": 33
+      "epoch": 21.12,
+      "grad_norm": 34.93315505981445,
+      "learning_rate": 4.767025089605735e-05,
+      "loss": 0.5494,
+      "step": 660
     },
     {
-      "epoch": 23.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4135932922363281,
-      "eval_runtime": 0.23,
-      "eval_samples_per_second": 34.776,
-      "eval_steps_per_second": 4.347,
-      "step": 35
+      "epoch": 21.44,
+      "grad_norm": 29.807035446166992,
+      "learning_rate": 4.755077658303465e-05,
+      "loss": 0.5279,
+      "step": 670
+    },
+    {
+      "epoch": 21.76,
+      "grad_norm": 17.350618362426758,
+      "learning_rate": 4.743130227001195e-05,
+      "loss": 0.5016,
+      "step": 680
+    },
+    {
+      "epoch": 21.984,
+      "eval_accuracy": 0.8198198198198198,
+      "eval_loss": 0.6610586047172546,
+      "eval_runtime": 9.2263,
+      "eval_samples_per_second": 48.123,
+      "eval_steps_per_second": 1.517,
+      "step": 687
+    },
+    {
+      "epoch": 22.08,
+      "grad_norm": 16.8582706451416,
+      "learning_rate": 4.731182795698925e-05,
+      "loss": 0.5321,
+      "step": 690
+    },
+    {
+      "epoch": 22.4,
+      "grad_norm": 22.713829040527344,
+      "learning_rate": 4.7192353643966546e-05,
+      "loss": 0.523,
+      "step": 700
+    },
+    {
+      "epoch": 22.72,
+      "grad_norm": 11.667048454284668,
+      "learning_rate": 4.707287933094385e-05,
+      "loss": 0.4996,
+      "step": 710
+    },
+    {
+      "epoch": 22.976,
+      "eval_accuracy": 0.8153153153153153,
+      "eval_loss": 0.6110232472419739,
+      "eval_runtime": 9.2454,
+      "eval_samples_per_second": 48.024,
+      "eval_steps_per_second": 1.514,
+      "step": 718
+    },
+    {
+      "epoch": 23.04,
+      "grad_norm": 22.822635650634766,
+      "learning_rate": 4.695340501792115e-05,
+      "loss": 0.5179,
+      "step": 720
+    },
+    {
+      "epoch": 23.36,
+      "grad_norm": 18.14311981201172,
+      "learning_rate": 4.683393070489845e-05,
+      "loss": 0.4879,
+      "step": 730
+    },
+    {
+      "epoch": 23.68,
+      "grad_norm": 20.045101165771484,
+      "learning_rate": 4.671445639187575e-05,
+      "loss": 0.4288,
+      "step": 740
     },
     {
       "epoch": 24.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4238104820251465,
-      "eval_runtime": 0.2362,
-      "eval_samples_per_second": 33.867,
-      "eval_steps_per_second": 4.233,
-      "step": 36
-    },
-    {
-      "epoch": 25.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.4286460876464844,
-      "eval_runtime": 0.2278,
-      "eval_samples_per_second": 35.125,
-      "eval_steps_per_second": 4.391,
-      "step": 37
-    },
-    {
-      "epoch": 26.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.42312753200531,
-      "eval_runtime": 0.2304,
-      "eval_samples_per_second": 34.725,
-      "eval_steps_per_second": 4.341,
-      "step": 39
-    },
-    {
-      "epoch": 26.666666666666668,
-      "grad_norm": 7.896138668060303,
-      "learning_rate": 3.846153846153846e-05,
-      "loss": 0.479,
-      "step": 40
+      "grad_norm": 16.844778060913086,
+      "learning_rate": 4.659498207885305e-05,
+      "loss": 0.4825,
+      "step": 750
+    },
+    {
+      "epoch": 24.0,
+      "eval_accuracy": 0.8063063063063063,
+      "eval_loss": 0.6475990414619446,
+      "eval_runtime": 9.3325,
+      "eval_samples_per_second": 47.576,
+      "eval_steps_per_second": 1.5,
+      "step": 750
+    },
+    {
+      "epoch": 24.32,
+      "grad_norm": 20.355302810668945,
+      "learning_rate": 4.647550776583035e-05,
+      "loss": 0.423,
+      "step": 760
+    },
+    {
+      "epoch": 24.64,
+      "grad_norm": 14.827963829040527,
+      "learning_rate": 4.635603345280765e-05,
+      "loss": 0.4105,
+      "step": 770
+    },
+    {
+      "epoch": 24.96,
+      "grad_norm": 22.720539093017578,
+      "learning_rate": 4.6236559139784944e-05,
+      "loss": 0.434,
+      "step": 780
+    },
+    {
+      "epoch": 24.992,
+      "eval_accuracy": 0.8040540540540541,
+      "eval_loss": 0.6792961359024048,
+      "eval_runtime": 9.238,
+      "eval_samples_per_second": 48.062,
+      "eval_steps_per_second": 1.515,
+      "step": 781
+    },
+    {
+      "epoch": 25.28,
+      "grad_norm": 18.849292755126953,
+      "learning_rate": 4.6117084826762245e-05,
+      "loss": 0.3863,
+      "step": 790
     },
     {
-      "epoch": 27.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.3963532447814941,
-      "eval_runtime": 0.225,
-      "eval_samples_per_second": 35.554,
-      "eval_steps_per_second": 4.444,
-      "step": 41
+      "epoch": 25.6,
+      "grad_norm": 10.710004806518555,
+      "learning_rate": 4.5997610513739546e-05,
+      "loss": 0.4023,
+      "step": 800
+    },
+    {
+      "epoch": 25.92,
+      "grad_norm": 17.30459976196289,
+      "learning_rate": 4.5878136200716846e-05,
+      "loss": 0.4296,
+      "step": 810
+    },
+    {
+      "epoch": 25.984,
+      "eval_accuracy": 0.8018018018018018,
+      "eval_loss": 0.6015193462371826,
+      "eval_runtime": 9.4018,
+      "eval_samples_per_second": 47.225,
+      "eval_steps_per_second": 1.489,
+      "step": 812
+    },
+    {
+      "epoch": 26.24,
+      "grad_norm": 23.36750030517578,
+      "learning_rate": 4.575866188769415e-05,
+      "loss": 0.3688,
+      "step": 820
+    },
+    {
+      "epoch": 26.56,
+      "grad_norm": 25.308101654052734,
+      "learning_rate": 4.563918757467145e-05,
+      "loss": 0.4054,
+      "step": 830
+    },
+    {
+      "epoch": 26.88,
+      "grad_norm": 37.7635383605957,
+      "learning_rate": 4.551971326164875e-05,
+      "loss": 0.36,
+      "step": 840
+    },
+    {
+      "epoch": 26.976,
+      "eval_accuracy": 0.8063063063063063,
+      "eval_loss": 0.6614833474159241,
+      "eval_runtime": 9.3569,
+      "eval_samples_per_second": 47.451,
+      "eval_steps_per_second": 1.496,
+      "step": 843
+    },
+    {
+      "epoch": 27.2,
+      "grad_norm": 16.466156005859375,
+      "learning_rate": 4.540023894862604e-05,
+      "loss": 0.3464,
+      "step": 850
+    },
+    {
+      "epoch": 27.52,
+      "grad_norm": 21.421571731567383,
+      "learning_rate": 4.528076463560334e-05,
+      "loss": 0.3376,
+      "step": 860
+    },
+    {
+      "epoch": 27.84,
+      "grad_norm": 15.365056037902832,
+      "learning_rate": 4.516129032258064e-05,
+      "loss": 0.3646,
+      "step": 870
     },
     {
       "epoch": 28.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.375662088394165,
-      "eval_runtime": 0.2285,
-      "eval_samples_per_second": 35.009,
-      "eval_steps_per_second": 4.376,
-      "step": 42
+      "eval_accuracy": 0.8220720720720721,
+      "eval_loss": 0.6059438586235046,
+      "eval_runtime": 9.3484,
+      "eval_samples_per_second": 47.495,
+      "eval_steps_per_second": 1.498,
+      "step": 875
     },
     {
-      "epoch": 29.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.350050687789917,
-      "eval_runtime": 0.2264,
-      "eval_samples_per_second": 35.335,
-      "eval_steps_per_second": 4.417,
-      "step": 43
+      "epoch": 28.16,
+      "grad_norm": 23.176517486572266,
+      "learning_rate": 4.5041816009557944e-05,
+      "loss": 0.3498,
+      "step": 880
     },
     {
-      "epoch": 30.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.277871012687683,
-      "eval_runtime": 0.2258,
-      "eval_samples_per_second": 35.433,
-      "eval_steps_per_second": 4.429,
-      "step": 45
+      "epoch": 28.48,
+      "grad_norm": 18.081356048583984,
+      "learning_rate": 4.4922341696535244e-05,
+      "loss": 0.3433,
+      "step": 890
     },
     {
-      "epoch": 31.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2359542846679688,
-      "eval_runtime": 0.2324,
-      "eval_samples_per_second": 34.427,
-      "eval_steps_per_second": 4.303,
-      "step": 47
+      "epoch": 28.8,
+      "grad_norm": 17.244199752807617,
+      "learning_rate": 4.4802867383512545e-05,
+      "loss": 0.3542,
+      "step": 900
     },
     {
-      "epoch": 32.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.218457579612732,
-      "eval_runtime": 0.2295,
-      "eval_samples_per_second": 34.863,
-      "eval_steps_per_second": 4.358,
-      "step": 48
-    },
-    {
-      "epoch": 33.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1920298337936401,
-      "eval_runtime": 0.2285,
-      "eval_samples_per_second": 35.005,
-      "eval_steps_per_second": 4.376,
-      "step": 49
-    },
-    {
-      "epoch": 33.333333333333336,
-      "grad_norm": 39.576210021972656,
-      "learning_rate": 3.418803418803419e-05,
-      "loss": 0.3504,
-      "step": 50
+      "epoch": 28.992,
+      "eval_accuracy": 0.7927927927927928,
+      "eval_loss": 0.6972593069076538,
+      "eval_runtime": 9.4382,
+      "eval_samples_per_second": 47.043,
+      "eval_steps_per_second": 1.483,
+      "step": 906
     },
     {
-      "epoch": 34.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1325857639312744,
-      "eval_runtime": 0.2278,
-      "eval_samples_per_second": 35.122,
-      "eval_steps_per_second": 4.39,
-      "step": 51
+      "epoch": 29.12,
+      "grad_norm": 18.359115600585938,
+      "learning_rate": 4.4683393070489845e-05,
+      "loss": 0.331,
+      "step": 910
     },
     {
-      "epoch": 35.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1017863750457764,
-      "eval_runtime": 0.2263,
-      "eval_samples_per_second": 35.352,
-      "eval_steps_per_second": 4.419,
-      "step": 53
+      "epoch": 29.44,
+      "grad_norm": 22.027292251586914,
+      "learning_rate": 4.4563918757467146e-05,
+      "loss": 0.3245,
+      "step": 920
     },
     {
-      "epoch": 36.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.0970098972320557,
-      "eval_runtime": 0.2308,
-      "eval_samples_per_second": 34.656,
-      "eval_steps_per_second": 4.332,
-      "step": 54
+      "epoch": 29.76,
+      "grad_norm": 23.590885162353516,
+      "learning_rate": 4.4444444444444447e-05,
+      "loss": 0.3091,
+      "step": 930
     },
     {
-      "epoch": 37.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1029536724090576,
-      "eval_runtime": 0.2758,
-      "eval_samples_per_second": 29.01,
-      "eval_steps_per_second": 3.626,
-      "step": 55
+      "epoch": 29.984,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.6400348544120789,
+      "eval_runtime": 9.2623,
+      "eval_samples_per_second": 47.936,
+      "eval_steps_per_second": 1.511,
+      "step": 937
     },
     {
-      "epoch": 38.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1377930641174316,
-      "eval_runtime": 0.226,
-      "eval_samples_per_second": 35.391,
-      "eval_steps_per_second": 4.424,
-      "step": 57
+      "epoch": 30.08,
+      "grad_norm": 10.66484260559082,
+      "learning_rate": 4.432497013142175e-05,
+      "loss": 0.2961,
+      "step": 940
     },
     {
-      "epoch": 39.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1719943284988403,
-      "eval_runtime": 0.2317,
-      "eval_samples_per_second": 34.529,
-      "eval_steps_per_second": 4.316,
-      "step": 59
+      "epoch": 30.4,
+      "grad_norm": 16.36039924621582,
+      "learning_rate": 4.420549581839905e-05,
+      "loss": 0.2871,
+      "step": 950
     },
     {
-      "epoch": 40.0,
-      "grad_norm": 31.4616756439209,
-      "learning_rate": 2.9914529914529915e-05,
-      "loss": 0.2864,
-      "step": 60
+      "epoch": 30.72,
+      "grad_norm": 17.09930419921875,
+      "learning_rate": 4.408602150537635e-05,
+      "loss": 0.2774,
+      "step": 960
     },
     {
-      "epoch": 40.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1867077350616455,
-      "eval_runtime": 0.2264,
-      "eval_samples_per_second": 35.336,
-      "eval_steps_per_second": 4.417,
-      "step": 60
+      "epoch": 30.976,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.5798152685165405,
+      "eval_runtime": 9.2742,
+      "eval_samples_per_second": 47.875,
+      "eval_steps_per_second": 1.51,
+      "step": 968
     },
     {
-      "epoch": 41.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1960291862487793,
-      "eval_runtime": 0.2297,
-      "eval_samples_per_second": 34.825,
-      "eval_steps_per_second": 4.353,
-      "step": 61
+      "epoch": 31.04,
+      "grad_norm": 10.91196060180664,
+      "learning_rate": 4.396654719235365e-05,
+      "loss": 0.2825,
+      "step": 970
     },
     {
-      "epoch": 42.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1959271430969238,
-      "eval_runtime": 0.2272,
-      "eval_samples_per_second": 35.217,
-      "eval_steps_per_second": 4.402,
-      "step": 63
+      "epoch": 31.36,
+      "grad_norm": 18.016260147094727,
+      "learning_rate": 4.384707287933095e-05,
+      "loss": 0.2692,
+      "step": 980
     },
     {
-      "epoch": 43.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1726518869400024,
-      "eval_runtime": 0.2279,
-      "eval_samples_per_second": 35.11,
-      "eval_steps_per_second": 4.389,
-      "step": 65
+      "epoch": 31.68,
+      "grad_norm": 24.916288375854492,
+      "learning_rate": 4.372759856630825e-05,
+      "loss": 0.2859,
+      "step": 990
     },
     {
-      "epoch": 44.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.165253758430481,
-      "eval_runtime": 0.227,
-      "eval_samples_per_second": 35.239,
-      "eval_steps_per_second": 4.405,
-      "step": 66
-    },
-    {
-      "epoch": 45.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1643680334091187,
-      "eval_runtime": 0.2339,
-      "eval_samples_per_second": 34.204,
-      "eval_steps_per_second": 4.276,
-      "step": 67
-    },
-    {
-      "epoch": 46.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1808971166610718,
-      "eval_runtime": 0.2293,
-      "eval_samples_per_second": 34.886,
-      "eval_steps_per_second": 4.361,
-      "step": 69
-    },
-    {
-      "epoch": 46.666666666666664,
-      "grad_norm": 18.457019805908203,
-      "learning_rate": 2.564102564102564e-05,
-      "loss": 0.2357,
-      "step": 70
+      "epoch": 32.0,
+      "grad_norm": 22.427671432495117,
+      "learning_rate": 4.360812425328555e-05,
+      "loss": 0.3166,
+      "step": 1000
     },
     {
-      "epoch": 47.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1901675462722778,
-      "eval_runtime": 0.2261,
-      "eval_samples_per_second": 35.381,
-      "eval_steps_per_second": 4.423,
-      "step": 71
+      "epoch": 32.0,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.6134085655212402,
+      "eval_runtime": 9.5628,
+      "eval_samples_per_second": 46.43,
+      "eval_steps_per_second": 1.464,
+      "step": 1000
     },
     {
-      "epoch": 48.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1872090101242065,
-      "eval_runtime": 0.2272,
-      "eval_samples_per_second": 35.206,
-      "eval_steps_per_second": 4.401,
-      "step": 72
+      "epoch": 32.32,
+      "grad_norm": 15.972029685974121,
+      "learning_rate": 4.3488649940262845e-05,
+      "loss": 0.242,
+      "step": 1010
     },
     {
-      "epoch": 49.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.189396858215332,
-      "eval_runtime": 0.2263,
-      "eval_samples_per_second": 35.345,
-      "eval_steps_per_second": 4.418,
-      "step": 73
+      "epoch": 32.64,
+      "grad_norm": 12.7739896774292,
+      "learning_rate": 4.3369175627240145e-05,
+      "loss": 0.2501,
+      "step": 1020
     },
     {
-      "epoch": 50.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1982437372207642,
-      "eval_runtime": 0.2428,
-      "eval_samples_per_second": 32.942,
-      "eval_steps_per_second": 4.118,
-      "step": 75
+      "epoch": 32.96,
+      "grad_norm": 31.982995986938477,
+      "learning_rate": 4.3249701314217446e-05,
+      "loss": 0.2878,
+      "step": 1030
     },
     {
-      "epoch": 51.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2417709827423096,
-      "eval_runtime": 0.2272,
-      "eval_samples_per_second": 35.217,
-      "eval_steps_per_second": 4.402,
-      "step": 77
+      "epoch": 32.992,
+      "eval_accuracy": 0.8063063063063063,
+      "eval_loss": 0.6353267431259155,
+      "eval_runtime": 9.3031,
+      "eval_samples_per_second": 47.726,
+      "eval_steps_per_second": 1.505,
+      "step": 1031
     },
     {
-      "epoch": 52.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2575104236602783,
-      "eval_runtime": 0.2279,
-      "eval_samples_per_second": 35.099,
-      "eval_steps_per_second": 4.387,
-      "step": 78
-    },
-    {
-      "epoch": 53.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2707685232162476,
-      "eval_runtime": 0.2281,
-      "eval_samples_per_second": 35.068,
-      "eval_steps_per_second": 4.384,
-      "step": 79
-    },
-    {
-      "epoch": 53.333333333333336,
-      "grad_norm": 4.976161479949951,
-      "learning_rate": 2.1367521367521368e-05,
-      "loss": 0.1561,
-      "step": 80
+      "epoch": 33.28,
+      "grad_norm": 16.256759643554688,
+      "learning_rate": 4.3130227001194746e-05,
+      "loss": 0.255,
+      "step": 1040
     },
     {
-      "epoch": 54.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2665772438049316,
-      "eval_runtime": 0.2284,
-      "eval_samples_per_second": 35.019,
-      "eval_steps_per_second": 4.377,
-      "step": 81
+      "epoch": 33.6,
+      "grad_norm": 12.84078311920166,
+      "learning_rate": 4.301075268817205e-05,
+      "loss": 0.2319,
+      "step": 1050
     },
     {
-      "epoch": 55.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2240548133850098,
-      "eval_runtime": 0.2281,
-      "eval_samples_per_second": 35.073,
-      "eval_steps_per_second": 4.384,
-      "step": 83
+      "epoch": 33.92,
+      "grad_norm": 21.021757125854492,
+      "learning_rate": 4.289127837514935e-05,
+      "loss": 0.2529,
+      "step": 1060
     },
     {
-      "epoch": 56.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2088531255722046,
-      "eval_runtime": 0.227,
-      "eval_samples_per_second": 35.239,
-      "eval_steps_per_second": 4.405,
-      "step": 84
+      "epoch": 33.984,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.6627711653709412,
+      "eval_runtime": 9.3278,
+      "eval_samples_per_second": 47.6,
+      "eval_steps_per_second": 1.501,
+      "step": 1062
     },
     {
-      "epoch": 57.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1913509368896484,
-      "eval_runtime": 0.2251,
-      "eval_samples_per_second": 35.546,
-      "eval_steps_per_second": 4.443,
-      "step": 85
+      "epoch": 34.24,
+      "grad_norm": 17.96116065979004,
+      "learning_rate": 4.277180406212665e-05,
+      "loss": 0.2474,
+      "step": 1070
     },
     {
-      "epoch": 58.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1558996438980103,
-      "eval_runtime": 0.231,
-      "eval_samples_per_second": 34.632,
-      "eval_steps_per_second": 4.329,
-      "step": 87
+      "epoch": 34.56,
+      "grad_norm": 18.702285766601562,
+      "learning_rate": 4.265232974910394e-05,
+      "loss": 0.208,
+      "step": 1080
     },
     {
-      "epoch": 59.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1386878490447998,
-      "eval_runtime": 0.2302,
-      "eval_samples_per_second": 34.748,
-      "eval_steps_per_second": 4.344,
-      "step": 89
+      "epoch": 34.88,
+      "grad_norm": 20.62737274169922,
+      "learning_rate": 4.253285543608124e-05,
+      "loss": 0.2601,
+      "step": 1090
     },
     {
-      "epoch": 60.0,
-      "grad_norm": 19.57244300842285,
-      "learning_rate": 1.7094017094017095e-05,
-      "loss": 0.1453,
-      "step": 90
+      "epoch": 34.976,
+      "eval_accuracy": 0.8040540540540541,
+      "eval_loss": 0.6367062330245972,
+      "eval_runtime": 9.3069,
+      "eval_samples_per_second": 47.707,
+      "eval_steps_per_second": 1.504,
+      "step": 1093
     },
     {
-      "epoch": 60.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1336504220962524,
-      "eval_runtime": 0.2275,
-      "eval_samples_per_second": 35.169,
-      "eval_steps_per_second": 4.396,
-      "step": 90
+      "epoch": 35.2,
+      "grad_norm": 16.595108032226562,
+      "learning_rate": 4.241338112305854e-05,
+      "loss": 0.2344,
+      "step": 1100
     },
     {
-      "epoch": 61.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1289597749710083,
-      "eval_runtime": 0.2329,
-      "eval_samples_per_second": 34.356,
-      "eval_steps_per_second": 4.295,
-      "step": 91
+      "epoch": 35.52,
+      "grad_norm": 23.414844512939453,
+      "learning_rate": 4.2293906810035844e-05,
+      "loss": 0.2383,
+      "step": 1110
     },
     {
-      "epoch": 62.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1369203329086304,
-      "eval_runtime": 0.2293,
-      "eval_samples_per_second": 34.894,
-      "eval_steps_per_second": 4.362,
-      "step": 93
+      "epoch": 35.84,
+      "grad_norm": 13.001286506652832,
+      "learning_rate": 4.2174432497013144e-05,
+      "loss": 0.2208,
+      "step": 1120
     },
     {
-      "epoch": 63.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1438777446746826,
-      "eval_runtime": 0.2263,
-      "eval_samples_per_second": 35.359,
-      "eval_steps_per_second": 4.42,
-      "step": 95
+      "epoch": 36.0,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.6312922835350037,
+      "eval_runtime": 9.351,
+      "eval_samples_per_second": 47.482,
+      "eval_steps_per_second": 1.497,
+      "step": 1125
     },
     {
-      "epoch": 64.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.14479660987854,
-      "eval_runtime": 0.2307,
-      "eval_samples_per_second": 34.673,
-      "eval_steps_per_second": 4.334,
-      "step": 96
-    },
-    {
-      "epoch": 65.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.153009057044983,
-      "eval_runtime": 0.2302,
-      "eval_samples_per_second": 34.748,
-      "eval_steps_per_second": 4.344,
-      "step": 97
-    },
-    {
-      "epoch": 66.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1718435287475586,
-      "eval_runtime": 0.2286,
-      "eval_samples_per_second": 35.003,
-      "eval_steps_per_second": 4.375,
-      "step": 99
-    },
-    {
-      "epoch": 66.66666666666667,
-      "grad_norm": 5.6247944831848145,
-      "learning_rate": 1.282051282051282e-05,
-      "loss": 0.1271,
-      "step": 100
+      "epoch": 36.16,
+      "grad_norm": 19.46753692626953,
+      "learning_rate": 4.2054958183990445e-05,
+      "loss": 0.1937,
+      "step": 1130
     },
     {
-      "epoch": 67.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.1965450048446655,
-      "eval_runtime": 0.2246,
-      "eval_samples_per_second": 35.611,
-      "eval_steps_per_second": 4.451,
-      "step": 101
+      "epoch": 36.48,
+      "grad_norm": 18.740930557250977,
+      "learning_rate": 4.1935483870967746e-05,
+      "loss": 0.2337,
+      "step": 1140
     },
     {
-      "epoch": 68.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2091799974441528,
-      "eval_runtime": 0.2266,
-      "eval_samples_per_second": 35.297,
-      "eval_steps_per_second": 4.412,
-      "step": 102
+      "epoch": 36.8,
+      "grad_norm": 16.52130889892578,
+      "learning_rate": 4.1816009557945046e-05,
+      "loss": 0.2342,
+      "step": 1150
     },
     {
-      "epoch": 69.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2176190614700317,
-      "eval_runtime": 0.226,
-      "eval_samples_per_second": 35.4,
-      "eval_steps_per_second": 4.425,
-      "step": 103
+      "epoch": 36.992,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.5969250798225403,
+      "eval_runtime": 9.5016,
+      "eval_samples_per_second": 46.729,
+      "eval_steps_per_second": 1.473,
+      "step": 1156
     },
     {
-      "epoch": 70.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2336797714233398,
-      "eval_runtime": 0.2275,
-      "eval_samples_per_second": 35.172,
-      "eval_steps_per_second": 4.397,
-      "step": 105
+      "epoch": 37.12,
+      "grad_norm": 20.486103057861328,
+      "learning_rate": 4.169653524492234e-05,
+      "loss": 0.2097,
+      "step": 1160
     },
     {
-      "epoch": 71.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2376230955123901,
-      "eval_runtime": 0.2255,
-      "eval_samples_per_second": 35.481,
-      "eval_steps_per_second": 4.435,
-      "step": 107
+      "epoch": 37.44,
+      "grad_norm": 27.214502334594727,
+      "learning_rate": 4.157706093189964e-05,
+      "loss": 0.2257,
+      "step": 1170
     },
     {
-      "epoch": 72.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.238419532775879,
-      "eval_runtime": 0.2258,
-      "eval_samples_per_second": 35.428,
-      "eval_steps_per_second": 4.429,
-      "step": 108
-    },
-    {
-      "epoch": 73.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2378357648849487,
-      "eval_runtime": 0.2254,
-      "eval_samples_per_second": 35.491,
-      "eval_steps_per_second": 4.436,
-      "step": 109
-    },
-    {
-      "epoch": 73.33333333333333,
-      "grad_norm": 16.670150756835938,
-      "learning_rate": 8.547008547008548e-06,
-      "loss": 0.1153,
-      "step": 110
+      "epoch": 37.76,
+      "grad_norm": 13.391148567199707,
+      "learning_rate": 4.145758661887694e-05,
+      "loss": 0.2122,
+      "step": 1180
     },
     {
-      "epoch": 74.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.238478183746338,
-      "eval_runtime": 0.2784,
-      "eval_samples_per_second": 28.737,
-      "eval_steps_per_second": 3.592,
-      "step": 111
+      "epoch": 37.984,
+      "eval_accuracy": 0.8198198198198198,
+      "eval_loss": 0.6390910148620605,
+      "eval_runtime": 9.1793,
+      "eval_samples_per_second": 48.37,
+      "eval_steps_per_second": 1.525,
+      "step": 1187
     },
     {
-      "epoch": 75.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2316068410873413,
-      "eval_runtime": 0.2262,
-      "eval_samples_per_second": 35.368,
-      "eval_steps_per_second": 4.421,
-      "step": 113
+      "epoch": 38.08,
+      "grad_norm": 14.237875938415527,
+      "learning_rate": 4.133811230585424e-05,
+      "loss": 0.1959,
+      "step": 1190
     },
     {
-      "epoch": 76.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2274171113967896,
-      "eval_runtime": 0.226,
-      "eval_samples_per_second": 35.392,
-      "eval_steps_per_second": 4.424,
-      "step": 114
+      "epoch": 38.4,
+      "grad_norm": 15.39142894744873,
+      "learning_rate": 4.121863799283154e-05,
+      "loss": 0.2144,
+      "step": 1200
     },
     {
-      "epoch": 77.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2251871824264526,
-      "eval_runtime": 0.2311,
-      "eval_samples_per_second": 34.618,
-      "eval_steps_per_second": 4.327,
-      "step": 115
+      "epoch": 38.72,
+      "grad_norm": 20.199790954589844,
+      "learning_rate": 4.109916367980884e-05,
+      "loss": 0.1791,
+      "step": 1210
     },
     {
-      "epoch": 78.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2195590734481812,
-      "eval_runtime": 0.2344,
-      "eval_samples_per_second": 34.126,
-      "eval_steps_per_second": 4.266,
-      "step": 117
+      "epoch": 38.976,
+      "eval_accuracy": 0.8108108108108109,
+      "eval_loss": 0.6770870685577393,
+      "eval_runtime": 9.2464,
+      "eval_samples_per_second": 48.018,
+      "eval_steps_per_second": 1.514,
+      "step": 1218
     },
     {
-      "epoch": 79.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2144805192947388,
-      "eval_runtime": 0.2276,
-      "eval_samples_per_second": 35.145,
-      "eval_steps_per_second": 4.393,
-      "step": 119
+      "epoch": 39.04,
+      "grad_norm": 8.080174446105957,
+      "learning_rate": 4.0979689366786144e-05,
+      "loss": 0.2185,
+      "step": 1220
     },
     {
-      "epoch": 80.0,
-      "grad_norm": 18.376235961914062,
-      "learning_rate": 4.273504273504274e-06,
-      "loss": 0.0882,
-      "step": 120
+      "epoch": 39.36,
+      "grad_norm": 35.202110290527344,
+      "learning_rate": 4.0860215053763444e-05,
+      "loss": 0.2096,
+      "step": 1230
     },
     {
-      "epoch": 80.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.213006615638733,
-      "eval_runtime": 0.2356,
-      "eval_samples_per_second": 33.955,
-      "eval_steps_per_second": 4.244,
-      "step": 120
+      "epoch": 39.68,
+      "grad_norm": 14.537675857543945,
+      "learning_rate": 4.074074074074074e-05,
+      "loss": 0.2078,
+      "step": 1240
     },
     {
-      "epoch": 81.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2117276191711426,
-      "eval_runtime": 0.2323,
-      "eval_samples_per_second": 34.443,
-      "eval_steps_per_second": 4.305,
-      "step": 121
+      "epoch": 40.0,
+      "grad_norm": 14.412237167358398,
+      "learning_rate": 4.062126642771804e-05,
+      "loss": 0.2113,
+      "step": 1250
     },
     {
-      "epoch": 82.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2097160816192627,
-      "eval_runtime": 0.2308,
-      "eval_samples_per_second": 34.655,
-      "eval_steps_per_second": 4.332,
-      "step": 123
+      "epoch": 40.0,
+      "eval_accuracy": 0.8085585585585585,
+      "eval_loss": 0.7035388946533203,
+      "eval_runtime": 9.3229,
+      "eval_samples_per_second": 47.625,
+      "eval_steps_per_second": 1.502,
+      "step": 1250
     },
     {
-      "epoch": 83.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2075334787368774,
-      "eval_runtime": 0.2252,
-      "eval_samples_per_second": 35.532,
-      "eval_steps_per_second": 4.441,
-      "step": 125
+      "epoch": 40.32,
+      "grad_norm": 12.491999626159668,
+      "learning_rate": 4.050179211469534e-05,
+      "loss": 0.1753,
+      "step": 1260
     },
     {
-      "epoch": 84.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.205414056777954,
-      "eval_runtime": 0.2269,
-      "eval_samples_per_second": 35.264,
-      "eval_steps_per_second": 4.408,
-      "step": 126
-    },
-    {
-      "epoch": 85.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2038739919662476,
-      "eval_runtime": 0.2445,
-      "eval_samples_per_second": 32.718,
-      "eval_steps_per_second": 4.09,
-      "step": 127
-    },
-    {
-      "epoch": 86.0,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2025001049041748,
-      "eval_runtime": 0.244,
-      "eval_samples_per_second": 32.782,
-      "eval_steps_per_second": 4.098,
-      "step": 129
-    },
-    {
-      "epoch": 86.66666666666667,
-      "grad_norm": 6.174490928649902,
-      "learning_rate": 0.0,
-      "loss": 0.0987,
-      "step": 130
+      "epoch": 40.64,
+      "grad_norm": 16.06132698059082,
+      "learning_rate": 4.038231780167264e-05,
+      "loss": 0.1666,
+      "step": 1270
     },
     {
-      "epoch": 86.66666666666667,
-      "eval_accuracy": 0.625,
-      "eval_loss": 1.2021381855010986,
-      "eval_runtime": 0.2265,
-      "eval_samples_per_second": 35.317,
-      "eval_steps_per_second": 4.415,
-      "step": 130
+      "epoch": 40.96,
+      "grad_norm": 23.15831184387207,
+      "learning_rate": 4.026284348864994e-05,
+      "loss": 0.1703,
+      "step": 1280
+    },
+    {
+      "epoch": 40.992,
+      "eval_accuracy": 0.8153153153153153,
+      "eval_loss": 0.7095688581466675,
+      "eval_runtime": 9.4284,
+      "eval_samples_per_second": 47.092,
+      "eval_steps_per_second": 1.485,
+      "step": 1281
+    },
+    {
+      "epoch": 41.28,
+      "grad_norm": 20.108991622924805,
+      "learning_rate": 4.014336917562724e-05,
+      "loss": 0.1522,
+      "step": 1290
+    },
+    {
+      "epoch": 41.6,
+      "grad_norm": 24.87325096130371,
+      "learning_rate": 4.002389486260454e-05,
+      "loss": 0.194,
+      "step": 1300
+    },
+    {
+      "epoch": 41.92,
+      "grad_norm": 15.418107032775879,
+      "learning_rate": 3.990442054958184e-05,
+      "loss": 0.1751,
+      "step": 1310
+    },
+    {
+      "epoch": 41.984,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.596449613571167,
+      "eval_runtime": 9.1487,
+      "eval_samples_per_second": 48.532,
+      "eval_steps_per_second": 1.53,
+      "step": 1312
+    },
+    {
+      "epoch": 42.24,
+      "grad_norm": 6.889113903045654,
+      "learning_rate": 3.978494623655914e-05,
+      "loss": 0.1619,
+      "step": 1320
+    },
+    {
+      "epoch": 42.56,
+      "grad_norm": 16.027725219726562,
+      "learning_rate": 3.9665471923536444e-05,
+      "loss": 0.1663,
+      "step": 1330
+    },
+    {
+      "epoch": 42.88,
+      "grad_norm": 24.438074111938477,
+      "learning_rate": 3.9545997610513744e-05,
+      "loss": 0.1889,
+      "step": 1340
+    },
+    {
+      "epoch": 42.976,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.6606948971748352,
+      "eval_runtime": 9.3714,
+      "eval_samples_per_second": 47.378,
+      "eval_steps_per_second": 1.494,
+      "step": 1343
+    },
+    {
+      "epoch": 43.2,
+      "grad_norm": 16.80255889892578,
+      "learning_rate": 3.9426523297491045e-05,
+      "loss": 0.1968,
+      "step": 1350
+    },
+    {
+      "epoch": 43.52,
+      "grad_norm": 16.66586685180664,
+      "learning_rate": 3.9307048984468345e-05,
+      "loss": 0.1971,
+      "step": 1360
+    },
+    {
+      "epoch": 43.84,
+      "grad_norm": 13.07772445678711,
+      "learning_rate": 3.9187574671445646e-05,
+      "loss": 0.1791,
+      "step": 1370
+    },
+    {
+      "epoch": 44.0,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.6999737024307251,
+      "eval_runtime": 9.2212,
+      "eval_samples_per_second": 48.15,
+      "eval_steps_per_second": 1.518,
+      "step": 1375
+    },
+    {
+      "epoch": 44.16,
+      "grad_norm": 11.09182071685791,
+      "learning_rate": 3.906810035842295e-05,
+      "loss": 0.1932,
+      "step": 1380
+    },
+    {
+      "epoch": 44.48,
+      "grad_norm": 12.857522964477539,
+      "learning_rate": 3.894862604540024e-05,
+      "loss": 0.1753,
+      "step": 1390
+    },
+    {
+      "epoch": 44.8,
+      "grad_norm": 13.396793365478516,
+      "learning_rate": 3.882915173237754e-05,
+      "loss": 0.1372,
+      "step": 1400
+    },
+    {
+      "epoch": 44.992,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.6866002678871155,
+      "eval_runtime": 9.2589,
+      "eval_samples_per_second": 47.954,
+      "eval_steps_per_second": 1.512,
+      "step": 1406
+    },
+    {
+      "epoch": 45.12,
+      "grad_norm": 13.315982818603516,
+      "learning_rate": 3.870967741935484e-05,
+      "loss": 0.169,
+      "step": 1410
+    },
+    {
+      "epoch": 45.44,
+      "grad_norm": 15.068939208984375,
+      "learning_rate": 3.859020310633214e-05,
+      "loss": 0.1658,
+      "step": 1420
+    },
+    {
+      "epoch": 45.76,
+      "grad_norm": 12.234659194946289,
+      "learning_rate": 3.847072879330944e-05,
+      "loss": 0.1785,
+      "step": 1430
+    },
+    {
+      "epoch": 45.984,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.6620847582817078,
+      "eval_runtime": 9.1604,
+      "eval_samples_per_second": 48.469,
+      "eval_steps_per_second": 1.528,
+      "step": 1437
+    },
+    {
+      "epoch": 46.08,
+      "grad_norm": 15.436200141906738,
+      "learning_rate": 3.8351254480286743e-05,
+      "loss": 0.159,
+      "step": 1440
+    },
+    {
+      "epoch": 46.4,
+      "grad_norm": 10.734991073608398,
+      "learning_rate": 3.8231780167264044e-05,
+      "loss": 0.1294,
+      "step": 1450
+    },
+    {
+      "epoch": 46.72,
+      "grad_norm": 13.887167930603027,
+      "learning_rate": 3.8112305854241345e-05,
+      "loss": 0.1469,
+      "step": 1460
+    },
+    {
+      "epoch": 46.976,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.6390572190284729,
+      "eval_runtime": 9.1369,
+      "eval_samples_per_second": 48.594,
+      "eval_steps_per_second": 1.532,
+      "step": 1468
+    },
+    {
+      "epoch": 47.04,
+      "grad_norm": 16.58067512512207,
+      "learning_rate": 3.799283154121864e-05,
+      "loss": 0.162,
+      "step": 1470
+    },
+    {
+      "epoch": 47.36,
+      "grad_norm": 11.742984771728516,
+      "learning_rate": 3.787335722819594e-05,
+      "loss": 0.1564,
+      "step": 1480
+    },
+    {
+      "epoch": 47.68,
+      "grad_norm": 13.77131175994873,
+      "learning_rate": 3.775388291517324e-05,
+      "loss": 0.1489,
+      "step": 1490
+    },
+    {
+      "epoch": 48.0,
+      "grad_norm": 19.7138671875,
+      "learning_rate": 3.763440860215054e-05,
+      "loss": 0.1628,
+      "step": 1500
+    },
+    {
+      "epoch": 48.0,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.6623378396034241,
+      "eval_runtime": 9.2787,
+      "eval_samples_per_second": 47.852,
+      "eval_steps_per_second": 1.509,
+      "step": 1500
+    },
+    {
+      "epoch": 48.32,
+      "grad_norm": 19.31183433532715,
+      "learning_rate": 3.751493428912784e-05,
+      "loss": 0.1831,
+      "step": 1510
+    },
+    {
+      "epoch": 48.64,
+      "grad_norm": 10.484661102294922,
+      "learning_rate": 3.739545997610514e-05,
+      "loss": 0.1434,
+      "step": 1520
+    },
+    {
+      "epoch": 48.96,
+      "grad_norm": 14.728410720825195,
+      "learning_rate": 3.727598566308244e-05,
+      "loss": 0.1425,
+      "step": 1530
+    },
+    {
+      "epoch": 48.992,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.6442891955375671,
+      "eval_runtime": 9.2746,
+      "eval_samples_per_second": 47.873,
+      "eval_steps_per_second": 1.509,
+      "step": 1531
+    },
+    {
+      "epoch": 49.28,
+      "grad_norm": 14.076499938964844,
+      "learning_rate": 3.715651135005974e-05,
+      "loss": 0.1476,
+      "step": 1540
+    },
+    {
+      "epoch": 49.6,
+      "grad_norm": 29.458377838134766,
+      "learning_rate": 3.7037037037037037e-05,
+      "loss": 0.1621,
+      "step": 1550
+    },
+    {
+      "epoch": 49.92,
+      "grad_norm": 11.280054092407227,
+      "learning_rate": 3.691756272401434e-05,
+      "loss": 0.1727,
+      "step": 1560
+    },
+    {
+      "epoch": 49.984,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.6360700726509094,
+      "eval_runtime": 9.1824,
+      "eval_samples_per_second": 48.353,
+      "eval_steps_per_second": 1.525,
+      "step": 1562
+    },
+    {
+      "epoch": 50.24,
+      "grad_norm": 14.625510215759277,
+      "learning_rate": 3.679808841099164e-05,
+      "loss": 0.137,
+      "step": 1570
+    },
+    {
+      "epoch": 50.56,
+      "grad_norm": 15.493489265441895,
+      "learning_rate": 3.667861409796894e-05,
+      "loss": 0.1319,
+      "step": 1580
+    },
+    {
+      "epoch": 50.88,
+      "grad_norm": 8.854424476623535,
+      "learning_rate": 3.655913978494624e-05,
+      "loss": 0.1442,
+      "step": 1590
+    },
+    {
+      "epoch": 50.976,
+      "eval_accuracy": 0.8490990990990991,
+      "eval_loss": 0.6397358179092407,
+      "eval_runtime": 9.322,
+      "eval_samples_per_second": 47.629,
+      "eval_steps_per_second": 1.502,
+      "step": 1593
+    },
+    {
+      "epoch": 51.2,
+      "grad_norm": 10.89920425415039,
+      "learning_rate": 3.643966547192354e-05,
+      "loss": 0.1272,
+      "step": 1600
+    },
+    {
+      "epoch": 51.52,
+      "grad_norm": 11.777154922485352,
+      "learning_rate": 3.632019115890084e-05,
+      "loss": 0.1019,
+      "step": 1610
+    },
+    {
+      "epoch": 51.84,
+      "grad_norm": 10.171144485473633,
+      "learning_rate": 3.6200716845878134e-05,
+      "loss": 0.1386,
+      "step": 1620
+    },
+    {
+      "epoch": 52.0,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.6835172772407532,
+      "eval_runtime": 9.2477,
+      "eval_samples_per_second": 48.012,
+      "eval_steps_per_second": 1.514,
+      "step": 1625
+    },
+    {
+      "epoch": 52.16,
+      "grad_norm": 16.571006774902344,
+      "learning_rate": 3.6081242532855435e-05,
+      "loss": 0.1235,
+      "step": 1630
+    },
+    {
+      "epoch": 52.48,
+      "grad_norm": 6.069274425506592,
+      "learning_rate": 3.5961768219832735e-05,
+      "loss": 0.1715,
+      "step": 1640
+    },
+    {
+      "epoch": 52.8,
+      "grad_norm": 21.08418083190918,
+      "learning_rate": 3.5842293906810036e-05,
+      "loss": 0.1564,
+      "step": 1650
+    },
+    {
+      "epoch": 52.992,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.7071699500083923,
+      "eval_runtime": 9.4809,
+      "eval_samples_per_second": 46.831,
+      "eval_steps_per_second": 1.477,
+      "step": 1656
+    },
+    {
+      "epoch": 53.12,
+      "grad_norm": 15.996872901916504,
+      "learning_rate": 3.5722819593787336e-05,
+      "loss": 0.1518,
+      "step": 1660
+    },
+    {
+      "epoch": 53.44,
+      "grad_norm": 11.845250129699707,
+      "learning_rate": 3.560334528076464e-05,
+      "loss": 0.1328,
+      "step": 1670
+    },
+    {
+      "epoch": 53.76,
+      "grad_norm": 7.606972694396973,
+      "learning_rate": 3.548387096774194e-05,
+      "loss": 0.1151,
+      "step": 1680
+    },
+    {
+      "epoch": 53.984,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.6834642291069031,
+      "eval_runtime": 9.5012,
+      "eval_samples_per_second": 46.731,
+      "eval_steps_per_second": 1.473,
+      "step": 1687
+    },
+    {
+      "epoch": 54.08,
+      "grad_norm": 10.345972061157227,
+      "learning_rate": 3.536439665471924e-05,
+      "loss": 0.1107,
+      "step": 1690
+    },
+    {
+      "epoch": 54.4,
+      "grad_norm": 11.632840156555176,
+      "learning_rate": 3.524492234169653e-05,
+      "loss": 0.1504,
+      "step": 1700
+    },
+    {
+      "epoch": 54.72,
+      "grad_norm": 10.270337104797363,
+      "learning_rate": 3.512544802867383e-05,
+      "loss": 0.1446,
+      "step": 1710
+    },
+    {
+      "epoch": 54.976,
+      "eval_accuracy": 0.8198198198198198,
+      "eval_loss": 0.7347476482391357,
+      "eval_runtime": 9.0983,
+      "eval_samples_per_second": 48.8,
+      "eval_steps_per_second": 1.539,
+      "step": 1718
+    },
+    {
+      "epoch": 55.04,
+      "grad_norm": 9.00815486907959,
+      "learning_rate": 3.500597371565113e-05,
+      "loss": 0.116,
+      "step": 1720
+    },
+    {
+      "epoch": 55.36,
+      "grad_norm": 3.8150763511657715,
+      "learning_rate": 3.4886499402628434e-05,
+      "loss": 0.1329,
+      "step": 1730
+    },
+    {
+      "epoch": 55.68,
+      "grad_norm": 10.648646354675293,
+      "learning_rate": 3.4767025089605734e-05,
+      "loss": 0.1214,
+      "step": 1740
+    },
+    {
+      "epoch": 56.0,
+      "grad_norm": 13.604965209960938,
+      "learning_rate": 3.4647550776583035e-05,
+      "loss": 0.1353,
+      "step": 1750
+    },
+    {
+      "epoch": 56.0,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.6935142278671265,
+      "eval_runtime": 9.1929,
+      "eval_samples_per_second": 48.298,
+      "eval_steps_per_second": 1.523,
+      "step": 1750
+    },
+    {
+      "epoch": 56.32,
+      "grad_norm": 16.767179489135742,
+      "learning_rate": 3.4528076463560336e-05,
+      "loss": 0.1196,
+      "step": 1760
+    },
+    {
+      "epoch": 56.64,
+      "grad_norm": 9.230161666870117,
+      "learning_rate": 3.4408602150537636e-05,
+      "loss": 0.125,
+      "step": 1770
+    },
+    {
+      "epoch": 56.96,
+      "grad_norm": 17.74578857421875,
+      "learning_rate": 3.428912783751494e-05,
+      "loss": 0.13,
+      "step": 1780
+    },
+    {
+      "epoch": 56.992,
+      "eval_accuracy": 0.8198198198198198,
+      "eval_loss": 0.7337386608123779,
+      "eval_runtime": 9.5019,
+      "eval_samples_per_second": 46.728,
+      "eval_steps_per_second": 1.473,
+      "step": 1781
+    },
+    {
+      "epoch": 57.28,
+      "grad_norm": 8.565692901611328,
+      "learning_rate": 3.416965352449224e-05,
+      "loss": 0.112,
+      "step": 1790
+    },
+    {
+      "epoch": 57.6,
+      "grad_norm": 9.264266014099121,
+      "learning_rate": 3.405017921146954e-05,
+      "loss": 0.1194,
+      "step": 1800
+    },
+    {
+      "epoch": 57.92,
+      "grad_norm": 10.091086387634277,
+      "learning_rate": 3.393070489844684e-05,
+      "loss": 0.1312,
+      "step": 1810
+    },
+    {
+      "epoch": 57.984,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.6625228524208069,
+      "eval_runtime": 9.3477,
+      "eval_samples_per_second": 47.498,
+      "eval_steps_per_second": 1.498,
+      "step": 1812
+    },
+    {
+      "epoch": 58.24,
+      "grad_norm": 11.256436347961426,
+      "learning_rate": 3.381123058542414e-05,
+      "loss": 0.1376,
+      "step": 1820
+    },
+    {
+      "epoch": 58.56,
+      "grad_norm": 11.114587783813477,
+      "learning_rate": 3.369175627240144e-05,
+      "loss": 0.1232,
+      "step": 1830
+    },
+    {
+      "epoch": 58.88,
+      "grad_norm": 16.502622604370117,
+      "learning_rate": 3.357228195937874e-05,
+      "loss": 0.1201,
+      "step": 1840
+    },
+    {
+      "epoch": 58.976,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.6955697536468506,
+      "eval_runtime": 9.3809,
+      "eval_samples_per_second": 47.33,
+      "eval_steps_per_second": 1.492,
+      "step": 1843
+    },
+    {
+      "epoch": 59.2,
+      "grad_norm": 15.201342582702637,
+      "learning_rate": 3.3452807646356034e-05,
+      "loss": 0.1237,
+      "step": 1850
+    },
+    {
+      "epoch": 59.52,
+      "grad_norm": 9.248207092285156,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.0885,
+      "step": 1860
+    },
+    {
+      "epoch": 59.84,
+      "grad_norm": 3.726451873779297,
+      "learning_rate": 3.3213859020310636e-05,
+      "loss": 0.1411,
+      "step": 1870
+    },
+    {
+      "epoch": 60.0,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.7289673686027527,
+      "eval_runtime": 9.594,
+      "eval_samples_per_second": 46.279,
+      "eval_steps_per_second": 1.459,
+      "step": 1875
+    },
+    {
+      "epoch": 60.16,
+      "grad_norm": 14.892931938171387,
+      "learning_rate": 3.3094384707287936e-05,
+      "loss": 0.1327,
+      "step": 1880
+    },
+    {
+      "epoch": 60.48,
+      "grad_norm": 13.88679313659668,
+      "learning_rate": 3.297491039426524e-05,
+      "loss": 0.1031,
+      "step": 1890
+    },
+    {
+      "epoch": 60.8,
+      "grad_norm": 10.291778564453125,
+      "learning_rate": 3.285543608124254e-05,
+      "loss": 0.1116,
+      "step": 1900
+    },
+    {
+      "epoch": 60.992,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7051534056663513,
+      "eval_runtime": 9.235,
+      "eval_samples_per_second": 48.078,
+      "eval_steps_per_second": 1.516,
+      "step": 1906
+    },
+    {
+      "epoch": 61.12,
+      "grad_norm": 10.923468589782715,
+      "learning_rate": 3.273596176821984e-05,
+      "loss": 0.0994,
+      "step": 1910
+    },
+    {
+      "epoch": 61.44,
+      "grad_norm": 23.09986114501953,
+      "learning_rate": 3.261648745519714e-05,
+      "loss": 0.1304,
+      "step": 1920
+    },
+    {
+      "epoch": 61.76,
+      "grad_norm": 13.718985557556152,
+      "learning_rate": 3.249701314217443e-05,
+      "loss": 0.1251,
+      "step": 1930
+    },
+    {
+      "epoch": 61.984,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.6914505362510681,
+      "eval_runtime": 9.385,
+      "eval_samples_per_second": 47.31,
+      "eval_steps_per_second": 1.492,
+      "step": 1937
+    },
+    {
+      "epoch": 62.08,
+      "grad_norm": 11.085161209106445,
+      "learning_rate": 3.237753882915173e-05,
+      "loss": 0.1472,
+      "step": 1940
+    },
+    {
+      "epoch": 62.4,
+      "grad_norm": 7.979318618774414,
+      "learning_rate": 3.2258064516129034e-05,
+      "loss": 0.1042,
+      "step": 1950
+    },
+    {
+      "epoch": 62.72,
+      "grad_norm": 7.216133117675781,
+      "learning_rate": 3.2138590203106334e-05,
+      "loss": 0.1101,
+      "step": 1960
+    },
+    {
+      "epoch": 62.976,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.6457498669624329,
+      "eval_runtime": 9.3484,
+      "eval_samples_per_second": 47.495,
+      "eval_steps_per_second": 1.498,
+      "step": 1968
+    },
+    {
+      "epoch": 63.04,
+      "grad_norm": 16.771268844604492,
+      "learning_rate": 3.2019115890083635e-05,
+      "loss": 0.1168,
+      "step": 1970
+    },
+    {
+      "epoch": 63.36,
+      "grad_norm": 42.29957962036133,
+      "learning_rate": 3.1899641577060935e-05,
+      "loss": 0.1213,
+      "step": 1980
+    },
+    {
+      "epoch": 63.68,
+      "grad_norm": 15.948644638061523,
+      "learning_rate": 3.1780167264038236e-05,
+      "loss": 0.0991,
+      "step": 1990
+    },
+    {
+      "epoch": 64.0,
+      "grad_norm": 6.990386962890625,
+      "learning_rate": 3.1660692951015537e-05,
+      "loss": 0.0883,
+      "step": 2000
+    },
+    {
+      "epoch": 64.0,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.6553166508674622,
+      "eval_runtime": 9.3629,
+      "eval_samples_per_second": 47.421,
+      "eval_steps_per_second": 1.495,
+      "step": 2000
+    },
+    {
+      "epoch": 64.32,
+      "grad_norm": 10.886075019836426,
+      "learning_rate": 3.154121863799283e-05,
+      "loss": 0.1273,
+      "step": 2010
+    },
+    {
+      "epoch": 64.64,
+      "grad_norm": 7.905044078826904,
+      "learning_rate": 3.142174432497013e-05,
+      "loss": 0.1052,
+      "step": 2020
+    },
+    {
+      "epoch": 64.96,
+      "grad_norm": 4.7507734298706055,
+      "learning_rate": 3.130227001194743e-05,
+      "loss": 0.1225,
+      "step": 2030
+    },
+    {
+      "epoch": 64.992,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.6454054713249207,
+      "eval_runtime": 9.374,
+      "eval_samples_per_second": 47.365,
+      "eval_steps_per_second": 1.493,
+      "step": 2031
+    },
+    {
+      "epoch": 65.28,
+      "grad_norm": 13.874588012695312,
+      "learning_rate": 3.118279569892473e-05,
+      "loss": 0.096,
+      "step": 2040
+    },
+    {
+      "epoch": 65.6,
+      "grad_norm": 13.141563415527344,
+      "learning_rate": 3.106332138590203e-05,
+      "loss": 0.1033,
+      "step": 2050
+    },
+    {
+      "epoch": 65.92,
+      "grad_norm": 9.248854637145996,
+      "learning_rate": 3.0943847072879333e-05,
+      "loss": 0.1135,
+      "step": 2060
+    },
+    {
+      "epoch": 65.984,
+      "eval_accuracy": 0.8513513513513513,
+      "eval_loss": 0.6616309881210327,
+      "eval_runtime": 9.2232,
+      "eval_samples_per_second": 48.14,
+      "eval_steps_per_second": 1.518,
+      "step": 2062
+    },
+    {
+      "epoch": 66.24,
+      "grad_norm": 8.99073600769043,
+      "learning_rate": 3.0824372759856634e-05,
+      "loss": 0.1237,
+      "step": 2070
+    },
+    {
+      "epoch": 66.56,
+      "grad_norm": 14.106173515319824,
+      "learning_rate": 3.070489844683393e-05,
+      "loss": 0.12,
+      "step": 2080
+    },
+    {
+      "epoch": 66.88,
+      "grad_norm": 17.03522300720215,
+      "learning_rate": 3.058542413381123e-05,
+      "loss": 0.1009,
+      "step": 2090
+    },
+    {
+      "epoch": 66.976,
+      "eval_accuracy": 0.8536036036036037,
+      "eval_loss": 0.6374746561050415,
+      "eval_runtime": 9.2564,
+      "eval_samples_per_second": 47.967,
+      "eval_steps_per_second": 1.512,
+      "step": 2093
+    },
+    {
+      "epoch": 67.2,
+      "grad_norm": 15.714427947998047,
+      "learning_rate": 3.046594982078853e-05,
+      "loss": 0.1002,
+      "step": 2100
+    },
+    {
+      "epoch": 67.52,
+      "grad_norm": 8.479767799377441,
+      "learning_rate": 3.034647550776583e-05,
+      "loss": 0.0863,
+      "step": 2110
+    },
+    {
+      "epoch": 67.84,
+      "grad_norm": 20.34163475036621,
+      "learning_rate": 3.022700119474313e-05,
+      "loss": 0.1027,
+      "step": 2120
+    },
+    {
+      "epoch": 68.0,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.6754458546638489,
+      "eval_runtime": 9.3779,
+      "eval_samples_per_second": 47.346,
+      "eval_steps_per_second": 1.493,
+      "step": 2125
+    },
+    {
+      "epoch": 68.16,
+      "grad_norm": 6.332460403442383,
+      "learning_rate": 3.010752688172043e-05,
+      "loss": 0.0972,
+      "step": 2130
+    },
+    {
+      "epoch": 68.48,
+      "grad_norm": 19.568702697753906,
+      "learning_rate": 2.998805256869773e-05,
+      "loss": 0.0891,
+      "step": 2140
+    },
+    {
+      "epoch": 68.8,
+      "grad_norm": 13.815423011779785,
+      "learning_rate": 2.9868578255675032e-05,
+      "loss": 0.0925,
+      "step": 2150
+    },
+    {
+      "epoch": 68.992,
+      "eval_accuracy": 0.8175675675675675,
+      "eval_loss": 0.7497189044952393,
+      "eval_runtime": 9.2003,
+      "eval_samples_per_second": 48.259,
+      "eval_steps_per_second": 1.522,
+      "step": 2156
+    },
+    {
+      "epoch": 69.12,
+      "grad_norm": 12.500412940979004,
+      "learning_rate": 2.974910394265233e-05,
+      "loss": 0.0903,
+      "step": 2160
+    },
+    {
+      "epoch": 69.44,
+      "grad_norm": 16.62474250793457,
+      "learning_rate": 2.962962962962963e-05,
+      "loss": 0.099,
+      "step": 2170
+    },
+    {
+      "epoch": 69.76,
+      "grad_norm": 5.302067279815674,
+      "learning_rate": 2.951015531660693e-05,
+      "loss": 0.0878,
+      "step": 2180
+    },
+    {
+      "epoch": 69.984,
+      "eval_accuracy": 0.8490990990990991,
+      "eval_loss": 0.6572933197021484,
+      "eval_runtime": 9.2909,
+      "eval_samples_per_second": 47.789,
+      "eval_steps_per_second": 1.507,
+      "step": 2187
+    },
+    {
+      "epoch": 70.08,
+      "grad_norm": 5.19707727432251,
+      "learning_rate": 2.939068100358423e-05,
+      "loss": 0.0701,
+      "step": 2190
+    },
+    {
+      "epoch": 70.4,
+      "grad_norm": 18.371549606323242,
+      "learning_rate": 2.9271206690561532e-05,
+      "loss": 0.0992,
+      "step": 2200
+    },
+    {
+      "epoch": 70.72,
+      "grad_norm": 13.812338829040527,
+      "learning_rate": 2.9151732377538832e-05,
+      "loss": 0.1093,
+      "step": 2210
+    },
+    {
+      "epoch": 70.976,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.701532781124115,
+      "eval_runtime": 9.2401,
+      "eval_samples_per_second": 48.051,
+      "eval_steps_per_second": 1.515,
+      "step": 2218
+    },
+    {
+      "epoch": 71.04,
+      "grad_norm": 7.2299580574035645,
+      "learning_rate": 2.9032258064516133e-05,
+      "loss": 0.0814,
+      "step": 2220
+    },
+    {
+      "epoch": 71.36,
+      "grad_norm": 9.509716987609863,
+      "learning_rate": 2.8912783751493434e-05,
+      "loss": 0.0957,
+      "step": 2230
+    },
+    {
+      "epoch": 71.68,
+      "grad_norm": 21.156930923461914,
+      "learning_rate": 2.8793309438470727e-05,
+      "loss": 0.1,
+      "step": 2240
+    },
+    {
+      "epoch": 72.0,
+      "grad_norm": 9.478891372680664,
+      "learning_rate": 2.8673835125448028e-05,
+      "loss": 0.1024,
+      "step": 2250
+    },
+    {
+      "epoch": 72.0,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.6907037496566772,
+      "eval_runtime": 9.2757,
+      "eval_samples_per_second": 47.867,
+      "eval_steps_per_second": 1.509,
+      "step": 2250
+    },
+    {
+      "epoch": 72.32,
+      "grad_norm": 20.353267669677734,
+      "learning_rate": 2.855436081242533e-05,
+      "loss": 0.1022,
+      "step": 2260
+    },
+    {
+      "epoch": 72.64,
+      "grad_norm": 13.162679672241211,
+      "learning_rate": 2.843488649940263e-05,
+      "loss": 0.1155,
+      "step": 2270
+    },
+    {
+      "epoch": 72.96,
+      "grad_norm": 11.615581512451172,
+      "learning_rate": 2.831541218637993e-05,
+      "loss": 0.0934,
+      "step": 2280
+    },
+    {
+      "epoch": 72.992,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7059099078178406,
+      "eval_runtime": 9.2867,
+      "eval_samples_per_second": 47.81,
+      "eval_steps_per_second": 1.508,
+      "step": 2281
+    },
+    {
+      "epoch": 73.28,
+      "grad_norm": 22.532894134521484,
+      "learning_rate": 2.819593787335723e-05,
+      "loss": 0.0751,
+      "step": 2290
+    },
+    {
+      "epoch": 73.6,
+      "grad_norm": 17.28580093383789,
+      "learning_rate": 2.807646356033453e-05,
+      "loss": 0.1037,
+      "step": 2300
+    },
+    {
+      "epoch": 73.92,
+      "grad_norm": 9.774279594421387,
+      "learning_rate": 2.7956989247311828e-05,
+      "loss": 0.103,
+      "step": 2310
+    },
+    {
+      "epoch": 73.984,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.715863823890686,
+      "eval_runtime": 9.2962,
+      "eval_samples_per_second": 47.762,
+      "eval_steps_per_second": 1.506,
+      "step": 2312
+    },
+    {
+      "epoch": 74.24,
+      "grad_norm": 11.014548301696777,
+      "learning_rate": 2.783751493428913e-05,
+      "loss": 0.0769,
+      "step": 2320
+    },
+    {
+      "epoch": 74.56,
+      "grad_norm": 12.803121566772461,
+      "learning_rate": 2.771804062126643e-05,
+      "loss": 0.0728,
+      "step": 2330
+    },
+    {
+      "epoch": 74.88,
+      "grad_norm": 11.47424602508545,
+      "learning_rate": 2.759856630824373e-05,
+      "loss": 0.0974,
+      "step": 2340
+    },
+    {
+      "epoch": 74.976,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.7323743104934692,
+      "eval_runtime": 9.3389,
+      "eval_samples_per_second": 47.543,
+      "eval_steps_per_second": 1.499,
+      "step": 2343
+    },
+    {
+      "epoch": 75.2,
+      "grad_norm": 14.975933074951172,
+      "learning_rate": 2.747909199522103e-05,
+      "loss": 0.0854,
+      "step": 2350
+    },
+    {
+      "epoch": 75.52,
+      "grad_norm": 10.35051441192627,
+      "learning_rate": 2.735961768219833e-05,
+      "loss": 0.0975,
+      "step": 2360
+    },
+    {
+      "epoch": 75.84,
+      "grad_norm": 22.704221725463867,
+      "learning_rate": 2.7240143369175632e-05,
+      "loss": 0.1049,
+      "step": 2370
+    },
+    {
+      "epoch": 76.0,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.7397353649139404,
+      "eval_runtime": 9.3333,
+      "eval_samples_per_second": 47.572,
+      "eval_steps_per_second": 1.5,
+      "step": 2375
+    },
+    {
+      "epoch": 76.16,
+      "grad_norm": 5.02980375289917,
+      "learning_rate": 2.7120669056152932e-05,
+      "loss": 0.1032,
+      "step": 2380
+    },
+    {
+      "epoch": 76.48,
+      "grad_norm": 17.770723342895508,
+      "learning_rate": 2.7001194743130226e-05,
+      "loss": 0.0938,
+      "step": 2390
+    },
+    {
+      "epoch": 76.8,
+      "grad_norm": 11.298961639404297,
+      "learning_rate": 2.6881720430107527e-05,
+      "loss": 0.097,
+      "step": 2400
+    },
+    {
+      "epoch": 76.992,
+      "eval_accuracy": 0.8175675675675675,
+      "eval_loss": 0.7529385089874268,
+      "eval_runtime": 9.1812,
+      "eval_samples_per_second": 48.36,
+      "eval_steps_per_second": 1.525,
+      "step": 2406
+    },
+    {
+      "epoch": 77.12,
+      "grad_norm": 10.798528671264648,
+      "learning_rate": 2.6762246117084827e-05,
+      "loss": 0.0675,
+      "step": 2410
+    },
+    {
+      "epoch": 77.44,
+      "grad_norm": 10.555415153503418,
+      "learning_rate": 2.6642771804062128e-05,
+      "loss": 0.0894,
+      "step": 2420
+    },
+    {
+      "epoch": 77.76,
+      "grad_norm": 5.5223612785339355,
+      "learning_rate": 2.652329749103943e-05,
+      "loss": 0.0816,
+      "step": 2430
+    },
+    {
+      "epoch": 77.984,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7174968719482422,
+      "eval_runtime": 9.2611,
+      "eval_samples_per_second": 47.943,
+      "eval_steps_per_second": 1.512,
+      "step": 2437
+    },
+    {
+      "epoch": 78.08,
+      "grad_norm": 7.4646382331848145,
+      "learning_rate": 2.640382317801673e-05,
+      "loss": 0.0791,
+      "step": 2440
+    },
+    {
+      "epoch": 78.4,
+      "grad_norm": 3.034799337387085,
+      "learning_rate": 2.628434886499403e-05,
+      "loss": 0.0722,
+      "step": 2450
+    },
+    {
+      "epoch": 78.72,
+      "grad_norm": 10.572321891784668,
+      "learning_rate": 2.616487455197133e-05,
+      "loss": 0.0902,
+      "step": 2460
+    },
+    {
+      "epoch": 78.976,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.7745317816734314,
+      "eval_runtime": 9.4539,
+      "eval_samples_per_second": 46.965,
+      "eval_steps_per_second": 1.481,
+      "step": 2468
+    },
+    {
+      "epoch": 79.04,
+      "grad_norm": 10.482852935791016,
+      "learning_rate": 2.6045400238948624e-05,
+      "loss": 0.0787,
+      "step": 2470
+    },
+    {
+      "epoch": 79.36,
+      "grad_norm": 8.270842552185059,
+      "learning_rate": 2.5925925925925925e-05,
+      "loss": 0.078,
+      "step": 2480
+    },
+    {
+      "epoch": 79.68,
+      "grad_norm": 7.286362171173096,
+      "learning_rate": 2.5806451612903226e-05,
+      "loss": 0.0676,
+      "step": 2490
+    },
+    {
+      "epoch": 80.0,
+      "grad_norm": 15.592032432556152,
+      "learning_rate": 2.5686977299880526e-05,
+      "loss": 0.0827,
+      "step": 2500
+    },
+    {
+      "epoch": 80.0,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.701685905456543,
+      "eval_runtime": 9.2982,
+      "eval_samples_per_second": 47.751,
+      "eval_steps_per_second": 1.506,
+      "step": 2500
+    },
+    {
+      "epoch": 80.32,
+      "grad_norm": 4.7289814949035645,
+      "learning_rate": 2.5567502986857827e-05,
+      "loss": 0.0594,
+      "step": 2510
+    },
+    {
+      "epoch": 80.64,
+      "grad_norm": 10.917496681213379,
+      "learning_rate": 2.5448028673835127e-05,
+      "loss": 0.073,
+      "step": 2520
+    },
+    {
+      "epoch": 80.96,
+      "grad_norm": 9.24520492553711,
+      "learning_rate": 2.5328554360812428e-05,
+      "loss": 0.0818,
+      "step": 2530
+    },
+    {
+      "epoch": 80.992,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.7711703777313232,
+      "eval_runtime": 9.3451,
+      "eval_samples_per_second": 47.511,
+      "eval_steps_per_second": 1.498,
+      "step": 2531
+    },
+    {
+      "epoch": 81.28,
+      "grad_norm": 7.302574634552002,
+      "learning_rate": 2.5209080047789725e-05,
+      "loss": 0.074,
+      "step": 2540
+    },
+    {
+      "epoch": 81.6,
+      "grad_norm": 6.845890998840332,
+      "learning_rate": 2.5089605734767026e-05,
+      "loss": 0.0649,
+      "step": 2550
+    },
+    {
+      "epoch": 81.92,
+      "grad_norm": 13.442338943481445,
+      "learning_rate": 2.4970131421744326e-05,
+      "loss": 0.076,
+      "step": 2560
+    },
+    {
+      "epoch": 81.984,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7341312766075134,
+      "eval_runtime": 9.2594,
+      "eval_samples_per_second": 47.951,
+      "eval_steps_per_second": 1.512,
+      "step": 2562
+    },
+    {
+      "epoch": 82.24,
+      "grad_norm": 11.702536582946777,
+      "learning_rate": 2.4850657108721627e-05,
+      "loss": 0.086,
+      "step": 2570
+    },
+    {
+      "epoch": 82.56,
+      "grad_norm": 10.26176643371582,
+      "learning_rate": 2.4731182795698928e-05,
+      "loss": 0.079,
+      "step": 2580
+    },
+    {
+      "epoch": 82.88,
+      "grad_norm": 6.915826320648193,
+      "learning_rate": 2.4611708482676228e-05,
+      "loss": 0.0837,
+      "step": 2590
+    },
+    {
+      "epoch": 82.976,
+      "eval_accuracy": 0.8490990990990991,
+      "eval_loss": 0.7242028713226318,
+      "eval_runtime": 9.2096,
+      "eval_samples_per_second": 48.211,
+      "eval_steps_per_second": 1.52,
+      "step": 2593
+    },
+    {
+      "epoch": 83.2,
+      "grad_norm": 11.702289581298828,
+      "learning_rate": 2.4492234169653525e-05,
+      "loss": 0.0889,
+      "step": 2600
+    },
+    {
+      "epoch": 83.52,
+      "grad_norm": 19.560651779174805,
+      "learning_rate": 2.4372759856630826e-05,
+      "loss": 0.0814,
+      "step": 2610
+    },
+    {
+      "epoch": 83.84,
+      "grad_norm": 5.7238874435424805,
+      "learning_rate": 2.4253285543608127e-05,
+      "loss": 0.0743,
+      "step": 2620
+    },
+    {
+      "epoch": 84.0,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.6999171376228333,
+      "eval_runtime": 9.5395,
+      "eval_samples_per_second": 46.543,
+      "eval_steps_per_second": 1.468,
+      "step": 2625
+    },
+    {
+      "epoch": 84.16,
+      "grad_norm": 13.587686538696289,
+      "learning_rate": 2.4133811230585427e-05,
+      "loss": 0.0888,
+      "step": 2630
+    },
+    {
+      "epoch": 84.48,
+      "grad_norm": 5.998437881469727,
+      "learning_rate": 2.4014336917562724e-05,
+      "loss": 0.0689,
+      "step": 2640
+    },
+    {
+      "epoch": 84.8,
+      "grad_norm": 12.043107986450195,
+      "learning_rate": 2.3894862604540025e-05,
+      "loss": 0.0552,
+      "step": 2650
+    },
+    {
+      "epoch": 84.992,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.6875202059745789,
+      "eval_runtime": 9.3028,
+      "eval_samples_per_second": 47.727,
+      "eval_steps_per_second": 1.505,
+      "step": 2656
+    },
+    {
+      "epoch": 85.12,
+      "grad_norm": 18.890920639038086,
+      "learning_rate": 2.3775388291517326e-05,
+      "loss": 0.101,
+      "step": 2660
+    },
+    {
+      "epoch": 85.44,
+      "grad_norm": 17.034523010253906,
+      "learning_rate": 2.3655913978494626e-05,
+      "loss": 0.0996,
+      "step": 2670
+    },
+    {
+      "epoch": 85.76,
+      "grad_norm": 16.122133255004883,
+      "learning_rate": 2.3536439665471923e-05,
+      "loss": 0.0762,
+      "step": 2680
+    },
+    {
+      "epoch": 85.984,
+      "eval_accuracy": 0.8581081081081081,
+      "eval_loss": 0.6743137836456299,
+      "eval_runtime": 9.3345,
+      "eval_samples_per_second": 47.566,
+      "eval_steps_per_second": 1.5,
+      "step": 2687
+    },
+    {
+      "epoch": 86.08,
+      "grad_norm": 26.113550186157227,
+      "learning_rate": 2.3416965352449224e-05,
+      "loss": 0.0827,
+      "step": 2690
+    },
+    {
+      "epoch": 86.4,
+      "grad_norm": 15.370433807373047,
+      "learning_rate": 2.3297491039426525e-05,
+      "loss": 0.0855,
+      "step": 2700
+    },
+    {
+      "epoch": 86.72,
+      "grad_norm": 7.994245529174805,
+      "learning_rate": 2.3178016726403825e-05,
+      "loss": 0.0742,
+      "step": 2710
+    },
+    {
+      "epoch": 86.976,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.7026851177215576,
+      "eval_runtime": 9.4608,
+      "eval_samples_per_second": 46.931,
+      "eval_steps_per_second": 1.48,
+      "step": 2718
+    },
+    {
+      "epoch": 87.04,
+      "grad_norm": 9.935162544250488,
+      "learning_rate": 2.3058542413381122e-05,
+      "loss": 0.08,
+      "step": 2720
+    },
+    {
+      "epoch": 87.36,
+      "grad_norm": 9.019417762756348,
+      "learning_rate": 2.2939068100358423e-05,
+      "loss": 0.0784,
+      "step": 2730
+    },
+    {
+      "epoch": 87.68,
+      "grad_norm": 15.763988494873047,
+      "learning_rate": 2.2819593787335724e-05,
+      "loss": 0.073,
+      "step": 2740
+    },
+    {
+      "epoch": 88.0,
+      "grad_norm": 13.025769233703613,
+      "learning_rate": 2.270011947431302e-05,
+      "loss": 0.0708,
+      "step": 2750
+    },
+    {
+      "epoch": 88.0,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7366507649421692,
+      "eval_runtime": 9.2515,
+      "eval_samples_per_second": 47.992,
+      "eval_steps_per_second": 1.513,
+      "step": 2750
+    },
+    {
+      "epoch": 88.32,
+      "grad_norm": 5.605508804321289,
+      "learning_rate": 2.258064516129032e-05,
+      "loss": 0.0684,
+      "step": 2760
+    },
+    {
+      "epoch": 88.64,
+      "grad_norm": 11.24120044708252,
+      "learning_rate": 2.2461170848267622e-05,
+      "loss": 0.0747,
+      "step": 2770
+    },
+    {
+      "epoch": 88.96,
+      "grad_norm": 12.07764720916748,
+      "learning_rate": 2.2341696535244923e-05,
+      "loss": 0.086,
+      "step": 2780
+    },
+    {
+      "epoch": 88.992,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.6904602646827698,
+      "eval_runtime": 9.2419,
+      "eval_samples_per_second": 48.042,
+      "eval_steps_per_second": 1.515,
+      "step": 2781
+    },
+    {
+      "epoch": 89.28,
+      "grad_norm": 14.17684555053711,
+      "learning_rate": 2.2222222222222223e-05,
+      "loss": 0.0653,
+      "step": 2790
+    },
+    {
+      "epoch": 89.6,
+      "grad_norm": 7.844547271728516,
+      "learning_rate": 2.2102747909199524e-05,
+      "loss": 0.0679,
+      "step": 2800
+    },
+    {
+      "epoch": 89.92,
+      "grad_norm": 17.304725646972656,
+      "learning_rate": 2.1983273596176824e-05,
+      "loss": 0.0575,
+      "step": 2810
+    },
+    {
+      "epoch": 89.984,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7041297554969788,
+      "eval_runtime": 9.165,
+      "eval_samples_per_second": 48.445,
+      "eval_steps_per_second": 1.528,
+      "step": 2812
+    },
+    {
+      "epoch": 90.24,
+      "grad_norm": 13.358567237854004,
+      "learning_rate": 2.1863799283154125e-05,
+      "loss": 0.0676,
+      "step": 2820
+    },
+    {
+      "epoch": 90.56,
+      "grad_norm": 15.546914100646973,
+      "learning_rate": 2.1744324970131422e-05,
+      "loss": 0.0569,
+      "step": 2830
+    },
+    {
+      "epoch": 90.88,
+      "grad_norm": 14.0934476852417,
+      "learning_rate": 2.1624850657108723e-05,
+      "loss": 0.0733,
+      "step": 2840
+    },
+    {
+      "epoch": 90.976,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.6464592814445496,
+      "eval_runtime": 9.22,
+      "eval_samples_per_second": 48.156,
+      "eval_steps_per_second": 1.518,
+      "step": 2843
+    },
+    {
+      "epoch": 91.2,
+      "grad_norm": 5.8752007484436035,
+      "learning_rate": 2.1505376344086024e-05,
+      "loss": 0.0577,
+      "step": 2850
+    },
+    {
+      "epoch": 91.52,
+      "grad_norm": 6.653694152832031,
+      "learning_rate": 2.1385902031063324e-05,
+      "loss": 0.0755,
+      "step": 2860
+    },
+    {
+      "epoch": 91.84,
+      "grad_norm": 6.564297676086426,
+      "learning_rate": 2.126642771804062e-05,
+      "loss": 0.0701,
+      "step": 2870
+    },
+    {
+      "epoch": 92.0,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.7065722942352295,
+      "eval_runtime": 9.308,
+      "eval_samples_per_second": 47.701,
+      "eval_steps_per_second": 1.504,
+      "step": 2875
+    },
+    {
+      "epoch": 92.16,
+      "grad_norm": 6.29132080078125,
+      "learning_rate": 2.1146953405017922e-05,
+      "loss": 0.0719,
+      "step": 2880
+    },
+    {
+      "epoch": 92.48,
+      "grad_norm": 12.695429801940918,
+      "learning_rate": 2.1027479091995223e-05,
+      "loss": 0.102,
+      "step": 2890
+    },
+    {
+      "epoch": 92.8,
+      "grad_norm": 5.133272171020508,
+      "learning_rate": 2.0908004778972523e-05,
+      "loss": 0.0782,
+      "step": 2900
+    },
+    {
+      "epoch": 92.992,
+      "eval_accuracy": 0.8243243243243243,
+      "eval_loss": 0.6954546570777893,
+      "eval_runtime": 9.1127,
+      "eval_samples_per_second": 48.723,
+      "eval_steps_per_second": 1.536,
+      "step": 2906
+    },
+    {
+      "epoch": 93.12,
+      "grad_norm": 3.4611129760742188,
+      "learning_rate": 2.078853046594982e-05,
+      "loss": 0.0674,
+      "step": 2910
+    },
+    {
+      "epoch": 93.44,
+      "grad_norm": 11.246231079101562,
+      "learning_rate": 2.066905615292712e-05,
+      "loss": 0.0818,
+      "step": 2920
+    },
+    {
+      "epoch": 93.76,
+      "grad_norm": 5.470245838165283,
+      "learning_rate": 2.054958183990442e-05,
+      "loss": 0.0754,
+      "step": 2930
+    },
+    {
+      "epoch": 93.984,
+      "eval_accuracy": 0.8468468468468469,
+      "eval_loss": 0.6836279630661011,
+      "eval_runtime": 9.1935,
+      "eval_samples_per_second": 48.295,
+      "eval_steps_per_second": 1.523,
+      "step": 2937
+    },
+    {
+      "epoch": 94.08,
+      "grad_norm": 21.865903854370117,
+      "learning_rate": 2.0430107526881722e-05,
+      "loss": 0.0647,
+      "step": 2940
+    },
+    {
+      "epoch": 94.4,
+      "grad_norm": 12.577595710754395,
+      "learning_rate": 2.031063321385902e-05,
+      "loss": 0.0657,
+      "step": 2950
+    },
+    {
+      "epoch": 94.72,
+      "grad_norm": 5.114993572235107,
+      "learning_rate": 2.019115890083632e-05,
+      "loss": 0.0545,
+      "step": 2960
+    },
+    {
+      "epoch": 94.976,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.7289859056472778,
+      "eval_runtime": 9.4039,
+      "eval_samples_per_second": 47.215,
+      "eval_steps_per_second": 1.489,
+      "step": 2968
+    },
+    {
+      "epoch": 95.04,
+      "grad_norm": 11.622882843017578,
+      "learning_rate": 2.007168458781362e-05,
+      "loss": 0.0762,
+      "step": 2970
+    },
+    {
+      "epoch": 95.36,
+      "grad_norm": 11.010184288024902,
+      "learning_rate": 1.995221027479092e-05,
+      "loss": 0.0546,
+      "step": 2980
+    },
+    {
+      "epoch": 95.68,
+      "grad_norm": 12.926688194274902,
+      "learning_rate": 1.9832735961768222e-05,
+      "loss": 0.0712,
+      "step": 2990
+    },
+    {
+      "epoch": 96.0,
+      "grad_norm": 9.490360260009766,
+      "learning_rate": 1.9713261648745522e-05,
+      "loss": 0.0913,
+      "step": 3000
+    },
+    {
+      "epoch": 96.0,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.7665247917175293,
+      "eval_runtime": 9.312,
+      "eval_samples_per_second": 47.68,
+      "eval_steps_per_second": 1.503,
+      "step": 3000
+    },
+    {
+      "epoch": 96.32,
+      "grad_norm": 15.037102699279785,
+      "learning_rate": 1.9593787335722823e-05,
+      "loss": 0.0602,
+      "step": 3010
+    },
+    {
+      "epoch": 96.64,
+      "grad_norm": 9.528903007507324,
+      "learning_rate": 1.947431302270012e-05,
+      "loss": 0.0691,
+      "step": 3020
+    },
+    {
+      "epoch": 96.96,
+      "grad_norm": 15.416064262390137,
+      "learning_rate": 1.935483870967742e-05,
+      "loss": 0.0816,
+      "step": 3030
+    },
+    {
+      "epoch": 96.992,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.7661374807357788,
+      "eval_runtime": 9.2186,
+      "eval_samples_per_second": 48.164,
+      "eval_steps_per_second": 1.519,
+      "step": 3031
+    },
+    {
+      "epoch": 97.28,
+      "grad_norm": 9.588016510009766,
+      "learning_rate": 1.923536439665472e-05,
+      "loss": 0.0699,
+      "step": 3040
+    },
+    {
+      "epoch": 97.6,
+      "grad_norm": 8.645281791687012,
+      "learning_rate": 1.9115890083632022e-05,
+      "loss": 0.0645,
+      "step": 3050
+    },
+    {
+      "epoch": 97.92,
+      "grad_norm": 9.152593612670898,
+      "learning_rate": 1.899641577060932e-05,
+      "loss": 0.0696,
+      "step": 3060
+    },
+    {
+      "epoch": 97.984,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.6920885443687439,
+      "eval_runtime": 9.5718,
+      "eval_samples_per_second": 46.386,
+      "eval_steps_per_second": 1.463,
+      "step": 3062
+    },
+    {
+      "epoch": 98.24,
+      "grad_norm": 11.67238712310791,
+      "learning_rate": 1.887694145758662e-05,
+      "loss": 0.0661,
+      "step": 3070
+    },
+    {
+      "epoch": 98.56,
+      "grad_norm": 9.657654762268066,
+      "learning_rate": 1.875746714456392e-05,
+      "loss": 0.078,
+      "step": 3080
+    },
+    {
+      "epoch": 98.88,
+      "grad_norm": 10.411086082458496,
+      "learning_rate": 1.863799283154122e-05,
+      "loss": 0.0627,
+      "step": 3090
+    },
+    {
+      "epoch": 98.976,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.7070387005805969,
+      "eval_runtime": 9.2361,
+      "eval_samples_per_second": 48.072,
+      "eval_steps_per_second": 1.516,
+      "step": 3093
+    },
+    {
+      "epoch": 99.2,
+      "grad_norm": 9.71650505065918,
+      "learning_rate": 1.8518518518518518e-05,
+      "loss": 0.0686,
+      "step": 3100
+    },
+    {
+      "epoch": 99.52,
+      "grad_norm": 6.4877519607543945,
+      "learning_rate": 1.839904420549582e-05,
+      "loss": 0.0587,
+      "step": 3110
+    },
+    {
+      "epoch": 99.84,
+      "grad_norm": 6.228066444396973,
+      "learning_rate": 1.827956989247312e-05,
+      "loss": 0.0562,
+      "step": 3120
+    },
+    {
+      "epoch": 100.0,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.7441924214363098,
+      "eval_runtime": 9.3695,
+      "eval_samples_per_second": 47.388,
+      "eval_steps_per_second": 1.494,
+      "step": 3125
+    },
+    {
+      "epoch": 100.16,
+      "grad_norm": 12.939260482788086,
+      "learning_rate": 1.816009557945042e-05,
+      "loss": 0.0581,
+      "step": 3130
+    },
+    {
+      "epoch": 100.48,
+      "grad_norm": 7.985968589782715,
+      "learning_rate": 1.8040621266427717e-05,
+      "loss": 0.0618,
+      "step": 3140
+    },
+    {
+      "epoch": 100.8,
+      "grad_norm": 8.757869720458984,
+      "learning_rate": 1.7921146953405018e-05,
+      "loss": 0.0742,
+      "step": 3150
+    },
+    {
+      "epoch": 100.992,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.699988842010498,
+      "eval_runtime": 9.324,
+      "eval_samples_per_second": 47.619,
+      "eval_steps_per_second": 1.501,
+      "step": 3156
+    },
+    {
+      "epoch": 101.12,
+      "grad_norm": 13.87547779083252,
+      "learning_rate": 1.780167264038232e-05,
+      "loss": 0.0611,
+      "step": 3160
+    },
+    {
+      "epoch": 101.44,
+      "grad_norm": 10.741342544555664,
+      "learning_rate": 1.768219832735962e-05,
+      "loss": 0.0636,
+      "step": 3170
+    },
+    {
+      "epoch": 101.76,
+      "grad_norm": 14.43264389038086,
+      "learning_rate": 1.7562724014336916e-05,
+      "loss": 0.0545,
+      "step": 3180
+    },
+    {
+      "epoch": 101.984,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.7311636209487915,
+      "eval_runtime": 9.1417,
+      "eval_samples_per_second": 48.569,
+      "eval_steps_per_second": 1.531,
+      "step": 3187
+    },
+    {
+      "epoch": 102.08,
+      "grad_norm": 4.225771903991699,
+      "learning_rate": 1.7443249701314217e-05,
+      "loss": 0.0597,
+      "step": 3190
+    },
+    {
+      "epoch": 102.4,
+      "grad_norm": 9.73902416229248,
+      "learning_rate": 1.7323775388291518e-05,
+      "loss": 0.0646,
+      "step": 3200
+    },
+    {
+      "epoch": 102.72,
+      "grad_norm": 6.130120277404785,
+      "learning_rate": 1.7204301075268818e-05,
+      "loss": 0.0635,
+      "step": 3210
+    },
+    {
+      "epoch": 102.976,
+      "eval_accuracy": 0.8490990990990991,
+      "eval_loss": 0.7231103181838989,
+      "eval_runtime": 9.4119,
+      "eval_samples_per_second": 47.174,
+      "eval_steps_per_second": 1.487,
+      "step": 3218
+    },
+    {
+      "epoch": 103.04,
+      "grad_norm": 7.3819451332092285,
+      "learning_rate": 1.708482676224612e-05,
+      "loss": 0.0519,
+      "step": 3220
+    },
+    {
+      "epoch": 103.36,
+      "grad_norm": 11.180880546569824,
+      "learning_rate": 1.696535244922342e-05,
+      "loss": 0.0719,
+      "step": 3230
+    },
+    {
+      "epoch": 103.68,
+      "grad_norm": 11.228340148925781,
+      "learning_rate": 1.684587813620072e-05,
+      "loss": 0.0625,
+      "step": 3240
+    },
+    {
+      "epoch": 104.0,
+      "grad_norm": 7.932921409606934,
+      "learning_rate": 1.6726403823178017e-05,
+      "loss": 0.0608,
+      "step": 3250
+    },
+    {
+      "epoch": 104.0,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.7332068085670471,
+      "eval_runtime": 9.1899,
+      "eval_samples_per_second": 48.314,
+      "eval_steps_per_second": 1.523,
+      "step": 3250
+    },
+    {
+      "epoch": 104.32,
+      "grad_norm": 2.481245756149292,
+      "learning_rate": 1.6606929510155318e-05,
+      "loss": 0.0753,
+      "step": 3260
+    },
+    {
+      "epoch": 104.64,
+      "grad_norm": 13.661800384521484,
+      "learning_rate": 1.648745519713262e-05,
+      "loss": 0.0717,
+      "step": 3270
+    },
+    {
+      "epoch": 104.96,
+      "grad_norm": 11.668190002441406,
+      "learning_rate": 1.636798088410992e-05,
+      "loss": 0.0769,
+      "step": 3280
+    },
+    {
+      "epoch": 104.992,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7328325510025024,
+      "eval_runtime": 9.1902,
+      "eval_samples_per_second": 48.312,
+      "eval_steps_per_second": 1.523,
+      "step": 3281
+    },
+    {
+      "epoch": 105.28,
+      "grad_norm": 12.411203384399414,
+      "learning_rate": 1.6248506571087216e-05,
+      "loss": 0.0596,
+      "step": 3290
+    },
+    {
+      "epoch": 105.6,
+      "grad_norm": 7.294625759124756,
+      "learning_rate": 1.6129032258064517e-05,
+      "loss": 0.0528,
+      "step": 3300
+    },
+    {
+      "epoch": 105.92,
+      "grad_norm": 6.775350093841553,
+      "learning_rate": 1.6009557945041817e-05,
+      "loss": 0.057,
+      "step": 3310
+    },
+    {
+      "epoch": 105.984,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.695361852645874,
+      "eval_runtime": 9.2092,
+      "eval_samples_per_second": 48.213,
+      "eval_steps_per_second": 1.52,
+      "step": 3312
+    },
+    {
+      "epoch": 106.24,
+      "grad_norm": 9.063501358032227,
+      "learning_rate": 1.5890083632019118e-05,
+      "loss": 0.0515,
+      "step": 3320
+    },
+    {
+      "epoch": 106.56,
+      "grad_norm": 7.814884185791016,
+      "learning_rate": 1.5770609318996415e-05,
+      "loss": 0.0465,
+      "step": 3330
+    },
+    {
+      "epoch": 106.88,
+      "grad_norm": 7.107222080230713,
+      "learning_rate": 1.5651135005973716e-05,
+      "loss": 0.0447,
+      "step": 3340
+    },
+    {
+      "epoch": 106.976,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7006074786186218,
+      "eval_runtime": 9.2139,
+      "eval_samples_per_second": 48.188,
+      "eval_steps_per_second": 1.519,
+      "step": 3343
+    },
+    {
+      "epoch": 107.2,
+      "grad_norm": 9.964568138122559,
+      "learning_rate": 1.5531660692951016e-05,
+      "loss": 0.0646,
+      "step": 3350
+    },
+    {
+      "epoch": 107.52,
+      "grad_norm": 5.107904434204102,
+      "learning_rate": 1.5412186379928317e-05,
+      "loss": 0.0513,
+      "step": 3360
+    },
+    {
+      "epoch": 107.84,
+      "grad_norm": 5.2939453125,
+      "learning_rate": 1.5292712066905614e-05,
+      "loss": 0.0629,
+      "step": 3370
+    },
+    {
+      "epoch": 108.0,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7149015069007874,
+      "eval_runtime": 9.5241,
+      "eval_samples_per_second": 46.619,
+      "eval_steps_per_second": 1.47,
+      "step": 3375
+    },
+    {
+      "epoch": 108.16,
+      "grad_norm": 12.05675220489502,
+      "learning_rate": 1.5173237753882915e-05,
+      "loss": 0.0627,
+      "step": 3380
+    },
+    {
+      "epoch": 108.48,
+      "grad_norm": 5.3016791343688965,
+      "learning_rate": 1.5053763440860215e-05,
+      "loss": 0.0502,
+      "step": 3390
+    },
+    {
+      "epoch": 108.8,
+      "grad_norm": 17.884672164916992,
+      "learning_rate": 1.4934289127837516e-05,
+      "loss": 0.0394,
+      "step": 3400
+    },
+    {
+      "epoch": 108.992,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.74688321352005,
+      "eval_runtime": 9.3012,
+      "eval_samples_per_second": 47.736,
+      "eval_steps_per_second": 1.505,
+      "step": 3406
+    },
+    {
+      "epoch": 109.12,
+      "grad_norm": 13.95982551574707,
+      "learning_rate": 1.4814814814814815e-05,
+      "loss": 0.0472,
+      "step": 3410
+    },
+    {
+      "epoch": 109.44,
+      "grad_norm": 7.970043659210205,
+      "learning_rate": 1.4695340501792116e-05,
+      "loss": 0.0587,
+      "step": 3420
+    },
+    {
+      "epoch": 109.76,
+      "grad_norm": 11.51051139831543,
+      "learning_rate": 1.4575866188769416e-05,
+      "loss": 0.0602,
+      "step": 3430
+    },
+    {
+      "epoch": 109.984,
+      "eval_accuracy": 0.8468468468468469,
+      "eval_loss": 0.7274497151374817,
+      "eval_runtime": 9.2259,
+      "eval_samples_per_second": 48.126,
+      "eval_steps_per_second": 1.517,
+      "step": 3437
+    },
+    {
+      "epoch": 110.08,
+      "grad_norm": 8.181411743164062,
+      "learning_rate": 1.4456391875746717e-05,
+      "loss": 0.062,
+      "step": 3440
+    },
+    {
+      "epoch": 110.4,
+      "grad_norm": 13.90931224822998,
+      "learning_rate": 1.4336917562724014e-05,
+      "loss": 0.0608,
+      "step": 3450
+    },
+    {
+      "epoch": 110.72,
+      "grad_norm": 8.419418334960938,
+      "learning_rate": 1.4217443249701315e-05,
+      "loss": 0.0635,
+      "step": 3460
+    },
+    {
+      "epoch": 110.976,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.7494940757751465,
+      "eval_runtime": 9.4779,
+      "eval_samples_per_second": 46.846,
+      "eval_steps_per_second": 1.477,
+      "step": 3468
+    },
+    {
+      "epoch": 111.04,
+      "grad_norm": 13.364533424377441,
+      "learning_rate": 1.4097968936678615e-05,
+      "loss": 0.0781,
+      "step": 3470
+    },
+    {
+      "epoch": 111.36,
+      "grad_norm": 5.620299339294434,
+      "learning_rate": 1.3978494623655914e-05,
+      "loss": 0.0526,
+      "step": 3480
+    },
+    {
+      "epoch": 111.68,
+      "grad_norm": 14.518438339233398,
+      "learning_rate": 1.3859020310633215e-05,
+      "loss": 0.0452,
+      "step": 3490
+    },
+    {
+      "epoch": 112.0,
+      "grad_norm": 3.248464822769165,
+      "learning_rate": 1.3739545997610515e-05,
+      "loss": 0.0565,
+      "step": 3500
+    },
+    {
+      "epoch": 112.0,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.78851717710495,
+      "eval_runtime": 9.2483,
+      "eval_samples_per_second": 48.009,
+      "eval_steps_per_second": 1.514,
+      "step": 3500
+    },
+    {
+      "epoch": 112.32,
+      "grad_norm": 9.410392761230469,
+      "learning_rate": 1.3620071684587816e-05,
+      "loss": 0.0411,
+      "step": 3510
+    },
+    {
+      "epoch": 112.64,
+      "grad_norm": 3.4295830726623535,
+      "learning_rate": 1.3500597371565113e-05,
+      "loss": 0.0465,
+      "step": 3520
+    },
+    {
+      "epoch": 112.96,
+      "grad_norm": 6.670384407043457,
+      "learning_rate": 1.3381123058542414e-05,
+      "loss": 0.035,
+      "step": 3530
+    },
+    {
+      "epoch": 112.992,
+      "eval_accuracy": 0.8468468468468469,
+      "eval_loss": 0.7178235650062561,
+      "eval_runtime": 9.1762,
+      "eval_samples_per_second": 48.386,
+      "eval_steps_per_second": 1.526,
+      "step": 3531
+    },
+    {
+      "epoch": 113.28,
+      "grad_norm": 6.2322468757629395,
+      "learning_rate": 1.3261648745519714e-05,
+      "loss": 0.0626,
+      "step": 3540
+    },
+    {
+      "epoch": 113.6,
+      "grad_norm": 5.694119453430176,
+      "learning_rate": 1.3142174432497015e-05,
+      "loss": 0.0638,
+      "step": 3550
+    },
+    {
+      "epoch": 113.92,
+      "grad_norm": 5.017783164978027,
+      "learning_rate": 1.3022700119474312e-05,
+      "loss": 0.0604,
+      "step": 3560
+    },
+    {
+      "epoch": 113.984,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7574167251586914,
+      "eval_runtime": 9.3469,
+      "eval_samples_per_second": 47.502,
+      "eval_steps_per_second": 1.498,
+      "step": 3562
+    },
+    {
+      "epoch": 114.24,
+      "grad_norm": 9.012121200561523,
+      "learning_rate": 1.2903225806451613e-05,
+      "loss": 0.0628,
+      "step": 3570
+    },
+    {
+      "epoch": 114.56,
+      "grad_norm": 6.070770740509033,
+      "learning_rate": 1.2783751493428913e-05,
+      "loss": 0.0559,
+      "step": 3580
+    },
+    {
+      "epoch": 114.88,
+      "grad_norm": 6.4133477210998535,
+      "learning_rate": 1.2664277180406214e-05,
+      "loss": 0.0507,
+      "step": 3590
+    },
+    {
+      "epoch": 114.976,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.790080726146698,
+      "eval_runtime": 9.284,
+      "eval_samples_per_second": 47.824,
+      "eval_steps_per_second": 1.508,
+      "step": 3593
+    },
+    {
+      "epoch": 115.2,
+      "grad_norm": 10.017669677734375,
+      "learning_rate": 1.2544802867383513e-05,
+      "loss": 0.0737,
+      "step": 3600
+    },
+    {
+      "epoch": 115.52,
+      "grad_norm": 11.685331344604492,
+      "learning_rate": 1.2425328554360813e-05,
+      "loss": 0.0551,
+      "step": 3610
+    },
+    {
+      "epoch": 115.84,
+      "grad_norm": 6.028275489807129,
+      "learning_rate": 1.2305854241338114e-05,
+      "loss": 0.05,
+      "step": 3620
+    },
+    {
+      "epoch": 116.0,
+      "eval_accuracy": 0.8198198198198198,
+      "eval_loss": 0.7729606032371521,
+      "eval_runtime": 9.2626,
+      "eval_samples_per_second": 47.935,
+      "eval_steps_per_second": 1.511,
+      "step": 3625
+    },
+    {
+      "epoch": 116.16,
+      "grad_norm": 20.972734451293945,
+      "learning_rate": 1.2186379928315413e-05,
+      "loss": 0.053,
+      "step": 3630
+    },
+    {
+      "epoch": 116.48,
+      "grad_norm": 21.527067184448242,
+      "learning_rate": 1.2066905615292714e-05,
+      "loss": 0.061,
+      "step": 3640
+    },
+    {
+      "epoch": 116.8,
+      "grad_norm": 2.8793394565582275,
+      "learning_rate": 1.1947431302270013e-05,
+      "loss": 0.0465,
+      "step": 3650
+    },
+    {
+      "epoch": 116.992,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.7966709136962891,
+      "eval_runtime": 9.2609,
+      "eval_samples_per_second": 47.944,
+      "eval_steps_per_second": 1.512,
+      "step": 3656
+    },
+    {
+      "epoch": 117.12,
+      "grad_norm": 10.335177421569824,
+      "learning_rate": 1.1827956989247313e-05,
+      "loss": 0.057,
+      "step": 3660
+    },
+    {
+      "epoch": 117.44,
+      "grad_norm": 10.481749534606934,
+      "learning_rate": 1.1708482676224612e-05,
+      "loss": 0.0417,
+      "step": 3670
+    },
+    {
+      "epoch": 117.76,
+      "grad_norm": 8.608804702758789,
+      "learning_rate": 1.1589008363201913e-05,
+      "loss": 0.042,
+      "step": 3680
+    },
+    {
+      "epoch": 117.984,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7766792178153992,
+      "eval_runtime": 9.2816,
+      "eval_samples_per_second": 47.837,
+      "eval_steps_per_second": 1.508,
+      "step": 3687
+    },
+    {
+      "epoch": 118.08,
+      "grad_norm": 10.435688972473145,
+      "learning_rate": 1.1469534050179212e-05,
+      "loss": 0.0474,
+      "step": 3690
+    },
+    {
+      "epoch": 118.4,
+      "grad_norm": 3.1348280906677246,
+      "learning_rate": 1.135005973715651e-05,
+      "loss": 0.0528,
+      "step": 3700
+    },
+    {
+      "epoch": 118.72,
+      "grad_norm": 17.36944007873535,
+      "learning_rate": 1.1230585424133811e-05,
+      "loss": 0.0609,
+      "step": 3710
+    },
+    {
+      "epoch": 118.976,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.7871721982955933,
+      "eval_runtime": 9.3483,
+      "eval_samples_per_second": 47.495,
+      "eval_steps_per_second": 1.498,
+      "step": 3718
+    },
+    {
+      "epoch": 119.04,
+      "grad_norm": 5.15559720993042,
+      "learning_rate": 1.1111111111111112e-05,
+      "loss": 0.0544,
+      "step": 3720
+    },
+    {
+      "epoch": 119.36,
+      "grad_norm": 9.22418212890625,
+      "learning_rate": 1.0991636798088412e-05,
+      "loss": 0.0358,
+      "step": 3730
+    },
+    {
+      "epoch": 119.68,
+      "grad_norm": 9.791460037231445,
+      "learning_rate": 1.0872162485065711e-05,
+      "loss": 0.0579,
+      "step": 3740
+    },
+    {
+      "epoch": 120.0,
+      "grad_norm": 8.853157997131348,
+      "learning_rate": 1.0752688172043012e-05,
+      "loss": 0.0379,
+      "step": 3750
+    },
+    {
+      "epoch": 120.0,
+      "eval_accuracy": 0.8513513513513513,
+      "eval_loss": 0.7684924602508545,
+      "eval_runtime": 9.2729,
+      "eval_samples_per_second": 47.882,
+      "eval_steps_per_second": 1.51,
+      "step": 3750
+    },
+    {
+      "epoch": 120.32,
+      "grad_norm": 11.307464599609375,
+      "learning_rate": 1.063321385902031e-05,
+      "loss": 0.0509,
+      "step": 3760
+    },
+    {
+      "epoch": 120.64,
+      "grad_norm": 7.135359287261963,
+      "learning_rate": 1.0513739545997611e-05,
+      "loss": 0.0359,
+      "step": 3770
+    },
+    {
+      "epoch": 120.96,
+      "grad_norm": 8.934365272521973,
+      "learning_rate": 1.039426523297491e-05,
+      "loss": 0.0579,
+      "step": 3780
+    },
+    {
+      "epoch": 120.992,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7708752751350403,
+      "eval_runtime": 9.1957,
+      "eval_samples_per_second": 48.283,
+      "eval_steps_per_second": 1.522,
+      "step": 3781
+    },
+    {
+      "epoch": 121.28,
+      "grad_norm": 10.474456787109375,
+      "learning_rate": 1.027479091995221e-05,
+      "loss": 0.0457,
+      "step": 3790
+    },
+    {
+      "epoch": 121.6,
+      "grad_norm": 7.360742092132568,
+      "learning_rate": 1.015531660692951e-05,
+      "loss": 0.0542,
+      "step": 3800
+    },
+    {
+      "epoch": 121.92,
+      "grad_norm": 8.417169570922852,
+      "learning_rate": 1.003584229390681e-05,
+      "loss": 0.0471,
+      "step": 3810
+    },
+    {
+      "epoch": 121.984,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7601270079612732,
+      "eval_runtime": 9.6072,
+      "eval_samples_per_second": 46.215,
+      "eval_steps_per_second": 1.457,
+      "step": 3812
+    },
+    {
+      "epoch": 122.24,
+      "grad_norm": 3.6518542766571045,
+      "learning_rate": 9.916367980884111e-06,
+      "loss": 0.0537,
+      "step": 3820
+    },
+    {
+      "epoch": 122.56,
+      "grad_norm": 1.70374596118927,
+      "learning_rate": 9.796893667861412e-06,
+      "loss": 0.0381,
+      "step": 3830
+    },
+    {
+      "epoch": 122.88,
+      "grad_norm": 6.339093208312988,
+      "learning_rate": 9.67741935483871e-06,
+      "loss": 0.0488,
+      "step": 3840
+    },
+    {
+      "epoch": 122.976,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.8230583667755127,
+      "eval_runtime": 9.2102,
+      "eval_samples_per_second": 48.208,
+      "eval_steps_per_second": 1.52,
+      "step": 3843
+    },
+    {
+      "epoch": 123.2,
+      "grad_norm": 7.372219562530518,
+      "learning_rate": 9.557945041816011e-06,
+      "loss": 0.0487,
+      "step": 3850
+    },
+    {
+      "epoch": 123.52,
+      "grad_norm": 10.117424964904785,
+      "learning_rate": 9.43847072879331e-06,
+      "loss": 0.0617,
+      "step": 3860
+    },
+    {
+      "epoch": 123.84,
+      "grad_norm": 8.506390571594238,
+      "learning_rate": 9.31899641577061e-06,
+      "loss": 0.0531,
+      "step": 3870
+    },
+    {
+      "epoch": 124.0,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.8015686869621277,
+      "eval_runtime": 9.327,
+      "eval_samples_per_second": 47.604,
+      "eval_steps_per_second": 1.501,
+      "step": 3875
+    },
+    {
+      "epoch": 124.16,
+      "grad_norm": 2.586430788040161,
+      "learning_rate": 9.19952210274791e-06,
+      "loss": 0.0484,
+      "step": 3880
+    },
+    {
+      "epoch": 124.48,
+      "grad_norm": 21.505414962768555,
+      "learning_rate": 9.08004778972521e-06,
+      "loss": 0.0432,
+      "step": 3890
+    },
+    {
+      "epoch": 124.8,
+      "grad_norm": 6.578880786895752,
+      "learning_rate": 8.960573476702509e-06,
+      "loss": 0.0446,
+      "step": 3900
+    },
+    {
+      "epoch": 124.992,
+      "eval_accuracy": 0.8423423423423423,
+      "eval_loss": 0.7805784940719604,
+      "eval_runtime": 9.2689,
+      "eval_samples_per_second": 47.902,
+      "eval_steps_per_second": 1.51,
+      "step": 3906
+    },
+    {
+      "epoch": 125.12,
+      "grad_norm": 14.002837181091309,
+      "learning_rate": 8.84109916367981e-06,
+      "loss": 0.0499,
+      "step": 3910
+    },
+    {
+      "epoch": 125.44,
+      "grad_norm": 9.619620323181152,
+      "learning_rate": 8.721624850657108e-06,
+      "loss": 0.0635,
+      "step": 3920
+    },
+    {
+      "epoch": 125.76,
+      "grad_norm": 10.778247833251953,
+      "learning_rate": 8.602150537634409e-06,
+      "loss": 0.0479,
+      "step": 3930
+    },
+    {
+      "epoch": 125.984,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.7668151259422302,
+      "eval_runtime": 9.1034,
+      "eval_samples_per_second": 48.773,
+      "eval_steps_per_second": 1.538,
+      "step": 3937
+    },
+    {
+      "epoch": 126.08,
+      "grad_norm": 7.485611438751221,
+      "learning_rate": 8.48267622461171e-06,
+      "loss": 0.0602,
+      "step": 3940
+    },
+    {
+      "epoch": 126.4,
+      "grad_norm": 4.371471405029297,
+      "learning_rate": 8.363201911589009e-06,
+      "loss": 0.0607,
+      "step": 3950
+    },
+    {
+      "epoch": 126.72,
+      "grad_norm": 21.91805648803711,
+      "learning_rate": 8.24372759856631e-06,
+      "loss": 0.0525,
+      "step": 3960
+    },
+    {
+      "epoch": 126.976,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.7874143719673157,
+      "eval_runtime": 9.4361,
+      "eval_samples_per_second": 47.053,
+      "eval_steps_per_second": 1.484,
+      "step": 3968
+    },
+    {
+      "epoch": 127.04,
+      "grad_norm": 17.411008834838867,
+      "learning_rate": 8.124253285543608e-06,
+      "loss": 0.0577,
+      "step": 3970
+    },
+    {
+      "epoch": 127.36,
+      "grad_norm": 3.1272220611572266,
+      "learning_rate": 8.004778972520909e-06,
+      "loss": 0.0504,
+      "step": 3980
+    },
+    {
+      "epoch": 127.68,
+      "grad_norm": 2.4175899028778076,
+      "learning_rate": 7.885304659498208e-06,
+      "loss": 0.053,
+      "step": 3990
+    },
+    {
+      "epoch": 128.0,
+      "grad_norm": 10.839136123657227,
+      "learning_rate": 7.765830346475508e-06,
+      "loss": 0.0512,
+      "step": 4000
+    },
+    {
+      "epoch": 128.0,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.7651507258415222,
+      "eval_runtime": 9.2854,
+      "eval_samples_per_second": 47.817,
+      "eval_steps_per_second": 1.508,
+      "step": 4000
+    },
+    {
+      "epoch": 128.32,
+      "grad_norm": 8.561548233032227,
+      "learning_rate": 7.646356033452807e-06,
+      "loss": 0.0456,
+      "step": 4010
+    },
+    {
+      "epoch": 128.64,
+      "grad_norm": 6.735084533691406,
+      "learning_rate": 7.526881720430108e-06,
+      "loss": 0.0403,
+      "step": 4020
+    },
+    {
+      "epoch": 128.96,
+      "grad_norm": 10.706457138061523,
+      "learning_rate": 7.4074074074074075e-06,
+      "loss": 0.0473,
+      "step": 4030
+    },
+    {
+      "epoch": 128.992,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7721133828163147,
+      "eval_runtime": 9.1988,
+      "eval_samples_per_second": 48.267,
+      "eval_steps_per_second": 1.522,
+      "step": 4031
+    },
+    {
+      "epoch": 129.28,
+      "grad_norm": 5.523827075958252,
+      "learning_rate": 7.287933094384708e-06,
+      "loss": 0.0368,
+      "step": 4040
+    },
+    {
+      "epoch": 129.6,
+      "grad_norm": 7.4960103034973145,
+      "learning_rate": 7.168458781362007e-06,
+      "loss": 0.0439,
+      "step": 4050
+    },
+    {
+      "epoch": 129.92,
+      "grad_norm": 12.937583923339844,
+      "learning_rate": 7.048984468339308e-06,
+      "loss": 0.0579,
+      "step": 4060
+    },
+    {
+      "epoch": 129.984,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7606847286224365,
+      "eval_runtime": 9.3067,
+      "eval_samples_per_second": 47.708,
+      "eval_steps_per_second": 1.504,
+      "step": 4062
+    },
+    {
+      "epoch": 130.24,
+      "grad_norm": 10.424357414245605,
+      "learning_rate": 6.929510155316607e-06,
+      "loss": 0.0328,
+      "step": 4070
+    },
+    {
+      "epoch": 130.56,
+      "grad_norm": 4.310163974761963,
+      "learning_rate": 6.810035842293908e-06,
+      "loss": 0.0356,
+      "step": 4080
+    },
+    {
+      "epoch": 130.88,
+      "grad_norm": 3.945925235748291,
+      "learning_rate": 6.690561529271207e-06,
+      "loss": 0.0444,
+      "step": 4090
+    },
+    {
+      "epoch": 130.976,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.7916550040245056,
+      "eval_runtime": 9.2498,
+      "eval_samples_per_second": 48.001,
+      "eval_steps_per_second": 1.514,
+      "step": 4093
+    },
+    {
+      "epoch": 131.2,
+      "grad_norm": 8.806529998779297,
+      "learning_rate": 6.5710872162485075e-06,
+      "loss": 0.0383,
+      "step": 4100
+    },
+    {
+      "epoch": 131.52,
+      "grad_norm": 4.7513322830200195,
+      "learning_rate": 6.451612903225806e-06,
+      "loss": 0.0536,
+      "step": 4110
+    },
+    {
+      "epoch": 131.84,
+      "grad_norm": 8.66804313659668,
+      "learning_rate": 6.332138590203107e-06,
+      "loss": 0.0462,
+      "step": 4120
+    },
+    {
+      "epoch": 132.0,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.7877137660980225,
+      "eval_runtime": 9.4541,
+      "eval_samples_per_second": 46.964,
+      "eval_steps_per_second": 1.481,
+      "step": 4125
+    },
+    {
+      "epoch": 132.16,
+      "grad_norm": 4.396764278411865,
+      "learning_rate": 6.212664277180407e-06,
+      "loss": 0.0365,
+      "step": 4130
+    },
+    {
+      "epoch": 132.48,
+      "grad_norm": 5.487627983093262,
+      "learning_rate": 6.0931899641577065e-06,
+      "loss": 0.0361,
+      "step": 4140
+    },
+    {
+      "epoch": 132.8,
+      "grad_norm": 9.7935791015625,
+      "learning_rate": 5.973715651135006e-06,
+      "loss": 0.0483,
+      "step": 4150
+    },
+    {
+      "epoch": 132.992,
+      "eval_accuracy": 0.8400900900900901,
+      "eval_loss": 0.8122358322143555,
+      "eval_runtime": 9.4201,
+      "eval_samples_per_second": 47.133,
+      "eval_steps_per_second": 1.486,
+      "step": 4156
+    },
+    {
+      "epoch": 133.12,
+      "grad_norm": 24.21430206298828,
+      "learning_rate": 5.854241338112306e-06,
+      "loss": 0.0441,
+      "step": 4160
+    },
+    {
+      "epoch": 133.44,
+      "grad_norm": 10.01777458190918,
+      "learning_rate": 5.734767025089606e-06,
+      "loss": 0.0517,
+      "step": 4170
+    },
+    {
+      "epoch": 133.76,
+      "grad_norm": 8.408638000488281,
+      "learning_rate": 5.6152927120669055e-06,
+      "loss": 0.042,
+      "step": 4180
+    },
+    {
+      "epoch": 133.984,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.7956117987632751,
+      "eval_runtime": 9.2257,
+      "eval_samples_per_second": 48.126,
+      "eval_steps_per_second": 1.518,
+      "step": 4187
+    },
+    {
+      "epoch": 134.08,
+      "grad_norm": 11.456778526306152,
+      "learning_rate": 5.495818399044206e-06,
+      "loss": 0.0511,
+      "step": 4190
+    },
+    {
+      "epoch": 134.4,
+      "grad_norm": 9.832137107849121,
+      "learning_rate": 5.376344086021506e-06,
+      "loss": 0.0402,
+      "step": 4200
+    },
+    {
+      "epoch": 134.72,
+      "grad_norm": 6.971923828125,
+      "learning_rate": 5.256869772998806e-06,
+      "loss": 0.0439,
+      "step": 4210
+    },
+    {
+      "epoch": 134.976,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.8281213641166687,
+      "eval_runtime": 9.3773,
+      "eval_samples_per_second": 47.348,
+      "eval_steps_per_second": 1.493,
+      "step": 4218
+    },
+    {
+      "epoch": 135.04,
+      "grad_norm": 9.771930694580078,
+      "learning_rate": 5.137395459976105e-06,
+      "loss": 0.053,
+      "step": 4220
+    },
+    {
+      "epoch": 135.36,
+      "grad_norm": 17.273054122924805,
+      "learning_rate": 5.017921146953405e-06,
+      "loss": 0.0351,
+      "step": 4230
+    },
+    {
+      "epoch": 135.68,
+      "grad_norm": 16.49187660217285,
+      "learning_rate": 4.898446833930706e-06,
+      "loss": 0.0375,
+      "step": 4240
+    },
+    {
+      "epoch": 136.0,
+      "grad_norm": 4.234256744384766,
+      "learning_rate": 4.7789725209080055e-06,
+      "loss": 0.0458,
+      "step": 4250
+    },
+    {
+      "epoch": 136.0,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.7723496556282043,
+      "eval_runtime": 9.1454,
+      "eval_samples_per_second": 48.549,
+      "eval_steps_per_second": 1.531,
+      "step": 4250
+    },
+    {
+      "epoch": 136.32,
+      "grad_norm": 6.829943656921387,
+      "learning_rate": 4.659498207885305e-06,
+      "loss": 0.0605,
+      "step": 4260
+    },
+    {
+      "epoch": 136.64,
+      "grad_norm": 6.178986072540283,
+      "learning_rate": 4.540023894862605e-06,
+      "loss": 0.0387,
+      "step": 4270
+    },
+    {
+      "epoch": 136.96,
+      "grad_norm": 6.282145023345947,
+      "learning_rate": 4.420549581839905e-06,
+      "loss": 0.0307,
+      "step": 4280
+    },
+    {
+      "epoch": 136.992,
+      "eval_accuracy": 0.8445945945945946,
+      "eval_loss": 0.7685997486114502,
+      "eval_runtime": 9.3166,
+      "eval_samples_per_second": 47.657,
+      "eval_steps_per_second": 1.503,
+      "step": 4281
+    },
+    {
+      "epoch": 137.28,
+      "grad_norm": 2.165778636932373,
+      "learning_rate": 4.3010752688172045e-06,
+      "loss": 0.0423,
+      "step": 4290
+    },
+    {
+      "epoch": 137.6,
+      "grad_norm": 2.3030285835266113,
+      "learning_rate": 4.181600955794504e-06,
+      "loss": 0.0381,
+      "step": 4300
+    },
+    {
+      "epoch": 137.92,
+      "grad_norm": 13.007665634155273,
+      "learning_rate": 4.062126642771804e-06,
+      "loss": 0.0481,
+      "step": 4310
+    },
+    {
+      "epoch": 137.984,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.7834463715553284,
+      "eval_runtime": 9.1579,
+      "eval_samples_per_second": 48.483,
+      "eval_steps_per_second": 1.529,
+      "step": 4312
+    },
+    {
+      "epoch": 138.24,
+      "grad_norm": 6.034422874450684,
+      "learning_rate": 3.942652329749104e-06,
+      "loss": 0.0428,
+      "step": 4320
+    },
+    {
+      "epoch": 138.56,
+      "grad_norm": 5.156213283538818,
+      "learning_rate": 3.8231780167264036e-06,
+      "loss": 0.0524,
+      "step": 4330
+    },
+    {
+      "epoch": 138.88,
+      "grad_norm": 7.5509443283081055,
+      "learning_rate": 3.7037037037037037e-06,
+      "loss": 0.0503,
+      "step": 4340
+    },
+    {
+      "epoch": 138.976,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.7986668944358826,
+      "eval_runtime": 9.2264,
+      "eval_samples_per_second": 48.123,
+      "eval_steps_per_second": 1.517,
+      "step": 4343
+    },
+    {
+      "epoch": 139.2,
+      "grad_norm": 3.5894615650177,
+      "learning_rate": 3.5842293906810035e-06,
+      "loss": 0.0367,
+      "step": 4350
+    },
+    {
+      "epoch": 139.52,
+      "grad_norm": 10.714777946472168,
+      "learning_rate": 3.4647550776583037e-06,
+      "loss": 0.0451,
+      "step": 4360
+    },
+    {
+      "epoch": 139.84,
+      "grad_norm": 20.11847496032715,
+      "learning_rate": 3.3452807646356034e-06,
+      "loss": 0.038,
+      "step": 4370
+    },
+    {
+      "epoch": 140.0,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.8156119585037231,
+      "eval_runtime": 9.2868,
+      "eval_samples_per_second": 47.81,
+      "eval_steps_per_second": 1.508,
+      "step": 4375
+    },
+    {
+      "epoch": 140.16,
+      "grad_norm": 5.215194225311279,
+      "learning_rate": 3.225806451612903e-06,
+      "loss": 0.0475,
+      "step": 4380
+    },
+    {
+      "epoch": 140.48,
+      "grad_norm": 7.803160667419434,
+      "learning_rate": 3.1063321385902034e-06,
+      "loss": 0.0391,
+      "step": 4390
+    },
+    {
+      "epoch": 140.8,
+      "grad_norm": 1.6925849914550781,
+      "learning_rate": 2.986857825567503e-06,
+      "loss": 0.0472,
+      "step": 4400
+    },
+    {
+      "epoch": 140.992,
+      "eval_accuracy": 0.8355855855855856,
+      "eval_loss": 0.802955150604248,
+      "eval_runtime": 9.2568,
+      "eval_samples_per_second": 47.965,
+      "eval_steps_per_second": 1.512,
+      "step": 4406
+    },
+    {
+      "epoch": 141.12,
+      "grad_norm": 6.969939231872559,
+      "learning_rate": 2.867383512544803e-06,
+      "loss": 0.0448,
+      "step": 4410
+    },
+    {
+      "epoch": 141.44,
+      "grad_norm": 5.957333087921143,
+      "learning_rate": 2.747909199522103e-06,
+      "loss": 0.0499,
+      "step": 4420
+    },
+    {
+      "epoch": 141.76,
+      "grad_norm": 6.240506172180176,
+      "learning_rate": 2.628434886499403e-06,
+      "loss": 0.0282,
+      "step": 4430
+    },
+    {
+      "epoch": 141.984,
+      "eval_accuracy": 0.8378378378378378,
+      "eval_loss": 0.7883952260017395,
+      "eval_runtime": 9.4327,
+      "eval_samples_per_second": 47.07,
+      "eval_steps_per_second": 1.484,
+      "step": 4437
+    },
+    {
+      "epoch": 142.08,
+      "grad_norm": 3.7007412910461426,
+      "learning_rate": 2.5089605734767026e-06,
+      "loss": 0.0417,
+      "step": 4440
+    },
+    {
+      "epoch": 142.4,
+      "grad_norm": 9.855685234069824,
+      "learning_rate": 2.3894862604540028e-06,
+      "loss": 0.0439,
+      "step": 4450
+    },
+    {
+      "epoch": 142.72,
+      "grad_norm": 3.489447832107544,
+      "learning_rate": 2.2700119474313025e-06,
+      "loss": 0.0541,
+      "step": 4460
+    },
+    {
+      "epoch": 142.976,
+      "eval_accuracy": 0.831081081081081,
+      "eval_loss": 0.7968920469284058,
+      "eval_runtime": 9.2527,
+      "eval_samples_per_second": 47.986,
+      "eval_steps_per_second": 1.513,
+      "step": 4468
+    },
+    {
+      "epoch": 143.04,
+      "grad_norm": 3.9319980144500732,
+      "learning_rate": 2.1505376344086023e-06,
+      "loss": 0.0368,
+      "step": 4470
+    },
+    {
+      "epoch": 143.36,
+      "grad_norm": 12.418336868286133,
+      "learning_rate": 2.031063321385902e-06,
+      "loss": 0.0279,
+      "step": 4480
+    },
+    {
+      "epoch": 143.68,
+      "grad_norm": 1.1635448932647705,
+      "learning_rate": 1.9115890083632018e-06,
+      "loss": 0.037,
+      "step": 4490
+    },
+    {
+      "epoch": 144.0,
+      "grad_norm": 6.279101371765137,
+      "learning_rate": 1.7921146953405017e-06,
+      "loss": 0.0415,
+      "step": 4500
+    },
+    {
+      "epoch": 144.0,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.789898693561554,
+      "eval_runtime": 9.2202,
+      "eval_samples_per_second": 48.155,
+      "eval_steps_per_second": 1.518,
+      "step": 4500
+    },
+    {
+      "epoch": 144.32,
+      "grad_norm": 1.137380838394165,
+      "learning_rate": 1.6726403823178017e-06,
+      "loss": 0.0449,
+      "step": 4510
+    },
+    {
+      "epoch": 144.64,
+      "grad_norm": 11.844816207885742,
+      "learning_rate": 1.5531660692951017e-06,
+      "loss": 0.0322,
+      "step": 4520
+    },
+    {
+      "epoch": 144.96,
+      "grad_norm": 10.018691062927246,
+      "learning_rate": 1.4336917562724014e-06,
+      "loss": 0.0579,
+      "step": 4530
+    },
+    {
+      "epoch": 144.992,
+      "eval_accuracy": 0.8265765765765766,
+      "eval_loss": 0.7978772521018982,
+      "eval_runtime": 9.4369,
+      "eval_samples_per_second": 47.05,
+      "eval_steps_per_second": 1.484,
+      "step": 4531
+    },
+    {
+      "epoch": 145.28,
+      "grad_norm": 10.856122016906738,
+      "learning_rate": 1.3142174432497014e-06,
+      "loss": 0.0534,
+      "step": 4540
+    },
+    {
+      "epoch": 145.6,
+      "grad_norm": 12.98085880279541,
+      "learning_rate": 1.1947431302270014e-06,
+      "loss": 0.045,
+      "step": 4550
+    },
+    {
+      "epoch": 145.92,
+      "grad_norm": 9.042325019836426,
+      "learning_rate": 1.0752688172043011e-06,
+      "loss": 0.048,
+      "step": 4560
+    },
+    {
+      "epoch": 145.984,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.793506383895874,
+      "eval_runtime": 9.2712,
+      "eval_samples_per_second": 47.89,
+      "eval_steps_per_second": 1.51,
+      "step": 4562
+    },
+    {
+      "epoch": 146.24,
+      "grad_norm": 6.015984535217285,
+      "learning_rate": 9.557945041816009e-07,
+      "loss": 0.0561,
+      "step": 4570
+    },
+    {
+      "epoch": 146.56,
+      "grad_norm": 13.623724937438965,
+      "learning_rate": 8.363201911589009e-07,
+      "loss": 0.0435,
+      "step": 4580
+    },
+    {
+      "epoch": 146.88,
+      "grad_norm": 7.4142842292785645,
+      "learning_rate": 7.168458781362007e-07,
+      "loss": 0.0353,
+      "step": 4590
+    },
+    {
+      "epoch": 146.976,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.793285608291626,
+      "eval_runtime": 9.135,
+      "eval_samples_per_second": 48.604,
+      "eval_steps_per_second": 1.533,
+      "step": 4593
+    },
+    {
+      "epoch": 147.2,
+      "grad_norm": 15.34078598022461,
+      "learning_rate": 5.973715651135007e-07,
+      "loss": 0.0294,
+      "step": 4600
+    },
+    {
+      "epoch": 147.52,
+      "grad_norm": 5.058185577392578,
+      "learning_rate": 4.778972520908004e-07,
+      "loss": 0.0447,
+      "step": 4610
+    },
+    {
+      "epoch": 147.84,
+      "grad_norm": 14.668283462524414,
+      "learning_rate": 3.5842293906810036e-07,
+      "loss": 0.0438,
+      "step": 4620
+    },
+    {
+      "epoch": 148.0,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.7915613651275635,
+      "eval_runtime": 9.1582,
+      "eval_samples_per_second": 48.481,
+      "eval_steps_per_second": 1.529,
+      "step": 4625
+    },
+    {
+      "epoch": 148.16,
+      "grad_norm": 8.498198509216309,
+      "learning_rate": 2.389486260454002e-07,
+      "loss": 0.0458,
+      "step": 4630
+    },
+    {
+      "epoch": 148.48,
+      "grad_norm": 3.9683709144592285,
+      "learning_rate": 1.194743130227001e-07,
+      "loss": 0.0298,
+      "step": 4640
+    },
+    {
+      "epoch": 148.8,
+      "grad_norm": 21.15007209777832,
+      "learning_rate": 0.0,
+      "loss": 0.0487,
+      "step": 4650
+    },
+    {
+      "epoch": 148.8,
+      "eval_accuracy": 0.8288288288288288,
+      "eval_loss": 0.792281985282898,
+      "eval_runtime": 9.2874,
+      "eval_samples_per_second": 47.806,
+      "eval_steps_per_second": 1.507,
+      "step": 4650
     },
     {
-      "epoch": 86.66666666666667,
-      "step": 130,
-      "total_flos": 1.574865655328932e+17,
-      "train_loss": 0.3546037518061124,
-      "train_runtime": 225.362,
-      "train_samples_per_second": 41.533,
-      "train_steps_per_second": 0.577
+      "epoch": 148.8,
+      "step": 4650,
+      "total_flos": 1.4980450073050055e+19,
+      "train_loss": 0.38013411539216196,
+      "train_runtime": 17110.8333,
+      "train_samples_per_second": 35.03,
+      "train_steps_per_second": 0.272
     }
   ],
   "logging_steps": 10,
-  "max_steps": 130,
+  "max_steps": 4650,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 130,
+  "num_train_epochs": 150,
   "save_steps": 500,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -909,7 +4631,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.574865655328932e+17,
+  "total_flos": 1.4980450073050055e+19,
   "train_batch_size": 32,
   "trial_name": null,
   "trial_params": null