diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4745 @@
+{
+  "best_metric": 0.4519657492637634,
+  "best_model_checkpoint": "Action_model_ViT_384/checkpoint-1400",
+  "epoch": 20.0,
+  "eval_steps": 100,
+  "global_step": 10720,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.04,
+      "grad_norm": 9.464336395263672,
+      "learning_rate": 9.983208955223881e-05,
+      "loss": 2.253,
+      "step": 20
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 11.0652437210083,
+      "learning_rate": 9.964552238805972e-05,
+      "loss": 1.725,
+      "step": 40
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 10.256282806396484,
+      "learning_rate": 9.945895522388061e-05,
+      "loss": 1.2015,
+      "step": 60
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 8.653861045837402,
+      "learning_rate": 9.927238805970149e-05,
+      "loss": 0.9538,
+      "step": 80
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 8.860688209533691,
+      "learning_rate": 9.908582089552239e-05,
+      "loss": 0.946,
+      "step": 100
+    },
+    {
+      "epoch": 0.19,
+      "eval_accuracy": 0.7803163444639719,
+      "eval_loss": 0.7539544105529785,
+      "eval_runtime": 21.8778,
+      "eval_samples_per_second": 26.008,
+      "eval_steps_per_second": 3.291,
+      "step": 100
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 8.424888610839844,
+      "learning_rate": 9.889925373134328e-05,
+      "loss": 0.985,
+      "step": 120
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 9.244832038879395,
+      "learning_rate": 9.871268656716418e-05,
+      "loss": 0.9033,
+      "step": 140
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 10.58477783203125,
+      "learning_rate": 9.852611940298508e-05,
+      "loss": 0.9291,
+      "step": 160
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 9.450798034667969,
+      "learning_rate": 9.833955223880598e-05,
+      "loss": 0.7907,
+      "step": 180
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 10.530333518981934,
+      "learning_rate": 9.815298507462687e-05,
+      "loss": 0.9248,
+      "step": 200
+    },
+    {
+      "epoch": 0.37,
+      "eval_accuracy": 0.7961335676625659,
+      "eval_loss": 0.6282442212104797,
+      "eval_runtime": 17.0774,
+      "eval_samples_per_second": 33.319,
+      "eval_steps_per_second": 4.216,
+      "step": 200
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 10.158860206604004,
+      "learning_rate": 9.796641791044776e-05,
+      "loss": 0.9931,
+      "step": 220
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.629036903381348,
+      "learning_rate": 9.777985074626867e-05,
+      "loss": 0.7287,
+      "step": 240
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 10.852154731750488,
+      "learning_rate": 9.759328358208956e-05,
+      "loss": 0.8955,
+      "step": 260
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 4.645742416381836,
+      "learning_rate": 9.740671641791045e-05,
+      "loss": 0.8936,
+      "step": 280
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 8.988308906555176,
+      "learning_rate": 9.722014925373136e-05,
+      "loss": 0.7968,
+      "step": 300
+    },
+    {
+      "epoch": 0.56,
+      "eval_accuracy": 0.8101933216168717,
+      "eval_loss": 0.5833593606948853,
+      "eval_runtime": 17.2737,
+      "eval_samples_per_second": 32.94,
+      "eval_steps_per_second": 4.168,
+      "step": 300
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.654094696044922,
+      "learning_rate": 9.703358208955225e-05,
+      "loss": 0.6716,
+      "step": 320
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 10.295234680175781,
+      "learning_rate": 9.684701492537314e-05,
+      "loss": 0.8094,
+      "step": 340
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.079020977020264,
+      "learning_rate": 9.666044776119403e-05,
+      "loss": 0.7198,
+      "step": 360
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 8.492669105529785,
+      "learning_rate": 9.647388059701493e-05,
+      "loss": 0.9095,
+      "step": 380
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.336089611053467,
+      "learning_rate": 9.628731343283582e-05,
+      "loss": 0.6992,
+      "step": 400
+    },
+    {
+      "epoch": 0.75,
+      "eval_accuracy": 0.8330404217926186,
+      "eval_loss": 0.564700186252594,
+      "eval_runtime": 17.0405,
+      "eval_samples_per_second": 33.391,
+      "eval_steps_per_second": 4.225,
+      "step": 400
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 8.37447452545166,
+      "learning_rate": 9.610074626865671e-05,
+      "loss": 0.6298,
+      "step": 420
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 16.46346092224121,
+      "learning_rate": 9.592350746268657e-05,
+      "loss": 0.849,
+      "step": 440
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 11.869742393493652,
+      "learning_rate": 9.573694029850747e-05,
+      "loss": 0.7196,
+      "step": 460
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 5.501863956451416,
+      "learning_rate": 9.555037313432836e-05,
+      "loss": 0.6862,
+      "step": 480
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 3.11259388923645,
+      "learning_rate": 9.536380597014925e-05,
+      "loss": 0.7331,
+      "step": 500
+    },
+    {
+      "epoch": 0.93,
+      "eval_accuracy": 0.8295254833040422,
+      "eval_loss": 0.5430368781089783,
+      "eval_runtime": 17.5273,
+      "eval_samples_per_second": 32.464,
+      "eval_steps_per_second": 4.108,
+      "step": 500
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 10.489935874938965,
+      "learning_rate": 9.517723880597016e-05,
+      "loss": 0.7952,
+      "step": 520
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 5.573906898498535,
+      "learning_rate": 9.499067164179105e-05,
+      "loss": 0.5974,
+      "step": 540
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 8.94275188446045,
+      "learning_rate": 9.480410447761194e-05,
+      "loss": 0.5629,
+      "step": 560
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 7.229942321777344,
+      "learning_rate": 9.461753731343283e-05,
+      "loss": 0.4051,
+      "step": 580
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 20.218753814697266,
+      "learning_rate": 9.443097014925374e-05,
+      "loss": 0.5822,
+      "step": 600
+    },
+    {
+      "epoch": 1.12,
+      "eval_accuracy": 0.8172231985940246,
+      "eval_loss": 0.589419960975647,
+      "eval_runtime": 17.0675,
+      "eval_samples_per_second": 33.338,
+      "eval_steps_per_second": 4.219,
+      "step": 600
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 14.375499725341797,
+      "learning_rate": 9.424440298507463e-05,
+      "loss": 0.5939,
+      "step": 620
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 4.248530864715576,
+      "learning_rate": 9.405783582089552e-05,
+      "loss": 0.6569,
+      "step": 640
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 10.027786254882812,
+      "learning_rate": 9.387126865671643e-05,
+      "loss": 0.6345,
+      "step": 660
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 9.075023651123047,
+      "learning_rate": 9.368470149253732e-05,
+      "loss": 0.4834,
+      "step": 680
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 11.023707389831543,
+      "learning_rate": 9.349813432835822e-05,
+      "loss": 0.5906,
+      "step": 700
+    },
+    {
+      "epoch": 1.31,
+      "eval_accuracy": 0.7908611599297012,
+      "eval_loss": 0.6861735582351685,
+      "eval_runtime": 17.0305,
+      "eval_samples_per_second": 33.411,
+      "eval_steps_per_second": 4.228,
+      "step": 700
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 5.9840216636657715,
+      "learning_rate": 9.331156716417911e-05,
+      "loss": 0.7206,
+      "step": 720
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 11.753890991210938,
+      "learning_rate": 9.3125e-05,
+      "loss": 0.6115,
+      "step": 740
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 1.2064365148544312,
+      "learning_rate": 9.293843283582089e-05,
+      "loss": 0.4046,
+      "step": 760
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 7.887212753295898,
+      "learning_rate": 9.275186567164179e-05,
+      "loss": 0.6207,
+      "step": 780
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 16.808399200439453,
+      "learning_rate": 9.256529850746269e-05,
+      "loss": 0.5911,
+      "step": 800
+    },
+    {
+      "epoch": 1.49,
+      "eval_accuracy": 0.8312829525483304,
+      "eval_loss": 0.536864697933197,
+      "eval_runtime": 17.1181,
+      "eval_samples_per_second": 33.24,
+      "eval_steps_per_second": 4.206,
+      "step": 800
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 9.534643173217773,
+      "learning_rate": 9.237873134328358e-05,
+      "loss": 0.521,
+      "step": 820
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 8.557387351989746,
+      "learning_rate": 9.219216417910448e-05,
+      "loss": 0.8094,
+      "step": 840
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 10.042027473449707,
+      "learning_rate": 9.200559701492538e-05,
+      "loss": 0.5499,
+      "step": 860
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 8.789073944091797,
+      "learning_rate": 9.181902985074627e-05,
+      "loss": 0.4584,
+      "step": 880
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 1.6369884014129639,
+      "learning_rate": 9.163246268656717e-05,
+      "loss": 0.4564,
+      "step": 900
+    },
+    {
+      "epoch": 1.68,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.4656906723976135,
+      "eval_runtime": 17.1286,
+      "eval_samples_per_second": 33.219,
+      "eval_steps_per_second": 4.204,
+      "step": 900
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 11.877409934997559,
+      "learning_rate": 9.144589552238806e-05,
+      "loss": 0.558,
+      "step": 920
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 8.138932228088379,
+      "learning_rate": 9.125932835820897e-05,
+      "loss": 0.4452,
+      "step": 940
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 13.710027694702148,
+      "learning_rate": 9.107276119402986e-05,
+      "loss": 0.5451,
+      "step": 960
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 7.554599285125732,
+      "learning_rate": 9.088619402985075e-05,
+      "loss": 0.6241,
+      "step": 980
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 10.94112777709961,
+      "learning_rate": 9.069962686567166e-05,
+      "loss": 0.6416,
+      "step": 1000
+    },
+    {
+      "epoch": 1.87,
+      "eval_accuracy": 0.8189806678383128,
+      "eval_loss": 0.5697047114372253,
+      "eval_runtime": 17.1833,
+      "eval_samples_per_second": 33.114,
+      "eval_steps_per_second": 4.19,
+      "step": 1000
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 7.1410651206970215,
+      "learning_rate": 9.051305970149255e-05,
+      "loss": 0.5184,
+      "step": 1020
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 9.702417373657227,
+      "learning_rate": 9.032649253731344e-05,
+      "loss": 0.4046,
+      "step": 1040
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 12.749421119689941,
+      "learning_rate": 9.013992537313433e-05,
+      "loss": 0.614,
+      "step": 1060
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 7.126010417938232,
+      "learning_rate": 8.995335820895523e-05,
+      "loss": 0.7246,
+      "step": 1080
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 6.920120716094971,
+      "learning_rate": 8.976679104477612e-05,
+      "loss": 0.5653,
+      "step": 1100
+    },
+    {
+      "epoch": 2.05,
+      "eval_accuracy": 0.8101933216168717,
+      "eval_loss": 0.6152170896530151,
+      "eval_runtime": 17.0836,
+      "eval_samples_per_second": 33.307,
+      "eval_steps_per_second": 4.215,
+      "step": 1100
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 8.713808059692383,
+      "learning_rate": 8.958022388059701e-05,
+      "loss": 0.611,
+      "step": 1120
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 1.4127197265625,
+      "learning_rate": 8.939365671641792e-05,
+      "loss": 0.4275,
+      "step": 1140
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 5.731436729431152,
+      "learning_rate": 8.920708955223881e-05,
+      "loss": 0.5003,
+      "step": 1160
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.9889954328536987,
+      "learning_rate": 8.90205223880597e-05,
+      "loss": 0.4496,
+      "step": 1180
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 8.190879821777344,
+      "learning_rate": 8.883395522388061e-05,
+      "loss": 0.4145,
+      "step": 1200
+    },
+    {
+      "epoch": 2.24,
+      "eval_accuracy": 0.8224956063268892,
+      "eval_loss": 0.5792708992958069,
+      "eval_runtime": 16.9838,
+      "eval_samples_per_second": 33.503,
+      "eval_steps_per_second": 4.239,
+      "step": 1200
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 6.807558059692383,
+      "learning_rate": 8.86473880597015e-05,
+      "loss": 0.4217,
+      "step": 1220
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 10.359559059143066,
+      "learning_rate": 8.846082089552239e-05,
+      "loss": 0.4424,
+      "step": 1240
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 4.082479000091553,
+      "learning_rate": 8.82742537313433e-05,
+      "loss": 0.3953,
+      "step": 1260
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 4.206302642822266,
+      "learning_rate": 8.808768656716419e-05,
+      "loss": 0.2739,
+      "step": 1280
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 9.99765682220459,
+      "learning_rate": 8.790111940298508e-05,
+      "loss": 0.4743,
+      "step": 1300
+    },
+    {
+      "epoch": 2.43,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.46424412727355957,
+      "eval_runtime": 17.06,
+      "eval_samples_per_second": 33.353,
+      "eval_steps_per_second": 4.22,
+      "step": 1300
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 1.8531745672225952,
+      "learning_rate": 8.771455223880597e-05,
+      "loss": 0.519,
+      "step": 1320
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.8780223727226257,
+      "learning_rate": 8.752798507462687e-05,
+      "loss": 0.3979,
+      "step": 1340
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 8.526026725769043,
+      "learning_rate": 8.734141791044776e-05,
+      "loss": 0.4342,
+      "step": 1360
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 4.428990840911865,
+      "learning_rate": 8.715485074626865e-05,
+      "loss": 0.5077,
+      "step": 1380
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.5647484064102173,
+      "learning_rate": 8.696828358208956e-05,
+      "loss": 0.4908,
+      "step": 1400
+    },
+    {
+      "epoch": 2.61,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.4519657492637634,
+      "eval_runtime": 17.0325,
+      "eval_samples_per_second": 33.407,
+      "eval_steps_per_second": 4.227,
+      "step": 1400
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 7.867344379425049,
+      "learning_rate": 8.678171641791045e-05,
+      "loss": 0.581,
+      "step": 1420
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 4.0395307540893555,
+      "learning_rate": 8.659514925373134e-05,
+      "loss": 0.3672,
+      "step": 1440
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 7.903719425201416,
+      "learning_rate": 8.640858208955225e-05,
+      "loss": 0.5442,
+      "step": 1460
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 17.981998443603516,
+      "learning_rate": 8.622201492537314e-05,
+      "loss": 0.4823,
+      "step": 1480
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 11.074972152709961,
+      "learning_rate": 8.603544776119403e-05,
+      "loss": 0.523,
+      "step": 1500
+    },
+    {
+      "epoch": 2.8,
+      "eval_accuracy": 0.8453427065026362,
+      "eval_loss": 0.49891671538352966,
+      "eval_runtime": 17.1346,
+      "eval_samples_per_second": 33.208,
+      "eval_steps_per_second": 4.202,
+      "step": 1500
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 9.141737937927246,
+      "learning_rate": 8.584888059701493e-05,
+      "loss": 0.4788,
+      "step": 1520
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 6.037851333618164,
+      "learning_rate": 8.566231343283583e-05,
+      "loss": 0.5995,
+      "step": 1540
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 7.742834568023682,
+      "learning_rate": 8.547574626865672e-05,
+      "loss": 0.445,
+      "step": 1560
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 9.633835792541504,
+      "learning_rate": 8.528917910447762e-05,
+      "loss": 0.4413,
+      "step": 1580
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 4.5287766456604,
+      "learning_rate": 8.510261194029851e-05,
+      "loss": 0.3315,
+      "step": 1600
+    },
+    {
+      "epoch": 2.99,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.4785976707935333,
+      "eval_runtime": 17.2215,
+      "eval_samples_per_second": 33.04,
+      "eval_steps_per_second": 4.181,
+      "step": 1600
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 4.111365795135498,
+      "learning_rate": 8.49160447761194e-05,
+      "loss": 0.3112,
+      "step": 1620
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 7.275732517242432,
+      "learning_rate": 8.47294776119403e-05,
+      "loss": 0.2702,
+      "step": 1640
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 10.048295974731445,
+      "learning_rate": 8.45429104477612e-05,
+      "loss": 0.315,
+      "step": 1660
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.4275367856025696,
+      "learning_rate": 8.435634328358209e-05,
+      "loss": 0.4266,
+      "step": 1680
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 10.006832122802734,
+      "learning_rate": 8.416977611940298e-05,
+      "loss": 0.2779,
+      "step": 1700
+    },
+    {
+      "epoch": 3.17,
+      "eval_accuracy": 0.8523725834797891,
+      "eval_loss": 0.5545657277107239,
+      "eval_runtime": 17.1132,
+      "eval_samples_per_second": 33.249,
+      "eval_steps_per_second": 4.207,
+      "step": 1700
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 6.872674942016602,
+      "learning_rate": 8.398320895522388e-05,
+      "loss": 0.3257,
+      "step": 1720
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 0.44901320338249207,
+      "learning_rate": 8.379664179104478e-05,
+      "loss": 0.3833,
+      "step": 1740
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 20.12860679626465,
+      "learning_rate": 8.361007462686568e-05,
+      "loss": 0.535,
+      "step": 1760
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 0.7962126731872559,
+      "learning_rate": 8.342350746268657e-05,
+      "loss": 0.4591,
+      "step": 1780
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 0.08330605179071426,
+      "learning_rate": 8.323694029850747e-05,
+      "loss": 0.2984,
+      "step": 1800
+    },
+    {
+      "epoch": 3.36,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.49768441915512085,
+      "eval_runtime": 17.1627,
+      "eval_samples_per_second": 33.153,
+      "eval_steps_per_second": 4.195,
+      "step": 1800
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 7.635754585266113,
+      "learning_rate": 8.305037313432837e-05,
+      "loss": 0.3303,
+      "step": 1820
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 11.634245872497559,
+      "learning_rate": 8.286380597014926e-05,
+      "loss": 0.2729,
+      "step": 1840
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.621952533721924,
+      "learning_rate": 8.267723880597016e-05,
+      "loss": 0.3225,
+      "step": 1860
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 3.5288310050964355,
+      "learning_rate": 8.249067164179106e-05,
+      "loss": 0.4136,
+      "step": 1880
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 12.683182716369629,
+      "learning_rate": 8.230410447761195e-05,
+      "loss": 0.5914,
+      "step": 1900
+    },
+    {
+      "epoch": 3.54,
+      "eval_accuracy": 0.8224956063268892,
+      "eval_loss": 0.6296223402023315,
+      "eval_runtime": 17.5053,
+      "eval_samples_per_second": 32.504,
+      "eval_steps_per_second": 4.113,
+      "step": 1900
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 8.898299217224121,
+      "learning_rate": 8.211753731343284e-05,
+      "loss": 0.2644,
+      "step": 1920
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 12.375215530395508,
+      "learning_rate": 8.193097014925373e-05,
+      "loss": 0.5328,
+      "step": 1940
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 11.185277938842773,
+      "learning_rate": 8.174440298507463e-05,
+      "loss": 0.4125,
+      "step": 1960
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 0.18444909155368805,
+      "learning_rate": 8.155783582089552e-05,
+      "loss": 0.2861,
+      "step": 1980
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.5073598623275757,
+      "learning_rate": 8.137126865671643e-05,
+      "loss": 0.3236,
+      "step": 2000
+    },
+    {
+      "epoch": 3.73,
+      "eval_accuracy": 0.8172231985940246,
+      "eval_loss": 0.7224897742271423,
+      "eval_runtime": 17.2411,
+      "eval_samples_per_second": 33.002,
+      "eval_steps_per_second": 4.176,
+      "step": 2000
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 9.637005805969238,
+      "learning_rate": 8.118470149253732e-05,
+      "loss": 0.5131,
+      "step": 2020
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 4.124598026275635,
+      "learning_rate": 8.099813432835821e-05,
+      "loss": 0.396,
+      "step": 2040
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.063644528388977,
+      "learning_rate": 8.081156716417912e-05,
+      "loss": 0.383,
+      "step": 2060
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.7862141132354736,
+      "learning_rate": 8.062500000000001e-05,
+      "loss": 0.3412,
+      "step": 2080
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 12.799152374267578,
+      "learning_rate": 8.04384328358209e-05,
+      "loss": 0.6194,
+      "step": 2100
+    },
+    {
+      "epoch": 3.92,
+      "eval_accuracy": 0.8506151142355008,
+      "eval_loss": 0.57828688621521,
+      "eval_runtime": 17.301,
+      "eval_samples_per_second": 32.888,
+      "eval_steps_per_second": 4.162,
+      "step": 2100
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 2.788487434387207,
+      "learning_rate": 8.025186567164179e-05,
+      "loss": 0.289,
+      "step": 2120
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 15.94080924987793,
+      "learning_rate": 8.00652985074627e-05,
+      "loss": 0.3143,
+      "step": 2140
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.9247996807098389,
+      "learning_rate": 7.987873134328359e-05,
+      "loss": 0.4614,
+      "step": 2160
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 8.811433792114258,
+      "learning_rate": 7.969216417910448e-05,
+      "loss": 0.2502,
+      "step": 2180
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 8.800023078918457,
+      "learning_rate": 7.950559701492538e-05,
+      "loss": 0.5066,
+      "step": 2200
+    },
+    {
+      "epoch": 4.1,
+      "eval_accuracy": 0.8260105448154658,
+      "eval_loss": 0.5824709534645081,
+      "eval_runtime": 17.0204,
+      "eval_samples_per_second": 33.43,
+      "eval_steps_per_second": 4.23,
+      "step": 2200
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 4.279669284820557,
+      "learning_rate": 7.931902985074627e-05,
+      "loss": 0.2544,
+      "step": 2220
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 0.03218838945031166,
+      "learning_rate": 7.913246268656716e-05,
+      "loss": 0.2884,
+      "step": 2240
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 0.6493469476699829,
+      "learning_rate": 7.894589552238805e-05,
+      "loss": 0.2128,
+      "step": 2260
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 9.412866592407227,
+      "learning_rate": 7.875932835820896e-05,
+      "loss": 0.3686,
+      "step": 2280
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 10.611538887023926,
+      "learning_rate": 7.857276119402985e-05,
+      "loss": 0.3532,
+      "step": 2300
+    },
+    {
+      "epoch": 4.29,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.56058269739151,
+      "eval_runtime": 17.1224,
+      "eval_samples_per_second": 33.231,
+      "eval_steps_per_second": 4.205,
+      "step": 2300
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 5.238863468170166,
+      "learning_rate": 7.838619402985074e-05,
+      "loss": 0.5088,
+      "step": 2320
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 4.093738079071045,
+      "learning_rate": 7.819962686567165e-05,
+      "loss": 0.3845,
+      "step": 2340
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 12.486943244934082,
+      "learning_rate": 7.801305970149254e-05,
+      "loss": 0.2818,
+      "step": 2360
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 5.873698711395264,
+      "learning_rate": 7.782649253731343e-05,
+      "loss": 0.2372,
+      "step": 2380
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 3.8764278888702393,
+      "learning_rate": 7.763992537313434e-05,
+      "loss": 0.3531,
+      "step": 2400
+    },
+    {
+      "epoch": 4.48,
+      "eval_accuracy": 0.8699472759226714,
+      "eval_loss": 0.5068355202674866,
+      "eval_runtime": 17.4707,
+      "eval_samples_per_second": 32.569,
+      "eval_steps_per_second": 4.121,
+      "step": 2400
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 2.714694023132324,
+      "learning_rate": 7.745335820895523e-05,
+      "loss": 0.4361,
+      "step": 2420
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 7.682305812835693,
+      "learning_rate": 7.726679104477613e-05,
+      "loss": 0.2619,
+      "step": 2440
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 3.9922590255737305,
+      "learning_rate": 7.708022388059702e-05,
+      "loss": 0.25,
+      "step": 2460
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.3995153903961182,
+      "learning_rate": 7.689365671641791e-05,
+      "loss": 0.399,
+      "step": 2480
+    },
+    {
+      "epoch": 4.66,
+      "grad_norm": 0.5660693645477295,
+      "learning_rate": 7.67070895522388e-05,
+      "loss": 0.2573,
+      "step": 2500
+    },
+    {
+      "epoch": 4.66,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.563208818435669,
+      "eval_runtime": 17.2254,
+      "eval_samples_per_second": 33.033,
+      "eval_steps_per_second": 4.18,
+      "step": 2500
+    },
+    {
+      "epoch": 4.7,
+      "grad_norm": 4.324345111846924,
+      "learning_rate": 7.65205223880597e-05,
+      "loss": 0.2071,
+      "step": 2520
+    },
+    {
+      "epoch": 4.74,
+      "grad_norm": 13.19472885131836,
+      "learning_rate": 7.63339552238806e-05,
+      "loss": 0.4581,
+      "step": 2540
+    },
+    {
+      "epoch": 4.78,
+      "grad_norm": 5.003000736236572,
+      "learning_rate": 7.61473880597015e-05,
+      "loss": 0.4357,
+      "step": 2560
+    },
+    {
+      "epoch": 4.81,
+      "grad_norm": 3.413198471069336,
+      "learning_rate": 7.596082089552239e-05,
+      "loss": 0.4419,
+      "step": 2580
+    },
+    {
+      "epoch": 4.85,
+      "grad_norm": 3.7665114402770996,
+      "learning_rate": 7.577425373134329e-05,
+      "loss": 0.2713,
+      "step": 2600
+    },
+    {
+      "epoch": 4.85,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.5046983957290649,
+      "eval_runtime": 17.1723,
+      "eval_samples_per_second": 33.135,
+      "eval_steps_per_second": 4.193,
+      "step": 2600
+    },
+    {
+      "epoch": 4.89,
+      "grad_norm": 10.373221397399902,
+      "learning_rate": 7.558768656716418e-05,
+      "loss": 0.2644,
+      "step": 2620
+    },
+    {
+      "epoch": 4.93,
+      "grad_norm": 4.17717981338501,
+      "learning_rate": 7.540111940298508e-05,
+      "loss": 0.229,
+      "step": 2640
+    },
+    {
+      "epoch": 4.96,
+      "grad_norm": 4.677199363708496,
+      "learning_rate": 7.521455223880597e-05,
+      "loss": 0.3751,
+      "step": 2660
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 4.891878128051758,
+      "learning_rate": 7.502798507462688e-05,
+      "loss": 0.2869,
+      "step": 2680
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 1.341995120048523,
+      "learning_rate": 7.484141791044777e-05,
+      "loss": 0.3538,
+      "step": 2700
+    },
+    {
+      "epoch": 5.04,
+      "eval_accuracy": 0.8471001757469244,
+      "eval_loss": 0.5988207459449768,
+      "eval_runtime": 17.1614,
+      "eval_samples_per_second": 33.156,
+      "eval_steps_per_second": 4.195,
+      "step": 2700
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.11995257437229156,
+      "learning_rate": 7.465485074626866e-05,
+      "loss": 0.3559,
+      "step": 2720
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 6.199445724487305,
+      "learning_rate": 7.446828358208957e-05,
+      "loss": 0.2866,
+      "step": 2740
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 3.587822198867798,
+      "learning_rate": 7.428171641791046e-05,
+      "loss": 0.348,
+      "step": 2760
+    },
+    {
+      "epoch": 5.19,
+      "grad_norm": 9.572598457336426,
+      "learning_rate": 7.409514925373135e-05,
+      "loss": 0.267,
+      "step": 2780
+    },
+    {
+      "epoch": 5.22,
+      "grad_norm": 9.213473320007324,
+      "learning_rate": 7.390858208955224e-05,
+      "loss": 0.2291,
+      "step": 2800
+    },
+    {
+      "epoch": 5.22,
+      "eval_accuracy": 0.8453427065026362,
+      "eval_loss": 0.5750865340232849,
+      "eval_runtime": 17.1217,
+      "eval_samples_per_second": 33.233,
+      "eval_steps_per_second": 4.205,
+      "step": 2800
+    },
+    {
+      "epoch": 5.26,
+      "grad_norm": 7.156803131103516,
+      "learning_rate": 7.372201492537314e-05,
+      "loss": 0.2964,
+      "step": 2820
+    },
+    {
+      "epoch": 5.3,
+      "grad_norm": 17.43887710571289,
+      "learning_rate": 7.353544776119403e-05,
+      "loss": 0.197,
+      "step": 2840
+    },
+    {
+      "epoch": 5.34,
+      "grad_norm": 0.24130187928676605,
+      "learning_rate": 7.334888059701492e-05,
+      "loss": 0.3145,
+      "step": 2860
+    },
+    {
+      "epoch": 5.37,
+      "grad_norm": 0.0535406693816185,
+      "learning_rate": 7.316231343283583e-05,
+      "loss": 0.2218,
+      "step": 2880
+    },
+    {
+      "epoch": 5.41,
+      "grad_norm": 12.81238079071045,
+      "learning_rate": 7.297574626865672e-05,
+      "loss": 0.2976,
+      "step": 2900
+    },
+    {
+      "epoch": 5.41,
+      "eval_accuracy": 0.8558875219683656,
+      "eval_loss": 0.5780738592147827,
+      "eval_runtime": 17.0963,
+      "eval_samples_per_second": 33.282,
+      "eval_steps_per_second": 4.211,
+      "step": 2900
+    },
+    {
+      "epoch": 5.45,
+      "grad_norm": 2.626818895339966,
+      "learning_rate": 7.278917910447761e-05,
+      "loss": 0.531,
+      "step": 2920
+    },
+    {
+      "epoch": 5.49,
+      "grad_norm": 6.691227912902832,
+      "learning_rate": 7.260261194029852e-05,
+      "loss": 0.2512,
+      "step": 2940
+    },
+    {
+      "epoch": 5.52,
+      "grad_norm": 4.872196674346924,
+      "learning_rate": 7.241604477611941e-05,
+      "loss": 0.2453,
+      "step": 2960
+    },
+    {
+      "epoch": 5.56,
+      "grad_norm": 17.621322631835938,
+      "learning_rate": 7.22294776119403e-05,
+      "loss": 0.3524,
+      "step": 2980
+    },
+    {
+      "epoch": 5.6,
+      "grad_norm": 1.6360100507736206,
+      "learning_rate": 7.204291044776121e-05,
+      "loss": 0.296,
+      "step": 3000
+    },
+    {
+      "epoch": 5.6,
+      "eval_accuracy": 0.8664323374340949,
+      "eval_loss": 0.5499468445777893,
+      "eval_runtime": 17.1322,
+      "eval_samples_per_second": 33.212,
+      "eval_steps_per_second": 4.203,
+      "step": 3000
+    },
+    {
+      "epoch": 5.63,
+      "grad_norm": 8.740306854248047,
+      "learning_rate": 7.18563432835821e-05,
+      "loss": 0.3207,
+      "step": 3020
+    },
+    {
+      "epoch": 5.67,
+      "grad_norm": 6.447021484375,
+      "learning_rate": 7.166977611940299e-05,
+      "loss": 0.2655,
+      "step": 3040
+    },
+    {
+      "epoch": 5.71,
+      "grad_norm": 0.5030248761177063,
+      "learning_rate": 7.148320895522389e-05,
+      "loss": 0.2813,
+      "step": 3060
+    },
+    {
+      "epoch": 5.75,
+      "grad_norm": 8.626925468444824,
+      "learning_rate": 7.129664179104478e-05,
+      "loss": 0.3507,
+      "step": 3080
+    },
+    {
+      "epoch": 5.78,
+      "grad_norm": 0.14797502756118774,
+      "learning_rate": 7.111007462686567e-05,
+      "loss": 0.3776,
+      "step": 3100
+    },
+    {
+      "epoch": 5.78,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.5718467831611633,
+      "eval_runtime": 17.3454,
+      "eval_samples_per_second": 32.804,
+      "eval_steps_per_second": 4.151,
+      "step": 3100
+    },
+    {
+      "epoch": 5.82,
+      "grad_norm": 11.719199180603027,
+      "learning_rate": 7.092350746268656e-05,
+      "loss": 0.3139,
+      "step": 3120
+    },
+    {
+      "epoch": 5.86,
+      "grad_norm": 5.869837284088135,
+      "learning_rate": 7.073694029850747e-05,
+      "loss": 0.3174,
+      "step": 3140
+    },
+    {
+      "epoch": 5.9,
+      "grad_norm": 6.535881042480469,
+      "learning_rate": 7.055037313432836e-05,
+      "loss": 0.3381,
+      "step": 3160
+    },
+    {
+      "epoch": 5.93,
+      "grad_norm": 0.16534599661827087,
+      "learning_rate": 7.036380597014925e-05,
+      "loss": 0.4368,
+      "step": 3180
+    },
+    {
+      "epoch": 5.97,
+      "grad_norm": 5.685976505279541,
+      "learning_rate": 7.017723880597016e-05,
+      "loss": 0.2213,
+      "step": 3200
+    },
+    {
+      "epoch": 5.97,
+      "eval_accuracy": 0.8681898066783831,
+      "eval_loss": 0.542100727558136,
+      "eval_runtime": 17.2482,
+      "eval_samples_per_second": 32.989,
+      "eval_steps_per_second": 4.174,
+      "step": 3200
+    },
+    {
+      "epoch": 6.01,
+      "grad_norm": 14.393926620483398,
+      "learning_rate": 6.999067164179105e-05,
+      "loss": 0.3303,
+      "step": 3220
+    },
+    {
+      "epoch": 6.04,
+      "grad_norm": 4.087838172912598,
+      "learning_rate": 6.980410447761194e-05,
+      "loss": 0.3331,
+      "step": 3240
+    },
+    {
+      "epoch": 6.08,
+      "grad_norm": 2.141066789627075,
+      "learning_rate": 6.961753731343284e-05,
+      "loss": 0.2789,
+      "step": 3260
+    },
+    {
+      "epoch": 6.12,
+      "grad_norm": 11.98432445526123,
+      "learning_rate": 6.943097014925374e-05,
+      "loss": 0.3372,
+      "step": 3280
+    },
+    {
+      "epoch": 6.16,
+      "grad_norm": 3.774228811264038,
+      "learning_rate": 6.924440298507463e-05,
+      "loss": 0.325,
+      "step": 3300
+    },
+    {
+      "epoch": 6.16,
+      "eval_accuracy": 0.8453427065026362,
+      "eval_loss": 0.6452943682670593,
+      "eval_runtime": 17.0316,
+      "eval_samples_per_second": 33.409,
+      "eval_steps_per_second": 4.227,
+      "step": 3300
+    },
+    {
+      "epoch": 6.19,
+      "grad_norm": 8.641297340393066,
+      "learning_rate": 6.905783582089553e-05,
+      "loss": 0.239,
+      "step": 3320
+    },
+    {
+      "epoch": 6.23,
+      "grad_norm": 8.221296310424805,
+      "learning_rate": 6.887126865671642e-05,
+      "loss": 0.2069,
+      "step": 3340
+    },
+    {
+      "epoch": 6.27,
+      "grad_norm": 7.5672526359558105,
+      "learning_rate": 6.868470149253731e-05,
+      "loss": 0.2353,
+      "step": 3360
+    },
+    {
+      "epoch": 6.31,
+      "grad_norm": 5.756315231323242,
+      "learning_rate": 6.84981343283582e-05,
+      "loss": 0.2395,
+      "step": 3380
+    },
+    {
+      "epoch": 6.34,
+      "grad_norm": 9.21039867401123,
+      "learning_rate": 6.831156716417911e-05,
+      "loss": 0.1594,
+      "step": 3400
+    },
+    {
+      "epoch": 6.34,
+      "eval_accuracy": 0.8646748681898067,
+      "eval_loss": 0.5557887554168701,
+      "eval_runtime": 17.2531,
+      "eval_samples_per_second": 32.98,
+      "eval_steps_per_second": 4.173,
+      "step": 3400
+    },
+    {
+      "epoch": 6.38,
+      "grad_norm": 5.185396671295166,
+      "learning_rate": 6.8125e-05,
+      "loss": 0.341,
+      "step": 3420
+    },
+    {
+      "epoch": 6.42,
+      "grad_norm": 7.673208713531494,
+      "learning_rate": 6.79384328358209e-05,
+      "loss": 0.2607,
+      "step": 3440
+    },
+    {
+      "epoch": 6.46,
+      "grad_norm": 6.5733232498168945,
+      "learning_rate": 6.775186567164179e-05,
+      "loss": 0.3484,
+      "step": 3460
+    },
+    {
+      "epoch": 6.49,
+      "grad_norm": 8.386521339416504,
+      "learning_rate": 6.75652985074627e-05,
+      "loss": 0.3434,
+      "step": 3480
+    },
+    {
+      "epoch": 6.53,
+      "grad_norm": 0.4976311922073364,
+      "learning_rate": 6.737873134328359e-05,
+      "loss": 0.3377,
+      "step": 3500
+    },
+    {
+      "epoch": 6.53,
+      "eval_accuracy": 0.8418277680140598,
+      "eval_loss": 0.6618914008140564,
+      "eval_runtime": 17.3191,
+      "eval_samples_per_second": 32.854,
+      "eval_steps_per_second": 4.157,
+      "step": 3500
+    },
+    {
+      "epoch": 6.57,
+      "grad_norm": 6.325254440307617,
+      "learning_rate": 6.719216417910448e-05,
+      "loss": 0.4631,
+      "step": 3520
+    },
+    {
+      "epoch": 6.6,
+      "grad_norm": 9.157130241394043,
+      "learning_rate": 6.700559701492538e-05,
+      "loss": 0.2828,
+      "step": 3540
+    },
+    {
+      "epoch": 6.64,
+      "grad_norm": 5.364408493041992,
+      "learning_rate": 6.681902985074628e-05,
+      "loss": 0.252,
+      "step": 3560
+    },
+    {
+      "epoch": 6.68,
+      "grad_norm": 0.147202268242836,
+      "learning_rate": 6.663246268656717e-05,
+      "loss": 0.2205,
+      "step": 3580
+    },
+    {
+      "epoch": 6.72,
+      "grad_norm": 3.279395580291748,
+      "learning_rate": 6.644589552238806e-05,
+      "loss": 0.3743,
+      "step": 3600
+    },
+    {
+      "epoch": 6.72,
+      "eval_accuracy": 0.8717047451669596,
+      "eval_loss": 0.5445847511291504,
+      "eval_runtime": 17.445,
+      "eval_samples_per_second": 32.617,
+      "eval_steps_per_second": 4.127,
+      "step": 3600
+    },
+    {
+      "epoch": 6.75,
+      "grad_norm": 0.26571547985076904,
+      "learning_rate": 6.625932835820897e-05,
+      "loss": 0.285,
+      "step": 3620
+    },
+    {
+      "epoch": 6.79,
+      "grad_norm": 6.474475383758545,
+      "learning_rate": 6.607276119402986e-05,
+      "loss": 0.2974,
+      "step": 3640
+    },
+    {
+      "epoch": 6.83,
+      "grad_norm": 0.4200984835624695,
+      "learning_rate": 6.588619402985074e-05,
+      "loss": 0.1905,
+      "step": 3660
+    },
+    {
+      "epoch": 6.87,
+      "grad_norm": 2.3140206336975098,
+      "learning_rate": 6.569962686567164e-05,
+      "loss": 0.2045,
+      "step": 3680
+    },
+    {
+      "epoch": 6.9,
+      "grad_norm": 0.08412935584783554,
+      "learning_rate": 6.551305970149254e-05,
+      "loss": 0.2327,
+      "step": 3700
+    },
+    {
+      "epoch": 6.9,
+      "eval_accuracy": 0.8734622144112478,
+      "eval_loss": 0.5483921766281128,
+      "eval_runtime": 17.1364,
+      "eval_samples_per_second": 33.204,
+      "eval_steps_per_second": 4.202,
+      "step": 3700
+    },
+    {
+      "epoch": 6.94,
+      "grad_norm": 2.7536630630493164,
+      "learning_rate": 6.532649253731343e-05,
+      "loss": 0.1663,
+      "step": 3720
+    },
+    {
+      "epoch": 6.98,
+      "grad_norm": 0.2596806585788727,
+      "learning_rate": 6.513992537313434e-05,
+      "loss": 0.3343,
+      "step": 3740
+    },
+    {
+      "epoch": 7.01,
+      "grad_norm": 4.579122066497803,
+      "learning_rate": 6.495335820895523e-05,
+      "loss": 0.2173,
+      "step": 3760
+    },
+    {
+      "epoch": 7.05,
+      "grad_norm": 0.9458146691322327,
+      "learning_rate": 6.476679104477612e-05,
+      "loss": 0.1829,
+      "step": 3780
+    },
+    {
+      "epoch": 7.09,
+      "grad_norm": 0.1997598111629486,
+      "learning_rate": 6.458022388059701e-05,
+      "loss": 0.1659,
+      "step": 3800
+    },
+    {
+      "epoch": 7.09,
+      "eval_accuracy": 0.8471001757469244,
+      "eval_loss": 0.6629351377487183,
+      "eval_runtime": 17.0239,
+      "eval_samples_per_second": 33.424,
+      "eval_steps_per_second": 4.229,
+      "step": 3800
+    },
+    {
+      "epoch": 7.13,
+      "grad_norm": 6.405377388000488,
+      "learning_rate": 6.439365671641792e-05,
+      "loss": 0.133,
+      "step": 3820
+    },
+    {
+      "epoch": 7.16,
+      "grad_norm": 0.3477083146572113,
+      "learning_rate": 6.420708955223881e-05,
+      "loss": 0.3308,
+      "step": 3840
+    },
+    {
+      "epoch": 7.2,
+      "grad_norm": 14.130784034729004,
+      "learning_rate": 6.40205223880597e-05,
+      "loss": 0.3193,
+      "step": 3860
+    },
+    {
+      "epoch": 7.24,
+      "grad_norm": 10.30834674835205,
+      "learning_rate": 6.383395522388061e-05,
+      "loss": 0.2314,
+      "step": 3880
+    },
+    {
+      "epoch": 7.28,
+      "grad_norm": 0.131292462348938,
+      "learning_rate": 6.36473880597015e-05,
+      "loss": 0.4036,
+      "step": 3900
+    },
+    {
+      "epoch": 7.28,
+      "eval_accuracy": 0.8330404217926186,
+      "eval_loss": 0.651033878326416,
+      "eval_runtime": 17.0735,
+      "eval_samples_per_second": 33.327,
+      "eval_steps_per_second": 4.217,
+      "step": 3900
+    },
+    {
+      "epoch": 7.31,
+      "grad_norm": 8.34850788116455,
+      "learning_rate": 6.34608208955224e-05,
+      "loss": 0.2416,
+      "step": 3920
+    },
+    {
+      "epoch": 7.35,
+      "grad_norm": 15.084284782409668,
+      "learning_rate": 6.327425373134329e-05,
+      "loss": 0.283,
+      "step": 3940
+    },
+    {
+      "epoch": 7.39,
+      "grad_norm": 0.007386388722807169,
+      "learning_rate": 6.308768656716418e-05,
+      "loss": 0.2168,
+      "step": 3960
+    },
+    {
+      "epoch": 7.43,
+      "grad_norm": 2.0279037952423096,
+      "learning_rate": 6.290111940298507e-05,
+      "loss": 0.2764,
+      "step": 3980
+    },
+    {
+      "epoch": 7.46,
+      "grad_norm": 4.890338897705078,
+      "learning_rate": 6.271455223880596e-05,
+      "loss": 0.2084,
+      "step": 4000
+    },
+    {
+      "epoch": 7.46,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.5640222430229187,
+      "eval_runtime": 17.1832,
+      "eval_samples_per_second": 33.114,
+      "eval_steps_per_second": 4.19,
+      "step": 4000
+    },
+    {
+      "epoch": 7.5,
+      "grad_norm": 0.015988588333129883,
+      "learning_rate": 6.252798507462687e-05,
+      "loss": 0.1223,
+      "step": 4020
+    },
+    {
+      "epoch": 7.54,
+      "grad_norm": 12.72655200958252,
+      "learning_rate": 6.234141791044776e-05,
+      "loss": 0.3431,
+      "step": 4040
+    },
+    {
+      "epoch": 7.57,
+      "grad_norm": 2.34236478805542,
+      "learning_rate": 6.215485074626865e-05,
+      "loss": 0.3444,
+      "step": 4060
+    },
+    {
+      "epoch": 7.61,
+      "grad_norm": 6.3446245193481445,
+      "learning_rate": 6.196828358208956e-05,
+      "loss": 0.3299,
+      "step": 4080
+    },
+    {
+      "epoch": 7.65,
+      "grad_norm": 0.03079419955611229,
+      "learning_rate": 6.178171641791045e-05,
+      "loss": 0.2251,
+      "step": 4100
+    },
+    {
+      "epoch": 7.65,
+      "eval_accuracy": 0.8541300527240774,
+      "eval_loss": 0.6378800272941589,
+      "eval_runtime": 17.2854,
+      "eval_samples_per_second": 32.918,
+      "eval_steps_per_second": 4.165,
+      "step": 4100
+    },
+    {
+      "epoch": 7.69,
+      "grad_norm": 5.350043773651123,
+      "learning_rate": 6.159514925373135e-05,
+      "loss": 0.2121,
+      "step": 4120
+    },
+    {
+      "epoch": 7.72,
+      "grad_norm": 7.147230625152588,
+      "learning_rate": 6.140858208955225e-05,
+      "loss": 0.2923,
+      "step": 4140
+    },
+    {
+      "epoch": 7.76,
+      "grad_norm": 8.773134231567383,
+      "learning_rate": 6.122201492537314e-05,
+      "loss": 0.2331,
+      "step": 4160
+    },
+    {
+      "epoch": 7.8,
+      "grad_norm": 4.792509078979492,
+      "learning_rate": 6.103544776119404e-05,
+      "loss": 0.4707,
+      "step": 4180
+    },
+    {
+      "epoch": 7.84,
+      "grad_norm": 1.0023802518844604,
+      "learning_rate": 6.084888059701492e-05,
+      "loss": 0.192,
+      "step": 4200
+    },
+    {
+      "epoch": 7.84,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.5896880626678467,
+      "eval_runtime": 17.2123,
+      "eval_samples_per_second": 33.058,
+      "eval_steps_per_second": 4.183,
+      "step": 4200
+    },
+    {
+      "epoch": 7.87,
+      "grad_norm": 0.8236982822418213,
+      "learning_rate": 6.066231343283583e-05,
+      "loss": 0.19,
+      "step": 4220
+    },
+    {
+      "epoch": 7.91,
+      "grad_norm": 0.19682720303535461,
+      "learning_rate": 6.047574626865672e-05,
+      "loss": 0.1853,
+      "step": 4240
+    },
+    {
+      "epoch": 7.95,
+      "grad_norm": 0.022599322721362114,
+      "learning_rate": 6.028917910447761e-05,
+      "loss": 0.2141,
+      "step": 4260
+    },
+    {
+      "epoch": 7.99,
+      "grad_norm": 0.11651401221752167,
+      "learning_rate": 6.010261194029851e-05,
+      "loss": 0.2876,
+      "step": 4280
+    },
+    {
+      "epoch": 8.02,
+      "grad_norm": 0.10463010519742966,
+      "learning_rate": 5.9916044776119404e-05,
+      "loss": 0.1956,
+      "step": 4300
+    },
+    {
+      "epoch": 8.02,
+      "eval_accuracy": 0.8699472759226714,
+      "eval_loss": 0.5874119997024536,
+      "eval_runtime": 17.1395,
+      "eval_samples_per_second": 33.198,
+      "eval_steps_per_second": 4.201,
+      "step": 4300
+    },
+    {
+      "epoch": 8.06,
+      "grad_norm": 4.419534206390381,
+      "learning_rate": 5.9729477611940296e-05,
+      "loss": 0.1994,
+      "step": 4320
+    },
+    {
+      "epoch": 8.1,
+      "grad_norm": 0.05122959241271019,
+      "learning_rate": 5.95429104477612e-05,
+      "loss": 0.1357,
+      "step": 4340
+    },
+    {
+      "epoch": 8.13,
+      "grad_norm": 2.251021146774292,
+      "learning_rate": 5.9356343283582095e-05,
+      "loss": 0.2917,
+      "step": 4360
+    },
+    {
+      "epoch": 8.17,
+      "grad_norm": 0.021823499351739883,
+      "learning_rate": 5.916977611940299e-05,
+      "loss": 0.2279,
+      "step": 4380
+    },
+    {
+      "epoch": 8.21,
+      "grad_norm": 2.368696928024292,
+      "learning_rate": 5.898320895522388e-05,
+      "loss": 0.1446,
+      "step": 4400
+    },
+    {
+      "epoch": 8.21,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.6461585760116577,
+      "eval_runtime": 17.2259,
+      "eval_samples_per_second": 33.032,
+      "eval_steps_per_second": 4.18,
+      "step": 4400
+    },
+    {
+      "epoch": 8.25,
+      "grad_norm": 6.274438858032227,
+      "learning_rate": 5.8796641791044785e-05,
+      "loss": 0.3054,
+      "step": 4420
+    },
+    {
+      "epoch": 8.28,
+      "grad_norm": 0.005037074442952871,
+      "learning_rate": 5.861007462686567e-05,
+      "loss": 0.1994,
+      "step": 4440
+    },
+    {
+      "epoch": 8.32,
+      "grad_norm": 0.9350100159645081,
+      "learning_rate": 5.842350746268656e-05,
+      "loss": 0.1253,
+      "step": 4460
+    },
+    {
+      "epoch": 8.36,
+      "grad_norm": 0.42407700419425964,
+      "learning_rate": 5.823694029850747e-05,
+      "loss": 0.2319,
+      "step": 4480
+    },
+    {
+      "epoch": 8.4,
+      "grad_norm": 0.04089637100696564,
+      "learning_rate": 5.805037313432836e-05,
+      "loss": 0.2971,
+      "step": 4500
+    },
+    {
+      "epoch": 8.4,
+      "eval_accuracy": 0.8734622144112478,
+      "eval_loss": 0.5908673405647278,
+      "eval_runtime": 16.9747,
+      "eval_samples_per_second": 33.52,
+      "eval_steps_per_second": 4.242,
+      "step": 4500
+    },
+    {
+      "epoch": 8.43,
+      "grad_norm": 0.35463401675224304,
+      "learning_rate": 5.7863805970149254e-05,
+      "loss": 0.1764,
+      "step": 4520
+    },
+    {
+      "epoch": 8.47,
+      "grad_norm": 0.1451406329870224,
+      "learning_rate": 5.767723880597016e-05,
+      "loss": 0.2318,
+      "step": 4540
+    },
+    {
+      "epoch": 8.51,
+      "grad_norm": 17.04181671142578,
+      "learning_rate": 5.749067164179105e-05,
+      "loss": 0.1935,
+      "step": 4560
+    },
+    {
+      "epoch": 8.54,
+      "grad_norm": 12.285507202148438,
+      "learning_rate": 5.7304104477611945e-05,
+      "loss": 0.2639,
+      "step": 4580
+    },
+    {
+      "epoch": 8.58,
+      "grad_norm": 0.009779583662748337,
+      "learning_rate": 5.711753731343284e-05,
+      "loss": 0.2665,
+      "step": 4600
+    },
+    {
+      "epoch": 8.58,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.676869809627533,
+      "eval_runtime": 16.8952,
+      "eval_samples_per_second": 33.678,
+      "eval_steps_per_second": 4.262,
+      "step": 4600
+    },
+    {
+      "epoch": 8.62,
+      "grad_norm": 2.7441444396972656,
+      "learning_rate": 5.6930970149253736e-05,
+      "loss": 0.2199,
+      "step": 4620
+    },
+    {
+      "epoch": 8.66,
+      "grad_norm": 4.1195573806762695,
+      "learning_rate": 5.674440298507463e-05,
+      "loss": 0.3658,
+      "step": 4640
+    },
+    {
+      "epoch": 8.69,
+      "grad_norm": 2.8499536514282227,
+      "learning_rate": 5.655783582089552e-05,
+      "loss": 0.1778,
+      "step": 4660
+    },
+    {
+      "epoch": 8.73,
+      "grad_norm": 0.07212776690721512,
+      "learning_rate": 5.637126865671643e-05,
+      "loss": 0.2054,
+      "step": 4680
+    },
+    {
+      "epoch": 8.77,
+      "grad_norm": 15.808812141418457,
+      "learning_rate": 5.618470149253732e-05,
+      "loss": 0.2937,
+      "step": 4700
+    },
+    {
+      "epoch": 8.77,
+      "eval_accuracy": 0.8506151142355008,
+      "eval_loss": 0.6759567260742188,
+      "eval_runtime": 17.1704,
+      "eval_samples_per_second": 33.138,
+      "eval_steps_per_second": 4.193,
+      "step": 4700
+    },
+    {
+      "epoch": 8.81,
+      "grad_norm": 3.0490574836730957,
+      "learning_rate": 5.599813432835821e-05,
+      "loss": 0.1529,
+      "step": 4720
+    },
+    {
+      "epoch": 8.84,
+      "grad_norm": 0.009331068024039268,
+      "learning_rate": 5.581156716417911e-05,
+      "loss": 0.2329,
+      "step": 4740
+    },
+    {
+      "epoch": 8.88,
+      "grad_norm": 1.2966663837432861,
+      "learning_rate": 5.5625000000000004e-05,
+      "loss": 0.4503,
+      "step": 4760
+    },
+    {
+      "epoch": 8.92,
+      "grad_norm": 16.463085174560547,
+      "learning_rate": 5.5438432835820896e-05,
+      "loss": 0.2412,
+      "step": 4780
+    },
+    {
+      "epoch": 8.96,
+      "grad_norm": 0.01172882504761219,
+      "learning_rate": 5.525186567164179e-05,
+      "loss": 0.1437,
+      "step": 4800
+    },
+    {
+      "epoch": 8.96,
+      "eval_accuracy": 0.8488576449912126,
+      "eval_loss": 0.6566036343574524,
+      "eval_runtime": 17.0018,
+      "eval_samples_per_second": 33.467,
+      "eval_steps_per_second": 4.235,
+      "step": 4800
+    },
+    {
+      "epoch": 8.99,
+      "grad_norm": 0.008390693925321102,
+      "learning_rate": 5.5065298507462694e-05,
+      "loss": 0.2241,
+      "step": 4820
+    },
+    {
+      "epoch": 9.03,
+      "grad_norm": 11.137565612792969,
+      "learning_rate": 5.487873134328359e-05,
+      "loss": 0.1879,
+      "step": 4840
+    },
+    {
+      "epoch": 9.07,
+      "grad_norm": 15.202289581298828,
+      "learning_rate": 5.4701492537313435e-05,
+      "loss": 0.1546,
+      "step": 4860
+    },
+    {
+      "epoch": 9.1,
+      "grad_norm": 0.29179397225379944,
+      "learning_rate": 5.451492537313433e-05,
+      "loss": 0.2482,
+      "step": 4880
+    },
+    {
+      "epoch": 9.14,
+      "grad_norm": 0.15197451412677765,
+      "learning_rate": 5.432835820895522e-05,
+      "loss": 0.1433,
+      "step": 4900
+    },
+    {
+      "epoch": 9.14,
+      "eval_accuracy": 0.8418277680140598,
+      "eval_loss": 0.6659190058708191,
+      "eval_runtime": 17.102,
+      "eval_samples_per_second": 33.271,
+      "eval_steps_per_second": 4.21,
+      "step": 4900
+    },
+    {
+      "epoch": 9.18,
+      "grad_norm": 2.3589694499969482,
+      "learning_rate": 5.4141791044776126e-05,
+      "loss": 0.2289,
+      "step": 4920
+    },
+    {
+      "epoch": 9.22,
+      "grad_norm": 0.27095305919647217,
+      "learning_rate": 5.395522388059702e-05,
+      "loss": 0.2409,
+      "step": 4940
+    },
+    {
+      "epoch": 9.25,
+      "grad_norm": 24.258569717407227,
+      "learning_rate": 5.376865671641791e-05,
+      "loss": 0.2386,
+      "step": 4960
+    },
+    {
+      "epoch": 9.29,
+      "grad_norm": 0.35387206077575684,
+      "learning_rate": 5.358208955223881e-05,
+      "loss": 0.2534,
+      "step": 4980
+    },
+    {
+      "epoch": 9.33,
+      "grad_norm": 0.04066289961338043,
+      "learning_rate": 5.33955223880597e-05,
+      "loss": 0.2069,
+      "step": 5000
+    },
+    {
+      "epoch": 9.33,
+      "eval_accuracy": 0.8541300527240774,
+      "eval_loss": 0.6824544668197632,
+      "eval_runtime": 17.0072,
+      "eval_samples_per_second": 33.456,
+      "eval_steps_per_second": 4.234,
+      "step": 5000
+    },
+    {
+      "epoch": 9.37,
+      "grad_norm": 1.950838327407837,
+      "learning_rate": 5.3208955223880595e-05,
+      "loss": 0.1991,
+      "step": 5020
+    },
+    {
+      "epoch": 9.4,
+      "grad_norm": 6.514630317687988,
+      "learning_rate": 5.30223880597015e-05,
+      "loss": 0.2723,
+      "step": 5040
+    },
+    {
+      "epoch": 9.44,
+      "grad_norm": 13.146520614624023,
+      "learning_rate": 5.283582089552239e-05,
+      "loss": 0.2475,
+      "step": 5060
+    },
+    {
+      "epoch": 9.48,
+      "grad_norm": 0.04828513413667679,
+      "learning_rate": 5.2649253731343286e-05,
+      "loss": 0.2883,
+      "step": 5080
+    },
+    {
+      "epoch": 9.51,
+      "grad_norm": 4.672499656677246,
+      "learning_rate": 5.246268656716418e-05,
+      "loss": 0.2095,
+      "step": 5100
+    },
+    {
+      "epoch": 9.51,
+      "eval_accuracy": 0.8664323374340949,
+      "eval_loss": 0.6156749725341797,
+      "eval_runtime": 17.0363,
+      "eval_samples_per_second": 33.399,
+      "eval_steps_per_second": 4.226,
+      "step": 5100
+    },
+    {
+      "epoch": 9.55,
+      "grad_norm": 1.7158946990966797,
+      "learning_rate": 5.227611940298508e-05,
+      "loss": 0.1681,
+      "step": 5120
+    },
+    {
+      "epoch": 9.59,
+      "grad_norm": 0.04500845819711685,
+      "learning_rate": 5.208955223880597e-05,
+      "loss": 0.3107,
+      "step": 5140
+    },
+    {
+      "epoch": 9.63,
+      "grad_norm": 6.066332817077637,
+      "learning_rate": 5.190298507462686e-05,
+      "loss": 0.2578,
+      "step": 5160
+    },
+    {
+      "epoch": 9.66,
+      "grad_norm": 4.105052471160889,
+      "learning_rate": 5.171641791044777e-05,
+      "loss": 0.1026,
+      "step": 5180
+    },
+    {
+      "epoch": 9.7,
+      "grad_norm": 0.7249622344970703,
+      "learning_rate": 5.152985074626866e-05,
+      "loss": 0.1579,
+      "step": 5200
+    },
+    {
+      "epoch": 9.7,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.6692686080932617,
+      "eval_runtime": 17.2004,
+      "eval_samples_per_second": 33.081,
+      "eval_steps_per_second": 4.186,
+      "step": 5200
+    },
+    {
+      "epoch": 9.74,
+      "grad_norm": 0.0021988856606185436,
+      "learning_rate": 5.134328358208955e-05,
+      "loss": 0.2816,
+      "step": 5220
+    },
+    {
+      "epoch": 9.78,
+      "grad_norm": 0.006726569030433893,
+      "learning_rate": 5.115671641791045e-05,
+      "loss": 0.1922,
+      "step": 5240
+    },
+    {
+      "epoch": 9.81,
+      "grad_norm": 7.35833740234375,
+      "learning_rate": 5.0970149253731344e-05,
+      "loss": 0.209,
+      "step": 5260
+    },
+    {
+      "epoch": 9.85,
+      "grad_norm": 1.70156729221344,
+      "learning_rate": 5.078358208955224e-05,
+      "loss": 0.0755,
+      "step": 5280
+    },
+    {
+      "epoch": 9.89,
+      "grad_norm": 8.165057182312012,
+      "learning_rate": 5.059701492537313e-05,
+      "loss": 0.1962,
+      "step": 5300
+    },
+    {
+      "epoch": 9.89,
+      "eval_accuracy": 0.8523725834797891,
+      "eval_loss": 0.6911476254463196,
+      "eval_runtime": 17.1764,
+      "eval_samples_per_second": 33.127,
+      "eval_steps_per_second": 4.192,
+      "step": 5300
+    },
+    {
+      "epoch": 9.93,
+      "grad_norm": 4.652767181396484,
+      "learning_rate": 5.0410447761194035e-05,
+      "loss": 0.1933,
+      "step": 5320
+    },
+    {
+      "epoch": 9.96,
+      "grad_norm": 0.02002258598804474,
+      "learning_rate": 5.022388059701493e-05,
+      "loss": 0.1235,
+      "step": 5340
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 3.154204845428467,
+      "learning_rate": 5.003731343283582e-05,
+      "loss": 0.2349,
+      "step": 5360
+    },
+    {
+      "epoch": 10.04,
+      "grad_norm": 0.7576759457588196,
+      "learning_rate": 4.985074626865672e-05,
+      "loss": 0.3175,
+      "step": 5380
+    },
+    {
+      "epoch": 10.07,
+      "grad_norm": 3.5345828533172607,
+      "learning_rate": 4.966417910447762e-05,
+      "loss": 0.3149,
+      "step": 5400
+    },
+    {
+      "epoch": 10.07,
+      "eval_accuracy": 0.8558875219683656,
+      "eval_loss": 0.6259763240814209,
+      "eval_runtime": 16.9438,
+      "eval_samples_per_second": 33.582,
+      "eval_steps_per_second": 4.249,
+      "step": 5400
+    },
+    {
+      "epoch": 10.11,
+      "grad_norm": 0.37791475653648376,
+      "learning_rate": 4.9477611940298504e-05,
+      "loss": 0.0893,
+      "step": 5420
+    },
+    {
+      "epoch": 10.15,
+      "grad_norm": 2.0877885818481445,
+      "learning_rate": 4.92910447761194e-05,
+      "loss": 0.1127,
+      "step": 5440
+    },
+    {
+      "epoch": 10.19,
+      "grad_norm": 0.013441379182040691,
+      "learning_rate": 4.91044776119403e-05,
+      "loss": 0.2152,
+      "step": 5460
+    },
+    {
+      "epoch": 10.22,
+      "grad_norm": 2.4906938076019287,
+      "learning_rate": 4.8917910447761195e-05,
+      "loss": 0.2481,
+      "step": 5480
+    },
+    {
+      "epoch": 10.26,
+      "grad_norm": 1.7139744758605957,
+      "learning_rate": 4.8731343283582094e-05,
+      "loss": 0.2166,
+      "step": 5500
+    },
+    {
+      "epoch": 10.26,
+      "eval_accuracy": 0.8769771528998243,
+      "eval_loss": 0.6199991106987,
+      "eval_runtime": 17.0131,
+      "eval_samples_per_second": 33.445,
+      "eval_steps_per_second": 4.232,
+      "step": 5500
+    },
+    {
+      "epoch": 10.3,
+      "grad_norm": 0.022912869229912758,
+      "learning_rate": 4.8544776119402986e-05,
+      "loss": 0.165,
+      "step": 5520
+    },
+    {
+      "epoch": 10.34,
+      "grad_norm": 0.055580299347639084,
+      "learning_rate": 4.8358208955223885e-05,
+      "loss": 0.2697,
+      "step": 5540
+    },
+    {
+      "epoch": 10.37,
+      "grad_norm": 6.861047267913818,
+      "learning_rate": 4.817164179104478e-05,
+      "loss": 0.2951,
+      "step": 5560
+    },
+    {
+      "epoch": 10.41,
+      "grad_norm": 10.092063903808594,
+      "learning_rate": 4.798507462686567e-05,
+      "loss": 0.1826,
+      "step": 5580
+    },
+    {
+      "epoch": 10.45,
+      "grad_norm": 6.707045078277588,
+      "learning_rate": 4.779850746268657e-05,
+      "loss": 0.1259,
+      "step": 5600
+    },
+    {
+      "epoch": 10.45,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.7164254188537598,
+      "eval_runtime": 17.0979,
+      "eval_samples_per_second": 33.279,
+      "eval_steps_per_second": 4.211,
+      "step": 5600
+    },
+    {
+      "epoch": 10.49,
+      "grad_norm": 14.931378364562988,
+      "learning_rate": 4.761194029850746e-05,
+      "loss": 0.1782,
+      "step": 5620
+    },
+    {
+      "epoch": 10.52,
+      "grad_norm": 1.0374107360839844,
+      "learning_rate": 4.742537313432836e-05,
+      "loss": 0.2298,
+      "step": 5640
+    },
+    {
+      "epoch": 10.56,
+      "grad_norm": 8.950643539428711,
+      "learning_rate": 4.723880597014926e-05,
+      "loss": 0.2259,
+      "step": 5660
+    },
+    {
+      "epoch": 10.6,
+      "grad_norm": 0.1816747635602951,
+      "learning_rate": 4.705223880597015e-05,
+      "loss": 0.1792,
+      "step": 5680
+    },
+    {
+      "epoch": 10.63,
+      "grad_norm": 3.34356427192688,
+      "learning_rate": 4.6865671641791045e-05,
+      "loss": 0.1892,
+      "step": 5700
+    },
+    {
+      "epoch": 10.63,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.7181826233863831,
+      "eval_runtime": 17.172,
+      "eval_samples_per_second": 33.135,
+      "eval_steps_per_second": 4.193,
+      "step": 5700
+    },
+    {
+      "epoch": 10.67,
+      "grad_norm": 5.057746887207031,
+      "learning_rate": 4.667910447761194e-05,
+      "loss": 0.2591,
+      "step": 5720
+    },
+    {
+      "epoch": 10.71,
+      "grad_norm": 0.08504645526409149,
+      "learning_rate": 4.6492537313432837e-05,
+      "loss": 0.1952,
+      "step": 5740
+    },
+    {
+      "epoch": 10.75,
+      "grad_norm": 9.373453140258789,
+      "learning_rate": 4.6305970149253736e-05,
+      "loss": 0.2723,
+      "step": 5760
+    },
+    {
+      "epoch": 10.78,
+      "grad_norm": 7.880430221557617,
+      "learning_rate": 4.611940298507463e-05,
+      "loss": 0.1482,
+      "step": 5780
+    },
+    {
+      "epoch": 10.82,
+      "grad_norm": 9.705270767211914,
+      "learning_rate": 4.593283582089553e-05,
+      "loss": 0.1953,
+      "step": 5800
+    },
+    {
+      "epoch": 10.82,
+      "eval_accuracy": 0.8418277680140598,
+      "eval_loss": 0.7192606925964355,
+      "eval_runtime": 17.1747,
+      "eval_samples_per_second": 33.13,
+      "eval_steps_per_second": 4.192,
+      "step": 5800
+    },
+    {
+      "epoch": 10.86,
+      "grad_norm": 4.473215103149414,
+      "learning_rate": 4.574626865671642e-05,
+      "loss": 0.2285,
+      "step": 5820
+    },
+    {
+      "epoch": 10.9,
+      "grad_norm": 1.655419945716858,
+      "learning_rate": 4.555970149253732e-05,
+      "loss": 0.1583,
+      "step": 5840
+    },
+    {
+      "epoch": 10.93,
+      "grad_norm": 21.153766632080078,
+      "learning_rate": 4.537313432835821e-05,
+      "loss": 0.1773,
+      "step": 5860
+    },
+    {
+      "epoch": 10.97,
+      "grad_norm": 1.8961058855056763,
+      "learning_rate": 4.5186567164179104e-05,
+      "loss": 0.2536,
+      "step": 5880
+    },
+    {
+      "epoch": 11.01,
+      "grad_norm": 0.11447744816541672,
+      "learning_rate": 4.5e-05,
+      "loss": 0.2392,
+      "step": 5900
+    },
+    {
+      "epoch": 11.01,
+      "eval_accuracy": 0.8664323374340949,
+      "eval_loss": 0.6621362566947937,
+      "eval_runtime": 16.8941,
+      "eval_samples_per_second": 33.68,
+      "eval_steps_per_second": 4.262,
+      "step": 5900
+    },
+    {
+      "epoch": 11.04,
+      "grad_norm": 33.42607116699219,
+      "learning_rate": 4.4813432835820895e-05,
+      "loss": 0.1171,
+      "step": 5920
+    },
+    {
+      "epoch": 11.08,
+      "grad_norm": 0.45287203788757324,
+      "learning_rate": 4.4626865671641794e-05,
+      "loss": 0.1849,
+      "step": 5940
+    },
+    {
+      "epoch": 11.12,
+      "grad_norm": 0.021968552842736244,
+      "learning_rate": 4.4440298507462694e-05,
+      "loss": 0.1649,
+      "step": 5960
+    },
+    {
+      "epoch": 11.16,
+      "grad_norm": 2.502751350402832,
+      "learning_rate": 4.4253731343283586e-05,
+      "loss": 0.1502,
+      "step": 5980
+    },
+    {
+      "epoch": 11.19,
+      "grad_norm": 0.021447136998176575,
+      "learning_rate": 4.406716417910448e-05,
+      "loss": 0.1594,
+      "step": 6000
+    },
+    {
+      "epoch": 11.19,
+      "eval_accuracy": 0.8488576449912126,
+      "eval_loss": 0.7470901012420654,
+      "eval_runtime": 17.0306,
+      "eval_samples_per_second": 33.411,
+      "eval_steps_per_second": 4.228,
+      "step": 6000
+    },
+    {
+      "epoch": 11.23,
+      "grad_norm": 6.504509449005127,
+      "learning_rate": 4.388059701492537e-05,
+      "loss": 0.2011,
+      "step": 6020
+    },
+    {
+      "epoch": 11.27,
+      "grad_norm": 0.0932011529803276,
+      "learning_rate": 4.369402985074627e-05,
+      "loss": 0.0886,
+      "step": 6040
+    },
+    {
+      "epoch": 11.31,
+      "grad_norm": 5.7348856925964355,
+      "learning_rate": 4.350746268656717e-05,
+      "loss": 0.3577,
+      "step": 6060
+    },
+    {
+      "epoch": 11.34,
+      "grad_norm": 5.531877517700195,
+      "learning_rate": 4.332089552238806e-05,
+      "loss": 0.2637,
+      "step": 6080
+    },
+    {
+      "epoch": 11.38,
+      "grad_norm": 12.052019119262695,
+      "learning_rate": 4.313432835820896e-05,
+      "loss": 0.2156,
+      "step": 6100
+    },
+    {
+      "epoch": 11.38,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.7316247820854187,
+      "eval_runtime": 16.9543,
+      "eval_samples_per_second": 33.561,
+      "eval_steps_per_second": 4.247,
+      "step": 6100
+    },
+    {
+      "epoch": 11.42,
+      "grad_norm": 0.02054465003311634,
+      "learning_rate": 4.294776119402985e-05,
+      "loss": 0.1609,
+      "step": 6120
+    },
+    {
+      "epoch": 11.46,
+      "grad_norm": 0.03626035898923874,
+      "learning_rate": 4.2761194029850746e-05,
+      "loss": 0.182,
+      "step": 6140
+    },
+    {
+      "epoch": 11.49,
+      "grad_norm": 17.250499725341797,
+      "learning_rate": 4.2574626865671645e-05,
+      "loss": 0.2388,
+      "step": 6160
+    },
+    {
+      "epoch": 11.53,
+      "grad_norm": 0.04123261943459511,
+      "learning_rate": 4.238805970149254e-05,
+      "loss": 0.1635,
+      "step": 6180
+    },
+    {
+      "epoch": 11.57,
+      "grad_norm": 5.6128034591674805,
+      "learning_rate": 4.2201492537313436e-05,
+      "loss": 0.137,
+      "step": 6200
+    },
+    {
+      "epoch": 11.57,
+      "eval_accuracy": 0.8699472759226714,
+      "eval_loss": 0.6836700439453125,
+      "eval_runtime": 17.1524,
+      "eval_samples_per_second": 33.173,
+      "eval_steps_per_second": 4.198,
+      "step": 6200
+    },
+    {
+      "epoch": 11.6,
+      "grad_norm": 9.23444652557373,
+      "learning_rate": 4.201492537313433e-05,
+      "loss": 0.1014,
+      "step": 6220
+    },
+    {
+      "epoch": 11.64,
+      "grad_norm": 0.0037231920287013054,
+      "learning_rate": 4.182835820895523e-05,
+      "loss": 0.1507,
+      "step": 6240
+    },
+    {
+      "epoch": 11.68,
+      "grad_norm": 0.021425573155283928,
+      "learning_rate": 4.164179104477613e-05,
+      "loss": 0.1692,
+      "step": 6260
+    },
+    {
+      "epoch": 11.72,
+      "grad_norm": 0.01996900700032711,
+      "learning_rate": 4.145522388059702e-05,
+      "loss": 0.1702,
+      "step": 6280
+    },
+    {
+      "epoch": 11.75,
+      "grad_norm": 11.529648780822754,
+      "learning_rate": 4.126865671641791e-05,
+      "loss": 0.181,
+      "step": 6300
+    },
+    {
+      "epoch": 11.75,
+      "eval_accuracy": 0.8646748681898067,
+      "eval_loss": 0.6594767570495605,
+      "eval_runtime": 17.0989,
+      "eval_samples_per_second": 33.277,
+      "eval_steps_per_second": 4.211,
+      "step": 6300
+    },
+    {
+      "epoch": 11.79,
+      "grad_norm": 11.14013671875,
+      "learning_rate": 4.1082089552238804e-05,
+      "loss": 0.1513,
+      "step": 6320
+    },
+    {
+      "epoch": 11.83,
+      "grad_norm": 9.364405632019043,
+      "learning_rate": 4.0895522388059703e-05,
+      "loss": 0.3018,
+      "step": 6340
+    },
+    {
+      "epoch": 11.87,
+      "grad_norm": 9.847725868225098,
+      "learning_rate": 4.07089552238806e-05,
+      "loss": 0.2924,
+      "step": 6360
+    },
+    {
+      "epoch": 11.9,
+      "grad_norm": 8.17735767364502,
+      "learning_rate": 4.0522388059701495e-05,
+      "loss": 0.2258,
+      "step": 6380
+    },
+    {
+      "epoch": 11.94,
+      "grad_norm": 0.046075087040662766,
+      "learning_rate": 4.0335820895522394e-05,
+      "loss": 0.2049,
+      "step": 6400
+    },
+    {
+      "epoch": 11.94,
+      "eval_accuracy": 0.8506151142355008,
+      "eval_loss": 0.6982370018959045,
+      "eval_runtime": 17.0751,
+      "eval_samples_per_second": 33.323,
+      "eval_steps_per_second": 4.217,
+      "step": 6400
+    },
+    {
+      "epoch": 11.98,
+      "grad_norm": 0.8415078520774841,
+      "learning_rate": 4.014925373134329e-05,
+      "loss": 0.2845,
+      "step": 6420
+    },
+    {
+      "epoch": 12.01,
+      "grad_norm": 8.314268112182617,
+      "learning_rate": 3.996268656716418e-05,
+      "loss": 0.0558,
+      "step": 6440
+    },
+    {
+      "epoch": 12.05,
+      "grad_norm": 11.97716999053955,
+      "learning_rate": 3.977611940298508e-05,
+      "loss": 0.1807,
+      "step": 6460
+    },
+    {
+      "epoch": 12.09,
+      "grad_norm": 18.473499298095703,
+      "learning_rate": 3.958955223880597e-05,
+      "loss": 0.1878,
+      "step": 6480
+    },
+    {
+      "epoch": 12.13,
+      "grad_norm": 0.020656852051615715,
+      "learning_rate": 3.940298507462687e-05,
+      "loss": 0.1028,
+      "step": 6500
+    },
+    {
+      "epoch": 12.13,
+      "eval_accuracy": 0.8681898066783831,
+      "eval_loss": 0.6771050691604614,
+      "eval_runtime": 16.9939,
+      "eval_samples_per_second": 33.483,
+      "eval_steps_per_second": 4.237,
+      "step": 6500
+    },
+    {
+      "epoch": 12.16,
+      "grad_norm": 8.395733833312988,
+      "learning_rate": 3.921641791044776e-05,
+      "loss": 0.1958,
+      "step": 6520
+    },
+    {
+      "epoch": 12.2,
+      "grad_norm": 10.099215507507324,
+      "learning_rate": 3.902985074626866e-05,
+      "loss": 0.3174,
+      "step": 6540
+    },
+    {
+      "epoch": 12.24,
+      "grad_norm": 0.006109456531703472,
+      "learning_rate": 3.8843283582089554e-05,
+      "loss": 0.1271,
+      "step": 6560
+    },
+    {
+      "epoch": 12.28,
+      "grad_norm": 3.6905906200408936,
+      "learning_rate": 3.8656716417910446e-05,
+      "loss": 0.1594,
+      "step": 6580
+    },
+    {
+      "epoch": 12.31,
+      "grad_norm": 2.759265899658203,
+      "learning_rate": 3.8470149253731345e-05,
+      "loss": 0.1347,
+      "step": 6600
+    },
+    {
+      "epoch": 12.31,
+      "eval_accuracy": 0.8699472759226714,
+      "eval_loss": 0.6840968728065491,
+      "eval_runtime": 17.0679,
+      "eval_samples_per_second": 33.337,
+      "eval_steps_per_second": 4.218,
+      "step": 6600
+    },
+    {
+      "epoch": 12.35,
+      "grad_norm": 7.070450782775879,
+      "learning_rate": 3.828358208955224e-05,
+      "loss": 0.2115,
+      "step": 6620
+    },
+    {
+      "epoch": 12.39,
+      "grad_norm": 2.8939919471740723,
+      "learning_rate": 3.809701492537314e-05,
+      "loss": 0.1757,
+      "step": 6640
+    },
+    {
+      "epoch": 12.43,
+      "grad_norm": 1.2387713193893433,
+      "learning_rate": 3.791044776119403e-05,
+      "loss": 0.1546,
+      "step": 6660
+    },
+    {
+      "epoch": 12.46,
+      "grad_norm": 1.7357149124145508,
+      "learning_rate": 3.772388059701493e-05,
+      "loss": 0.1793,
+      "step": 6680
+    },
+    {
+      "epoch": 12.5,
+      "grad_norm": 2.236783981323242,
+      "learning_rate": 3.753731343283583e-05,
+      "loss": 0.1269,
+      "step": 6700
+    },
+    {
+      "epoch": 12.5,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.7225701808929443,
+      "eval_runtime": 17.0652,
+      "eval_samples_per_second": 33.343,
+      "eval_steps_per_second": 4.219,
+      "step": 6700
+    },
+    {
+      "epoch": 12.54,
+      "grad_norm": 8.641979217529297,
+      "learning_rate": 3.735074626865671e-05,
+      "loss": 0.206,
+      "step": 6720
+    },
+    {
+      "epoch": 12.57,
+      "grad_norm": 4.790677070617676,
+      "learning_rate": 3.716417910447761e-05,
+      "loss": 0.0905,
+      "step": 6740
+    },
+    {
+      "epoch": 12.61,
+      "grad_norm": 0.05156024172902107,
+      "learning_rate": 3.6977611940298505e-05,
+      "loss": 0.2714,
+      "step": 6760
+    },
+    {
+      "epoch": 12.65,
+      "grad_norm": 0.008736786432564259,
+      "learning_rate": 3.6791044776119404e-05,
+      "loss": 0.1353,
+      "step": 6780
+    },
+    {
+      "epoch": 12.69,
+      "grad_norm": 9.966650009155273,
+      "learning_rate": 3.66044776119403e-05,
+      "loss": 0.2288,
+      "step": 6800
+    },
+    {
+      "epoch": 12.69,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.7082507014274597,
+      "eval_runtime": 17.0545,
+      "eval_samples_per_second": 33.364,
+      "eval_steps_per_second": 4.222,
+      "step": 6800
+    },
+    {
+      "epoch": 12.72,
+      "grad_norm": 9.128669738769531,
+      "learning_rate": 3.6417910447761196e-05,
+      "loss": 0.2364,
+      "step": 6820
+    },
+    {
+      "epoch": 12.76,
+      "grad_norm": 13.353703498840332,
+      "learning_rate": 3.6231343283582095e-05,
+      "loss": 0.1751,
+      "step": 6840
+    },
+    {
+      "epoch": 12.8,
+      "grad_norm": 0.0016426431247964501,
+      "learning_rate": 3.604477611940299e-05,
+      "loss": 0.1218,
+      "step": 6860
+    },
+    {
+      "epoch": 12.84,
+      "grad_norm": 21.353153228759766,
+      "learning_rate": 3.585820895522388e-05,
+      "loss": 0.2285,
+      "step": 6880
+    },
+    {
+      "epoch": 12.87,
+      "grad_norm": 0.005000925622880459,
+      "learning_rate": 3.567164179104478e-05,
+      "loss": 0.1094,
+      "step": 6900
+    },
+    {
+      "epoch": 12.87,
+      "eval_accuracy": 0.8471001757469244,
+      "eval_loss": 0.7455118298530579,
+      "eval_runtime": 16.989,
+      "eval_samples_per_second": 33.492,
+      "eval_steps_per_second": 4.238,
+      "step": 6900
+    },
+    {
+      "epoch": 12.91,
+      "grad_norm": 0.06411085277795792,
+      "learning_rate": 3.548507462686567e-05,
+      "loss": 0.0722,
+      "step": 6920
+    },
+    {
+      "epoch": 12.95,
+      "grad_norm": 0.025857912376523018,
+      "learning_rate": 3.529850746268657e-05,
+      "loss": 0.1329,
+      "step": 6940
+    },
+    {
+      "epoch": 12.99,
+      "grad_norm": 0.33732709288597107,
+      "learning_rate": 3.511194029850746e-05,
+      "loss": 0.3328,
+      "step": 6960
+    },
+    {
+      "epoch": 13.02,
+      "grad_norm": 5.887298583984375,
+      "learning_rate": 3.492537313432836e-05,
+      "loss": 0.231,
+      "step": 6980
+    },
+    {
+      "epoch": 13.06,
+      "grad_norm": 0.004036621656268835,
+      "learning_rate": 3.4738805970149254e-05,
+      "loss": 0.0661,
+      "step": 7000
+    },
+    {
+      "epoch": 13.06,
+      "eval_accuracy": 0.8541300527240774,
+      "eval_loss": 0.7329704165458679,
+      "eval_runtime": 17.0078,
+      "eval_samples_per_second": 33.455,
+      "eval_steps_per_second": 4.233,
+      "step": 7000
+    },
+    {
+      "epoch": 13.1,
+      "grad_norm": 15.431544303894043,
+      "learning_rate": 3.455223880597015e-05,
+      "loss": 0.201,
+      "step": 7020
+    },
+    {
+      "epoch": 13.13,
+      "grad_norm": 5.1752448081970215,
+      "learning_rate": 3.4365671641791046e-05,
+      "loss": 0.0862,
+      "step": 7040
+    },
+    {
+      "epoch": 13.17,
+      "grad_norm": 1.1051908731460571,
+      "learning_rate": 3.417910447761194e-05,
+      "loss": 0.265,
+      "step": 7060
+    },
+    {
+      "epoch": 13.21,
+      "grad_norm": 11.935373306274414,
+      "learning_rate": 3.399253731343284e-05,
+      "loss": 0.1464,
+      "step": 7080
+    },
+    {
+      "epoch": 13.25,
+      "grad_norm": 0.6996492743492126,
+      "learning_rate": 3.380597014925374e-05,
+      "loss": 0.1811,
+      "step": 7100
+    },
+    {
+      "epoch": 13.25,
+      "eval_accuracy": 0.843585237258348,
+      "eval_loss": 0.7363202571868896,
+      "eval_runtime": 17.0328,
+      "eval_samples_per_second": 33.406,
+      "eval_steps_per_second": 4.227,
+      "step": 7100
+    },
+    {
+      "epoch": 13.28,
+      "grad_norm": 6.990137577056885,
+      "learning_rate": 3.361940298507463e-05,
+      "loss": 0.1367,
+      "step": 7120
+    },
+    {
+      "epoch": 13.32,
+      "grad_norm": 1.822852611541748,
+      "learning_rate": 3.343283582089553e-05,
+      "loss": 0.1357,
+      "step": 7140
+    },
+    {
+      "epoch": 13.36,
+      "grad_norm": 0.07311952859163284,
+      "learning_rate": 3.3246268656716414e-05,
+      "loss": 0.1033,
+      "step": 7160
+    },
+    {
+      "epoch": 13.4,
+      "grad_norm": 0.17573679983615875,
+      "learning_rate": 3.305970149253731e-05,
+      "loss": 0.1454,
+      "step": 7180
+    },
+    {
+      "epoch": 13.43,
+      "grad_norm": 17.635812759399414,
+      "learning_rate": 3.287313432835821e-05,
+      "loss": 0.2225,
+      "step": 7200
+    },
+    {
+      "epoch": 13.43,
+      "eval_accuracy": 0.8453427065026362,
+      "eval_loss": 0.7756754755973816,
+      "eval_runtime": 16.9088,
+      "eval_samples_per_second": 33.651,
+      "eval_steps_per_second": 4.258,
+      "step": 7200
+    },
+    {
+      "epoch": 13.47,
+      "grad_norm": 0.03351084142923355,
+      "learning_rate": 3.2686567164179105e-05,
+      "loss": 0.2215,
+      "step": 7220
+    },
+    {
+      "epoch": 13.51,
+      "grad_norm": 2.7680368423461914,
+      "learning_rate": 3.2500000000000004e-05,
+      "loss": 0.2434,
+      "step": 7240
+    },
+    {
+      "epoch": 13.54,
+      "grad_norm": 10.765890121459961,
+      "learning_rate": 3.2313432835820896e-05,
+      "loss": 0.171,
+      "step": 7260
+    },
+    {
+      "epoch": 13.58,
+      "grad_norm": 0.16040430963039398,
+      "learning_rate": 3.2126865671641796e-05,
+      "loss": 0.1097,
+      "step": 7280
+    },
+    {
+      "epoch": 13.62,
+      "grad_norm": 0.061374932527542114,
+      "learning_rate": 3.194029850746269e-05,
+      "loss": 0.1619,
+      "step": 7300
+    },
+    {
+      "epoch": 13.62,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.7361024618148804,
+      "eval_runtime": 17.1751,
+      "eval_samples_per_second": 33.129,
+      "eval_steps_per_second": 4.192,
+      "step": 7300
+    },
+    {
+      "epoch": 13.66,
+      "grad_norm": 1.8059743642807007,
+      "learning_rate": 3.175373134328358e-05,
+      "loss": 0.1061,
+      "step": 7320
+    },
+    {
+      "epoch": 13.69,
+      "grad_norm": 0.16721239686012268,
+      "learning_rate": 3.156716417910448e-05,
+      "loss": 0.1472,
+      "step": 7340
+    },
+    {
+      "epoch": 13.73,
+      "grad_norm": 0.06868579983711243,
+      "learning_rate": 3.138059701492537e-05,
+      "loss": 0.1798,
+      "step": 7360
+    },
+    {
+      "epoch": 13.77,
+      "grad_norm": 6.617785930633545,
+      "learning_rate": 3.119402985074627e-05,
+      "loss": 0.1685,
+      "step": 7380
+    },
+    {
+      "epoch": 13.81,
+      "grad_norm": 14.757624626159668,
+      "learning_rate": 3.100746268656717e-05,
+      "loss": 0.2032,
+      "step": 7400
+    },
+    {
+      "epoch": 13.81,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.7656068801879883,
+      "eval_runtime": 17.0207,
+      "eval_samples_per_second": 33.43,
+      "eval_steps_per_second": 4.23,
+      "step": 7400
+    },
+    {
+      "epoch": 13.84,
+      "grad_norm": 0.21257492899894714,
+      "learning_rate": 3.082089552238806e-05,
+      "loss": 0.1963,
+      "step": 7420
+    },
+    {
+      "epoch": 13.88,
+      "grad_norm": 5.779119491577148,
+      "learning_rate": 3.0634328358208955e-05,
+      "loss": 0.2829,
+      "step": 7440
+    },
+    {
+      "epoch": 13.92,
+      "grad_norm": 0.8466629385948181,
+      "learning_rate": 3.044776119402985e-05,
+      "loss": 0.1115,
+      "step": 7460
+    },
+    {
+      "epoch": 13.96,
+      "grad_norm": 11.776174545288086,
+      "learning_rate": 3.0261194029850747e-05,
+      "loss": 0.2296,
+      "step": 7480
+    },
+    {
+      "epoch": 13.99,
+      "grad_norm": 0.8353527784347534,
+      "learning_rate": 3.0074626865671646e-05,
+      "loss": 0.0216,
+      "step": 7500
+    },
+    {
+      "epoch": 13.99,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.776033878326416,
+      "eval_runtime": 17.075,
+      "eval_samples_per_second": 33.324,
+      "eval_steps_per_second": 4.217,
+      "step": 7500
+    },
+    {
+      "epoch": 14.03,
+      "grad_norm": 0.020064346492290497,
+      "learning_rate": 2.9888059701492538e-05,
+      "loss": 0.1374,
+      "step": 7520
+    },
+    {
+      "epoch": 14.07,
+      "grad_norm": 0.5000738501548767,
+      "learning_rate": 2.9701492537313434e-05,
+      "loss": 0.1624,
+      "step": 7540
+    },
+    {
+      "epoch": 14.1,
+      "grad_norm": 6.905271530151367,
+      "learning_rate": 2.9514925373134326e-05,
+      "loss": 0.1382,
+      "step": 7560
+    },
+    {
+      "epoch": 14.14,
+      "grad_norm": 0.021415790542960167,
+      "learning_rate": 2.9328358208955226e-05,
+      "loss": 0.063,
+      "step": 7580
+    },
+    {
+      "epoch": 14.18,
+      "grad_norm": 9.256654739379883,
+      "learning_rate": 2.9141791044776125e-05,
+      "loss": 0.2476,
+      "step": 7600
+    },
+    {
+      "epoch": 14.18,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.7723203897476196,
+      "eval_runtime": 17.11,
+      "eval_samples_per_second": 33.255,
+      "eval_steps_per_second": 4.208,
+      "step": 7600
+    },
+    {
+      "epoch": 14.22,
+      "grad_norm": 0.5965019464492798,
+      "learning_rate": 2.8955223880597017e-05,
+      "loss": 0.1495,
+      "step": 7620
+    },
+    {
+      "epoch": 14.25,
+      "grad_norm": 10.70635986328125,
+      "learning_rate": 2.8768656716417913e-05,
+      "loss": 0.2356,
+      "step": 7640
+    },
+    {
+      "epoch": 14.29,
+      "grad_norm": 0.5673186182975769,
+      "learning_rate": 2.8582089552238805e-05,
+      "loss": 0.1841,
+      "step": 7660
+    },
+    {
+      "epoch": 14.33,
+      "grad_norm": 0.03677751123905182,
+      "learning_rate": 2.8395522388059705e-05,
+      "loss": 0.0926,
+      "step": 7680
+    },
+    {
+      "epoch": 14.37,
+      "grad_norm": 6.326465606689453,
+      "learning_rate": 2.82089552238806e-05,
+      "loss": 0.1616,
+      "step": 7700
+    },
+    {
+      "epoch": 14.37,
+      "eval_accuracy": 0.8787346221441125,
+      "eval_loss": 0.724672794342041,
+      "eval_runtime": 17.0499,
+      "eval_samples_per_second": 33.373,
+      "eval_steps_per_second": 4.223,
+      "step": 7700
+    },
+    {
+      "epoch": 14.4,
+      "grad_norm": 3.3912713527679443,
+      "learning_rate": 2.8022388059701493e-05,
+      "loss": 0.1158,
+      "step": 7720
+    },
+    {
+      "epoch": 14.44,
+      "grad_norm": 0.4542391002178192,
+      "learning_rate": 2.7835820895522392e-05,
+      "loss": 0.1064,
+      "step": 7740
+    },
+    {
+      "epoch": 14.48,
+      "grad_norm": 0.004322316497564316,
+      "learning_rate": 2.7649253731343284e-05,
+      "loss": 0.0728,
+      "step": 7760
+    },
+    {
+      "epoch": 14.51,
+      "grad_norm": 0.16555258631706238,
+      "learning_rate": 2.746268656716418e-05,
+      "loss": 0.1623,
+      "step": 7780
+    },
+    {
+      "epoch": 14.55,
+      "grad_norm": 1.5429511070251465,
+      "learning_rate": 2.727611940298508e-05,
+      "loss": 0.1142,
+      "step": 7800
+    },
+    {
+      "epoch": 14.55,
+      "eval_accuracy": 0.8699472759226714,
+      "eval_loss": 0.7907147407531738,
+      "eval_runtime": 17.0225,
+      "eval_samples_per_second": 33.426,
+      "eval_steps_per_second": 4.23,
+      "step": 7800
+    },
+    {
+      "epoch": 14.59,
+      "grad_norm": 5.796298027038574,
+      "learning_rate": 2.7089552238805972e-05,
+      "loss": 0.2558,
+      "step": 7820
+    },
+    {
+      "epoch": 14.63,
+      "grad_norm": 0.09839625656604767,
+      "learning_rate": 2.6902985074626868e-05,
+      "loss": 0.1097,
+      "step": 7840
+    },
+    {
+      "epoch": 14.66,
+      "grad_norm": 5.781641483306885,
+      "learning_rate": 2.671641791044776e-05,
+      "loss": 0.1776,
+      "step": 7860
+    },
+    {
+      "epoch": 14.7,
+      "grad_norm": 0.6178101301193237,
+      "learning_rate": 2.652985074626866e-05,
+      "loss": 0.1488,
+      "step": 7880
+    },
+    {
+      "epoch": 14.74,
+      "grad_norm": 5.687034606933594,
+      "learning_rate": 2.6343283582089555e-05,
+      "loss": 0.0906,
+      "step": 7900
+    },
+    {
+      "epoch": 14.74,
+      "eval_accuracy": 0.8646748681898067,
+      "eval_loss": 0.78293776512146,
+      "eval_runtime": 16.9166,
+      "eval_samples_per_second": 33.636,
+      "eval_steps_per_second": 4.256,
+      "step": 7900
+    },
+    {
+      "epoch": 14.78,
+      "grad_norm": 9.950879096984863,
+      "learning_rate": 2.6156716417910447e-05,
+      "loss": 0.1565,
+      "step": 7920
+    },
+    {
+      "epoch": 14.81,
+      "grad_norm": 9.66186237335205,
+      "learning_rate": 2.5970149253731346e-05,
+      "loss": 0.1013,
+      "step": 7940
+    },
+    {
+      "epoch": 14.85,
+      "grad_norm": 10.813685417175293,
+      "learning_rate": 2.578358208955224e-05,
+      "loss": 0.1255,
+      "step": 7960
+    },
+    {
+      "epoch": 14.89,
+      "grad_norm": 11.837493896484375,
+      "learning_rate": 2.5597014925373135e-05,
+      "loss": 0.1535,
+      "step": 7980
+    },
+    {
+      "epoch": 14.93,
+      "grad_norm": 2.3248374462127686,
+      "learning_rate": 2.5419776119402987e-05,
+      "loss": 0.2199,
+      "step": 8000
+    },
+    {
+      "epoch": 14.93,
+      "eval_accuracy": 0.8717047451669596,
+      "eval_loss": 0.7426685690879822,
+      "eval_runtime": 16.9375,
+      "eval_samples_per_second": 33.594,
+      "eval_steps_per_second": 4.251,
+      "step": 8000
+    },
+    {
+      "epoch": 14.96,
+      "grad_norm": 1.272861123085022,
+      "learning_rate": 2.523320895522388e-05,
+      "loss": 0.1287,
+      "step": 8020
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 0.7936806082725525,
+      "learning_rate": 2.5046641791044778e-05,
+      "loss": 0.15,
+      "step": 8040
+    },
+    {
+      "epoch": 15.04,
+      "grad_norm": 14.286187171936035,
+      "learning_rate": 2.4860074626865674e-05,
+      "loss": 0.2127,
+      "step": 8060
+    },
+    {
+      "epoch": 15.07,
+      "grad_norm": 5.237728118896484,
+      "learning_rate": 2.467350746268657e-05,
+      "loss": 0.0859,
+      "step": 8080
+    },
+    {
+      "epoch": 15.11,
+      "grad_norm": 0.36676111817359924,
+      "learning_rate": 2.4486940298507462e-05,
+      "loss": 0.0643,
+      "step": 8100
+    },
+    {
+      "epoch": 15.11,
+      "eval_accuracy": 0.8699472759226714,
+      "eval_loss": 0.7280339002609253,
+      "eval_runtime": 17.1154,
+      "eval_samples_per_second": 33.245,
+      "eval_steps_per_second": 4.207,
+      "step": 8100
+    },
+    {
+      "epoch": 15.15,
+      "grad_norm": 6.870665073394775,
+      "learning_rate": 2.4300373134328358e-05,
+      "loss": 0.1615,
+      "step": 8120
+    },
+    {
+      "epoch": 15.19,
+      "grad_norm": 7.524236679077148,
+      "learning_rate": 2.4113805970149257e-05,
+      "loss": 0.146,
+      "step": 8140
+    },
+    {
+      "epoch": 15.22,
+      "grad_norm": 0.026915544643998146,
+      "learning_rate": 2.392723880597015e-05,
+      "loss": 0.122,
+      "step": 8160
+    },
+    {
+      "epoch": 15.26,
+      "grad_norm": 0.047888416796922684,
+      "learning_rate": 2.3740671641791045e-05,
+      "loss": 0.12,
+      "step": 8180
+    },
+    {
+      "epoch": 15.3,
+      "grad_norm": 0.0014809180283918977,
+      "learning_rate": 2.355410447761194e-05,
+      "loss": 0.1685,
+      "step": 8200
+    },
+    {
+      "epoch": 15.3,
+      "eval_accuracy": 0.8541300527240774,
+      "eval_loss": 0.8381420373916626,
+      "eval_runtime": 17.0832,
+      "eval_samples_per_second": 33.308,
+      "eval_steps_per_second": 4.215,
+      "step": 8200
+    },
+    {
+      "epoch": 15.34,
+      "grad_norm": 0.002554853679612279,
+      "learning_rate": 2.3367537313432837e-05,
+      "loss": 0.1104,
+      "step": 8220
+    },
+    {
+      "epoch": 15.37,
+      "grad_norm": 1.5922454595565796,
+      "learning_rate": 2.3180970149253733e-05,
+      "loss": 0.1104,
+      "step": 8240
+    },
+    {
+      "epoch": 15.41,
+      "grad_norm": 17.750452041625977,
+      "learning_rate": 2.299440298507463e-05,
+      "loss": 0.1665,
+      "step": 8260
+    },
+    {
+      "epoch": 15.45,
+      "grad_norm": 24.73293113708496,
+      "learning_rate": 2.2807835820895524e-05,
+      "loss": 0.0949,
+      "step": 8280
+    },
+    {
+      "epoch": 15.49,
+      "grad_norm": 10.69165325164795,
+      "learning_rate": 2.262126865671642e-05,
+      "loss": 0.1677,
+      "step": 8300
+    },
+    {
+      "epoch": 15.49,
+      "eval_accuracy": 0.8506151142355008,
+      "eval_loss": 0.8638322949409485,
+      "eval_runtime": 17.046,
+      "eval_samples_per_second": 33.38,
+      "eval_steps_per_second": 4.224,
+      "step": 8300
+    },
+    {
+      "epoch": 15.52,
+      "grad_norm": 0.0646175667643547,
+      "learning_rate": 2.2434701492537312e-05,
+      "loss": 0.1934,
+      "step": 8320
+    },
+    {
+      "epoch": 15.56,
+      "grad_norm": 24.913108825683594,
+      "learning_rate": 2.224813432835821e-05,
+      "loss": 0.1628,
+      "step": 8340
+    },
+    {
+      "epoch": 15.6,
+      "grad_norm": 0.015693532302975655,
+      "learning_rate": 2.2061567164179107e-05,
+      "loss": 0.2041,
+      "step": 8360
+    },
+    {
+      "epoch": 15.63,
+      "grad_norm": 0.037644147872924805,
+      "learning_rate": 2.1875e-05,
+      "loss": 0.0673,
+      "step": 8380
+    },
+    {
+      "epoch": 15.67,
+      "grad_norm": 0.4659203588962555,
+      "learning_rate": 2.1688432835820896e-05,
+      "loss": 0.1399,
+      "step": 8400
+    },
+    {
+      "epoch": 15.67,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.8422539234161377,
+      "eval_runtime": 17.0579,
+      "eval_samples_per_second": 33.357,
+      "eval_steps_per_second": 4.221,
+      "step": 8400
+    },
+    {
+      "epoch": 15.71,
+      "grad_norm": 0.547573983669281,
+      "learning_rate": 2.150186567164179e-05,
+      "loss": 0.0836,
+      "step": 8420
+    },
+    {
+      "epoch": 15.75,
+      "grad_norm": 2.5873095989227295,
+      "learning_rate": 2.1315298507462687e-05,
+      "loss": 0.2234,
+      "step": 8440
+    },
+    {
+      "epoch": 15.78,
+      "grad_norm": 0.00556514598429203,
+      "learning_rate": 2.1128731343283583e-05,
+      "loss": 0.1347,
+      "step": 8460
+    },
+    {
+      "epoch": 15.82,
+      "grad_norm": 10.745226860046387,
+      "learning_rate": 2.094216417910448e-05,
+      "loss": 0.115,
+      "step": 8480
+    },
+    {
+      "epoch": 15.86,
+      "grad_norm": 0.016960641369223595,
+      "learning_rate": 2.0755597014925375e-05,
+      "loss": 0.1041,
+      "step": 8500
+    },
+    {
+      "epoch": 15.86,
+      "eval_accuracy": 0.8541300527240774,
+      "eval_loss": 0.8050984144210815,
+      "eval_runtime": 17.0123,
+      "eval_samples_per_second": 33.446,
+      "eval_steps_per_second": 4.232,
+      "step": 8500
+    },
+    {
+      "epoch": 15.9,
+      "grad_norm": 12.282808303833008,
+      "learning_rate": 2.056902985074627e-05,
+      "loss": 0.1409,
+      "step": 8520
+    },
+    {
+      "epoch": 15.93,
+      "grad_norm": 12.106451034545898,
+      "learning_rate": 2.0382462686567166e-05,
+      "loss": 0.1204,
+      "step": 8540
+    },
+    {
+      "epoch": 15.97,
+      "grad_norm": 6.440662384033203,
+      "learning_rate": 2.0195895522388062e-05,
+      "loss": 0.082,
+      "step": 8560
+    },
+    {
+      "epoch": 16.01,
+      "grad_norm": 10.880598068237305,
+      "learning_rate": 2.0009328358208958e-05,
+      "loss": 0.1609,
+      "step": 8580
+    },
+    {
+      "epoch": 16.04,
+      "grad_norm": 4.452532768249512,
+      "learning_rate": 1.982276119402985e-05,
+      "loss": 0.2223,
+      "step": 8600
+    },
+    {
+      "epoch": 16.04,
+      "eval_accuracy": 0.8646748681898067,
+      "eval_loss": 0.7767916321754456,
+      "eval_runtime": 17.1443,
+      "eval_samples_per_second": 33.189,
+      "eval_steps_per_second": 4.2,
+      "step": 8600
+    },
+    {
+      "epoch": 16.08,
+      "grad_norm": 0.23833756148815155,
+      "learning_rate": 1.9636194029850746e-05,
+      "loss": 0.0848,
+      "step": 8620
+    },
+    {
+      "epoch": 16.12,
+      "grad_norm": 6.388860702514648,
+      "learning_rate": 1.9449626865671645e-05,
+      "loss": 0.1208,
+      "step": 8640
+    },
+    {
+      "epoch": 16.16,
+      "grad_norm": 5.347766399383545,
+      "learning_rate": 1.9263059701492537e-05,
+      "loss": 0.0911,
+      "step": 8660
+    },
+    {
+      "epoch": 16.19,
+      "grad_norm": 0.004135070834308863,
+      "learning_rate": 1.9076492537313433e-05,
+      "loss": 0.078,
+      "step": 8680
+    },
+    {
+      "epoch": 16.23,
+      "grad_norm": 0.5564485192298889,
+      "learning_rate": 1.888992537313433e-05,
+      "loss": 0.1016,
+      "step": 8700
+    },
+    {
+      "epoch": 16.23,
+      "eval_accuracy": 0.8646748681898067,
+      "eval_loss": 0.7965061068534851,
+      "eval_runtime": 17.0034,
+      "eval_samples_per_second": 33.464,
+      "eval_steps_per_second": 4.234,
+      "step": 8700
+    },
+    {
+      "epoch": 16.27,
+      "grad_norm": 0.3795854449272156,
+      "learning_rate": 1.8703358208955225e-05,
+      "loss": 0.1023,
+      "step": 8720
+    },
+    {
+      "epoch": 16.31,
+      "grad_norm": 0.0014576130779460073,
+      "learning_rate": 1.8516791044776117e-05,
+      "loss": 0.1396,
+      "step": 8740
+    },
+    {
+      "epoch": 16.34,
+      "grad_norm": 11.313010215759277,
+      "learning_rate": 1.8339552238805972e-05,
+      "loss": 0.2095,
+      "step": 8760
+    },
+    {
+      "epoch": 16.38,
+      "grad_norm": 10.762919425964355,
+      "learning_rate": 1.8152985074626865e-05,
+      "loss": 0.1846,
+      "step": 8780
+    },
+    {
+      "epoch": 16.42,
+      "grad_norm": 6.421539783477783,
+      "learning_rate": 1.796641791044776e-05,
+      "loss": 0.065,
+      "step": 8800
+    },
+    {
+      "epoch": 16.42,
+      "eval_accuracy": 0.8418277680140598,
+      "eval_loss": 0.8331155180931091,
+      "eval_runtime": 16.9631,
+      "eval_samples_per_second": 33.543,
+      "eval_steps_per_second": 4.245,
+      "step": 8800
+    },
+    {
+      "epoch": 16.46,
+      "grad_norm": 0.0021264124661684036,
+      "learning_rate": 1.777985074626866e-05,
+      "loss": 0.2048,
+      "step": 8820
+    },
+    {
+      "epoch": 16.49,
+      "grad_norm": 5.9840569496154785,
+      "learning_rate": 1.7593283582089552e-05,
+      "loss": 0.0313,
+      "step": 8840
+    },
+    {
+      "epoch": 16.53,
+      "grad_norm": 0.011331523768603802,
+      "learning_rate": 1.7406716417910448e-05,
+      "loss": 0.0916,
+      "step": 8860
+    },
+    {
+      "epoch": 16.57,
+      "grad_norm": 2.43318510055542,
+      "learning_rate": 1.7220149253731344e-05,
+      "loss": 0.1146,
+      "step": 8880
+    },
+    {
+      "epoch": 16.6,
+      "grad_norm": 3.961604118347168,
+      "learning_rate": 1.703358208955224e-05,
+      "loss": 0.1156,
+      "step": 8900
+    },
+    {
+      "epoch": 16.6,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.8023034334182739,
+      "eval_runtime": 17.1344,
+      "eval_samples_per_second": 33.208,
+      "eval_steps_per_second": 4.202,
+      "step": 8900
+    },
+    {
+      "epoch": 16.64,
+      "grad_norm": 0.04533189907670021,
+      "learning_rate": 1.6847014925373135e-05,
+      "loss": 0.1438,
+      "step": 8920
+    },
+    {
+      "epoch": 16.68,
+      "grad_norm": 2.201357841491699,
+      "learning_rate": 1.666044776119403e-05,
+      "loss": 0.1182,
+      "step": 8940
+    },
+    {
+      "epoch": 16.72,
+      "grad_norm": 0.2703014016151428,
+      "learning_rate": 1.6473880597014927e-05,
+      "loss": 0.0762,
+      "step": 8960
+    },
+    {
+      "epoch": 16.75,
+      "grad_norm": 0.0025351608637720346,
+      "learning_rate": 1.6287313432835823e-05,
+      "loss": 0.0937,
+      "step": 8980
+    },
+    {
+      "epoch": 16.79,
+      "grad_norm": 0.0020602380391210318,
+      "learning_rate": 1.6100746268656715e-05,
+      "loss": 0.2263,
+      "step": 9000
+    },
+    {
+      "epoch": 16.79,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.811631977558136,
+      "eval_runtime": 17.2889,
+      "eval_samples_per_second": 32.911,
+      "eval_steps_per_second": 4.165,
+      "step": 9000
+    },
+    {
+      "epoch": 16.83,
+      "grad_norm": 0.00895586609840393,
+      "learning_rate": 1.5914179104477614e-05,
+      "loss": 0.082,
+      "step": 9020
+    },
+    {
+      "epoch": 16.87,
+      "grad_norm": 4.494884014129639,
+      "learning_rate": 1.572761194029851e-05,
+      "loss": 0.0447,
+      "step": 9040
+    },
+    {
+      "epoch": 16.9,
+      "grad_norm": 18.793230056762695,
+      "learning_rate": 1.5541044776119403e-05,
+      "loss": 0.1469,
+      "step": 9060
+    },
+    {
+      "epoch": 16.94,
+      "grad_norm": 14.875761032104492,
+      "learning_rate": 1.53544776119403e-05,
+      "loss": 0.1156,
+      "step": 9080
+    },
+    {
+      "epoch": 16.98,
+      "grad_norm": 0.11905315518379211,
+      "learning_rate": 1.5167910447761194e-05,
+      "loss": 0.1197,
+      "step": 9100
+    },
+    {
+      "epoch": 16.98,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.8489682674407959,
+      "eval_runtime": 16.9217,
+      "eval_samples_per_second": 33.625,
+      "eval_steps_per_second": 4.255,
+      "step": 9100
+    },
+    {
+      "epoch": 17.01,
+      "grad_norm": 0.0026268402580171824,
+      "learning_rate": 1.4981343283582092e-05,
+      "loss": 0.0712,
+      "step": 9120
+    },
+    {
+      "epoch": 17.05,
+      "grad_norm": 9.214509963989258,
+      "learning_rate": 1.4794776119402986e-05,
+      "loss": 0.1138,
+      "step": 9140
+    },
+    {
+      "epoch": 17.09,
+      "grad_norm": 0.004354454111307859,
+      "learning_rate": 1.4608208955223882e-05,
+      "loss": 0.1392,
+      "step": 9160
+    },
+    {
+      "epoch": 17.13,
+      "grad_norm": 0.01067269779741764,
+      "learning_rate": 1.4421641791044777e-05,
+      "loss": 0.1146,
+      "step": 9180
+    },
+    {
+      "epoch": 17.16,
+      "grad_norm": 7.142704486846924,
+      "learning_rate": 1.4235074626865671e-05,
+      "loss": 0.1931,
+      "step": 9200
+    },
+    {
+      "epoch": 17.16,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.8193662166595459,
+      "eval_runtime": 17.0722,
+      "eval_samples_per_second": 33.329,
+      "eval_steps_per_second": 4.217,
+      "step": 9200
+    },
+    {
+      "epoch": 17.2,
+      "grad_norm": 0.0009618746698834002,
+      "learning_rate": 1.4048507462686567e-05,
+      "loss": 0.0797,
+      "step": 9220
+    },
+    {
+      "epoch": 17.24,
+      "grad_norm": 0.00881529413163662,
+      "learning_rate": 1.3861940298507465e-05,
+      "loss": 0.1337,
+      "step": 9240
+    },
+    {
+      "epoch": 17.28,
+      "grad_norm": 2.5846807956695557,
+      "learning_rate": 1.3675373134328359e-05,
+      "loss": 0.0745,
+      "step": 9260
+    },
+    {
+      "epoch": 17.31,
+      "grad_norm": 0.1657407432794571,
+      "learning_rate": 1.3488805970149255e-05,
+      "loss": 0.1334,
+      "step": 9280
+    },
+    {
+      "epoch": 17.35,
+      "grad_norm": 0.15658806264400482,
+      "learning_rate": 1.3302238805970149e-05,
+      "loss": 0.1289,
+      "step": 9300
+    },
+    {
+      "epoch": 17.35,
+      "eval_accuracy": 0.8488576449912126,
+      "eval_loss": 0.8352965712547302,
+      "eval_runtime": 17.0215,
+      "eval_samples_per_second": 33.428,
+      "eval_steps_per_second": 4.23,
+      "step": 9300
+    },
+    {
+      "epoch": 17.39,
+      "grad_norm": 0.22855500876903534,
+      "learning_rate": 1.3115671641791044e-05,
+      "loss": 0.0523,
+      "step": 9320
+    },
+    {
+      "epoch": 17.43,
+      "grad_norm": 0.02038070559501648,
+      "learning_rate": 1.2929104477611942e-05,
+      "loss": 0.1233,
+      "step": 9340
+    },
+    {
+      "epoch": 17.46,
+      "grad_norm": 6.642090797424316,
+      "learning_rate": 1.2742537313432836e-05,
+      "loss": 0.1139,
+      "step": 9360
+    },
+    {
+      "epoch": 17.5,
+      "grad_norm": 11.216026306152344,
+      "learning_rate": 1.2555970149253732e-05,
+      "loss": 0.1488,
+      "step": 9380
+    },
+    {
+      "epoch": 17.54,
+      "grad_norm": 0.03946173936128616,
+      "learning_rate": 1.2369402985074626e-05,
+      "loss": 0.2039,
+      "step": 9400
+    },
+    {
+      "epoch": 17.54,
+      "eval_accuracy": 0.8453427065026362,
+      "eval_loss": 0.8163497447967529,
+      "eval_runtime": 17.1586,
+      "eval_samples_per_second": 33.161,
+      "eval_steps_per_second": 4.196,
+      "step": 9400
+    },
+    {
+      "epoch": 17.57,
+      "grad_norm": 0.9558226466178894,
+      "learning_rate": 1.2182835820895523e-05,
+      "loss": 0.1998,
+      "step": 9420
+    },
+    {
+      "epoch": 17.61,
+      "grad_norm": 0.011750847101211548,
+      "learning_rate": 1.1996268656716418e-05,
+      "loss": 0.0569,
+      "step": 9440
+    },
+    {
+      "epoch": 17.65,
+      "grad_norm": 1.238696575164795,
+      "learning_rate": 1.1809701492537315e-05,
+      "loss": 0.148,
+      "step": 9460
+    },
+    {
+      "epoch": 17.69,
+      "grad_norm": 0.08940445631742477,
+      "learning_rate": 1.1623134328358209e-05,
+      "loss": 0.0476,
+      "step": 9480
+    },
+    {
+      "epoch": 17.72,
+      "grad_norm": 6.847604274749756,
+      "learning_rate": 1.1436567164179105e-05,
+      "loss": 0.0825,
+      "step": 9500
+    },
+    {
+      "epoch": 17.72,
+      "eval_accuracy": 0.8523725834797891,
+      "eval_loss": 0.7942213416099548,
+      "eval_runtime": 17.0648,
+      "eval_samples_per_second": 33.343,
+      "eval_steps_per_second": 4.219,
+      "step": 9500
+    },
+    {
+      "epoch": 17.76,
+      "grad_norm": 6.633146286010742,
+      "learning_rate": 1.125e-05,
+      "loss": 0.1269,
+      "step": 9520
+    },
+    {
+      "epoch": 17.8,
+      "grad_norm": 1.072378396987915,
+      "learning_rate": 1.1063432835820897e-05,
+      "loss": 0.1456,
+      "step": 9540
+    },
+    {
+      "epoch": 17.84,
+      "grad_norm": 0.1579461693763733,
+      "learning_rate": 1.0876865671641792e-05,
+      "loss": 0.1114,
+      "step": 9560
+    },
+    {
+      "epoch": 17.87,
+      "grad_norm": 0.007003374397754669,
+      "learning_rate": 1.0690298507462686e-05,
+      "loss": 0.1193,
+      "step": 9580
+    },
+    {
+      "epoch": 17.91,
+      "grad_norm": 0.08359462022781372,
+      "learning_rate": 1.0503731343283582e-05,
+      "loss": 0.0712,
+      "step": 9600
+    },
+    {
+      "epoch": 17.91,
+      "eval_accuracy": 0.8558875219683656,
+      "eval_loss": 0.8027055263519287,
+      "eval_runtime": 17.0834,
+      "eval_samples_per_second": 33.307,
+      "eval_steps_per_second": 4.215,
+      "step": 9600
+    },
+    {
+      "epoch": 17.95,
+      "grad_norm": 0.0008617080166004598,
+      "learning_rate": 1.0317164179104478e-05,
+      "loss": 0.0536,
+      "step": 9620
+    },
+    {
+      "epoch": 17.99,
+      "grad_norm": 14.401564598083496,
+      "learning_rate": 1.0130597014925374e-05,
+      "loss": 0.144,
+      "step": 9640
+    },
+    {
+      "epoch": 18.02,
+      "grad_norm": 0.7699527740478516,
+      "learning_rate": 9.94402985074627e-06,
+      "loss": 0.2091,
+      "step": 9660
+    },
+    {
+      "epoch": 18.06,
+      "grad_norm": 0.009821569547057152,
+      "learning_rate": 9.757462686567165e-06,
+      "loss": 0.1697,
+      "step": 9680
+    },
+    {
+      "epoch": 18.1,
+      "grad_norm": 0.0026506283320486546,
+      "learning_rate": 9.57089552238806e-06,
+      "loss": 0.244,
+      "step": 9700
+    },
+    {
+      "epoch": 18.1,
+      "eval_accuracy": 0.8664323374340949,
+      "eval_loss": 0.7803473472595215,
+      "eval_runtime": 17.1118,
+      "eval_samples_per_second": 33.252,
+      "eval_steps_per_second": 4.208,
+      "step": 9700
+    },
+    {
+      "epoch": 18.13,
+      "grad_norm": 4.237458229064941,
+      "learning_rate": 9.384328358208957e-06,
+      "loss": 0.0916,
+      "step": 9720
+    },
+    {
+      "epoch": 18.17,
+      "grad_norm": 0.013631208799779415,
+      "learning_rate": 9.197761194029851e-06,
+      "loss": 0.1047,
+      "step": 9740
+    },
+    {
+      "epoch": 18.21,
+      "grad_norm": 0.0034202809911221266,
+      "learning_rate": 9.011194029850747e-06,
+      "loss": 0.0155,
+      "step": 9760
+    },
+    {
+      "epoch": 18.25,
+      "grad_norm": 0.0012619507033377886,
+      "learning_rate": 8.824626865671643e-06,
+      "loss": 0.1541,
+      "step": 9780
+    },
+    {
+      "epoch": 18.28,
+      "grad_norm": 0.07258515805006027,
+      "learning_rate": 8.638059701492537e-06,
+      "loss": 0.1482,
+      "step": 9800
+    },
+    {
+      "epoch": 18.28,
+      "eval_accuracy": 0.8629173989455184,
+      "eval_loss": 0.77537602186203,
+      "eval_runtime": 17.1146,
+      "eval_samples_per_second": 33.247,
+      "eval_steps_per_second": 4.207,
+      "step": 9800
+    },
+    {
+      "epoch": 18.32,
+      "grad_norm": 0.02240763045847416,
+      "learning_rate": 8.451492537313434e-06,
+      "loss": 0.0417,
+      "step": 9820
+    },
+    {
+      "epoch": 18.36,
+      "grad_norm": 0.012936491519212723,
+      "learning_rate": 8.264925373134328e-06,
+      "loss": 0.1184,
+      "step": 9840
+    },
+    {
+      "epoch": 18.4,
+      "grad_norm": 0.24189649522304535,
+      "learning_rate": 8.078358208955224e-06,
+      "loss": 0.0735,
+      "step": 9860
+    },
+    {
+      "epoch": 18.43,
+      "grad_norm": 8.71208667755127,
+      "learning_rate": 7.89179104477612e-06,
+      "loss": 0.1064,
+      "step": 9880
+    },
+    {
+      "epoch": 18.47,
+      "grad_norm": 0.06968268752098083,
+      "learning_rate": 7.705223880597016e-06,
+      "loss": 0.1829,
+      "step": 9900
+    },
+    {
+      "epoch": 18.47,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.7809686064720154,
+      "eval_runtime": 17.0651,
+      "eval_samples_per_second": 33.343,
+      "eval_steps_per_second": 4.219,
+      "step": 9900
+    },
+    {
+      "epoch": 18.51,
+      "grad_norm": 0.2621960937976837,
+      "learning_rate": 7.5186567164179115e-06,
+      "loss": 0.0857,
+      "step": 9920
+    },
+    {
+      "epoch": 18.54,
+      "grad_norm": 0.046136654913425446,
+      "learning_rate": 7.332089552238806e-06,
+      "loss": 0.1365,
+      "step": 9940
+    },
+    {
+      "epoch": 18.58,
+      "grad_norm": 7.872718334197998,
+      "learning_rate": 7.145522388059701e-06,
+      "loss": 0.203,
+      "step": 9960
+    },
+    {
+      "epoch": 18.62,
+      "grad_norm": 2.0751500129699707,
+      "learning_rate": 6.958955223880598e-06,
+      "loss": 0.0651,
+      "step": 9980
+    },
+    {
+      "epoch": 18.66,
+      "grad_norm": 8.155353546142578,
+      "learning_rate": 6.772388059701493e-06,
+      "loss": 0.019,
+      "step": 10000
+    },
+    {
+      "epoch": 18.66,
+      "eval_accuracy": 0.8558875219683656,
+      "eval_loss": 0.7971689105033875,
+      "eval_runtime": 17.2413,
+      "eval_samples_per_second": 33.002,
+      "eval_steps_per_second": 4.176,
+      "step": 10000
+    },
+    {
+      "epoch": 18.69,
+      "grad_norm": 0.0101598696783185,
+      "learning_rate": 6.585820895522389e-06,
+      "loss": 0.1177,
+      "step": 10020
+    },
+    {
+      "epoch": 18.73,
+      "grad_norm": 14.113308906555176,
+      "learning_rate": 6.399253731343284e-06,
+      "loss": 0.1711,
+      "step": 10040
+    },
+    {
+      "epoch": 18.77,
+      "grad_norm": 0.21149051189422607,
+      "learning_rate": 6.2126865671641795e-06,
+      "loss": 0.0757,
+      "step": 10060
+    },
+    {
+      "epoch": 18.81,
+      "grad_norm": 0.002241868758574128,
+      "learning_rate": 6.026119402985074e-06,
+      "loss": 0.1009,
+      "step": 10080
+    },
+    {
+      "epoch": 18.84,
+      "grad_norm": 4.295605659484863,
+      "learning_rate": 5.83955223880597e-06,
+      "loss": 0.061,
+      "step": 10100
+    },
+    {
+      "epoch": 18.84,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.8180297613143921,
+      "eval_runtime": 17.031,
+      "eval_samples_per_second": 33.41,
+      "eval_steps_per_second": 4.228,
+      "step": 10100
+    },
+    {
+      "epoch": 18.88,
+      "grad_norm": 0.002981310011819005,
+      "learning_rate": 5.652985074626866e-06,
+      "loss": 0.1798,
+      "step": 10120
+    },
+    {
+      "epoch": 18.92,
+      "grad_norm": 1.8660227060317993,
+      "learning_rate": 5.466417910447762e-06,
+      "loss": 0.1254,
+      "step": 10140
+    },
+    {
+      "epoch": 18.96,
+      "grad_norm": 0.07149626314640045,
+      "learning_rate": 5.279850746268657e-06,
+      "loss": 0.1079,
+      "step": 10160
+    },
+    {
+      "epoch": 18.99,
+      "grad_norm": 18.285696029663086,
+      "learning_rate": 5.0932835820895525e-06,
+      "loss": 0.1836,
+      "step": 10180
+    },
+    {
+      "epoch": 19.03,
+      "grad_norm": 0.1172376275062561,
+      "learning_rate": 4.906716417910448e-06,
+      "loss": 0.117,
+      "step": 10200
+    },
+    {
+      "epoch": 19.03,
+      "eval_accuracy": 0.8558875219683656,
+      "eval_loss": 0.8318653106689453,
+      "eval_runtime": 17.0956,
+      "eval_samples_per_second": 33.283,
+      "eval_steps_per_second": 4.212,
+      "step": 10200
+    },
+    {
+      "epoch": 19.07,
+      "grad_norm": 11.238788604736328,
+      "learning_rate": 4.720149253731344e-06,
+      "loss": 0.1321,
+      "step": 10220
+    },
+    {
+      "epoch": 19.1,
+      "grad_norm": 0.0014608531491830945,
+      "learning_rate": 4.533582089552239e-06,
+      "loss": 0.0755,
+      "step": 10240
+    },
+    {
+      "epoch": 19.14,
+      "grad_norm": 10.21114444732666,
+      "learning_rate": 4.347014925373134e-06,
+      "loss": 0.1943,
+      "step": 10260
+    },
+    {
+      "epoch": 19.18,
+      "grad_norm": 5.378198146820068,
+      "learning_rate": 4.16044776119403e-06,
+      "loss": 0.1181,
+      "step": 10280
+    },
+    {
+      "epoch": 19.22,
+      "grad_norm": 0.7626181840896606,
+      "learning_rate": 3.9738805970149256e-06,
+      "loss": 0.1858,
+      "step": 10300
+    },
+    {
+      "epoch": 19.22,
+      "eval_accuracy": 0.8558875219683656,
+      "eval_loss": 0.8431510925292969,
+      "eval_runtime": 17.0075,
+      "eval_samples_per_second": 33.456,
+      "eval_steps_per_second": 4.233,
+      "step": 10300
+    },
+    {
+      "epoch": 19.25,
+      "grad_norm": 0.2383130043745041,
+      "learning_rate": 3.7873134328358214e-06,
+      "loss": 0.109,
+      "step": 10320
+    },
+    {
+      "epoch": 19.29,
+      "grad_norm": 17.503469467163086,
+      "learning_rate": 3.6007462686567163e-06,
+      "loss": 0.133,
+      "step": 10340
+    },
+    {
+      "epoch": 19.33,
+      "grad_norm": 0.002066264161840081,
+      "learning_rate": 3.414179104477612e-06,
+      "loss": 0.1026,
+      "step": 10360
+    },
+    {
+      "epoch": 19.37,
+      "grad_norm": 9.919361114501953,
+      "learning_rate": 3.2276119402985075e-06,
+      "loss": 0.1805,
+      "step": 10380
+    },
+    {
+      "epoch": 19.4,
+      "grad_norm": 0.002000547247007489,
+      "learning_rate": 3.041044776119403e-06,
+      "loss": 0.1087,
+      "step": 10400
+    },
+    {
+      "epoch": 19.4,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.8273361921310425,
+      "eval_runtime": 16.9239,
+      "eval_samples_per_second": 33.621,
+      "eval_steps_per_second": 4.254,
+      "step": 10400
+    },
+    {
+      "epoch": 19.44,
+      "grad_norm": 10.143733024597168,
+      "learning_rate": 2.8544776119402986e-06,
+      "loss": 0.099,
+      "step": 10420
+    },
+    {
+      "epoch": 19.48,
+      "grad_norm": 5.309617042541504,
+      "learning_rate": 2.6679104477611944e-06,
+      "loss": 0.0829,
+      "step": 10440
+    },
+    {
+      "epoch": 19.51,
+      "grad_norm": 0.5168270468711853,
+      "learning_rate": 2.4813432835820898e-06,
+      "loss": 0.0891,
+      "step": 10460
+    },
+    {
+      "epoch": 19.55,
+      "grad_norm": 5.676023483276367,
+      "learning_rate": 2.294776119402985e-06,
+      "loss": 0.1859,
+      "step": 10480
+    },
+    {
+      "epoch": 19.59,
+      "grad_norm": 0.0018354007042944431,
+      "learning_rate": 2.1082089552238805e-06,
+      "loss": 0.1983,
+      "step": 10500
+    },
+    {
+      "epoch": 19.59,
+      "eval_accuracy": 0.8611599297012302,
+      "eval_loss": 0.8257032036781311,
+      "eval_runtime": 17.014,
+      "eval_samples_per_second": 33.443,
+      "eval_steps_per_second": 4.232,
+      "step": 10500
+    },
+    {
+      "epoch": 19.63,
+      "grad_norm": 8.31838321685791,
+      "learning_rate": 1.9216417910447763e-06,
+      "loss": 0.1072,
+      "step": 10520
+    },
+    {
+      "epoch": 19.66,
+      "grad_norm": 13.230073928833008,
+      "learning_rate": 1.7350746268656717e-06,
+      "loss": 0.1747,
+      "step": 10540
+    },
+    {
+      "epoch": 19.7,
+      "grad_norm": 0.010875330306589603,
+      "learning_rate": 1.5485074626865673e-06,
+      "loss": 0.2028,
+      "step": 10560
+    },
+    {
+      "epoch": 19.74,
+      "grad_norm": 31.933826446533203,
+      "learning_rate": 1.3619402985074628e-06,
+      "loss": 0.1461,
+      "step": 10580
+    },
+    {
+      "epoch": 19.78,
+      "grad_norm": 0.025413936004042625,
+      "learning_rate": 1.1753731343283582e-06,
+      "loss": 0.2453,
+      "step": 10600
+    },
+    {
+      "epoch": 19.78,
+      "eval_accuracy": 0.8576449912126538,
+      "eval_loss": 0.8176620602607727,
+      "eval_runtime": 17.0802,
+      "eval_samples_per_second": 33.313,
+      "eval_steps_per_second": 4.215,
+      "step": 10600
+    },
+    {
+      "epoch": 19.81,
+      "grad_norm": 21.636062622070312,
+      "learning_rate": 9.888059701492538e-07,
+      "loss": 0.1882,
+      "step": 10620
+    },
+    {
+      "epoch": 19.85,
+      "grad_norm": 0.06387708336114883,
+      "learning_rate": 8.022388059701493e-07,
+      "loss": 0.0729,
+      "step": 10640
+    },
+    {
+      "epoch": 19.89,
+      "grad_norm": 0.0024608001112937927,
+      "learning_rate": 6.156716417910448e-07,
+      "loss": 0.0236,
+      "step": 10660
+    },
+    {
+      "epoch": 19.93,
+      "grad_norm": 0.1950324922800064,
+      "learning_rate": 4.291044776119403e-07,
+      "loss": 0.086,
+      "step": 10680
+    },
+    {
+      "epoch": 19.96,
+      "grad_norm": 0.0383419506251812,
+      "learning_rate": 2.4253731343283584e-07,
+      "loss": 0.1189,
+      "step": 10700
+    },
+    {
+      "epoch": 19.96,
+      "eval_accuracy": 0.859402460456942,
+      "eval_loss": 0.8201273679733276,
+      "eval_runtime": 17.0643,
+      "eval_samples_per_second": 33.344,
+      "eval_steps_per_second": 4.219,
+      "step": 10700
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 22.60196876525879,
+      "learning_rate": 5.597014925373135e-08,
+      "loss": 0.2808,
+      "step": 10720
+    },
+    {
+      "epoch": 20.0,
+      "step": 10720,
+      "total_flos": 1.9598089091159163e+19,
+      "train_loss": 0.2734005047119598,
+      "train_runtime": 8010.8383,
+      "train_samples_per_second": 10.705,
+      "train_steps_per_second": 1.338
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 10720,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 20,
+  "save_steps": 100,
+  "total_flos": 1.9598089091159163e+19,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}