diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7069 @@
+{
+  "best_metric": 0.46077683568000793,
+  "best_model_checkpoint": "./exper_batch_8_e8/checkpoint-9400",
+  "epoch": 8.0,
+  "global_step": 10216,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00019980422866092406,
+      "loss": 4.9654,
+      "step": 10
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0001996084573218481,
+      "loss": 4.8684,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00019941268598277215,
+      "loss": 4.8214,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00019921691464369617,
+      "loss": 4.6964,
+      "step": 40
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0001990211433046202,
+      "loss": 4.5588,
+      "step": 50
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00019882537196554424,
+      "loss": 4.5544,
+      "step": 60
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0001986296006264683,
+      "loss": 4.3677,
+      "step": 70
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00019843382928739234,
+      "loss": 4.2368,
+      "step": 80
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0001982380579483164,
+      "loss": 4.2028,
+      "step": 90
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00019804228660924044,
+      "loss": 4.2202,
+      "step": 100
+    },
+    {
+      "epoch": 0.08,
+      "eval_accuracy": 0.12374710871241326,
+      "eval_loss": 4.124450206756592,
+      "eval_runtime": 43.2831,
+      "eval_samples_per_second": 59.931,
+      "eval_steps_per_second": 7.509,
+      "step": 100
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00019784651527016446,
+      "loss": 4.0062,
+      "step": 110
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00019765074393108848,
+      "loss": 4.1862,
+      "step": 120
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00019745497259201253,
+      "loss": 4.0323,
+      "step": 130
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00019725920125293657,
+      "loss": 3.8437,
+      "step": 140
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00019706342991386062,
+      "loss": 3.8453,
+      "step": 150
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00019686765857478467,
+      "loss": 3.8992,
+      "step": 160
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00019667188723570872,
+      "loss": 3.8014,
+      "step": 170
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00019647611589663274,
+      "loss": 3.9764,
+      "step": 180
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00019628034455755676,
+      "loss": 3.6376,
+      "step": 190
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0001960845732184808,
+      "loss": 3.467,
+      "step": 200
+    },
+    {
+      "epoch": 0.16,
+      "eval_accuracy": 0.21434078643022358,
+      "eval_loss": 3.562227487564087,
+      "eval_runtime": 42.516,
+      "eval_samples_per_second": 61.012,
+      "eval_steps_per_second": 7.644,
+      "step": 200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00019588880187940486,
+      "loss": 3.4815,
+      "step": 210
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0001956930305403289,
+      "loss": 3.626,
+      "step": 220
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00019549725920125295,
+      "loss": 3.425,
+      "step": 230
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.000195301487862177,
+      "loss": 3.3261,
+      "step": 240
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00019510571652310102,
+      "loss": 3.3677,
+      "step": 250
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00019490994518402507,
+      "loss": 3.3689,
+      "step": 260
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0001947141738449491,
+      "loss": 3.2287,
+      "step": 270
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00019451840250587314,
+      "loss": 3.1849,
+      "step": 280
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00019432263116679719,
+      "loss": 3.1894,
+      "step": 290
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00019412685982772123,
+      "loss": 3.3469,
+      "step": 300
+    },
+    {
+      "epoch": 0.23,
+      "eval_accuracy": 0.26754047802621433,
+      "eval_loss": 3.1687722206115723,
+      "eval_runtime": 43.3412,
+      "eval_samples_per_second": 59.851,
+      "eval_steps_per_second": 7.499,
+      "step": 300
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00019393108848864528,
+      "loss": 3.1748,
+      "step": 310
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00019373531714956933,
+      "loss": 3.1283,
+      "step": 320
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00019353954581049335,
+      "loss": 2.9591,
+      "step": 330
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0001933437744714174,
+      "loss": 2.9965,
+      "step": 340
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00019314800313234142,
+      "loss": 3.0854,
+      "step": 350
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00019295223179326547,
+      "loss": 2.875,
+      "step": 360
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00019275646045418952,
+      "loss": 2.9558,
+      "step": 370
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00019256068911511356,
+      "loss": 2.7257,
+      "step": 380
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0001923649177760376,
+      "loss": 3.1288,
+      "step": 390
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00019216914643696163,
+      "loss": 2.8086,
+      "step": 400
+    },
+    {
+      "epoch": 0.31,
+      "eval_accuracy": 0.3033924441017733,
+      "eval_loss": 2.8964760303497314,
+      "eval_runtime": 42.7345,
+      "eval_samples_per_second": 60.7,
+      "eval_steps_per_second": 7.605,
+      "step": 400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00019197337509788568,
+      "loss": 2.8508,
+      "step": 410
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0001917776037588097,
+      "loss": 2.6375,
+      "step": 420
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00019158183241973375,
+      "loss": 2.6023,
+      "step": 430
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0001913860610806578,
+      "loss": 2.7481,
+      "step": 440
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00019119028974158185,
+      "loss": 2.8379,
+      "step": 450
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0001909945184025059,
+      "loss": 2.6653,
+      "step": 460
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00019079874706342992,
+      "loss": 2.4517,
+      "step": 470
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00019060297572435396,
+      "loss": 2.7008,
+      "step": 480
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.000190407204385278,
+      "loss": 2.4153,
+      "step": 490
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00019021143304620203,
+      "loss": 2.6291,
+      "step": 500
+    },
+    {
+      "epoch": 0.39,
+      "eval_accuracy": 0.40246723207401697,
+      "eval_loss": 2.5858304500579834,
+      "eval_runtime": 43.0916,
+      "eval_samples_per_second": 60.197,
+      "eval_steps_per_second": 7.542,
+      "step": 500
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00019001566170712608,
+      "loss": 2.6765,
+      "step": 510
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00018981989036805013,
+      "loss": 2.4549,
+      "step": 520
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00018962411902897418,
+      "loss": 2.5246,
+      "step": 530
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00018942834768989822,
+      "loss": 2.3387,
+      "step": 540
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00018923257635082225,
+      "loss": 2.5484,
+      "step": 550
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0001890368050117463,
+      "loss": 2.7829,
+      "step": 560
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00018884103367267031,
+      "loss": 2.2403,
+      "step": 570
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00018864526233359436,
+      "loss": 2.4251,
+      "step": 580
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0001884494909945184,
+      "loss": 2.1143,
+      "step": 590
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00018825371965544246,
+      "loss": 2.2382,
+      "step": 600
+    },
+    {
+      "epoch": 0.47,
+      "eval_accuracy": 0.41326137239784116,
+      "eval_loss": 2.2908029556274414,
+      "eval_runtime": 43.1132,
+      "eval_samples_per_second": 60.167,
+      "eval_steps_per_second": 7.538,
+      "step": 600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0001880579483163665,
+      "loss": 2.2726,
+      "step": 610
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00018786217697729053,
+      "loss": 2.3857,
+      "step": 620
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00018766640563821458,
+      "loss": 2.1657,
+      "step": 630
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00018747063429913862,
+      "loss": 2.3817,
+      "step": 640
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00018727486296006264,
+      "loss": 2.2828,
+      "step": 650
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0001870790916209867,
+      "loss": 2.2487,
+      "step": 660
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00018688332028191074,
+      "loss": 2.4604,
+      "step": 670
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0001866875489428348,
+      "loss": 2.1307,
+      "step": 680
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0001864917776037588,
+      "loss": 2.2876,
+      "step": 690
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00018629600626468286,
+      "loss": 1.9259,
+      "step": 700
+    },
+    {
+      "epoch": 0.55,
+      "eval_accuracy": 0.46761757902852735,
+      "eval_loss": 2.200721263885498,
+      "eval_runtime": 42.7003,
+      "eval_samples_per_second": 60.749,
+      "eval_steps_per_second": 7.611,
+      "step": 700
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0001861002349256069,
+      "loss": 2.0621,
+      "step": 710
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00018590446358653093,
+      "loss": 1.7867,
+      "step": 720
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00018570869224745497,
+      "loss": 2.1419,
+      "step": 730
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00018551292090837902,
+      "loss": 2.1075,
+      "step": 740
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00018531714956930307,
+      "loss": 2.0453,
+      "step": 750
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00018512137823022712,
+      "loss": 2.067,
+      "step": 760
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00018492560689115114,
+      "loss": 1.9266,
+      "step": 770
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001847298355520752,
+      "loss": 2.2861,
+      "step": 780
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00018453406421299924,
+      "loss": 1.978,
+      "step": 790
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00018433829287392326,
+      "loss": 1.8088,
+      "step": 800
+    },
+    {
+      "epoch": 0.63,
+      "eval_accuracy": 0.47417116422513494,
+      "eval_loss": 2.041884422302246,
+      "eval_runtime": 42.6812,
+      "eval_samples_per_second": 60.776,
+      "eval_steps_per_second": 7.615,
+      "step": 800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001841425215348473,
+      "loss": 1.8517,
+      "step": 810
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00018394675019577135,
+      "loss": 2.1789,
+      "step": 820
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001837509788566954,
+      "loss": 2.0317,
+      "step": 830
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00018355520751761942,
+      "loss": 1.964,
+      "step": 840
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00018335943617854347,
+      "loss": 1.8727,
+      "step": 850
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00018316366483946752,
+      "loss": 1.8722,
+      "step": 860
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00018296789350039154,
+      "loss": 1.8382,
+      "step": 870
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001827721221613156,
+      "loss": 1.7831,
+      "step": 880
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00018257635082223963,
+      "loss": 1.6627,
+      "step": 890
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00018238057948316368,
+      "loss": 1.9462,
+      "step": 900
+    },
+    {
+      "epoch": 0.7,
+      "eval_accuracy": 0.5578257517347726,
+      "eval_loss": 1.6793174743652344,
+      "eval_runtime": 42.5528,
+      "eval_samples_per_second": 60.96,
+      "eval_steps_per_second": 7.638,
+      "step": 900
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001821848081440877,
+      "loss": 1.7561,
+      "step": 910
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00018198903680501175,
+      "loss": 1.9231,
+      "step": 920
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001817932654659358,
+      "loss": 1.5815,
+      "step": 930
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00018159749412685985,
+      "loss": 1.8771,
+      "step": 940
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00018140172278778387,
+      "loss": 1.6328,
+      "step": 950
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00018120595144870792,
+      "loss": 1.747,
+      "step": 960
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00018101018010963196,
+      "loss": 1.847,
+      "step": 970
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00018081440877055599,
+      "loss": 1.6034,
+      "step": 980
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018061863743148003,
+      "loss": 1.5846,
+      "step": 990
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018042286609240408,
+      "loss": 1.5392,
+      "step": 1000
+    },
+    {
+      "epoch": 0.78,
+      "eval_accuracy": 0.6079414032382421,
+      "eval_loss": 1.5460304021835327,
+      "eval_runtime": 43.0303,
+      "eval_samples_per_second": 60.283,
+      "eval_steps_per_second": 7.553,
+      "step": 1000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018022709475332813,
+      "loss": 1.373,
+      "step": 1010
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018003132341425218,
+      "loss": 1.589,
+      "step": 1020
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001798355520751762,
+      "loss": 1.5769,
+      "step": 1030
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00017963978073610025,
+      "loss": 1.4197,
+      "step": 1040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0001794440093970243,
+      "loss": 1.6831,
+      "step": 1050
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00017924823805794832,
+      "loss": 1.6604,
+      "step": 1060
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017905246671887236,
+      "loss": 1.3292,
+      "step": 1070
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001788566953797964,
+      "loss": 1.6968,
+      "step": 1080
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017866092404072046,
+      "loss": 1.4353,
+      "step": 1090
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017846515270164448,
+      "loss": 1.561,
+      "step": 1100
+    },
+    {
+      "epoch": 0.86,
+      "eval_accuracy": 0.569005397070162,
+      "eval_loss": 1.5792826414108276,
+      "eval_runtime": 43.1399,
+      "eval_samples_per_second": 60.13,
+      "eval_steps_per_second": 7.534,
+      "step": 1100
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017826938136256853,
+      "loss": 1.6354,
+      "step": 1110
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017807361002349258,
+      "loss": 1.4021,
+      "step": 1120
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001778778386844166,
+      "loss": 1.3457,
+      "step": 1130
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017768206734534065,
+      "loss": 1.3459,
+      "step": 1140
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0001774862960062647,
+      "loss": 1.4383,
+      "step": 1150
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00017729052466718874,
+      "loss": 1.5853,
+      "step": 1160
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001770947533281128,
+      "loss": 1.27,
+      "step": 1170
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001768989819890368,
+      "loss": 1.197,
+      "step": 1180
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00017670321064996086,
+      "loss": 1.1779,
+      "step": 1190
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00017650743931088488,
+      "loss": 1.2135,
+      "step": 1200
+    },
+    {
+      "epoch": 0.94,
+      "eval_accuracy": 0.5929067077872012,
+      "eval_loss": 1.466301679611206,
+      "eval_runtime": 42.4934,
+      "eval_samples_per_second": 61.045,
+      "eval_steps_per_second": 7.648,
+      "step": 1200
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00017631166797180893,
+      "loss": 1.4053,
+      "step": 1210
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00017611589663273298,
+      "loss": 1.4909,
+      "step": 1220
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00017592012529365702,
+      "loss": 1.5044,
+      "step": 1230
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00017572435395458107,
+      "loss": 1.6701,
+      "step": 1240
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001755285826155051,
+      "loss": 1.6198,
+      "step": 1250
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00017533281127642914,
+      "loss": 1.3338,
+      "step": 1260
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001751370399373532,
+      "loss": 1.2655,
+      "step": 1270
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0001749412685982772,
+      "loss": 1.2768,
+      "step": 1280
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00017474549725920126,
+      "loss": 1.2836,
+      "step": 1290
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0001745497259201253,
+      "loss": 1.0725,
+      "step": 1300
+    },
+    {
+      "epoch": 1.02,
+      "eval_accuracy": 0.6534309946029299,
+      "eval_loss": 1.297423005104065,
+      "eval_runtime": 42.4541,
+      "eval_samples_per_second": 61.101,
+      "eval_steps_per_second": 7.655,
+      "step": 1300
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00017435395458104935,
+      "loss": 1.0784,
+      "step": 1310
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0001741581832419734,
+      "loss": 1.1167,
+      "step": 1320
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00017396241190289742,
+      "loss": 1.1686,
+      "step": 1330
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00017376664056382147,
+      "loss": 1.049,
+      "step": 1340
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.0001735708692247455,
+      "loss": 0.7355,
+      "step": 1350
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00017337509788566954,
+      "loss": 0.762,
+      "step": 1360
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0001731793265465936,
+      "loss": 1.1496,
+      "step": 1370
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00017298355520751764,
+      "loss": 1.0427,
+      "step": 1380
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00017278778386844168,
+      "loss": 0.9247,
+      "step": 1390
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001725920125293657,
+      "loss": 0.8696,
+      "step": 1400
+    },
+    {
+      "epoch": 1.1,
+      "eval_accuracy": 0.6569005397070162,
+      "eval_loss": 1.2405539751052856,
+      "eval_runtime": 42.8114,
+      "eval_samples_per_second": 60.591,
+      "eval_steps_per_second": 7.591,
+      "step": 1400
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00017239624119028975,
+      "loss": 1.1499,
+      "step": 1410
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00017220046985121377,
+      "loss": 0.9314,
+      "step": 1420
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00017200469851213782,
+      "loss": 0.8725,
+      "step": 1430
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00017180892717306187,
+      "loss": 1.1714,
+      "step": 1440
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017161315583398592,
+      "loss": 1.0311,
+      "step": 1450
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017141738449490997,
+      "loss": 0.8996,
+      "step": 1460
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00017122161315583401,
+      "loss": 1.0091,
+      "step": 1470
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00017102584181675804,
+      "loss": 1.1246,
+      "step": 1480
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00017083007047768206,
+      "loss": 1.063,
+      "step": 1490
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0001706342991386061,
+      "loss": 0.8758,
+      "step": 1500
+    },
+    {
+      "epoch": 1.17,
+      "eval_accuracy": 0.6622976098689283,
+      "eval_loss": 1.212697982788086,
+      "eval_runtime": 42.4924,
+      "eval_samples_per_second": 61.046,
+      "eval_steps_per_second": 7.648,
+      "step": 1500
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00017043852779953015,
+      "loss": 0.854,
+      "step": 1510
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0001702427564604542,
+      "loss": 0.9358,
+      "step": 1520
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00017004698512137825,
+      "loss": 0.8443,
+      "step": 1530
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0001698512137823023,
+      "loss": 1.1945,
+      "step": 1540
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00016965544244322632,
+      "loss": 1.182,
+      "step": 1550
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.00016945967110415037,
+      "loss": 0.7876,
+      "step": 1560
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00016926389976507439,
+      "loss": 0.7027,
+      "step": 1570
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00016906812842599843,
+      "loss": 1.0055,
+      "step": 1580
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00016887235708692248,
+      "loss": 0.886,
+      "step": 1590
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00016867658574784653,
+      "loss": 1.1737,
+      "step": 1600
+    },
+    {
+      "epoch": 1.25,
+      "eval_accuracy": 0.6549730146491904,
+      "eval_loss": 1.2243409156799316,
+      "eval_runtime": 42.418,
+      "eval_samples_per_second": 61.153,
+      "eval_steps_per_second": 7.662,
+      "step": 1600
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00016848081440877058,
+      "loss": 0.8519,
+      "step": 1610
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00016828504306969463,
+      "loss": 1.0416,
+      "step": 1620
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00016808927173061865,
+      "loss": 0.8192,
+      "step": 1630
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00016789350039154267,
+      "loss": 1.0579,
+      "step": 1640
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00016769772905246672,
+      "loss": 0.9239,
+      "step": 1650
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00016750195771339076,
+      "loss": 0.9535,
+      "step": 1660
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0001673061863743148,
+      "loss": 0.9319,
+      "step": 1670
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00016711041503523886,
+      "loss": 0.8143,
+      "step": 1680
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0001669146436961629,
+      "loss": 0.9921,
+      "step": 1690
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00016671887235708693,
+      "loss": 0.8242,
+      "step": 1700
+    },
+    {
+      "epoch": 1.33,
+      "eval_accuracy": 0.6734772552043177,
+      "eval_loss": 1.137111783027649,
+      "eval_runtime": 42.7018,
+      "eval_samples_per_second": 60.747,
+      "eval_steps_per_second": 7.611,
+      "step": 1700
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00016652310101801095,
+      "loss": 1.1058,
+      "step": 1710
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.000166327329678935,
+      "loss": 0.8331,
+      "step": 1720
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00016613155833985905,
+      "loss": 0.9411,
+      "step": 1730
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0001659357870007831,
+      "loss": 0.6935,
+      "step": 1740
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.00016574001566170714,
+      "loss": 0.7819,
+      "step": 1750
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0001655442443226312,
+      "loss": 0.8969,
+      "step": 1760
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0001653484729835552,
+      "loss": 0.6313,
+      "step": 1770
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.00016515270164447926,
+      "loss": 1.1179,
+      "step": 1780
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.00016495693030540328,
+      "loss": 0.8904,
+      "step": 1790
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.00016476115896632733,
+      "loss": 1.0141,
+      "step": 1800
+    },
+    {
+      "epoch": 1.41,
+      "eval_accuracy": 0.7023901310717039,
+      "eval_loss": 1.0536144971847534,
+      "eval_runtime": 42.962,
+      "eval_samples_per_second": 60.379,
+      "eval_steps_per_second": 7.565,
+      "step": 1800
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.00016456538762725138,
+      "loss": 0.8489,
+      "step": 1810
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00016436961628817542,
+      "loss": 0.7427,
+      "step": 1820
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00016417384494909947,
+      "loss": 0.9745,
+      "step": 1830
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.00016397807361002352,
+      "loss": 0.8825,
+      "step": 1840
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.00016378230227094754,
+      "loss": 0.7787,
+      "step": 1850
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.00016358653093187156,
+      "loss": 0.9332,
+      "step": 1860
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0001633907595927956,
+      "loss": 0.7362,
+      "step": 1870
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.00016319498825371966,
+      "loss": 0.9051,
+      "step": 1880
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0001629992169146437,
+      "loss": 0.7092,
+      "step": 1890
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00016280344557556775,
+      "loss": 0.9855,
+      "step": 1900
+    },
+    {
+      "epoch": 1.49,
+      "eval_accuracy": 0.720508866615266,
+      "eval_loss": 0.9885048866271973,
+      "eval_runtime": 42.7318,
+      "eval_samples_per_second": 60.704,
+      "eval_steps_per_second": 7.606,
+      "step": 1900
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0001626076742364918,
+      "loss": 0.7958,
+      "step": 1910
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.00016241190289741582,
+      "loss": 0.7243,
+      "step": 1920
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.00016221613155833984,
+      "loss": 0.7715,
+      "step": 1930
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0001620203602192639,
+      "loss": 0.8117,
+      "step": 1940
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.00016182458888018794,
+      "loss": 0.5676,
+      "step": 1950
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.000161628817541112,
+      "loss": 0.6431,
+      "step": 1960
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.00016143304620203604,
+      "loss": 0.7356,
+      "step": 1970
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.00016123727486296008,
+      "loss": 0.6934,
+      "step": 1980
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0001610415035238841,
+      "loss": 0.766,
+      "step": 1990
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00016084573218480815,
+      "loss": 0.805,
+      "step": 2000
+    },
+    {
+      "epoch": 1.57,
+      "eval_accuracy": 0.7478797224363917,
+      "eval_loss": 0.9048407077789307,
+      "eval_runtime": 43.838,
+      "eval_samples_per_second": 59.172,
+      "eval_steps_per_second": 7.414,
+      "step": 2000
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00016064996084573217,
+      "loss": 0.7299,
+      "step": 2010
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.00016045418950665622,
+      "loss": 0.7288,
+      "step": 2020
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.00016025841816758027,
+      "loss": 0.7434,
+      "step": 2030
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.00016006264682850432,
+      "loss": 0.7146,
+      "step": 2040
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00015986687548942837,
+      "loss": 0.6466,
+      "step": 2050
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00015967110415035241,
+      "loss": 0.7848,
+      "step": 2060
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.00015947533281127644,
+      "loss": 0.8508,
+      "step": 2070
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.00015927956147220046,
+      "loss": 0.5842,
+      "step": 2080
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0001590837901331245,
+      "loss": 0.7399,
+      "step": 2090
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.00015888801879404855,
+      "loss": 0.7207,
+      "step": 2100
+    },
+    {
+      "epoch": 1.64,
+      "eval_accuracy": 0.7490362374710872,
+      "eval_loss": 0.8842461109161377,
+      "eval_runtime": 42.7259,
+      "eval_samples_per_second": 60.713,
+      "eval_steps_per_second": 7.607,
+      "step": 2100
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0001586922474549726,
+      "loss": 0.7662,
+      "step": 2110
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.00015849647611589665,
+      "loss": 0.6789,
+      "step": 2120
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0001583007047768207,
+      "loss": 0.8986,
+      "step": 2130
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.00015810493343774472,
+      "loss": 0.7488,
+      "step": 2140
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.00015790916209866877,
+      "loss": 0.5487,
+      "step": 2150
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0001577133907595928,
+      "loss": 0.7602,
+      "step": 2160
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.00015751761942051683,
+      "loss": 0.5251,
+      "step": 2170
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.00015732184808144088,
+      "loss": 0.7298,
+      "step": 2180
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.00015712607674236493,
+      "loss": 0.7112,
+      "step": 2190
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00015693030540328898,
+      "loss": 0.7101,
+      "step": 2200
+    },
+    {
+      "epoch": 1.72,
+      "eval_accuracy": 0.743639167309175,
+      "eval_loss": 0.8954148292541504,
+      "eval_runtime": 43.1939,
+      "eval_samples_per_second": 60.055,
+      "eval_steps_per_second": 7.524,
+      "step": 2200
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.000156734534064213,
+      "loss": 0.6801,
+      "step": 2210
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.00015653876272513705,
+      "loss": 0.9109,
+      "step": 2220
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.00015634299138606107,
+      "loss": 0.6723,
+      "step": 2230
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.00015614722004698512,
+      "loss": 0.5826,
+      "step": 2240
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00015595144870790916,
+      "loss": 0.5993,
+      "step": 2250
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0001557556773688332,
+      "loss": 0.5554,
+      "step": 2260
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.00015555990602975726,
+      "loss": 0.5562,
+      "step": 2270
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0001553641346906813,
+      "loss": 0.5963,
+      "step": 2280
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.00015516836335160533,
+      "loss": 0.916,
+      "step": 2290
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.00015497259201252938,
+      "loss": 0.5946,
+      "step": 2300
+    },
+    {
+      "epoch": 1.8,
+      "eval_accuracy": 0.7386276021588281,
+      "eval_loss": 0.9174211025238037,
+      "eval_runtime": 42.4595,
+      "eval_samples_per_second": 61.093,
+      "eval_steps_per_second": 7.654,
+      "step": 2300
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0001547768206734534,
+      "loss": 0.6563,
+      "step": 2310
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.00015458104933437745,
+      "loss": 0.9754,
+      "step": 2320
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0001543852779953015,
+      "loss": 0.5358,
+      "step": 2330
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.00015418950665622554,
+      "loss": 0.8423,
+      "step": 2340
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0001539937353171496,
+      "loss": 0.5932,
+      "step": 2350
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0001537979639780736,
+      "loss": 0.4604,
+      "step": 2360
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00015360219263899766,
+      "loss": 0.7251,
+      "step": 2370
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0001534064212999217,
+      "loss": 0.6496,
+      "step": 2380
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.00015321064996084573,
+      "loss": 0.5118,
+      "step": 2390
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.00015301487862176978,
+      "loss": 0.6937,
+      "step": 2400
+    },
+    {
+      "epoch": 1.88,
+      "eval_accuracy": 0.7760215882806476,
+      "eval_loss": 0.7818014621734619,
+      "eval_runtime": 42.8365,
+      "eval_samples_per_second": 60.556,
+      "eval_steps_per_second": 7.587,
+      "step": 2400
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.00015281910728269383,
+      "loss": 0.759,
+      "step": 2410
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.00015262333594361787,
+      "loss": 1.029,
+      "step": 2420
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0001524275646045419,
+      "loss": 0.4319,
+      "step": 2430
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.00015223179326546594,
+      "loss": 0.6477,
+      "step": 2440
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.00015203602192639,
+      "loss": 0.7279,
+      "step": 2450
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.000151840250587314,
+      "loss": 0.7137,
+      "step": 2460
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.00015164447924823806,
+      "loss": 0.6099,
+      "step": 2470
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0001514487079091621,
+      "loss": 0.59,
+      "step": 2480
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.00015125293657008616,
+      "loss": 0.7626,
+      "step": 2490
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.00015105716523101018,
+      "loss": 0.5593,
+      "step": 2500
+    },
+    {
+      "epoch": 1.96,
+      "eval_accuracy": 0.7933693138010794,
+      "eval_loss": 0.7448990941047668,
+      "eval_runtime": 42.9077,
+      "eval_samples_per_second": 60.455,
+      "eval_steps_per_second": 7.574,
+      "step": 2500
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.00015086139389193422,
+      "loss": 0.7344,
+      "step": 2510
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.00015066562255285827,
+      "loss": 0.5355,
+      "step": 2520
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00015046985121378232,
+      "loss": 0.5933,
+      "step": 2530
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.00015027407987470634,
+      "loss": 0.5659,
+      "step": 2540
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0001500783085356304,
+      "loss": 0.6364,
+      "step": 2550
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.00014988253719655444,
+      "loss": 0.5475,
+      "step": 2560
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.00014968676585747849,
+      "loss": 0.3134,
+      "step": 2570
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0001494909945184025,
+      "loss": 0.2598,
+      "step": 2580
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.00014929522317932655,
+      "loss": 0.455,
+      "step": 2590
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0001490994518402506,
+      "loss": 0.4139,
+      "step": 2600
+    },
+    {
+      "epoch": 2.04,
+      "eval_accuracy": 0.7829606784888203,
+      "eval_loss": 0.7786519527435303,
+      "eval_runtime": 42.9538,
+      "eval_samples_per_second": 60.39,
+      "eval_steps_per_second": 7.566,
+      "step": 2600
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.00014890368050117462,
+      "loss": 0.274,
+      "step": 2610
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.00014870790916209867,
+      "loss": 0.385,
+      "step": 2620
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.00014851213782302272,
+      "loss": 0.4684,
+      "step": 2630
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.00014831636648394677,
+      "loss": 0.6072,
+      "step": 2640
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0001481205951448708,
+      "loss": 0.4207,
+      "step": 2650
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.00014792482380579484,
+      "loss": 0.299,
+      "step": 2660
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.00014772905246671888,
+      "loss": 0.4115,
+      "step": 2670
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.00014753328112764293,
+      "loss": 0.3697,
+      "step": 2680
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.00014733750978856695,
+      "loss": 0.2921,
+      "step": 2690
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.000147141738449491,
+      "loss": 0.2929,
+      "step": 2700
+    },
+    {
+      "epoch": 2.11,
+      "eval_accuracy": 0.7945258288357748,
+      "eval_loss": 0.7122095227241516,
+      "eval_runtime": 42.9472,
+      "eval_samples_per_second": 60.4,
+      "eval_steps_per_second": 7.567,
+      "step": 2700
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.00014694596711041505,
+      "loss": 0.5186,
+      "step": 2710
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.00014675019577133907,
+      "loss": 0.5239,
+      "step": 2720
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.00014655442443226312,
+      "loss": 0.4042,
+      "step": 2730
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.00014635865309318717,
+      "loss": 0.4424,
+      "step": 2740
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.00014616288175411121,
+      "loss": 0.5403,
+      "step": 2750
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.00014596711041503524,
+      "loss": 0.4172,
+      "step": 2760
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.00014577133907595928,
+      "loss": 0.3765,
+      "step": 2770
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.00014557556773688333,
+      "loss": 0.4216,
+      "step": 2780
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.00014537979639780738,
+      "loss": 0.4168,
+      "step": 2790
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0001451840250587314,
+      "loss": 0.4159,
+      "step": 2800
+    },
+    {
+      "epoch": 2.19,
+      "eval_accuracy": 0.7906707787201234,
+      "eval_loss": 0.7445757985115051,
+      "eval_runtime": 42.5269,
+      "eval_samples_per_second": 60.997,
+      "eval_steps_per_second": 7.642,
+      "step": 2800
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.00014498825371965545,
+      "loss": 0.4718,
+      "step": 2810
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0001447924823805795,
+      "loss": 0.2849,
+      "step": 2820
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.00014459671104150354,
+      "loss": 0.344,
+      "step": 2830
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.00014440093970242757,
+      "loss": 0.5219,
+      "step": 2840
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.00014420516836335161,
+      "loss": 0.3233,
+      "step": 2850
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.00014400939702427566,
+      "loss": 0.3051,
+      "step": 2860
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.00014381362568519968,
+      "loss": 0.1959,
+      "step": 2870
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.00014361785434612373,
+      "loss": 0.5148,
+      "step": 2880
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.00014342208300704778,
+      "loss": 0.4508,
+      "step": 2890
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.00014322631166797183,
+      "loss": 0.4079,
+      "step": 2900
+    },
+    {
+      "epoch": 2.27,
+      "eval_accuracy": 0.7937548188126445,
+      "eval_loss": 0.7353845834732056,
+      "eval_runtime": 42.6433,
+      "eval_samples_per_second": 60.83,
+      "eval_steps_per_second": 7.621,
+      "step": 2900
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.00014303054032889585,
+      "loss": 0.5128,
+      "step": 2910
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0001428347689898199,
+      "loss": 0.2881,
+      "step": 2920
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.00014263899765074394,
+      "loss": 0.2049,
+      "step": 2930
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.00014244322631166796,
+      "loss": 0.3678,
+      "step": 2940
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.000142247454972592,
+      "loss": 0.2487,
+      "step": 2950
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.00014205168363351606,
+      "loss": 0.3959,
+      "step": 2960
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0001418559122944401,
+      "loss": 0.3581,
+      "step": 2970
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.00014166014095536416,
+      "loss": 0.4362,
+      "step": 2980
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.00014146436961628818,
+      "loss": 0.3037,
+      "step": 2990
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.00014126859827721223,
+      "loss": 0.516,
+      "step": 3000
+    },
+    {
+      "epoch": 2.35,
+      "eval_accuracy": 0.8006939090208173,
+      "eval_loss": 0.7499482035636902,
+      "eval_runtime": 43.1894,
+      "eval_samples_per_second": 60.061,
+      "eval_steps_per_second": 7.525,
+      "step": 3000
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.00014107282693813625,
+      "loss": 0.3493,
+      "step": 3010
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0001408770555990603,
+      "loss": 0.2129,
+      "step": 3020
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.00014068128425998434,
+      "loss": 0.2848,
+      "step": 3030
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0001404855129209084,
+      "loss": 0.2469,
+      "step": 3040
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.00014028974158183244,
+      "loss": 0.3759,
+      "step": 3050
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0001400939702427565,
+      "loss": 0.6726,
+      "step": 3060
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0001398981989036805,
+      "loss": 0.7171,
+      "step": 3070
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.00013970242756460456,
+      "loss": 0.2932,
+      "step": 3080
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.00013950665622552858,
+      "loss": 0.5227,
+      "step": 3090
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.00013931088488645262,
+      "loss": 0.2728,
+      "step": 3100
+    },
+    {
+      "epoch": 2.43,
+      "eval_accuracy": 0.8060909791827294,
+      "eval_loss": 0.6850709319114685,
+      "eval_runtime": 42.4754,
+      "eval_samples_per_second": 61.071,
+      "eval_steps_per_second": 7.651,
+      "step": 3100
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.00013911511354737667,
+      "loss": 0.4258,
+      "step": 3110
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.00013891934220830072,
+      "loss": 0.4522,
+      "step": 3120
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.00013872357086922477,
+      "loss": 0.3043,
+      "step": 3130
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0001385277995301488,
+      "loss": 0.343,
+      "step": 3140
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.00013833202819107284,
+      "loss": 0.2787,
+      "step": 3150
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.00013813625685199686,
+      "loss": 0.3468,
+      "step": 3160
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0001379404855129209,
+      "loss": 0.3276,
+      "step": 3170
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.00013774471417384496,
+      "loss": 0.3304,
+      "step": 3180
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.000137548942834769,
+      "loss": 0.5513,
+      "step": 3190
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.00013735317149569305,
+      "loss": 0.4159,
+      "step": 3200
+    },
+    {
+      "epoch": 2.51,
+      "eval_accuracy": 0.799922898997687,
+      "eval_loss": 0.7258460521697998,
+      "eval_runtime": 42.4196,
+      "eval_samples_per_second": 61.151,
+      "eval_steps_per_second": 7.662,
+      "step": 3200
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0001371574001566171,
+      "loss": 0.3052,
+      "step": 3210
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.00013696162881754112,
+      "loss": 0.2511,
+      "step": 3220
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.00013676585747846514,
+      "loss": 0.2168,
+      "step": 3230
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0001365700861393892,
+      "loss": 0.3143,
+      "step": 3240
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.00013637431480031324,
+      "loss": 0.5571,
+      "step": 3250
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.00013617854346123729,
+      "loss": 0.3211,
+      "step": 3260
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.00013598277212216133,
+      "loss": 0.1281,
+      "step": 3270
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.00013578700078308538,
+      "loss": 0.3457,
+      "step": 3280
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0001355912294440094,
+      "loss": 0.3865,
+      "step": 3290
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.00013539545810493345,
+      "loss": 0.3396,
+      "step": 3300
+    },
+    {
+      "epoch": 2.58,
+      "eval_accuracy": 0.7972243639167309,
+      "eval_loss": 0.7454713582992554,
+      "eval_runtime": 43.0743,
+      "eval_samples_per_second": 60.222,
+      "eval_steps_per_second": 7.545,
+      "step": 3300
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.00013519968676585747,
+      "loss": 0.2844,
+      "step": 3310
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.00013500391542678152,
+      "loss": 0.3513,
+      "step": 3320
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.00013480814408770557,
+      "loss": 0.599,
+      "step": 3330
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.00013461237274862962,
+      "loss": 0.3572,
+      "step": 3340
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.00013441660140955366,
+      "loss": 0.3062,
+      "step": 3350
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0001342208300704777,
+      "loss": 0.4149,
+      "step": 3360
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.00013402505873140173,
+      "loss": 0.4975,
+      "step": 3370
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.00013382928739232575,
+      "loss": 0.4868,
+      "step": 3380
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0001336335160532498,
+      "loss": 0.4995,
+      "step": 3390
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.00013343774471417385,
+      "loss": 0.1918,
+      "step": 3400
+    },
+    {
+      "epoch": 2.66,
+      "eval_accuracy": 0.8118735543562067,
+      "eval_loss": 0.6793243885040283,
+      "eval_runtime": 43.5121,
+      "eval_samples_per_second": 59.616,
+      "eval_steps_per_second": 7.469,
+      "step": 3400
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0001332419733750979,
+      "loss": 0.2264,
+      "step": 3410
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.00013304620203602195,
+      "loss": 0.4274,
+      "step": 3420
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.000132850430696946,
+      "loss": 0.5721,
+      "step": 3430
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.00013265465935787001,
+      "loss": 0.3092,
+      "step": 3440
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.00013245888801879404,
+      "loss": 0.3641,
+      "step": 3450
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.00013226311667971808,
+      "loss": 0.3499,
+      "step": 3460
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.00013206734534064213,
+      "loss": 0.4357,
+      "step": 3470
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.00013187157400156618,
+      "loss": 0.4001,
+      "step": 3480
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.00013167580266249023,
+      "loss": 0.2878,
+      "step": 3490
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.00013148003132341428,
+      "loss": 0.1228,
+      "step": 3500
+    },
+    {
+      "epoch": 2.74,
+      "eval_accuracy": 0.8134155744024673,
+      "eval_loss": 0.6696260571479797,
+      "eval_runtime": 44.1835,
+      "eval_samples_per_second": 58.71,
+      "eval_steps_per_second": 7.356,
+      "step": 3500
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0001312842599843383,
+      "loss": 0.2127,
+      "step": 3510
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00013108848864526232,
+      "loss": 0.2146,
+      "step": 3520
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00013089271730618637,
+      "loss": 0.3265,
+      "step": 3530
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0001306969459671104,
+      "loss": 0.3343,
+      "step": 3540
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.00013050117462803446,
+      "loss": 0.2671,
+      "step": 3550
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0001303054032889585,
+      "loss": 0.5154,
+      "step": 3560
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.00013010963194988256,
+      "loss": 0.4433,
+      "step": 3570
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0001299138606108066,
+      "loss": 0.2071,
+      "step": 3580
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.00012971808927173063,
+      "loss": 0.4085,
+      "step": 3590
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.00012952231793265465,
+      "loss": 0.2671,
+      "step": 3600
+    },
+    {
+      "epoch": 2.82,
+      "eval_accuracy": 0.8284502698535081,
+      "eval_loss": 0.6305551528930664,
+      "eval_runtime": 42.7303,
+      "eval_samples_per_second": 60.706,
+      "eval_steps_per_second": 7.606,
+      "step": 3600
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0001293265465935787,
+      "loss": 0.3797,
+      "step": 3610
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.00012913077525450274,
+      "loss": 0.3847,
+      "step": 3620
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0001289350039154268,
+      "loss": 0.3102,
+      "step": 3630
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.00012873923257635084,
+      "loss": 0.245,
+      "step": 3640
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0001285434612372749,
+      "loss": 0.4126,
+      "step": 3650
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0001283476898981989,
+      "loss": 0.2856,
+      "step": 3660
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.00012815191855912293,
+      "loss": 0.5175,
+      "step": 3670
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.00012795614722004698,
+      "loss": 0.3468,
+      "step": 3680
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.00012776037588097103,
+      "loss": 0.5132,
+      "step": 3690
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.00012756460454189507,
+      "loss": 0.4986,
+      "step": 3700
+    },
+    {
+      "epoch": 2.9,
+      "eval_accuracy": 0.8296067848882035,
+      "eval_loss": 0.6111488342285156,
+      "eval_runtime": 43.0053,
+      "eval_samples_per_second": 60.318,
+      "eval_steps_per_second": 7.557,
+      "step": 3700
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.00012736883320281912,
+      "loss": 0.1616,
+      "step": 3710
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.00012717306186374317,
+      "loss": 0.2459,
+      "step": 3720
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0001269772905246672,
+      "loss": 0.2729,
+      "step": 3730
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.00012678151918559124,
+      "loss": 0.2623,
+      "step": 3740
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.00012658574784651526,
+      "loss": 0.4633,
+      "step": 3750
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0001263899765074393,
+      "loss": 0.3108,
+      "step": 3760
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.00012619420516836336,
+      "loss": 0.4072,
+      "step": 3770
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0001259984338292874,
+      "loss": 0.4505,
+      "step": 3780
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.00012580266249021145,
+      "loss": 0.2185,
+      "step": 3790
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0001256068911511355,
+      "loss": 0.3699,
+      "step": 3800
+    },
+    {
+      "epoch": 2.98,
+      "eval_accuracy": 0.8508095605242868,
+      "eval_loss": 0.5599544048309326,
+      "eval_runtime": 42.9456,
+      "eval_samples_per_second": 60.402,
+      "eval_steps_per_second": 7.568,
+      "step": 3800
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.00012541111981205952,
+      "loss": 0.2703,
+      "step": 3810
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.00012521534847298354,
+      "loss": 0.3731,
+      "step": 3820
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0001250195771339076,
+      "loss": 0.3624,
+      "step": 3830
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.00012482380579483164,
+      "loss": 0.2052,
+      "step": 3840
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.00012462803445575569,
+      "loss": 0.137,
+      "step": 3850
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.00012443226311667973,
+      "loss": 0.0728,
+      "step": 3860
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.00012423649177760378,
+      "loss": 0.1354,
+      "step": 3870
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0001240407204385278,
+      "loss": 0.1875,
+      "step": 3880
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.00012384494909945185,
+      "loss": 0.1052,
+      "step": 3890
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.00012364917776037587,
+      "loss": 0.0444,
+      "step": 3900
+    },
+    {
+      "epoch": 3.05,
+      "eval_accuracy": 0.8330763299922899,
+      "eval_loss": 0.6021418571472168,
+      "eval_runtime": 42.9708,
+      "eval_samples_per_second": 60.367,
+      "eval_steps_per_second": 7.563,
+      "step": 3900
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.00012345340642129992,
+      "loss": 0.0918,
+      "step": 3910
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.00012325763508222397,
+      "loss": 0.2504,
+      "step": 3920
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.00012306186374314802,
+      "loss": 0.0886,
+      "step": 3930
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.00012286609240407206,
+      "loss": 0.1183,
+      "step": 3940
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.00012267032106499608,
+      "loss": 0.1335,
+      "step": 3950
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.00012247454972592013,
+      "loss": 0.1226,
+      "step": 3960
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.00012227877838684415,
+      "loss": 0.121,
+      "step": 3970
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0001220830070477682,
+      "loss": 0.2394,
+      "step": 3980
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.00012188723570869225,
+      "loss": 0.144,
+      "step": 3990
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0001216914643696163,
+      "loss": 0.1489,
+      "step": 4000
+    },
+    {
+      "epoch": 3.13,
+      "eval_accuracy": 0.8515805705474171,
+      "eval_loss": 0.5599444508552551,
+      "eval_runtime": 42.5227,
+      "eval_samples_per_second": 61.003,
+      "eval_steps_per_second": 7.643,
+      "step": 4000
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.00012149569303054033,
+      "loss": 0.2079,
+      "step": 4010
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.00012129992169146438,
+      "loss": 0.1506,
+      "step": 4020
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.00012110415035238843,
+      "loss": 0.0703,
+      "step": 4030
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.00012090837901331246,
+      "loss": 0.1252,
+      "step": 4040
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.00012071260767423648,
+      "loss": 0.2067,
+      "step": 4050
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.00012051683633516053,
+      "loss": 0.0803,
+      "step": 4060
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.00012032106499608458,
+      "loss": 0.0704,
+      "step": 4070
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.00012012529365700861,
+      "loss": 0.1058,
+      "step": 4080
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.00011992952231793266,
+      "loss": 0.1987,
+      "step": 4090
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.00011973375097885671,
+      "loss": 0.15,
+      "step": 4100
+    },
+    {
+      "epoch": 3.21,
+      "eval_accuracy": 0.8365458750963762,
+      "eval_loss": 0.637698233127594,
+      "eval_runtime": 43.0914,
+      "eval_samples_per_second": 60.198,
+      "eval_steps_per_second": 7.542,
+      "step": 4100
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.00011953797963978075,
+      "loss": 0.0895,
+      "step": 4110
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.00011934220830070477,
+      "loss": 0.218,
+      "step": 4120
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.00011914643696162881,
+      "loss": 0.1983,
+      "step": 4130
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.00011895066562255286,
+      "loss": 0.3032,
+      "step": 4140
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0001187548942834769,
+      "loss": 0.2105,
+      "step": 4150
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.00011855912294440094,
+      "loss": 0.2574,
+      "step": 4160
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.00011836335160532499,
+      "loss": 0.1724,
+      "step": 4170
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.00011816758026624904,
+      "loss": 0.1104,
+      "step": 4180
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.00011797180892717308,
+      "loss": 0.2242,
+      "step": 4190
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0001177760375880971,
+      "loss": 0.2535,
+      "step": 4200
+    },
+    {
+      "epoch": 3.29,
+      "eval_accuracy": 0.8542791056283732,
+      "eval_loss": 0.5751731991767883,
+      "eval_runtime": 43.0642,
+      "eval_samples_per_second": 60.236,
+      "eval_steps_per_second": 7.547,
+      "step": 4200
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.00011758026624902114,
+      "loss": 0.168,
+      "step": 4210
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.00011738449490994519,
+      "loss": 0.1172,
+      "step": 4220
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.00011718872357086923,
+      "loss": 0.0686,
+      "step": 4230
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.00011699295223179327,
+      "loss": 0.1587,
+      "step": 4240
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.00011679718089271732,
+      "loss": 0.1698,
+      "step": 4250
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.00011660140955364136,
+      "loss": 0.1721,
+      "step": 4260
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.00011640563821456538,
+      "loss": 0.1456,
+      "step": 4270
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.00011620986687548943,
+      "loss": 0.1781,
+      "step": 4280
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.00011601409553641347,
+      "loss": 0.1216,
+      "step": 4290
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.00011581832419733751,
+      "loss": 0.2679,
+      "step": 4300
+    },
+    {
+      "epoch": 3.37,
+      "eval_accuracy": 0.8608326908249807,
+      "eval_loss": 0.5677012801170349,
+      "eval_runtime": 43.2544,
+      "eval_samples_per_second": 59.971,
+      "eval_steps_per_second": 7.514,
+      "step": 4300
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.00011562255285826156,
+      "loss": 0.1553,
+      "step": 4310
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.0001154267815191856,
+      "loss": 0.2593,
+      "step": 4320
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.00011523101018010964,
+      "loss": 0.1407,
+      "step": 4330
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.00011503523884103369,
+      "loss": 0.1669,
+      "step": 4340
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00011483946750195771,
+      "loss": 0.128,
+      "step": 4350
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00011464369616288176,
+      "loss": 0.2217,
+      "step": 4360
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.00011444792482380579,
+      "loss": 0.177,
+      "step": 4370
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.00011425215348472984,
+      "loss": 0.2244,
+      "step": 4380
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.00011405638214565389,
+      "loss": 0.1063,
+      "step": 4390
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.00011386061080657792,
+      "loss": 0.0989,
+      "step": 4400
+    },
+    {
+      "epoch": 3.45,
+      "eval_accuracy": 0.8396299151888974,
+      "eval_loss": 0.632458508014679,
+      "eval_runtime": 42.6541,
+      "eval_samples_per_second": 60.815,
+      "eval_steps_per_second": 7.619,
+      "step": 4400
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.00011366483946750197,
+      "loss": 0.0779,
+      "step": 4410
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.00011346906812842602,
+      "loss": 0.0968,
+      "step": 4420
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.00011327329678935004,
+      "loss": 0.2006,
+      "step": 4430
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.00011307752545027409,
+      "loss": 0.1425,
+      "step": 4440
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.00011288175411119812,
+      "loss": 0.1997,
+      "step": 4450
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.00011268598277212217,
+      "loss": 0.2291,
+      "step": 4460
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.00011249021143304622,
+      "loss": 0.1769,
+      "step": 4470
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.00011229444009397025,
+      "loss": 0.0921,
+      "step": 4480
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0001120986687548943,
+      "loss": 0.2156,
+      "step": 4490
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.00011190289741581832,
+      "loss": 0.0825,
+      "step": 4500
+    },
+    {
+      "epoch": 3.52,
+      "eval_accuracy": 0.8523515805705474,
+      "eval_loss": 0.5978727340698242,
+      "eval_runtime": 42.4722,
+      "eval_samples_per_second": 61.075,
+      "eval_steps_per_second": 7.652,
+      "step": 4500
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.00011170712607674237,
+      "loss": 0.0799,
+      "step": 4510
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0001115113547376664,
+      "loss": 0.1545,
+      "step": 4520
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.00011131558339859045,
+      "loss": 0.1169,
+      "step": 4530
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0001111198120595145,
+      "loss": 0.4212,
+      "step": 4540
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.00011092404072043853,
+      "loss": 0.2368,
+      "step": 4550
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.00011072826938136258,
+      "loss": 0.1994,
+      "step": 4560
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.00011053249804228663,
+      "loss": 0.0994,
+      "step": 4570
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.00011033672670321065,
+      "loss": 0.167,
+      "step": 4580
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.00011014095536413468,
+      "loss": 0.2424,
+      "step": 4590
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.00010994518402505873,
+      "loss": 0.0427,
+      "step": 4600
+    },
+    {
+      "epoch": 3.6,
+      "eval_accuracy": 0.8515805705474171,
+      "eval_loss": 0.5903280377388,
+      "eval_runtime": 43.224,
+      "eval_samples_per_second": 60.013,
+      "eval_steps_per_second": 7.519,
+      "step": 4600
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.00010974941268598278,
+      "loss": 0.0999,
+      "step": 4610
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.00010955364134690682,
+      "loss": 0.151,
+      "step": 4620
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.00010935787000783086,
+      "loss": 0.1038,
+      "step": 4630
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.00010916209866875491,
+      "loss": 0.1604,
+      "step": 4640
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.00010896632732967893,
+      "loss": 0.1713,
+      "step": 4650
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.00010877055599060297,
+      "loss": 0.2647,
+      "step": 4660
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.00010857478465152702,
+      "loss": 0.2392,
+      "step": 4670
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.00010837901331245106,
+      "loss": 0.1119,
+      "step": 4680
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.00010818324197337511,
+      "loss": 0.0985,
+      "step": 4690
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.00010798747063429915,
+      "loss": 0.1806,
+      "step": 4700
+    },
+    {
+      "epoch": 3.68,
+      "eval_accuracy": 0.8627602158828065,
+      "eval_loss": 0.532320499420166,
+      "eval_runtime": 42.434,
+      "eval_samples_per_second": 61.13,
+      "eval_steps_per_second": 7.659,
+      "step": 4700
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0001077916992952232,
+      "loss": 0.0366,
+      "step": 4710
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.00010759592795614724,
+      "loss": 0.2069,
+      "step": 4720
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.00010740015661707126,
+      "loss": 0.134,
+      "step": 4730
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0001072043852779953,
+      "loss": 0.2445,
+      "step": 4740
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.00010700861393891935,
+      "loss": 0.1275,
+      "step": 4750
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0001068128425998434,
+      "loss": 0.1135,
+      "step": 4760
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.00010661707126076743,
+      "loss": 0.1426,
+      "step": 4770
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.00010642129992169148,
+      "loss": 0.0767,
+      "step": 4780
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.00010622552858261552,
+      "loss": 0.0461,
+      "step": 4790
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.00010602975724353954,
+      "loss": 0.2672,
+      "step": 4800
+    },
+    {
+      "epoch": 3.76,
+      "eval_accuracy": 0.8604471858134156,
+      "eval_loss": 0.5687525272369385,
+      "eval_runtime": 43.0037,
+      "eval_samples_per_second": 60.32,
+      "eval_steps_per_second": 7.557,
+      "step": 4800
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.00010583398590446358,
+      "loss": 0.1244,
+      "step": 4810
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.00010563821456538763,
+      "loss": 0.0886,
+      "step": 4820
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.00010544244322631168,
+      "loss": 0.1926,
+      "step": 4830
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.00010524667188723571,
+      "loss": 0.203,
+      "step": 4840
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.00010505090054815976,
+      "loss": 0.1096,
+      "step": 4850
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.0001048551292090838,
+      "loss": 0.0341,
+      "step": 4860
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.00010465935787000784,
+      "loss": 0.061,
+      "step": 4870
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.00010446358653093186,
+      "loss": 0.1125,
+      "step": 4880
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.00010426781519185591,
+      "loss": 0.1129,
+      "step": 4890
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.00010407204385277996,
+      "loss": 0.2674,
+      "step": 4900
+    },
+    {
+      "epoch": 3.84,
+      "eval_accuracy": 0.8635312259059368,
+      "eval_loss": 0.5369026064872742,
+      "eval_runtime": 42.5487,
+      "eval_samples_per_second": 60.965,
+      "eval_steps_per_second": 7.638,
+      "step": 4900
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.00010387627251370399,
+      "loss": 0.204,
+      "step": 4910
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.00010368050117462804,
+      "loss": 0.1772,
+      "step": 4920
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.00010348472983555209,
+      "loss": 0.1255,
+      "step": 4930
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.00010328895849647614,
+      "loss": 0.3212,
+      "step": 4940
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.00010309318715740016,
+      "loss": 0.2109,
+      "step": 4950
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.00010289741581832419,
+      "loss": 0.0613,
+      "step": 4960
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.00010270164447924824,
+      "loss": 0.177,
+      "step": 4970
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.00010250587314017229,
+      "loss": 0.0931,
+      "step": 4980
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.00010231010180109632,
+      "loss": 0.1028,
+      "step": 4990
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.00010211433046202037,
+      "loss": 0.2185,
+      "step": 5000
+    },
+    {
+      "epoch": 3.92,
+      "eval_accuracy": 0.882035466461064,
+      "eval_loss": 0.4742802381515503,
+      "eval_runtime": 42.9765,
+      "eval_samples_per_second": 60.359,
+      "eval_steps_per_second": 7.562,
+      "step": 5000
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.00010191855912294442,
+      "loss": 0.0499,
+      "step": 5010
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.00010172278778386845,
+      "loss": 0.1724,
+      "step": 5020
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.00010152701644479247,
+      "loss": 0.1446,
+      "step": 5030
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.00010133124510571652,
+      "loss": 0.2122,
+      "step": 5040
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.00010113547376664057,
+      "loss": 0.2345,
+      "step": 5050
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0001009397024275646,
+      "loss": 0.1296,
+      "step": 5060
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.00010074393108848865,
+      "loss": 0.2116,
+      "step": 5070
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.0001005481597494127,
+      "loss": 0.0821,
+      "step": 5080
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.00010035238841033673,
+      "loss": 0.2605,
+      "step": 5090
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.00010015661707126078,
+      "loss": 0.2195,
+      "step": 5100
+    },
+    {
+      "epoch": 3.99,
+      "eval_accuracy": 0.8708558211256746,
+      "eval_loss": 0.5340372920036316,
+      "eval_runtime": 42.404,
+      "eval_samples_per_second": 61.173,
+      "eval_steps_per_second": 7.664,
+      "step": 5100
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 9.996084573218482e-05,
+      "loss": 0.0939,
+      "step": 5110
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 9.976507439310885e-05,
+      "loss": 0.021,
+      "step": 5120
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 9.956930305403289e-05,
+      "loss": 0.0161,
+      "step": 5130
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 9.937353171495693e-05,
+      "loss": 0.0938,
+      "step": 5140
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 9.917776037588098e-05,
+      "loss": 0.1866,
+      "step": 5150
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 9.898198903680502e-05,
+      "loss": 0.068,
+      "step": 5160
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 9.878621769772905e-05,
+      "loss": 0.018,
+      "step": 5170
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 9.85904463586531e-05,
+      "loss": 0.1678,
+      "step": 5180
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 9.839467501957715e-05,
+      "loss": 0.0251,
+      "step": 5190
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 9.819890368050118e-05,
+      "loss": 0.0049,
+      "step": 5200
+    },
+    {
+      "epoch": 4.07,
+      "eval_accuracy": 0.8608326908249807,
+      "eval_loss": 0.5882797837257385,
+      "eval_runtime": 42.9904,
+      "eval_samples_per_second": 60.339,
+      "eval_steps_per_second": 7.56,
+      "step": 5200
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 9.800313234142522e-05,
+      "loss": 0.0808,
+      "step": 5210
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 9.780736100234926e-05,
+      "loss": 0.131,
+      "step": 5220
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 9.761158966327331e-05,
+      "loss": 0.1858,
+      "step": 5230
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 9.741581832419733e-05,
+      "loss": 0.017,
+      "step": 5240
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 9.722004698512138e-05,
+      "loss": 0.0228,
+      "step": 5250
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 9.702427564604543e-05,
+      "loss": 0.0183,
+      "step": 5260
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 9.682850430696946e-05,
+      "loss": 0.0653,
+      "step": 5270
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 9.66327329678935e-05,
+      "loss": 0.0188,
+      "step": 5280
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 9.643696162881755e-05,
+      "loss": 0.0113,
+      "step": 5290
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 9.62411902897416e-05,
+      "loss": 0.0204,
+      "step": 5300
+    },
+    {
+      "epoch": 4.15,
+      "eval_accuracy": 0.853893600616808,
+      "eval_loss": 0.6102246642112732,
+      "eval_runtime": 43.196,
+      "eval_samples_per_second": 60.052,
+      "eval_steps_per_second": 7.524,
+      "step": 5300
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 9.604541895066563e-05,
+      "loss": 0.0485,
+      "step": 5310
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 9.584964761158966e-05,
+      "loss": 0.1008,
+      "step": 5320
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 9.565387627251371e-05,
+      "loss": 0.0887,
+      "step": 5330
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 9.545810493343776e-05,
+      "loss": 0.1583,
+      "step": 5340
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 9.526233359436178e-05,
+      "loss": 0.079,
+      "step": 5350
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 9.506656225528583e-05,
+      "loss": 0.0464,
+      "step": 5360
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 9.487079091620988e-05,
+      "loss": 0.102,
+      "step": 5370
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 9.467501957713391e-05,
+      "loss": 0.0247,
+      "step": 5380
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 9.447924823805795e-05,
+      "loss": 0.0791,
+      "step": 5390
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 9.4283476898982e-05,
+      "loss": 0.0652,
+      "step": 5400
+    },
+    {
+      "epoch": 4.23,
+      "eval_accuracy": 0.8670007710100232,
+      "eval_loss": 0.5658612847328186,
+      "eval_runtime": 43.1575,
+      "eval_samples_per_second": 60.105,
+      "eval_steps_per_second": 7.531,
+      "step": 5400
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 9.408770555990604e-05,
+      "loss": 0.0195,
+      "step": 5410
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 9.389193422083008e-05,
+      "loss": 0.0612,
+      "step": 5420
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 9.369616288175411e-05,
+      "loss": 0.0867,
+      "step": 5430
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 9.350039154267816e-05,
+      "loss": 0.0931,
+      "step": 5440
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 9.33046202036022e-05,
+      "loss": 0.0365,
+      "step": 5450
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 9.310884886452623e-05,
+      "loss": 0.0487,
+      "step": 5460
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 9.291307752545028e-05,
+      "loss": 0.0199,
+      "step": 5470
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 9.271730618637432e-05,
+      "loss": 0.1251,
+      "step": 5480
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 9.252153484729836e-05,
+      "loss": 0.0314,
+      "step": 5490
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 9.232576350822239e-05,
+      "loss": 0.028,
+      "step": 5500
+    },
+    {
+      "epoch": 4.31,
+      "eval_accuracy": 0.8839629915188898,
+      "eval_loss": 0.4916422665119171,
+      "eval_runtime": 42.5129,
+      "eval_samples_per_second": 61.017,
+      "eval_steps_per_second": 7.645,
+      "step": 5500
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 9.212999216914644e-05,
+      "loss": 0.0537,
+      "step": 5510
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 9.193422083007049e-05,
+      "loss": 0.0384,
+      "step": 5520
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 9.173844949099452e-05,
+      "loss": 0.0924,
+      "step": 5530
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 9.154267815191856e-05,
+      "loss": 0.0085,
+      "step": 5540
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 9.13469068128426e-05,
+      "loss": 0.0653,
+      "step": 5550
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 9.115113547376664e-05,
+      "loss": 0.1537,
+      "step": 5560
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 9.095536413469069e-05,
+      "loss": 0.0295,
+      "step": 5570
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 9.075959279561472e-05,
+      "loss": 0.0677,
+      "step": 5580
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 9.056382145653877e-05,
+      "loss": 0.0472,
+      "step": 5590
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 9.03680501174628e-05,
+      "loss": 0.0423,
+      "step": 5600
+    },
+    {
+      "epoch": 4.39,
+      "eval_accuracy": 0.8735543562066307,
+      "eval_loss": 0.5706442594528198,
+      "eval_runtime": 42.5535,
+      "eval_samples_per_second": 60.959,
+      "eval_steps_per_second": 7.637,
+      "step": 5600
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 9.017227877838685e-05,
+      "loss": 0.1299,
+      "step": 5610
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 8.997650743931089e-05,
+      "loss": 0.045,
+      "step": 5620
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 8.978073610023494e-05,
+      "loss": 0.0185,
+      "step": 5630
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 8.958496476115897e-05,
+      "loss": 0.0174,
+      "step": 5640
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 8.9389193422083e-05,
+      "loss": 0.0276,
+      "step": 5650
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 8.919342208300705e-05,
+      "loss": 0.0087,
+      "step": 5660
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 8.899765074393109e-05,
+      "loss": 0.0353,
+      "step": 5670
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 8.880187940485514e-05,
+      "loss": 0.0432,
+      "step": 5680
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 8.860610806577917e-05,
+      "loss": 0.0379,
+      "step": 5690
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 8.841033672670322e-05,
+      "loss": 0.0087,
+      "step": 5700
+    },
+    {
+      "epoch": 4.46,
+      "eval_accuracy": 0.8696993060909792,
+      "eval_loss": 0.5652517676353455,
+      "eval_runtime": 42.4374,
+      "eval_samples_per_second": 61.125,
+      "eval_steps_per_second": 7.658,
+      "step": 5700
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 8.821456538762725e-05,
+      "loss": 0.0528,
+      "step": 5710
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 8.80187940485513e-05,
+      "loss": 0.0121,
+      "step": 5720
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 8.782302270947533e-05,
+      "loss": 0.0709,
+      "step": 5730
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 8.762725137039938e-05,
+      "loss": 0.0972,
+      "step": 5740
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 8.743148003132342e-05,
+      "loss": 0.0282,
+      "step": 5750
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 8.723570869224747e-05,
+      "loss": 0.0089,
+      "step": 5760
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 8.70399373531715e-05,
+      "loss": 0.0967,
+      "step": 5770
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 8.684416601409553e-05,
+      "loss": 0.0987,
+      "step": 5780
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 8.664839467501958e-05,
+      "loss": 0.0433,
+      "step": 5790
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 8.645262333594362e-05,
+      "loss": 0.0964,
+      "step": 5800
+    },
+    {
+      "epoch": 4.54,
+      "eval_accuracy": 0.8754818812644565,
+      "eval_loss": 0.5423335433006287,
+      "eval_runtime": 42.6014,
+      "eval_samples_per_second": 60.89,
+      "eval_steps_per_second": 7.629,
+      "step": 5800
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 8.625685199686766e-05,
+      "loss": 0.0588,
+      "step": 5810
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 8.60610806577917e-05,
+      "loss": 0.0106,
+      "step": 5820
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 8.586530931871575e-05,
+      "loss": 0.028,
+      "step": 5830
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 8.566953797963978e-05,
+      "loss": 0.0728,
+      "step": 5840
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 8.547376664056383e-05,
+      "loss": 0.0151,
+      "step": 5850
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 8.527799530148786e-05,
+      "loss": 0.0728,
+      "step": 5860
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 8.508222396241191e-05,
+      "loss": 0.0515,
+      "step": 5870
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 8.488645262333595e-05,
+      "loss": 0.017,
+      "step": 5880
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 8.469068128425998e-05,
+      "loss": 0.0212,
+      "step": 5890
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 8.449490994518403e-05,
+      "loss": 0.0841,
+      "step": 5900
+    },
+    {
+      "epoch": 4.62,
+      "eval_accuracy": 0.874325366229761,
+      "eval_loss": 0.5160160660743713,
+      "eval_runtime": 42.4551,
+      "eval_samples_per_second": 61.1,
+      "eval_steps_per_second": 7.655,
+      "step": 5900
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 8.429913860610808e-05,
+      "loss": 0.0324,
+      "step": 5910
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 8.410336726703211e-05,
+      "loss": 0.1511,
+      "step": 5920
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 8.390759592795615e-05,
+      "loss": 0.0696,
+      "step": 5930
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 8.37118245888802e-05,
+      "loss": 0.0583,
+      "step": 5940
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 8.351605324980424e-05,
+      "loss": 0.2193,
+      "step": 5950
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 8.332028191072828e-05,
+      "loss": 0.0459,
+      "step": 5960
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 8.312451057165231e-05,
+      "loss": 0.0129,
+      "step": 5970
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 8.292873923257636e-05,
+      "loss": 0.0559,
+      "step": 5980
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 8.27329678935004e-05,
+      "loss": 0.0874,
+      "step": 5990
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 8.253719655442443e-05,
+      "loss": 0.0945,
+      "step": 6000
+    },
+    {
+      "epoch": 4.7,
+      "eval_accuracy": 0.8696993060909792,
+      "eval_loss": 0.5532014966011047,
+      "eval_runtime": 42.9383,
+      "eval_samples_per_second": 60.412,
+      "eval_steps_per_second": 7.569,
+      "step": 6000
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 8.234142521534848e-05,
+      "loss": 0.1306,
+      "step": 6010
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 8.214565387627252e-05,
+      "loss": 0.0446,
+      "step": 6020
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 8.194988253719656e-05,
+      "loss": 0.0247,
+      "step": 6030
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 8.17541111981206e-05,
+      "loss": 0.1395,
+      "step": 6040
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 8.155833985904464e-05,
+      "loss": 0.0221,
+      "step": 6050
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 8.136256851996869e-05,
+      "loss": 0.1725,
+      "step": 6060
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 8.116679718089272e-05,
+      "loss": 0.184,
+      "step": 6070
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 8.097102584181676e-05,
+      "loss": 0.0241,
+      "step": 6080
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 8.07752545027408e-05,
+      "loss": 0.0739,
+      "step": 6090
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 8.057948316366485e-05,
+      "loss": 0.0311,
+      "step": 6100
+    },
+    {
+      "epoch": 4.78,
+      "eval_accuracy": 0.8866615265998458,
+      "eval_loss": 0.49468106031417847,
+      "eval_runtime": 42.2707,
+      "eval_samples_per_second": 61.366,
+      "eval_steps_per_second": 7.689,
+      "step": 6100
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 8.038371182458888e-05,
+      "loss": 0.0041,
+      "step": 6110
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 8.018794048551292e-05,
+      "loss": 0.0041,
+      "step": 6120
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 7.999216914643697e-05,
+      "loss": 0.0471,
+      "step": 6130
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 7.9796397807361e-05,
+      "loss": 0.0728,
+      "step": 6140
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 7.960062646828504e-05,
+      "loss": 0.0837,
+      "step": 6150
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 7.940485512920909e-05,
+      "loss": 0.02,
+      "step": 6160
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 7.920908379013314e-05,
+      "loss": 0.0659,
+      "step": 6170
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 7.901331245105716e-05,
+      "loss": 0.0038,
+      "step": 6180
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 7.88175411119812e-05,
+      "loss": 0.034,
+      "step": 6190
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 7.862176977290525e-05,
+      "loss": 0.0423,
+      "step": 6200
+    },
+    {
+      "epoch": 4.86,
+      "eval_accuracy": 0.8843484965304549,
+      "eval_loss": 0.5063336491584778,
+      "eval_runtime": 42.8346,
+      "eval_samples_per_second": 60.558,
+      "eval_steps_per_second": 7.587,
+      "step": 6200
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 7.84259984338293e-05,
+      "loss": 0.0176,
+      "step": 6210
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 7.823022709475332e-05,
+      "loss": 0.0743,
+      "step": 6220
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 7.803445575567737e-05,
+      "loss": 0.0121,
+      "step": 6230
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 7.783868441660142e-05,
+      "loss": 0.0319,
+      "step": 6240
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 7.764291307752545e-05,
+      "loss": 0.0833,
+      "step": 6250
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 7.744714173844949e-05,
+      "loss": 0.0819,
+      "step": 6260
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 7.725137039937354e-05,
+      "loss": 0.0198,
+      "step": 6270
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 7.705559906029758e-05,
+      "loss": 0.1829,
+      "step": 6280
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 7.685982772122162e-05,
+      "loss": 0.0522,
+      "step": 6290
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 7.666405638214565e-05,
+      "loss": 0.1348,
+      "step": 6300
+    },
+    {
+      "epoch": 4.93,
+      "eval_accuracy": 0.874325366229761,
+      "eval_loss": 0.5619357824325562,
+      "eval_runtime": 43.1362,
+      "eval_samples_per_second": 60.135,
+      "eval_steps_per_second": 7.534,
+      "step": 6300
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 7.64682850430697e-05,
+      "loss": 0.0771,
+      "step": 6310
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 7.627251370399374e-05,
+      "loss": 0.0655,
+      "step": 6320
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 7.607674236491777e-05,
+      "loss": 0.0091,
+      "step": 6330
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 7.588097102584182e-05,
+      "loss": 0.0475,
+      "step": 6340
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 7.568519968676587e-05,
+      "loss": 0.0593,
+      "step": 6350
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 7.54894283476899e-05,
+      "loss": 0.0078,
+      "step": 6360
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 7.529365700861393e-05,
+      "loss": 0.0049,
+      "step": 6370
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 7.509788566953798e-05,
+      "loss": 0.0143,
+      "step": 6380
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 7.490211433046203e-05,
+      "loss": 0.1002,
+      "step": 6390
+    },
+    {
+      "epoch": 5.01,
+      "learning_rate": 7.470634299138607e-05,
+      "loss": 0.049,
+      "step": 6400
+    },
+    {
+      "epoch": 5.01,
+      "eval_accuracy": 0.8731688511950655,
+      "eval_loss": 0.5800209045410156,
+      "eval_runtime": 44.2948,
+      "eval_samples_per_second": 58.562,
+      "eval_steps_per_second": 7.337,
+      "step": 6400
+    },
+    {
+      "epoch": 5.02,
+      "learning_rate": 7.45105716523101e-05,
+      "loss": 0.012,
+      "step": 6410
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 7.431480031323415e-05,
+      "loss": 0.006,
+      "step": 6420
+    },
+    {
+      "epoch": 5.04,
+      "learning_rate": 7.411902897415818e-05,
+      "loss": 0.0024,
+      "step": 6430
+    },
+    {
+      "epoch": 5.04,
+      "learning_rate": 7.392325763508223e-05,
+      "loss": 0.0566,
+      "step": 6440
+    },
+    {
+      "epoch": 5.05,
+      "learning_rate": 7.372748629600627e-05,
+      "loss": 0.0055,
+      "step": 6450
+    },
+    {
+      "epoch": 5.06,
+      "learning_rate": 7.353171495693031e-05,
+      "loss": 0.0197,
+      "step": 6460
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 7.333594361785435e-05,
+      "loss": 0.0017,
+      "step": 6470
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 7.314017227877838e-05,
+      "loss": 0.0488,
+      "step": 6480
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 7.294440093970243e-05,
+      "loss": 0.1744,
+      "step": 6490
+    },
+    {
+      "epoch": 5.09,
+      "learning_rate": 7.274862960062648e-05,
+      "loss": 0.0053,
+      "step": 6500
+    },
+    {
+      "epoch": 5.09,
+      "eval_accuracy": 0.8770239013107171,
+      "eval_loss": 0.5499477982521057,
+      "eval_runtime": 42.4492,
+      "eval_samples_per_second": 61.108,
+      "eval_steps_per_second": 7.656,
+      "step": 6500
+    },
+    {
+      "epoch": 5.1,
+      "learning_rate": 7.255285826155051e-05,
+      "loss": 0.0047,
+      "step": 6510
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 7.235708692247455e-05,
+      "loss": 0.0056,
+      "step": 6520
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 7.21613155833986e-05,
+      "loss": 0.0018,
+      "step": 6530
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 7.196554424432263e-05,
+      "loss": 0.0026,
+      "step": 6540
+    },
+    {
+      "epoch": 5.13,
+      "learning_rate": 7.176977290524668e-05,
+      "loss": 0.0018,
+      "step": 6550
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 7.157400156617071e-05,
+      "loss": 0.0294,
+      "step": 6560
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 7.137823022709476e-05,
+      "loss": 0.0158,
+      "step": 6570
+    },
+    {
+      "epoch": 5.15,
+      "learning_rate": 7.11824588880188e-05,
+      "loss": 0.0116,
+      "step": 6580
+    },
+    {
+      "epoch": 5.16,
+      "learning_rate": 7.098668754894284e-05,
+      "loss": 0.0015,
+      "step": 6590
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 7.079091620986688e-05,
+      "loss": 0.0234,
+      "step": 6600
+    },
+    {
+      "epoch": 5.17,
+      "eval_accuracy": 0.887432536622976,
+      "eval_loss": 0.5102210640907288,
+      "eval_runtime": 42.9176,
+      "eval_samples_per_second": 60.441,
+      "eval_steps_per_second": 7.573,
+      "step": 6600
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 7.059514487079093e-05,
+      "loss": 0.0013,
+      "step": 6610
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 7.039937353171496e-05,
+      "loss": 0.0074,
+      "step": 6620
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 7.020360219263901e-05,
+      "loss": 0.0108,
+      "step": 6630
+    },
+    {
+      "epoch": 5.2,
+      "learning_rate": 7.000783085356304e-05,
+      "loss": 0.0025,
+      "step": 6640
+    },
+    {
+      "epoch": 5.21,
+      "learning_rate": 6.981205951448708e-05,
+      "loss": 0.0259,
+      "step": 6650
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 6.961628817541112e-05,
+      "loss": 0.0165,
+      "step": 6660
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 6.942051683633516e-05,
+      "loss": 0.0055,
+      "step": 6670
+    },
+    {
+      "epoch": 5.23,
+      "learning_rate": 6.922474549725921e-05,
+      "loss": 0.0042,
+      "step": 6680
+    },
+    {
+      "epoch": 5.24,
+      "learning_rate": 6.902897415818324e-05,
+      "loss": 0.0347,
+      "step": 6690
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 6.883320281910729e-05,
+      "loss": 0.0192,
+      "step": 6700
+    },
+    {
+      "epoch": 5.25,
+      "eval_accuracy": 0.8835774865073246,
+      "eval_loss": 0.5447149872779846,
+      "eval_runtime": 43.0165,
+      "eval_samples_per_second": 60.302,
+      "eval_steps_per_second": 7.555,
+      "step": 6700
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 6.863743148003132e-05,
+      "loss": 0.0683,
+      "step": 6710
+    },
+    {
+      "epoch": 5.26,
+      "learning_rate": 6.844166014095537e-05,
+      "loss": 0.0041,
+      "step": 6720
+    },
+    {
+      "epoch": 5.27,
+      "learning_rate": 6.824588880187941e-05,
+      "loss": 0.0575,
+      "step": 6730
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 6.805011746280345e-05,
+      "loss": 0.0031,
+      "step": 6740
+    },
+    {
+      "epoch": 5.29,
+      "learning_rate": 6.785434612372749e-05,
+      "loss": 0.002,
+      "step": 6750
+    },
+    {
+      "epoch": 5.29,
+      "learning_rate": 6.765857478465152e-05,
+      "loss": 0.0028,
+      "step": 6760
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 6.746280344557557e-05,
+      "loss": 0.0124,
+      "step": 6770
+    },
+    {
+      "epoch": 5.31,
+      "learning_rate": 6.726703210649962e-05,
+      "loss": 0.0019,
+      "step": 6780
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 6.707126076742365e-05,
+      "loss": 0.0016,
+      "step": 6790
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 6.687548942834769e-05,
+      "loss": 0.0029,
+      "step": 6800
+    },
+    {
+      "epoch": 5.32,
+      "eval_accuracy": 0.8936006168080185,
+      "eval_loss": 0.4787052571773529,
+      "eval_runtime": 42.3932,
+      "eval_samples_per_second": 61.189,
+      "eval_steps_per_second": 7.666,
+      "step": 6800
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 6.667971808927174e-05,
+      "loss": 0.0104,
+      "step": 6810
+    },
+    {
+      "epoch": 5.34,
+      "learning_rate": 6.648394675019577e-05,
+      "loss": 0.0096,
+      "step": 6820
+    },
+    {
+      "epoch": 5.35,
+      "learning_rate": 6.628817541111982e-05,
+      "loss": 0.0022,
+      "step": 6830
+    },
+    {
+      "epoch": 5.36,
+      "learning_rate": 6.609240407204385e-05,
+      "loss": 0.0072,
+      "step": 6840
+    },
+    {
+      "epoch": 5.36,
+      "learning_rate": 6.58966327329679e-05,
+      "loss": 0.0043,
+      "step": 6850
+    },
+    {
+      "epoch": 5.37,
+      "learning_rate": 6.570086139389194e-05,
+      "loss": 0.0204,
+      "step": 6860
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 6.550509005481597e-05,
+      "loss": 0.0127,
+      "step": 6870
+    },
+    {
+      "epoch": 5.39,
+      "learning_rate": 6.530931871574002e-05,
+      "loss": 0.0077,
+      "step": 6880
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 6.511354737666407e-05,
+      "loss": 0.0084,
+      "step": 6890
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 6.49177760375881e-05,
+      "loss": 0.0249,
+      "step": 6900
+    },
+    {
+      "epoch": 5.4,
+      "eval_accuracy": 0.887047031611411,
+      "eval_loss": 0.5231930017471313,
+      "eval_runtime": 42.7159,
+      "eval_samples_per_second": 60.727,
+      "eval_steps_per_second": 7.608,
+      "step": 6900
+    },
+    {
+      "epoch": 5.41,
+      "learning_rate": 6.472200469851214e-05,
+      "loss": 0.0032,
+      "step": 6910
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 6.452623335943618e-05,
+      "loss": 0.0027,
+      "step": 6920
+    },
+    {
+      "epoch": 5.43,
+      "learning_rate": 6.433046202036023e-05,
+      "loss": 0.0036,
+      "step": 6930
+    },
+    {
+      "epoch": 5.43,
+      "learning_rate": 6.413469068128425e-05,
+      "loss": 0.0077,
+      "step": 6940
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 6.39389193422083e-05,
+      "loss": 0.0783,
+      "step": 6950
+    },
+    {
+      "epoch": 5.45,
+      "learning_rate": 6.374314800313235e-05,
+      "loss": 0.0027,
+      "step": 6960
+    },
+    {
+      "epoch": 5.46,
+      "learning_rate": 6.35473766640564e-05,
+      "loss": 0.0012,
+      "step": 6970
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 6.335160532498042e-05,
+      "loss": 0.0022,
+      "step": 6980
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 6.315583398590447e-05,
+      "loss": 0.0021,
+      "step": 6990
+    },
+    {
+      "epoch": 5.48,
+      "learning_rate": 6.296006264682851e-05,
+      "loss": 0.0671,
+      "step": 7000
+    },
+    {
+      "epoch": 5.48,
+      "eval_accuracy": 0.89745566692367,
+      "eval_loss": 0.476556658744812,
+      "eval_runtime": 42.526,
+      "eval_samples_per_second": 60.998,
+      "eval_steps_per_second": 7.642,
+      "step": 7000
+    },
+    {
+      "epoch": 5.49,
+      "learning_rate": 6.276429130775255e-05,
+      "loss": 0.0072,
+      "step": 7010
+    },
+    {
+      "epoch": 5.5,
+      "learning_rate": 6.256851996867658e-05,
+      "loss": 0.0417,
+      "step": 7020
+    },
+    {
+      "epoch": 5.51,
+      "learning_rate": 6.237274862960063e-05,
+      "loss": 0.0029,
+      "step": 7030
+    },
+    {
+      "epoch": 5.51,
+      "learning_rate": 6.217697729052468e-05,
+      "loss": 0.0037,
+      "step": 7040
+    },
+    {
+      "epoch": 5.52,
+      "learning_rate": 6.19812059514487e-05,
+      "loss": 0.0043,
+      "step": 7050
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 6.178543461237275e-05,
+      "loss": 0.0443,
+      "step": 7060
+    },
+    {
+      "epoch": 5.54,
+      "learning_rate": 6.15896632732968e-05,
+      "loss": 0.0719,
+      "step": 7070
+    },
+    {
+      "epoch": 5.54,
+      "learning_rate": 6.139389193422083e-05,
+      "loss": 0.0917,
+      "step": 7080
+    },
+    {
+      "epoch": 5.55,
+      "learning_rate": 6.119812059514487e-05,
+      "loss": 0.0033,
+      "step": 7090
+    },
+    {
+      "epoch": 5.56,
+      "learning_rate": 6.100234925606891e-05,
+      "loss": 0.0056,
+      "step": 7100
+    },
+    {
+      "epoch": 5.56,
+      "eval_accuracy": 0.8893600616808018,
+      "eval_loss": 0.5135548710823059,
+      "eval_runtime": 42.8439,
+      "eval_samples_per_second": 60.545,
+      "eval_steps_per_second": 7.586,
+      "step": 7100
+    },
+    {
+      "epoch": 5.57,
+      "learning_rate": 6.0806577916992954e-05,
+      "loss": 0.0099,
+      "step": 7110
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 6.0610806577917e-05,
+      "loss": 0.0245,
+      "step": 7120
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 6.041503523884103e-05,
+      "loss": 0.0109,
+      "step": 7130
+    },
+    {
+      "epoch": 5.59,
+      "learning_rate": 6.021926389976508e-05,
+      "loss": 0.0145,
+      "step": 7140
+    },
+    {
+      "epoch": 5.6,
+      "learning_rate": 6.002349256068912e-05,
+      "loss": 0.0024,
+      "step": 7150
+    },
+    {
+      "epoch": 5.61,
+      "learning_rate": 5.9827721221613154e-05,
+      "loss": 0.0021,
+      "step": 7160
+    },
+    {
+      "epoch": 5.61,
+      "learning_rate": 5.9631949882537195e-05,
+      "loss": 0.0014,
+      "step": 7170
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 5.943617854346124e-05,
+      "loss": 0.0013,
+      "step": 7180
+    },
+    {
+      "epoch": 5.63,
+      "learning_rate": 5.9240407204385285e-05,
+      "loss": 0.0049,
+      "step": 7190
+    },
+    {
+      "epoch": 5.64,
+      "learning_rate": 5.904463586530932e-05,
+      "loss": 0.003,
+      "step": 7200
+    },
+    {
+      "epoch": 5.64,
+      "eval_accuracy": 0.8882035466461063,
+      "eval_loss": 0.5084769129753113,
+      "eval_runtime": 42.353,
+      "eval_samples_per_second": 61.247,
+      "eval_steps_per_second": 7.674,
+      "step": 7200
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 5.884886452623336e-05,
+      "loss": 0.0016,
+      "step": 7210
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 5.86530931871574e-05,
+      "loss": 0.0355,
+      "step": 7220
+    },
+    {
+      "epoch": 5.66,
+      "learning_rate": 5.845732184808145e-05,
+      "loss": 0.0793,
+      "step": 7230
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 5.826155050900548e-05,
+      "loss": 0.0582,
+      "step": 7240
+    },
+    {
+      "epoch": 5.68,
+      "learning_rate": 5.8065779169929525e-05,
+      "loss": 0.0028,
+      "step": 7250
+    },
+    {
+      "epoch": 5.69,
+      "learning_rate": 5.787000783085357e-05,
+      "loss": 0.003,
+      "step": 7260
+    },
+    {
+      "epoch": 5.69,
+      "learning_rate": 5.767423649177761e-05,
+      "loss": 0.0013,
+      "step": 7270
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 5.747846515270164e-05,
+      "loss": 0.0021,
+      "step": 7280
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 5.728269381362569e-05,
+      "loss": 0.0021,
+      "step": 7290
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 5.708692247454973e-05,
+      "loss": 0.0015,
+      "step": 7300
+    },
+    {
+      "epoch": 5.72,
+      "eval_accuracy": 0.8970701619121049,
+      "eval_loss": 0.4831724762916565,
+      "eval_runtime": 42.709,
+      "eval_samples_per_second": 60.737,
+      "eval_steps_per_second": 7.61,
+      "step": 7300
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 5.689115113547377e-05,
+      "loss": 0.0043,
+      "step": 7310
+    },
+    {
+      "epoch": 5.73,
+      "learning_rate": 5.669537979639781e-05,
+      "loss": 0.0014,
+      "step": 7320
+    },
+    {
+      "epoch": 5.74,
+      "learning_rate": 5.649960845732185e-05,
+      "loss": 0.0688,
+      "step": 7330
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 5.63038371182459e-05,
+      "loss": 0.0218,
+      "step": 7340
+    },
+    {
+      "epoch": 5.76,
+      "learning_rate": 5.6108065779169924e-05,
+      "loss": 0.0265,
+      "step": 7350
+    },
+    {
+      "epoch": 5.76,
+      "learning_rate": 5.591229444009397e-05,
+      "loss": 0.0135,
+      "step": 7360
+    },
+    {
+      "epoch": 5.77,
+      "learning_rate": 5.5716523101018014e-05,
+      "loss": 0.0013,
+      "step": 7370
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 5.5520751761942055e-05,
+      "loss": 0.0053,
+      "step": 7380
+    },
+    {
+      "epoch": 5.79,
+      "learning_rate": 5.532498042286609e-05,
+      "loss": 0.0043,
+      "step": 7390
+    },
+    {
+      "epoch": 5.79,
+      "learning_rate": 5.512920908379014e-05,
+      "loss": 0.0014,
+      "step": 7400
+    },
+    {
+      "epoch": 5.79,
+      "eval_accuracy": 0.899768696993061,
+      "eval_loss": 0.4648475646972656,
+      "eval_runtime": 42.8931,
+      "eval_samples_per_second": 60.476,
+      "eval_steps_per_second": 7.577,
+      "step": 7400
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 5.493343774471418e-05,
+      "loss": 0.041,
+      "step": 7410
+    },
+    {
+      "epoch": 5.81,
+      "learning_rate": 5.473766640563822e-05,
+      "loss": 0.0023,
+      "step": 7420
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 5.4541895066562255e-05,
+      "loss": 0.0121,
+      "step": 7430
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 5.4346123727486296e-05,
+      "loss": 0.0374,
+      "step": 7440
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 5.4150352388410344e-05,
+      "loss": 0.0411,
+      "step": 7450
+    },
+    {
+      "epoch": 5.84,
+      "learning_rate": 5.3954581049334385e-05,
+      "loss": 0.003,
+      "step": 7460
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 5.375880971025842e-05,
+      "loss": 0.0196,
+      "step": 7470
+    },
+    {
+      "epoch": 5.86,
+      "learning_rate": 5.356303837118246e-05,
+      "loss": 0.0024,
+      "step": 7480
+    },
+    {
+      "epoch": 5.87,
+      "learning_rate": 5.33672670321065e-05,
+      "loss": 0.0114,
+      "step": 7490
+    },
+    {
+      "epoch": 5.87,
+      "learning_rate": 5.317149569303054e-05,
+      "loss": 0.0065,
+      "step": 7500
+    },
+    {
+      "epoch": 5.87,
+      "eval_accuracy": 0.8978411719352352,
+      "eval_loss": 0.4739063084125519,
+      "eval_runtime": 42.5769,
+      "eval_samples_per_second": 60.925,
+      "eval_steps_per_second": 7.633,
+      "step": 7500
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 5.297572435395458e-05,
+      "loss": 0.0013,
+      "step": 7510
+    },
+    {
+      "epoch": 5.89,
+      "learning_rate": 5.2779953014878626e-05,
+      "loss": 0.032,
+      "step": 7520
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 5.258418167580267e-05,
+      "loss": 0.002,
+      "step": 7530
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 5.23884103367267e-05,
+      "loss": 0.0018,
+      "step": 7540
+    },
+    {
+      "epoch": 5.91,
+      "learning_rate": 5.219263899765074e-05,
+      "loss": 0.0334,
+      "step": 7550
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 5.199686765857479e-05,
+      "loss": 0.0019,
+      "step": 7560
+    },
+    {
+      "epoch": 5.93,
+      "learning_rate": 5.180109631949883e-05,
+      "loss": 0.0012,
+      "step": 7570
+    },
+    {
+      "epoch": 5.94,
+      "learning_rate": 5.160532498042287e-05,
+      "loss": 0.0201,
+      "step": 7580
+    },
+    {
+      "epoch": 5.94,
+      "learning_rate": 5.140955364134691e-05,
+      "loss": 0.0651,
+      "step": 7590
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 5.121378230227095e-05,
+      "loss": 0.0011,
+      "step": 7600
+    },
+    {
+      "epoch": 5.95,
+      "eval_accuracy": 0.8866615265998458,
+      "eval_loss": 0.5348610281944275,
+      "eval_runtime": 42.5481,
+      "eval_samples_per_second": 60.966,
+      "eval_steps_per_second": 7.638,
+      "step": 7600
+    },
+    {
+      "epoch": 5.96,
+      "learning_rate": 5.1018010963195e-05,
+      "loss": 0.0339,
+      "step": 7610
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 5.0822239624119025e-05,
+      "loss": 0.0269,
+      "step": 7620
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 5.062646828504307e-05,
+      "loss": 0.0175,
+      "step": 7630
+    },
+    {
+      "epoch": 5.98,
+      "learning_rate": 5.0430696945967114e-05,
+      "loss": 0.0098,
+      "step": 7640
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 5.0234925606891156e-05,
+      "loss": 0.0115,
+      "step": 7650
+    },
+    {
+      "epoch": 6.0,
+      "learning_rate": 5.003915426781519e-05,
+      "loss": 0.0017,
+      "step": 7660
+    },
+    {
+      "epoch": 6.01,
+      "learning_rate": 4.984338292873924e-05,
+      "loss": 0.0011,
+      "step": 7670
+    },
+    {
+      "epoch": 6.01,
+      "learning_rate": 4.964761158966327e-05,
+      "loss": 0.0034,
+      "step": 7680
+    },
+    {
+      "epoch": 6.02,
+      "learning_rate": 4.9451840250587314e-05,
+      "loss": 0.0085,
+      "step": 7690
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 4.9256068911511355e-05,
+      "loss": 0.0021,
+      "step": 7700
+    },
+    {
+      "epoch": 6.03,
+      "eval_accuracy": 0.8847340015420201,
+      "eval_loss": 0.5459861755371094,
+      "eval_runtime": 42.4623,
+      "eval_samples_per_second": 61.09,
+      "eval_steps_per_second": 7.654,
+      "step": 7700
+    },
+    {
+      "epoch": 6.04,
+      "learning_rate": 4.9060297572435396e-05,
+      "loss": 0.0011,
+      "step": 7710
+    },
+    {
+      "epoch": 6.05,
+      "learning_rate": 4.886452623335944e-05,
+      "loss": 0.0071,
+      "step": 7720
+    },
+    {
+      "epoch": 6.05,
+      "learning_rate": 4.866875489428348e-05,
+      "loss": 0.0667,
+      "step": 7730
+    },
+    {
+      "epoch": 6.06,
+      "learning_rate": 4.847298355520752e-05,
+      "loss": 0.001,
+      "step": 7740
+    },
+    {
+      "epoch": 6.07,
+      "learning_rate": 4.827721221613156e-05,
+      "loss": 0.112,
+      "step": 7750
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 4.80814408770556e-05,
+      "loss": 0.0251,
+      "step": 7760
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 4.7885669537979644e-05,
+      "loss": 0.0011,
+      "step": 7770
+    },
+    {
+      "epoch": 6.09,
+      "learning_rate": 4.7689898198903685e-05,
+      "loss": 0.001,
+      "step": 7780
+    },
+    {
+      "epoch": 6.1,
+      "learning_rate": 4.7494126859827726e-05,
+      "loss": 0.0055,
+      "step": 7790
+    },
+    {
+      "epoch": 6.11,
+      "learning_rate": 4.729835552075176e-05,
+      "loss": 0.0012,
+      "step": 7800
+    },
+    {
+      "epoch": 6.11,
+      "eval_accuracy": 0.8889745566692367,
+      "eval_loss": 0.5309260487556458,
+      "eval_runtime": 42.6745,
+      "eval_samples_per_second": 60.786,
+      "eval_steps_per_second": 7.616,
+      "step": 7800
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 4.710258418167581e-05,
+      "loss": 0.018,
+      "step": 7810
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 4.6906812842599844e-05,
+      "loss": 0.0009,
+      "step": 7820
+    },
+    {
+      "epoch": 6.13,
+      "learning_rate": 4.671104150352389e-05,
+      "loss": 0.0089,
+      "step": 7830
+    },
+    {
+      "epoch": 6.14,
+      "learning_rate": 4.6515270164447926e-05,
+      "loss": 0.005,
+      "step": 7840
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 4.631949882537197e-05,
+      "loss": 0.0006,
+      "step": 7850
+    },
+    {
+      "epoch": 6.16,
+      "learning_rate": 4.612372748629601e-05,
+      "loss": 0.0011,
+      "step": 7860
+    },
+    {
+      "epoch": 6.16,
+      "learning_rate": 4.592795614722005e-05,
+      "loss": 0.0019,
+      "step": 7870
+    },
+    {
+      "epoch": 6.17,
+      "learning_rate": 4.573218480814409e-05,
+      "loss": 0.0842,
+      "step": 7880
+    },
+    {
+      "epoch": 6.18,
+      "learning_rate": 4.5536413469068126e-05,
+      "loss": 0.0028,
+      "step": 7890
+    },
+    {
+      "epoch": 6.19,
+      "learning_rate": 4.5340642129992174e-05,
+      "loss": 0.0011,
+      "step": 7900
+    },
+    {
+      "epoch": 6.19,
+      "eval_accuracy": 0.899768696993061,
+      "eval_loss": 0.48518332839012146,
+      "eval_runtime": 42.8404,
+      "eval_samples_per_second": 60.55,
+      "eval_steps_per_second": 7.586,
+      "step": 7900
+    },
+    {
+      "epoch": 6.19,
+      "learning_rate": 4.514487079091621e-05,
+      "loss": 0.0011,
+      "step": 7910
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 4.4949099451840256e-05,
+      "loss": 0.0012,
+      "step": 7920
+    },
+    {
+      "epoch": 6.21,
+      "learning_rate": 4.475332811276429e-05,
+      "loss": 0.0008,
+      "step": 7930
+    },
+    {
+      "epoch": 6.22,
+      "learning_rate": 4.455755677368834e-05,
+      "loss": 0.001,
+      "step": 7940
+    },
+    {
+      "epoch": 6.23,
+      "learning_rate": 4.436178543461237e-05,
+      "loss": 0.001,
+      "step": 7950
+    },
+    {
+      "epoch": 6.23,
+      "learning_rate": 4.4166014095536414e-05,
+      "loss": 0.0014,
+      "step": 7960
+    },
+    {
+      "epoch": 6.24,
+      "learning_rate": 4.3970242756460456e-05,
+      "loss": 0.0013,
+      "step": 7970
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 4.37744714173845e-05,
+      "loss": 0.0015,
+      "step": 7980
+    },
+    {
+      "epoch": 6.26,
+      "learning_rate": 4.357870007830854e-05,
+      "loss": 0.0196,
+      "step": 7990
+    },
+    {
+      "epoch": 6.26,
+      "learning_rate": 4.338292873923258e-05,
+      "loss": 0.0093,
+      "step": 8000
+    },
+    {
+      "epoch": 6.26,
+      "eval_accuracy": 0.899768696993061,
+      "eval_loss": 0.47506049275398254,
+      "eval_runtime": 42.6262,
+      "eval_samples_per_second": 60.855,
+      "eval_steps_per_second": 7.624,
+      "step": 8000
+    },
+    {
+      "epoch": 6.27,
+      "learning_rate": 4.318715740015662e-05,
+      "loss": 0.0008,
+      "step": 8010
+    },
+    {
+      "epoch": 6.28,
+      "learning_rate": 4.2991386061080655e-05,
+      "loss": 0.0022,
+      "step": 8020
+    },
+    {
+      "epoch": 6.29,
+      "learning_rate": 4.27956147220047e-05,
+      "loss": 0.001,
+      "step": 8030
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 4.259984338292874e-05,
+      "loss": 0.0014,
+      "step": 8040
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 4.2404072043852786e-05,
+      "loss": 0.0007,
+      "step": 8050
+    },
+    {
+      "epoch": 6.31,
+      "learning_rate": 4.220830070477682e-05,
+      "loss": 0.0073,
+      "step": 8060
+    },
+    {
+      "epoch": 6.32,
+      "learning_rate": 4.201252936570086e-05,
+      "loss": 0.0007,
+      "step": 8070
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 4.18167580266249e-05,
+      "loss": 0.0014,
+      "step": 8080
+    },
+    {
+      "epoch": 6.34,
+      "learning_rate": 4.1620986687548944e-05,
+      "loss": 0.1095,
+      "step": 8090
+    },
+    {
+      "epoch": 6.34,
+      "learning_rate": 4.1425215348472985e-05,
+      "loss": 0.003,
+      "step": 8100
+    },
+    {
+      "epoch": 6.34,
+      "eval_accuracy": 0.8962991518889746,
+      "eval_loss": 0.49340829253196716,
+      "eval_runtime": 43.1125,
+      "eval_samples_per_second": 60.168,
+      "eval_steps_per_second": 7.538,
+      "step": 8100
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 4.1229444009397027e-05,
+      "loss": 0.0012,
+      "step": 8110
+    },
+    {
+      "epoch": 6.36,
+      "learning_rate": 4.103367267032107e-05,
+      "loss": 0.0009,
+      "step": 8120
+    },
+    {
+      "epoch": 6.37,
+      "learning_rate": 4.083790133124511e-05,
+      "loss": 0.001,
+      "step": 8130
+    },
+    {
+      "epoch": 6.37,
+      "learning_rate": 4.064212999216915e-05,
+      "loss": 0.0019,
+      "step": 8140
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 4.044635865309319e-05,
+      "loss": 0.0021,
+      "step": 8150
+    },
+    {
+      "epoch": 6.39,
+      "learning_rate": 4.025058731401723e-05,
+      "loss": 0.0009,
+      "step": 8160
+    },
+    {
+      "epoch": 6.4,
+      "learning_rate": 4.0054815974941274e-05,
+      "loss": 0.0007,
+      "step": 8170
+    },
+    {
+      "epoch": 6.41,
+      "learning_rate": 3.985904463586531e-05,
+      "loss": 0.001,
+      "step": 8180
+    },
+    {
+      "epoch": 6.41,
+      "learning_rate": 3.966327329678935e-05,
+      "loss": 0.0007,
+      "step": 8190
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 3.946750195771339e-05,
+      "loss": 0.0027,
+      "step": 8200
+    },
+    {
+      "epoch": 6.42,
+      "eval_accuracy": 0.9028527370855821,
+      "eval_loss": 0.4882272183895111,
+      "eval_runtime": 42.9231,
+      "eval_samples_per_second": 60.434,
+      "eval_steps_per_second": 7.572,
+      "step": 8200
+    },
+    {
+      "epoch": 6.43,
+      "learning_rate": 3.927173061863743e-05,
+      "loss": 0.0011,
+      "step": 8210
+    },
+    {
+      "epoch": 6.44,
+      "learning_rate": 3.9075959279561474e-05,
+      "loss": 0.0008,
+      "step": 8220
+    },
+    {
+      "epoch": 6.44,
+      "learning_rate": 3.8880187940485515e-05,
+      "loss": 0.0007,
+      "step": 8230
+    },
+    {
+      "epoch": 6.45,
+      "learning_rate": 3.8684416601409556e-05,
+      "loss": 0.0021,
+      "step": 8240
+    },
+    {
+      "epoch": 6.46,
+      "learning_rate": 3.84886452623336e-05,
+      "loss": 0.0471,
+      "step": 8250
+    },
+    {
+      "epoch": 6.47,
+      "learning_rate": 3.829287392325764e-05,
+      "loss": 0.062,
+      "step": 8260
+    },
+    {
+      "epoch": 6.48,
+      "learning_rate": 3.809710258418167e-05,
+      "loss": 0.0023,
+      "step": 8270
+    },
+    {
+      "epoch": 6.48,
+      "learning_rate": 3.790133124510572e-05,
+      "loss": 0.0181,
+      "step": 8280
+    },
+    {
+      "epoch": 6.49,
+      "learning_rate": 3.7705559906029756e-05,
+      "loss": 0.0007,
+      "step": 8290
+    },
+    {
+      "epoch": 6.5,
+      "learning_rate": 3.7509788566953804e-05,
+      "loss": 0.0009,
+      "step": 8300
+    },
+    {
+      "epoch": 6.5,
+      "eval_accuracy": 0.9020817270624518,
+      "eval_loss": 0.4806201756000519,
+      "eval_runtime": 42.7792,
+      "eval_samples_per_second": 60.637,
+      "eval_steps_per_second": 7.597,
+      "step": 8300
+    },
+    {
+      "epoch": 6.51,
+      "learning_rate": 3.731401722787784e-05,
+      "loss": 0.001,
+      "step": 8310
+    },
+    {
+      "epoch": 6.52,
+      "learning_rate": 3.7118245888801886e-05,
+      "loss": 0.001,
+      "step": 8320
+    },
+    {
+      "epoch": 6.52,
+      "learning_rate": 3.692247454972592e-05,
+      "loss": 0.0648,
+      "step": 8330
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 3.672670321064996e-05,
+      "loss": 0.0007,
+      "step": 8340
+    },
+    {
+      "epoch": 6.54,
+      "learning_rate": 3.6530931871574e-05,
+      "loss": 0.012,
+      "step": 8350
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 3.6335160532498045e-05,
+      "loss": 0.0011,
+      "step": 8360
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 3.6139389193422086e-05,
+      "loss": 0.001,
+      "step": 8370
+    },
+    {
+      "epoch": 6.56,
+      "learning_rate": 3.594361785434612e-05,
+      "loss": 0.0009,
+      "step": 8380
+    },
+    {
+      "epoch": 6.57,
+      "learning_rate": 3.574784651527017e-05,
+      "loss": 0.0009,
+      "step": 8390
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 3.55520751761942e-05,
+      "loss": 0.0009,
+      "step": 8400
+    },
+    {
+      "epoch": 6.58,
+      "eval_accuracy": 0.9028527370855821,
+      "eval_loss": 0.4974198639392853,
+      "eval_runtime": 42.7479,
+      "eval_samples_per_second": 60.681,
+      "eval_steps_per_second": 7.603,
+      "step": 8400
+    },
+    {
+      "epoch": 6.59,
+      "learning_rate": 3.535630383711825e-05,
+      "loss": 0.001,
+      "step": 8410
+    },
+    {
+      "epoch": 6.59,
+      "learning_rate": 3.5160532498042285e-05,
+      "loss": 0.001,
+      "step": 8420
+    },
+    {
+      "epoch": 6.6,
+      "learning_rate": 3.4964761158966333e-05,
+      "loss": 0.0009,
+      "step": 8430
+    },
+    {
+      "epoch": 6.61,
+      "learning_rate": 3.476898981989037e-05,
+      "loss": 0.002,
+      "step": 8440
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 3.457321848081441e-05,
+      "loss": 0.0007,
+      "step": 8450
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 3.437744714173845e-05,
+      "loss": 0.0115,
+      "step": 8460
+    },
+    {
+      "epoch": 6.63,
+      "learning_rate": 3.418167580266249e-05,
+      "loss": 0.0016,
+      "step": 8470
+    },
+    {
+      "epoch": 6.64,
+      "learning_rate": 3.398590446358653e-05,
+      "loss": 0.0007,
+      "step": 8480
+    },
+    {
+      "epoch": 6.65,
+      "learning_rate": 3.3790133124510574e-05,
+      "loss": 0.0006,
+      "step": 8490
+    },
+    {
+      "epoch": 6.66,
+      "learning_rate": 3.3594361785434616e-05,
+      "loss": 0.0009,
+      "step": 8500
+    },
+    {
+      "epoch": 6.66,
+      "eval_accuracy": 0.9074787972243639,
+      "eval_loss": 0.4748244285583496,
+      "eval_runtime": 42.4884,
+      "eval_samples_per_second": 61.052,
+      "eval_steps_per_second": 7.649,
+      "step": 8500
+    },
+    {
+      "epoch": 6.66,
+      "learning_rate": 3.339859044635866e-05,
+      "loss": 0.0008,
+      "step": 8510
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 3.32028191072827e-05,
+      "loss": 0.0012,
+      "step": 8520
+    },
+    {
+      "epoch": 6.68,
+      "learning_rate": 3.300704776820673e-05,
+      "loss": 0.0012,
+      "step": 8530
+    },
+    {
+      "epoch": 6.69,
+      "learning_rate": 3.281127642913078e-05,
+      "loss": 0.0012,
+      "step": 8540
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 3.2615505090054815e-05,
+      "loss": 0.0008,
+      "step": 8550
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 3.2419733750978856e-05,
+      "loss": 0.0008,
+      "step": 8560
+    },
+    {
+      "epoch": 6.71,
+      "learning_rate": 3.22239624119029e-05,
+      "loss": 0.0011,
+      "step": 8570
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 3.202819107282694e-05,
+      "loss": 0.0139,
+      "step": 8580
+    },
+    {
+      "epoch": 6.73,
+      "learning_rate": 3.183241973375098e-05,
+      "loss": 0.0007,
+      "step": 8590
+    },
+    {
+      "epoch": 6.73,
+      "learning_rate": 3.163664839467502e-05,
+      "loss": 0.0008,
+      "step": 8600
+    },
+    {
+      "epoch": 6.73,
+      "eval_accuracy": 0.9094063222821896,
+      "eval_loss": 0.47228720784187317,
+      "eval_runtime": 42.8729,
+      "eval_samples_per_second": 60.504,
+      "eval_steps_per_second": 7.581,
+      "step": 8600
+    },
+    {
+      "epoch": 6.74,
+      "learning_rate": 3.144087705559906e-05,
+      "loss": 0.0008,
+      "step": 8610
+    },
+    {
+      "epoch": 6.75,
+      "learning_rate": 3.1245105716523104e-05,
+      "loss": 0.0005,
+      "step": 8620
+    },
+    {
+      "epoch": 6.76,
+      "learning_rate": 3.1049334377447145e-05,
+      "loss": 0.0007,
+      "step": 8630
+    },
+    {
+      "epoch": 6.77,
+      "learning_rate": 3.0853563038371186e-05,
+      "loss": 0.0014,
+      "step": 8640
+    },
+    {
+      "epoch": 6.77,
+      "learning_rate": 3.065779169929522e-05,
+      "loss": 0.0006,
+      "step": 8650
+    },
+    {
+      "epoch": 6.78,
+      "learning_rate": 3.046202036021927e-05,
+      "loss": 0.0005,
+      "step": 8660
+    },
+    {
+      "epoch": 6.79,
+      "learning_rate": 3.0266249021143307e-05,
+      "loss": 0.0007,
+      "step": 8670
+    },
+    {
+      "epoch": 6.8,
+      "learning_rate": 3.0070477682067348e-05,
+      "loss": 0.0007,
+      "step": 8680
+    },
+    {
+      "epoch": 6.81,
+      "learning_rate": 2.9874706342991386e-05,
+      "loss": 0.0012,
+      "step": 8690
+    },
+    {
+      "epoch": 6.81,
+      "learning_rate": 2.9678935003915427e-05,
+      "loss": 0.001,
+      "step": 8700
+    },
+    {
+      "epoch": 6.81,
+      "eval_accuracy": 0.9097918272937549,
+      "eval_loss": 0.4691937565803528,
+      "eval_runtime": 42.807,
+      "eval_samples_per_second": 60.598,
+      "eval_steps_per_second": 7.592,
+      "step": 8700
+    },
+    {
+      "epoch": 6.82,
+      "learning_rate": 2.948316366483947e-05,
+      "loss": 0.0021,
+      "step": 8710
+    },
+    {
+      "epoch": 6.83,
+      "learning_rate": 2.9287392325763506e-05,
+      "loss": 0.0008,
+      "step": 8720
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 2.909162098668755e-05,
+      "loss": 0.0006,
+      "step": 8730
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 2.889584964761159e-05,
+      "loss": 0.0006,
+      "step": 8740
+    },
+    {
+      "epoch": 6.85,
+      "learning_rate": 2.8700078308535634e-05,
+      "loss": 0.0312,
+      "step": 8750
+    },
+    {
+      "epoch": 6.86,
+      "learning_rate": 2.850430696945967e-05,
+      "loss": 0.0009,
+      "step": 8760
+    },
+    {
+      "epoch": 6.87,
+      "learning_rate": 2.8308535630383716e-05,
+      "loss": 0.0006,
+      "step": 8770
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 2.8112764291307754e-05,
+      "loss": 0.001,
+      "step": 8780
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 2.7916992952231795e-05,
+      "loss": 0.0077,
+      "step": 8790
+    },
+    {
+      "epoch": 6.89,
+      "learning_rate": 2.7721221613155833e-05,
+      "loss": 0.0007,
+      "step": 8800
+    },
+    {
+      "epoch": 6.89,
+      "eval_accuracy": 0.9074787972243639,
+      "eval_loss": 0.4725954532623291,
+      "eval_runtime": 42.9848,
+      "eval_samples_per_second": 60.347,
+      "eval_steps_per_second": 7.561,
+      "step": 8800
+    },
+    {
+      "epoch": 6.9,
+      "learning_rate": 2.7525450274079878e-05,
+      "loss": 0.001,
+      "step": 8810
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 2.7329678935003916e-05,
+      "loss": 0.0006,
+      "step": 8820
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 2.713390759592796e-05,
+      "loss": 0.0007,
+      "step": 8830
+    },
+    {
+      "epoch": 6.92,
+      "learning_rate": 2.6938136256851998e-05,
+      "loss": 0.0496,
+      "step": 8840
+    },
+    {
+      "epoch": 6.93,
+      "learning_rate": 2.6742364917776043e-05,
+      "loss": 0.0008,
+      "step": 8850
+    },
+    {
+      "epoch": 6.94,
+      "learning_rate": 2.654659357870008e-05,
+      "loss": 0.0026,
+      "step": 8860
+    },
+    {
+      "epoch": 6.95,
+      "learning_rate": 2.635082223962412e-05,
+      "loss": 0.0007,
+      "step": 8870
+    },
+    {
+      "epoch": 6.95,
+      "learning_rate": 2.615505090054816e-05,
+      "loss": 0.0017,
+      "step": 8880
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 2.59592795614722e-05,
+      "loss": 0.0007,
+      "step": 8890
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 2.5763508222396242e-05,
+      "loss": 0.0011,
+      "step": 8900
+    },
+    {
+      "epoch": 6.97,
+      "eval_accuracy": 0.9067077872012336,
+      "eval_loss": 0.4685651659965515,
+      "eval_runtime": 43.4324,
+      "eval_samples_per_second": 59.725,
+      "eval_steps_per_second": 7.483,
+      "step": 8900
+    },
+    {
+      "epoch": 6.98,
+      "learning_rate": 2.556773688332028e-05,
+      "loss": 0.0013,
+      "step": 8910
+    },
+    {
+      "epoch": 6.99,
+      "learning_rate": 2.5371965544244325e-05,
+      "loss": 0.0022,
+      "step": 8920
+    },
+    {
+      "epoch": 6.99,
+      "learning_rate": 2.5176194205168363e-05,
+      "loss": 0.0005,
+      "step": 8930
+    },
+    {
+      "epoch": 7.0,
+      "learning_rate": 2.4980422866092404e-05,
+      "loss": 0.0006,
+      "step": 8940
+    },
+    {
+      "epoch": 7.01,
+      "learning_rate": 2.4784651527016445e-05,
+      "loss": 0.0008,
+      "step": 8950
+    },
+    {
+      "epoch": 7.02,
+      "learning_rate": 2.4588880187940486e-05,
+      "loss": 0.0008,
+      "step": 8960
+    },
+    {
+      "epoch": 7.02,
+      "learning_rate": 2.4393108848864528e-05,
+      "loss": 0.0006,
+      "step": 8970
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 2.419733750978857e-05,
+      "loss": 0.0008,
+      "step": 8980
+    },
+    {
+      "epoch": 7.04,
+      "learning_rate": 2.4001566170712607e-05,
+      "loss": 0.0008,
+      "step": 8990
+    },
+    {
+      "epoch": 7.05,
+      "learning_rate": 2.3805794831636648e-05,
+      "loss": 0.0006,
+      "step": 9000
+    },
+    {
+      "epoch": 7.05,
+      "eval_accuracy": 0.9055512721665382,
+      "eval_loss": 0.46525028347969055,
+      "eval_runtime": 43.4306,
+      "eval_samples_per_second": 59.728,
+      "eval_steps_per_second": 7.483,
+      "step": 9000
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 2.361002349256069e-05,
+      "loss": 0.0009,
+      "step": 9010
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 2.341425215348473e-05,
+      "loss": 0.0006,
+      "step": 9020
+    },
+    {
+      "epoch": 7.07,
+      "learning_rate": 2.3218480814408772e-05,
+      "loss": 0.0058,
+      "step": 9030
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 2.3022709475332813e-05,
+      "loss": 0.0006,
+      "step": 9040
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 2.2826938136256854e-05,
+      "loss": 0.0007,
+      "step": 9050
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 2.2631166797180896e-05,
+      "loss": 0.0007,
+      "step": 9060
+    },
+    {
+      "epoch": 7.1,
+      "learning_rate": 2.2435395458104934e-05,
+      "loss": 0.0006,
+      "step": 9070
+    },
+    {
+      "epoch": 7.11,
+      "learning_rate": 2.2239624119028975e-05,
+      "loss": 0.0005,
+      "step": 9080
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 2.2043852779953016e-05,
+      "loss": 0.0007,
+      "step": 9090
+    },
+    {
+      "epoch": 7.13,
+      "learning_rate": 2.1848081440877057e-05,
+      "loss": 0.0006,
+      "step": 9100
+    },
+    {
+      "epoch": 7.13,
+      "eval_accuracy": 0.9028527370855821,
+      "eval_loss": 0.47551360726356506,
+      "eval_runtime": 43.1363,
+      "eval_samples_per_second": 60.135,
+      "eval_steps_per_second": 7.534,
+      "step": 9100
+    },
+    {
+      "epoch": 7.13,
+      "learning_rate": 2.1652310101801095e-05,
+      "loss": 0.0007,
+      "step": 9110
+    },
+    {
+      "epoch": 7.14,
+      "learning_rate": 2.1456538762725137e-05,
+      "loss": 0.0151,
+      "step": 9120
+    },
+    {
+      "epoch": 7.15,
+      "learning_rate": 2.1260767423649178e-05,
+      "loss": 0.0005,
+      "step": 9130
+    },
+    {
+      "epoch": 7.16,
+      "learning_rate": 2.106499608457322e-05,
+      "loss": 0.0007,
+      "step": 9140
+    },
+    {
+      "epoch": 7.17,
+      "learning_rate": 2.086922474549726e-05,
+      "loss": 0.0006,
+      "step": 9150
+    },
+    {
+      "epoch": 7.17,
+      "learning_rate": 2.06734534064213e-05,
+      "loss": 0.0006,
+      "step": 9160
+    },
+    {
+      "epoch": 7.18,
+      "learning_rate": 2.0477682067345343e-05,
+      "loss": 0.0005,
+      "step": 9170
+    },
+    {
+      "epoch": 7.19,
+      "learning_rate": 2.028191072826938e-05,
+      "loss": 0.0005,
+      "step": 9180
+    },
+    {
+      "epoch": 7.2,
+      "learning_rate": 2.0086139389193422e-05,
+      "loss": 0.0006,
+      "step": 9190
+    },
+    {
+      "epoch": 7.2,
+      "learning_rate": 1.9890368050117463e-05,
+      "loss": 0.0007,
+      "step": 9200
+    },
+    {
+      "epoch": 7.2,
+      "eval_accuracy": 0.9036237471087124,
+      "eval_loss": 0.4632953405380249,
+      "eval_runtime": 42.785,
+      "eval_samples_per_second": 60.629,
+      "eval_steps_per_second": 7.596,
+      "step": 9200
+    },
+    {
+      "epoch": 7.21,
+      "learning_rate": 1.9694596711041505e-05,
+      "loss": 0.0006,
+      "step": 9210
+    },
+    {
+      "epoch": 7.22,
+      "learning_rate": 1.9498825371965546e-05,
+      "loss": 0.0005,
+      "step": 9220
+    },
+    {
+      "epoch": 7.23,
+      "learning_rate": 1.9303054032889587e-05,
+      "loss": 0.0006,
+      "step": 9230
+    },
+    {
+      "epoch": 7.24,
+      "learning_rate": 1.9107282693813628e-05,
+      "loss": 0.0005,
+      "step": 9240
+    },
+    {
+      "epoch": 7.24,
+      "learning_rate": 1.891151135473767e-05,
+      "loss": 0.0005,
+      "step": 9250
+    },
+    {
+      "epoch": 7.25,
+      "learning_rate": 1.8715740015661707e-05,
+      "loss": 0.0005,
+      "step": 9260
+    },
+    {
+      "epoch": 7.26,
+      "learning_rate": 1.851996867658575e-05,
+      "loss": 0.0005,
+      "step": 9270
+    },
+    {
+      "epoch": 7.27,
+      "learning_rate": 1.8324197337509787e-05,
+      "loss": 0.0007,
+      "step": 9280
+    },
+    {
+      "epoch": 7.27,
+      "learning_rate": 1.8128425998433828e-05,
+      "loss": 0.001,
+      "step": 9290
+    },
+    {
+      "epoch": 7.28,
+      "learning_rate": 1.793265465935787e-05,
+      "loss": 0.0067,
+      "step": 9300
+    },
+    {
+      "epoch": 7.28,
+      "eval_accuracy": 0.9036237471087124,
+      "eval_loss": 0.46110159158706665,
+      "eval_runtime": 43.0582,
+      "eval_samples_per_second": 60.244,
+      "eval_steps_per_second": 7.548,
+      "step": 9300
+    },
+    {
+      "epoch": 7.29,
+      "learning_rate": 1.773688332028191e-05,
+      "loss": 0.0005,
+      "step": 9310
+    },
+    {
+      "epoch": 7.3,
+      "learning_rate": 1.754111198120595e-05,
+      "loss": 0.0006,
+      "step": 9320
+    },
+    {
+      "epoch": 7.31,
+      "learning_rate": 1.7345340642129993e-05,
+      "loss": 0.0009,
+      "step": 9330
+    },
+    {
+      "epoch": 7.31,
+      "learning_rate": 1.7149569303054034e-05,
+      "loss": 0.0007,
+      "step": 9340
+    },
+    {
+      "epoch": 7.32,
+      "learning_rate": 1.6953797963978075e-05,
+      "loss": 0.0181,
+      "step": 9350
+    },
+    {
+      "epoch": 7.33,
+      "learning_rate": 1.6758026624902117e-05,
+      "loss": 0.0005,
+      "step": 9360
+    },
+    {
+      "epoch": 7.34,
+      "learning_rate": 1.6562255285826155e-05,
+      "loss": 0.0005,
+      "step": 9370
+    },
+    {
+      "epoch": 7.35,
+      "learning_rate": 1.6366483946750196e-05,
+      "loss": 0.0006,
+      "step": 9380
+    },
+    {
+      "epoch": 7.35,
+      "learning_rate": 1.6170712607674237e-05,
+      "loss": 0.0008,
+      "step": 9390
+    },
+    {
+      "epoch": 7.36,
+      "learning_rate": 1.597494126859828e-05,
+      "loss": 0.0007,
+      "step": 9400
+    },
+    {
+      "epoch": 7.36,
+      "eval_accuracy": 0.905165767154973,
+      "eval_loss": 0.46077683568000793,
+      "eval_runtime": 43.0899,
+      "eval_samples_per_second": 60.2,
+      "eval_steps_per_second": 7.542,
+      "step": 9400
+    },
+    {
+      "epoch": 7.37,
+      "learning_rate": 1.577916992952232e-05,
+      "loss": 0.0009,
+      "step": 9410
+    },
+    {
+      "epoch": 7.38,
+      "learning_rate": 1.558339859044636e-05,
+      "loss": 0.0005,
+      "step": 9420
+    },
+    {
+      "epoch": 7.38,
+      "learning_rate": 1.5387627251370402e-05,
+      "loss": 0.0006,
+      "step": 9430
+    },
+    {
+      "epoch": 7.39,
+      "learning_rate": 1.5191855912294442e-05,
+      "loss": 0.0005,
+      "step": 9440
+    },
+    {
+      "epoch": 7.4,
+      "learning_rate": 1.499608457321848e-05,
+      "loss": 0.0006,
+      "step": 9450
+    },
+    {
+      "epoch": 7.41,
+      "learning_rate": 1.480031323414252e-05,
+      "loss": 0.0004,
+      "step": 9460
+    },
+    {
+      "epoch": 7.42,
+      "learning_rate": 1.4604541895066562e-05,
+      "loss": 0.0006,
+      "step": 9470
+    },
+    {
+      "epoch": 7.42,
+      "learning_rate": 1.4408770555990603e-05,
+      "loss": 0.0007,
+      "step": 9480
+    },
+    {
+      "epoch": 7.43,
+      "learning_rate": 1.4212999216914643e-05,
+      "loss": 0.0005,
+      "step": 9490
+    },
+    {
+      "epoch": 7.44,
+      "learning_rate": 1.4017227877838684e-05,
+      "loss": 0.0007,
+      "step": 9500
+    },
+    {
+      "epoch": 7.44,
+      "eval_accuracy": 0.9043947571318427,
+      "eval_loss": 0.4622529447078705,
+      "eval_runtime": 43.1294,
+      "eval_samples_per_second": 60.145,
+      "eval_steps_per_second": 7.535,
+      "step": 9500
+    },
+    {
+      "epoch": 7.45,
+      "learning_rate": 1.3821456538762725e-05,
+      "loss": 0.0005,
+      "step": 9510
+    },
+    {
+      "epoch": 7.45,
+      "learning_rate": 1.3625685199686767e-05,
+      "loss": 0.0004,
+      "step": 9520
+    },
+    {
+      "epoch": 7.46,
+      "learning_rate": 1.3429913860610806e-05,
+      "loss": 0.0007,
+      "step": 9530
+    },
+    {
+      "epoch": 7.47,
+      "learning_rate": 1.3234142521534848e-05,
+      "loss": 0.0006,
+      "step": 9540
+    },
+    {
+      "epoch": 7.48,
+      "learning_rate": 1.3038371182458889e-05,
+      "loss": 0.0007,
+      "step": 9550
+    },
+    {
+      "epoch": 7.49,
+      "learning_rate": 1.284259984338293e-05,
+      "loss": 0.0005,
+      "step": 9560
+    },
+    {
+      "epoch": 7.49,
+      "learning_rate": 1.2646828504306971e-05,
+      "loss": 0.0005,
+      "step": 9570
+    },
+    {
+      "epoch": 7.5,
+      "learning_rate": 1.2451057165231011e-05,
+      "loss": 0.0005,
+      "step": 9580
+    },
+    {
+      "epoch": 7.51,
+      "learning_rate": 1.225528582615505e-05,
+      "loss": 0.0006,
+      "step": 9590
+    },
+    {
+      "epoch": 7.52,
+      "learning_rate": 1.2059514487079092e-05,
+      "loss": 0.0005,
+      "step": 9600
+    },
+    {
+      "epoch": 7.52,
+      "eval_accuracy": 0.9055512721665382,
+      "eval_loss": 0.46212539076805115,
+      "eval_runtime": 43.0566,
+      "eval_samples_per_second": 60.246,
+      "eval_steps_per_second": 7.548,
+      "step": 9600
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 1.1863743148003133e-05,
+      "loss": 0.0006,
+      "step": 9610
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 1.1667971808927174e-05,
+      "loss": 0.0005,
+      "step": 9620
+    },
+    {
+      "epoch": 7.54,
+      "learning_rate": 1.1472200469851214e-05,
+      "loss": 0.0021,
+      "step": 9630
+    },
+    {
+      "epoch": 7.55,
+      "learning_rate": 1.1276429130775255e-05,
+      "loss": 0.0004,
+      "step": 9640
+    },
+    {
+      "epoch": 7.56,
+      "learning_rate": 1.1080657791699296e-05,
+      "loss": 0.0005,
+      "step": 9650
+    },
+    {
+      "epoch": 7.56,
+      "learning_rate": 1.0884886452623338e-05,
+      "loss": 0.0006,
+      "step": 9660
+    },
+    {
+      "epoch": 7.57,
+      "learning_rate": 1.0689115113547377e-05,
+      "loss": 0.0008,
+      "step": 9670
+    },
+    {
+      "epoch": 7.58,
+      "learning_rate": 1.0493343774471417e-05,
+      "loss": 0.0096,
+      "step": 9680
+    },
+    {
+      "epoch": 7.59,
+      "learning_rate": 1.0297572435395458e-05,
+      "loss": 0.0006,
+      "step": 9690
+    },
+    {
+      "epoch": 7.6,
+      "learning_rate": 1.01018010963195e-05,
+      "loss": 0.0005,
+      "step": 9700
+    },
+    {
+      "epoch": 7.6,
+      "eval_accuracy": 0.9055512721665382,
+      "eval_loss": 0.46149030327796936,
+      "eval_runtime": 42.8413,
+      "eval_samples_per_second": 60.549,
+      "eval_steps_per_second": 7.586,
+      "step": 9700
+    },
+    {
+      "epoch": 7.6,
+      "learning_rate": 9.90602975724354e-06,
+      "loss": 0.0004,
+      "step": 9710
+    },
+    {
+      "epoch": 7.61,
+      "learning_rate": 9.710258418167582e-06,
+      "loss": 0.0012,
+      "step": 9720
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 9.514487079091621e-06,
+      "loss": 0.0006,
+      "step": 9730
+    },
+    {
+      "epoch": 7.63,
+      "learning_rate": 9.318715740015663e-06,
+      "loss": 0.0005,
+      "step": 9740
+    },
+    {
+      "epoch": 7.64,
+      "learning_rate": 9.122944400939702e-06,
+      "loss": 0.0005,
+      "step": 9750
+    },
+    {
+      "epoch": 7.64,
+      "learning_rate": 8.927173061863743e-06,
+      "loss": 0.0005,
+      "step": 9760
+    },
+    {
+      "epoch": 7.65,
+      "learning_rate": 8.731401722787785e-06,
+      "loss": 0.0005,
+      "step": 9770
+    },
+    {
+      "epoch": 7.66,
+      "learning_rate": 8.535630383711824e-06,
+      "loss": 0.0007,
+      "step": 9780
+    },
+    {
+      "epoch": 7.67,
+      "learning_rate": 8.339859044635866e-06,
+      "loss": 0.0005,
+      "step": 9790
+    },
+    {
+      "epoch": 7.67,
+      "learning_rate": 8.144087705559907e-06,
+      "loss": 0.0005,
+      "step": 9800
+    },
+    {
+      "epoch": 7.67,
+      "eval_accuracy": 0.9059367771781033,
+      "eval_loss": 0.4611620604991913,
+      "eval_runtime": 43.2785,
+      "eval_samples_per_second": 59.937,
+      "eval_steps_per_second": 7.51,
+      "step": 9800
+    },
+    {
+      "epoch": 7.68,
+      "learning_rate": 7.948316366483948e-06,
+      "loss": 0.0005,
+      "step": 9810
+    },
+    {
+      "epoch": 7.69,
+      "learning_rate": 7.752545027407988e-06,
+      "loss": 0.0008,
+      "step": 9820
+    },
+    {
+      "epoch": 7.7,
+      "learning_rate": 7.556773688332029e-06,
+      "loss": 0.0005,
+      "step": 9830
+    },
+    {
+      "epoch": 7.71,
+      "learning_rate": 7.3610023492560685e-06,
+      "loss": 0.0004,
+      "step": 9840
+    },
+    {
+      "epoch": 7.71,
+      "learning_rate": 7.16523101018011e-06,
+      "loss": 0.0005,
+      "step": 9850
+    },
+    {
+      "epoch": 7.72,
+      "learning_rate": 6.96945967110415e-06,
+      "loss": 0.0005,
+      "step": 9860
+    },
+    {
+      "epoch": 7.73,
+      "learning_rate": 6.7736883320281914e-06,
+      "loss": 0.0005,
+      "step": 9870
+    },
+    {
+      "epoch": 7.74,
+      "learning_rate": 6.577916992952232e-06,
+      "loss": 0.0048,
+      "step": 9880
+    },
+    {
+      "epoch": 7.74,
+      "learning_rate": 6.382145653876273e-06,
+      "loss": 0.0006,
+      "step": 9890
+    },
+    {
+      "epoch": 7.75,
+      "learning_rate": 6.1863743148003135e-06,
+      "loss": 0.0005,
+      "step": 9900
+    },
+    {
+      "epoch": 7.75,
+      "eval_accuracy": 0.9074787972243639,
+      "eval_loss": 0.46262314915657043,
+      "eval_runtime": 43.0889,
+      "eval_samples_per_second": 60.201,
+      "eval_steps_per_second": 7.543,
+      "step": 9900
+    },
+    {
+      "epoch": 7.76,
+      "learning_rate": 5.990602975724354e-06,
+      "loss": 0.0006,
+      "step": 9910
+    },
+    {
+      "epoch": 7.77,
+      "learning_rate": 5.794831636648395e-06,
+      "loss": 0.0005,
+      "step": 9920
+    },
+    {
+      "epoch": 7.78,
+      "learning_rate": 5.599060297572436e-06,
+      "loss": 0.0005,
+      "step": 9930
+    },
+    {
+      "epoch": 7.78,
+      "learning_rate": 5.403288958496476e-06,
+      "loss": 0.0005,
+      "step": 9940
+    },
+    {
+      "epoch": 7.79,
+      "learning_rate": 5.207517619420517e-06,
+      "loss": 0.0004,
+      "step": 9950
+    },
+    {
+      "epoch": 7.8,
+      "learning_rate": 5.011746280344558e-06,
+      "loss": 0.0005,
+      "step": 9960
+    },
+    {
+      "epoch": 7.81,
+      "learning_rate": 4.815974941268599e-06,
+      "loss": 0.0005,
+      "step": 9970
+    },
+    {
+      "epoch": 7.82,
+      "learning_rate": 4.620203602192639e-06,
+      "loss": 0.0011,
+      "step": 9980
+    },
+    {
+      "epoch": 7.82,
+      "learning_rate": 4.42443226311668e-06,
+      "loss": 0.001,
+      "step": 9990
+    },
+    {
+      "epoch": 7.83,
+      "learning_rate": 4.22866092404072e-06,
+      "loss": 0.0004,
+      "step": 10000
+    },
+    {
+      "epoch": 7.83,
+      "eval_accuracy": 0.9074787972243639,
+      "eval_loss": 0.4625774025917053,
+      "eval_runtime": 42.7953,
+      "eval_samples_per_second": 60.614,
+      "eval_steps_per_second": 7.594,
+      "step": 10000
+    },
+    {
+      "epoch": 7.84,
+      "learning_rate": 4.0328895849647615e-06,
+      "loss": 0.0005,
+      "step": 10010
+    },
+    {
+      "epoch": 7.85,
+      "learning_rate": 3.837118245888802e-06,
+      "loss": 0.0007,
+      "step": 10020
+    },
+    {
+      "epoch": 7.85,
+      "learning_rate": 3.6413469068128423e-06,
+      "loss": 0.0005,
+      "step": 10030
+    },
+    {
+      "epoch": 7.86,
+      "learning_rate": 3.4455755677368836e-06,
+      "loss": 0.0005,
+      "step": 10040
+    },
+    {
+      "epoch": 7.87,
+      "learning_rate": 3.2498042286609244e-06,
+      "loss": 0.0006,
+      "step": 10050
+    },
+    {
+      "epoch": 7.88,
+      "learning_rate": 3.054032889584965e-06,
+      "loss": 0.0005,
+      "step": 10060
+    },
+    {
+      "epoch": 7.89,
+      "learning_rate": 2.8582615505090057e-06,
+      "loss": 0.0006,
+      "step": 10070
+    },
+    {
+      "epoch": 7.89,
+      "learning_rate": 2.662490211433046e-06,
+      "loss": 0.0007,
+      "step": 10080
+    },
+    {
+      "epoch": 7.9,
+      "learning_rate": 2.466718872357087e-06,
+      "loss": 0.0005,
+      "step": 10090
+    },
+    {
+      "epoch": 7.91,
+      "learning_rate": 2.2709475332811278e-06,
+      "loss": 0.0005,
+      "step": 10100
+    },
+    {
+      "epoch": 7.91,
+      "eval_accuracy": 0.9074787972243639,
+      "eval_loss": 0.4625736474990845,
+      "eval_runtime": 42.5583,
+      "eval_samples_per_second": 60.952,
+      "eval_steps_per_second": 7.637,
+      "step": 10100
+    },
+    {
+      "epoch": 7.92,
+      "learning_rate": 2.0751761942051686e-06,
+      "loss": 0.0004,
+      "step": 10110
+    },
+    {
+      "epoch": 7.92,
+      "learning_rate": 1.8794048551292093e-06,
+      "loss": 0.0007,
+      "step": 10120
+    },
+    {
+      "epoch": 7.93,
+      "learning_rate": 1.6836335160532499e-06,
+      "loss": 0.0006,
+      "step": 10130
+    },
+    {
+      "epoch": 7.94,
+      "learning_rate": 1.4878621769772905e-06,
+      "loss": 0.0005,
+      "step": 10140
+    },
+    {
+      "epoch": 7.95,
+      "learning_rate": 1.2920908379013313e-06,
+      "loss": 0.0005,
+      "step": 10150
+    },
+    {
+      "epoch": 7.96,
+      "learning_rate": 1.096319498825372e-06,
+      "loss": 0.0005,
+      "step": 10160
+    },
+    {
+      "epoch": 7.96,
+      "learning_rate": 9.005481597494127e-07,
+      "loss": 0.0005,
+      "step": 10170
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 7.047768206734534e-07,
+      "loss": 0.0007,
+      "step": 10180
+    },
+    {
+      "epoch": 7.98,
+      "learning_rate": 5.090054815974942e-07,
+      "loss": 0.0005,
+      "step": 10190
+    },
+    {
+      "epoch": 7.99,
+      "learning_rate": 3.1323414252153486e-07,
+      "loss": 0.0006,
+      "step": 10200
+    },
+    {
+      "epoch": 7.99,
+      "eval_accuracy": 0.9078643022359291,
+      "eval_loss": 0.46260592341423035,
+      "eval_runtime": 43.1376,
+      "eval_samples_per_second": 60.133,
+      "eval_steps_per_second": 7.534,
+      "step": 10200
+    },
+    {
+      "epoch": 8.0,
+      "learning_rate": 1.1746280344557558e-07,
+      "loss": 0.0454,
+      "step": 10210
+    },
+    {
+      "epoch": 8.0,
+      "step": 10216,
+      "total_flos": 6.337884979995771e+18,
+      "train_loss": 0.48898702390789517,
+      "train_runtime": 10019.7896,
+      "train_samples_per_second": 8.152,
+      "train_steps_per_second": 1.02
+    }
+  ],
+  "max_steps": 10216,
+  "num_train_epochs": 8,
+  "total_flos": 6.337884979995771e+18,
+  "trial_name": null,
+  "trial_params": null
+}