diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4561 @@
+{
+  "best_metric": 0.06347225606441498,
+  "best_model_checkpoint": "/kaggle/working/Model/google-vit-base-patch16-384-batch_16_epoch_4_classes_24/checkpoint-3600",
+  "epoch": 4.0,
+  "eval_steps": 100,
+  "global_step": 5744,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.608365058898926,
+      "learning_rate": 0.00019965181058495822,
+      "loss": 2.6683,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.99545955657959,
+      "learning_rate": 0.00019930362116991646,
+      "loss": 1.3199,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.903165817260742,
+      "learning_rate": 0.00019895543175487465,
+      "loss": 0.6993,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.235982418060303,
+      "learning_rate": 0.00019860724233983288,
+      "loss": 0.4142,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.177570819854736,
+      "learning_rate": 0.0001982590529247911,
+      "loss": 0.5028,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.7044296264648438,
+      "learning_rate": 0.00019791086350974934,
+      "loss": 0.3604,
+      "step": 60
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.362403392791748,
+      "learning_rate": 0.00019756267409470752,
+      "loss": 0.2577,
+      "step": 70
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.817968845367432,
+      "learning_rate": 0.00019721448467966573,
+      "loss": 0.1792,
+      "step": 80
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 8.561410903930664,
+      "learning_rate": 0.00019686629526462397,
+      "loss": 0.4319,
+      "step": 90
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.326521873474121,
+      "learning_rate": 0.00019651810584958218,
+      "loss": 0.2947,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "eval_accuracy": 0.9353448275862069,
+      "eval_loss": 0.24909386038780212,
+      "eval_runtime": 39.0896,
+      "eval_samples_per_second": 17.805,
+      "eval_steps_per_second": 2.226,
+      "step": 100
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.267539978027344,
+      "learning_rate": 0.0001961699164345404,
+      "loss": 0.2,
+      "step": 110
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.16868898272514343,
+      "learning_rate": 0.0001958217270194986,
+      "loss": 0.1645,
+      "step": 120
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.9674816131591797,
+      "learning_rate": 0.00019547353760445685,
+      "loss": 0.1851,
+      "step": 130
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.389801025390625,
+      "learning_rate": 0.00019512534818941506,
+      "loss": 0.1896,
+      "step": 140
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.4343678951263428,
+      "learning_rate": 0.00019477715877437327,
+      "loss": 0.302,
+      "step": 150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.64504861831665,
+      "learning_rate": 0.00019442896935933148,
+      "loss": 0.291,
+      "step": 160
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.4216318130493164,
+      "learning_rate": 0.0001940807799442897,
+      "loss": 0.1888,
+      "step": 170
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.917649269104004,
+      "learning_rate": 0.00019373259052924793,
+      "loss": 0.2216,
+      "step": 180
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.220677852630615,
+      "learning_rate": 0.00019338440111420614,
+      "loss": 0.2315,
+      "step": 190
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.801267147064209,
+      "learning_rate": 0.00019303621169916436,
+      "loss": 0.1703,
+      "step": 200
+    },
+    {
+      "epoch": 0.14,
+      "eval_accuracy": 0.9339080459770115,
+      "eval_loss": 0.23769310116767883,
+      "eval_runtime": 35.6997,
+      "eval_samples_per_second": 19.496,
+      "eval_steps_per_second": 2.437,
+      "step": 200
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.5499558448791504,
+      "learning_rate": 0.00019268802228412257,
+      "loss": 0.2146,
+      "step": 210
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.020380020141602,
+      "learning_rate": 0.0001923398328690808,
+      "loss": 0.4034,
+      "step": 220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.2459611892700195,
+      "learning_rate": 0.00019199164345403902,
+      "loss": 0.2545,
+      "step": 230
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.17939189076423645,
+      "learning_rate": 0.0001916434540389972,
+      "loss": 0.0967,
+      "step": 240
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.2918466329574585,
+      "learning_rate": 0.00019129526462395544,
+      "loss": 0.1047,
+      "step": 250
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.19346456229686737,
+      "learning_rate": 0.00019094707520891365,
+      "loss": 0.1653,
+      "step": 260
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.239362955093384,
+      "learning_rate": 0.0001905988857938719,
+      "loss": 0.2189,
+      "step": 270
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.44739365577697754,
+      "learning_rate": 0.00019025069637883008,
+      "loss": 0.3003,
+      "step": 280
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.6296844482421875,
+      "learning_rate": 0.00018990250696378832,
+      "loss": 0.1217,
+      "step": 290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.19472052156925201,
+      "learning_rate": 0.00018955431754874653,
+      "loss": 0.0797,
+      "step": 300
+    },
+    {
+      "epoch": 0.21,
+      "eval_accuracy": 0.9640804597701149,
+      "eval_loss": 0.14126166701316833,
+      "eval_runtime": 35.5893,
+      "eval_samples_per_second": 19.556,
+      "eval_steps_per_second": 2.445,
+      "step": 300
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.6157753467559814,
+      "learning_rate": 0.00018920612813370474,
+      "loss": 0.2083,
+      "step": 310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.2029290199279785,
+      "learning_rate": 0.00018885793871866295,
+      "loss": 0.0689,
+      "step": 320
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.600524663925171,
+      "learning_rate": 0.00018850974930362116,
+      "loss": 0.1281,
+      "step": 330
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.636768817901611,
+      "learning_rate": 0.0001881615598885794,
+      "loss": 0.1777,
+      "step": 340
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.280453085899353,
+      "learning_rate": 0.00018781337047353762,
+      "loss": 0.1184,
+      "step": 350
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 11.173064231872559,
+      "learning_rate": 0.00018746518105849583,
+      "loss": 0.1703,
+      "step": 360
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.2499542236328125,
+      "learning_rate": 0.00018711699164345404,
+      "loss": 0.1653,
+      "step": 370
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.3910412788391113,
+      "learning_rate": 0.00018676880222841225,
+      "loss": 0.4659,
+      "step": 380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.7442002296447754,
+      "learning_rate": 0.0001864206128133705,
+      "loss": 0.1462,
+      "step": 390
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.4805572032928467,
+      "learning_rate": 0.0001860724233983287,
+      "loss": 0.1035,
+      "step": 400
+    },
+    {
+      "epoch": 0.28,
+      "eval_accuracy": 0.9640804597701149,
+      "eval_loss": 0.10567597299814224,
+      "eval_runtime": 35.7816,
+      "eval_samples_per_second": 19.451,
+      "eval_steps_per_second": 2.431,
+      "step": 400
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.05641770362854,
+      "learning_rate": 0.00018572423398328691,
+      "loss": 0.2023,
+      "step": 410
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.117976188659668,
+      "learning_rate": 0.00018537604456824513,
+      "loss": 0.2038,
+      "step": 420
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.819580554962158,
+      "learning_rate": 0.00018502785515320337,
+      "loss": 0.1822,
+      "step": 430
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.43668073415756226,
+      "learning_rate": 0.00018467966573816158,
+      "loss": 0.142,
+      "step": 440
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.4008846282958984,
+      "learning_rate": 0.0001843314763231198,
+      "loss": 0.1302,
+      "step": 450
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.694020688533783,
+      "learning_rate": 0.000183983286908078,
+      "loss": 0.1471,
+      "step": 460
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.4394240379333496,
+      "learning_rate": 0.0001836350974930362,
+      "loss": 0.0524,
+      "step": 470
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.31817400455474854,
+      "learning_rate": 0.00018328690807799445,
+      "loss": 0.1666,
+      "step": 480
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.899374008178711,
+      "learning_rate": 0.00018293871866295264,
+      "loss": 0.3,
+      "step": 490
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.5271711349487305,
+      "learning_rate": 0.00018259052924791088,
+      "loss": 0.0532,
+      "step": 500
+    },
+    {
+      "epoch": 0.35,
+      "eval_accuracy": 0.9482758620689655,
+      "eval_loss": 0.17112933099269867,
+      "eval_runtime": 35.7523,
+      "eval_samples_per_second": 19.467,
+      "eval_steps_per_second": 2.433,
+      "step": 500
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 7.314517021179199,
+      "learning_rate": 0.0001822423398328691,
+      "loss": 0.1091,
+      "step": 510
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.5478276014328003,
+      "learning_rate": 0.00018189415041782733,
+      "loss": 0.2114,
+      "step": 520
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.8942975997924805,
+      "learning_rate": 0.0001815459610027855,
+      "loss": 0.1946,
+      "step": 530
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.3088340759277344,
+      "learning_rate": 0.00018119777158774372,
+      "loss": 0.1564,
+      "step": 540
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.6240744590759277,
+      "learning_rate": 0.00018084958217270196,
+      "loss": 0.0936,
+      "step": 550
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.7066690921783447,
+      "learning_rate": 0.00018050139275766017,
+      "loss": 0.0437,
+      "step": 560
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 10.780313491821289,
+      "learning_rate": 0.0001801532033426184,
+      "loss": 0.1595,
+      "step": 570
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.528925657272339,
+      "learning_rate": 0.0001798050139275766,
+      "loss": 0.1321,
+      "step": 580
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.18334412574768066,
+      "learning_rate": 0.00017945682451253484,
+      "loss": 0.1395,
+      "step": 590
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.294697284698486,
+      "learning_rate": 0.00017910863509749305,
+      "loss": 0.1004,
+      "step": 600
+    },
+    {
+      "epoch": 0.42,
+      "eval_accuracy": 0.9525862068965517,
+      "eval_loss": 0.174613818526268,
+      "eval_runtime": 36.2243,
+      "eval_samples_per_second": 19.214,
+      "eval_steps_per_second": 2.402,
+      "step": 600
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 2.1188316345214844,
+      "learning_rate": 0.00017876044568245126,
+      "loss": 0.2278,
+      "step": 610
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3354647755622864,
+      "learning_rate": 0.00017841225626740947,
+      "loss": 0.1387,
+      "step": 620
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.039503831416368484,
+      "learning_rate": 0.00017806406685236768,
+      "loss": 0.0835,
+      "step": 630
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.021547358483076096,
+      "learning_rate": 0.00017771587743732592,
+      "loss": 0.0939,
+      "step": 640
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.863506317138672,
+      "learning_rate": 0.00017736768802228414,
+      "loss": 0.1468,
+      "step": 650
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.39873701333999634,
+      "learning_rate": 0.00017701949860724235,
+      "loss": 0.0403,
+      "step": 660
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.014231318607926369,
+      "learning_rate": 0.00017667130919220056,
+      "loss": 0.2571,
+      "step": 670
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.528050899505615,
+      "learning_rate": 0.0001763231197771588,
+      "loss": 0.0652,
+      "step": 680
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.15766894817352295,
+      "learning_rate": 0.000175974930362117,
+      "loss": 0.1783,
+      "step": 690
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.2292269468307495,
+      "learning_rate": 0.0001756267409470752,
+      "loss": 0.0962,
+      "step": 700
+    },
+    {
+      "epoch": 0.49,
+      "eval_accuracy": 0.9554597701149425,
+      "eval_loss": 0.15980468690395355,
+      "eval_runtime": 35.7953,
+      "eval_samples_per_second": 19.444,
+      "eval_steps_per_second": 2.43,
+      "step": 700
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.254995584487915,
+      "learning_rate": 0.00017527855153203343,
+      "loss": 0.0357,
+      "step": 710
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.948057651519775,
+      "learning_rate": 0.00017493036211699165,
+      "loss": 0.1107,
+      "step": 720
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 9.733255386352539,
+      "learning_rate": 0.00017458217270194989,
+      "loss": 0.1582,
+      "step": 730
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.013814004138112068,
+      "learning_rate": 0.00017423398328690807,
+      "loss": 0.0915,
+      "step": 740
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.37093743681907654,
+      "learning_rate": 0.0001738857938718663,
+      "loss": 0.0646,
+      "step": 750
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.888106346130371,
+      "learning_rate": 0.00017353760445682452,
+      "loss": 0.0698,
+      "step": 760
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.17707830667495728,
+      "learning_rate": 0.00017318941504178273,
+      "loss": 0.2098,
+      "step": 770
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.0932948887348175,
+      "learning_rate": 0.00017284122562674097,
+      "loss": 0.0913,
+      "step": 780
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.9526076316833496,
+      "learning_rate": 0.00017249303621169916,
+      "loss": 0.0578,
+      "step": 790
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.7513306140899658,
+      "learning_rate": 0.0001721448467966574,
+      "loss": 0.1579,
+      "step": 800
+    },
+    {
+      "epoch": 0.56,
+      "eval_accuracy": 0.9439655172413793,
+      "eval_loss": 0.1741354912519455,
+      "eval_runtime": 35.7323,
+      "eval_samples_per_second": 19.478,
+      "eval_steps_per_second": 2.435,
+      "step": 800
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.912064790725708,
+      "learning_rate": 0.0001717966573816156,
+      "loss": 0.111,
+      "step": 810
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 4.864778995513916,
+      "learning_rate": 0.00017144846796657385,
+      "loss": 0.1486,
+      "step": 820
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.14718465507030487,
+      "learning_rate": 0.00017110027855153203,
+      "loss": 0.1166,
+      "step": 830
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.20318205654621124,
+      "learning_rate": 0.00017075208913649024,
+      "loss": 0.1693,
+      "step": 840
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.11161267757415771,
+      "learning_rate": 0.00017040389972144848,
+      "loss": 0.3625,
+      "step": 850
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.465303421020508,
+      "learning_rate": 0.0001700557103064067,
+      "loss": 0.1,
+      "step": 860
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1702408641576767,
+      "learning_rate": 0.0001697075208913649,
+      "loss": 0.1512,
+      "step": 870
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1287883222103119,
+      "learning_rate": 0.00016935933147632312,
+      "loss": 0.1398,
+      "step": 880
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 5.473878860473633,
+      "learning_rate": 0.00016901114206128136,
+      "loss": 0.1271,
+      "step": 890
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.7570542097091675,
+      "learning_rate": 0.00016866295264623957,
+      "loss": 0.0532,
+      "step": 900
+    },
+    {
+      "epoch": 0.63,
+      "eval_accuracy": 0.9669540229885057,
+      "eval_loss": 0.09742401540279388,
+      "eval_runtime": 35.6843,
+      "eval_samples_per_second": 19.504,
+      "eval_steps_per_second": 2.438,
+      "step": 900
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.47937777638435364,
+      "learning_rate": 0.00016831476323119778,
+      "loss": 0.0376,
+      "step": 910
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.07491227984428406,
+      "learning_rate": 0.000167966573816156,
+      "loss": 0.0845,
+      "step": 920
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.6370360851287842,
+      "learning_rate": 0.0001676183844011142,
+      "loss": 0.0426,
+      "step": 930
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.40192294120788574,
+      "learning_rate": 0.00016727019498607244,
+      "loss": 0.0985,
+      "step": 940
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.27094128727912903,
+      "learning_rate": 0.00016692200557103066,
+      "loss": 0.1015,
+      "step": 950
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.04987002909183502,
+      "learning_rate": 0.00016657381615598887,
+      "loss": 0.1305,
+      "step": 960
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 8.257625579833984,
+      "learning_rate": 0.00016622562674094708,
+      "loss": 0.359,
+      "step": 970
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.3513457775115967,
+      "learning_rate": 0.00016587743732590532,
+      "loss": 0.1196,
+      "step": 980
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.744544506072998,
+      "learning_rate": 0.00016552924791086353,
+      "loss": 0.1392,
+      "step": 990
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.7153432369232178,
+      "learning_rate": 0.00016518105849582171,
+      "loss": 0.1594,
+      "step": 1000
+    },
+    {
+      "epoch": 0.7,
+      "eval_accuracy": 0.9181034482758621,
+      "eval_loss": 0.28419315814971924,
+      "eval_runtime": 35.8256,
+      "eval_samples_per_second": 19.427,
+      "eval_steps_per_second": 2.428,
+      "step": 1000
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.401228666305542,
+      "learning_rate": 0.00016483286908077995,
+      "loss": 0.1808,
+      "step": 1010
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.6505862474441528,
+      "learning_rate": 0.00016448467966573817,
+      "loss": 0.135,
+      "step": 1020
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 3.2271783351898193,
+      "learning_rate": 0.0001641364902506964,
+      "loss": 0.0702,
+      "step": 1030
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.8183447122573853,
+      "learning_rate": 0.0001637883008356546,
+      "loss": 0.0763,
+      "step": 1040
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 11.863089561462402,
+      "learning_rate": 0.00016344011142061283,
+      "loss": 0.1261,
+      "step": 1050
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.03473610803484917,
+      "learning_rate": 0.00016309192200557104,
+      "loss": 0.118,
+      "step": 1060
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.3610630929470062,
+      "learning_rate": 0.00016274373259052925,
+      "loss": 0.0383,
+      "step": 1070
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.0649971142411232,
+      "learning_rate": 0.00016239554317548746,
+      "loss": 0.0616,
+      "step": 1080
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.01660127006471157,
+      "learning_rate": 0.00016204735376044568,
+      "loss": 0.1321,
+      "step": 1090
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.6202359199523926,
+      "learning_rate": 0.00016169916434540392,
+      "loss": 0.0488,
+      "step": 1100
+    },
+    {
+      "epoch": 0.77,
+      "eval_accuracy": 0.9224137931034483,
+      "eval_loss": 0.29275649785995483,
+      "eval_runtime": 35.6698,
+      "eval_samples_per_second": 19.512,
+      "eval_steps_per_second": 2.439,
+      "step": 1100
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.814648628234863,
+      "learning_rate": 0.00016135097493036213,
+      "loss": 0.2168,
+      "step": 1110
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.07795737683773041,
+      "learning_rate": 0.00016100278551532034,
+      "loss": 0.1589,
+      "step": 1120
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 9.246214866638184,
+      "learning_rate": 0.00016065459610027855,
+      "loss": 0.0448,
+      "step": 1130
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.6626620292663574,
+      "learning_rate": 0.00016030640668523676,
+      "loss": 0.0533,
+      "step": 1140
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2033206671476364,
+      "learning_rate": 0.000159958217270195,
+      "loss": 0.06,
+      "step": 1150
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.868011236190796,
+      "learning_rate": 0.00015961002785515321,
+      "loss": 0.1047,
+      "step": 1160
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.4080984592437744,
+      "learning_rate": 0.00015926183844011143,
+      "loss": 0.158,
+      "step": 1170
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.1310753971338272,
+      "learning_rate": 0.00015891364902506964,
+      "loss": 0.0284,
+      "step": 1180
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 5.160470008850098,
+      "learning_rate": 0.00015856545961002788,
+      "loss": 0.1274,
+      "step": 1190
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 11.978700637817383,
+      "learning_rate": 0.0001582172701949861,
+      "loss": 0.1122,
+      "step": 1200
+    },
+    {
+      "epoch": 0.84,
+      "eval_accuracy": 0.9137931034482759,
+      "eval_loss": 0.30952975153923035,
+      "eval_runtime": 35.658,
+      "eval_samples_per_second": 19.519,
+      "eval_steps_per_second": 2.44,
+      "step": 1200
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 3.5138704776763916,
+      "learning_rate": 0.0001578690807799443,
+      "loss": 0.1662,
+      "step": 1210
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.6575515270233154,
+      "learning_rate": 0.0001575208913649025,
+      "loss": 0.1419,
+      "step": 1220
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.337413787841797,
+      "learning_rate": 0.00015717270194986072,
+      "loss": 0.1191,
+      "step": 1230
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.0867462232708931,
+      "learning_rate": 0.00015682451253481896,
+      "loss": 0.1029,
+      "step": 1240
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 5.1576948165893555,
+      "learning_rate": 0.00015647632311977715,
+      "loss": 0.0307,
+      "step": 1250
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.009300777688622475,
+      "learning_rate": 0.0001561281337047354,
+      "loss": 0.0369,
+      "step": 1260
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.04234935715794563,
+      "learning_rate": 0.0001557799442896936,
+      "loss": 0.0172,
+      "step": 1270
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.0444934107363224,
+      "learning_rate": 0.00015543175487465184,
+      "loss": 0.0562,
+      "step": 1280
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.13354717195034027,
+      "learning_rate": 0.00015508356545961002,
+      "loss": 0.0312,
+      "step": 1290
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 4.039272785186768,
+      "learning_rate": 0.00015473537604456823,
+      "loss": 0.1252,
+      "step": 1300
+    },
+    {
+      "epoch": 0.91,
+      "eval_accuracy": 0.9568965517241379,
+      "eval_loss": 0.14112895727157593,
+      "eval_runtime": 35.6262,
+      "eval_samples_per_second": 19.536,
+      "eval_steps_per_second": 2.442,
+      "step": 1300
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.016441693529486656,
+      "learning_rate": 0.00015438718662952647,
+      "loss": 0.1007,
+      "step": 1310
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.12619511783123016,
+      "learning_rate": 0.00015403899721448469,
+      "loss": 0.1028,
+      "step": 1320
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.022222580388188362,
+      "learning_rate": 0.00015369080779944292,
+      "loss": 0.0635,
+      "step": 1330
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.950347900390625,
+      "learning_rate": 0.0001533426183844011,
+      "loss": 0.1184,
+      "step": 1340
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8607469201087952,
+      "learning_rate": 0.00015299442896935935,
+      "loss": 0.0982,
+      "step": 1350
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.6757292747497559,
+      "learning_rate": 0.00015264623955431756,
+      "loss": 0.0786,
+      "step": 1360
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 8.006093978881836,
+      "learning_rate": 0.00015229805013927577,
+      "loss": 0.1273,
+      "step": 1370
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.7346742153167725,
+      "learning_rate": 0.00015194986072423398,
+      "loss": 0.1616,
+      "step": 1380
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.2187693119049072,
+      "learning_rate": 0.0001516016713091922,
+      "loss": 0.0447,
+      "step": 1390
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.1204167604446411,
+      "learning_rate": 0.00015125348189415043,
+      "loss": 0.0517,
+      "step": 1400
+    },
+    {
+      "epoch": 0.97,
+      "eval_accuracy": 0.9683908045977011,
+      "eval_loss": 0.13779418170452118,
+      "eval_runtime": 35.6464,
+      "eval_samples_per_second": 19.525,
+      "eval_steps_per_second": 2.441,
+      "step": 1400
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.01675868220627308,
+      "learning_rate": 0.00015090529247910865,
+      "loss": 0.0548,
+      "step": 1410
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.027674928307533264,
+      "learning_rate": 0.00015055710306406686,
+      "loss": 0.0635,
+      "step": 1420
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.13874393701553345,
+      "learning_rate": 0.00015020891364902507,
+      "loss": 0.0405,
+      "step": 1430
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 4.346076488494873,
+      "learning_rate": 0.0001498607242339833,
+      "loss": 0.0398,
+      "step": 1440
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 1.2912051677703857,
+      "learning_rate": 0.00014951253481894152,
+      "loss": 0.0062,
+      "step": 1450
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.05856345221400261,
+      "learning_rate": 0.0001491643454038997,
+      "loss": 0.0065,
+      "step": 1460
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.254682779312134,
+      "learning_rate": 0.00014881615598885795,
+      "loss": 0.0765,
+      "step": 1470
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.031248953193426132,
+      "learning_rate": 0.00014846796657381616,
+      "loss": 0.0622,
+      "step": 1480
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.13116455078125,
+      "learning_rate": 0.0001481197771587744,
+      "loss": 0.0767,
+      "step": 1490
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.005425759591162205,
+      "learning_rate": 0.00014777158774373258,
+      "loss": 0.047,
+      "step": 1500
+    },
+    {
+      "epoch": 1.04,
+      "eval_accuracy": 0.9482758620689655,
+      "eval_loss": 0.2595459520816803,
+      "eval_runtime": 35.615,
+      "eval_samples_per_second": 19.542,
+      "eval_steps_per_second": 2.443,
+      "step": 1500
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.005179308354854584,
+      "learning_rate": 0.00014742339832869082,
+      "loss": 0.0226,
+      "step": 1510
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.0027408453170210123,
+      "learning_rate": 0.00014707520891364903,
+      "loss": 0.0027,
+      "step": 1520
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.004645737819373608,
+      "learning_rate": 0.00014672701949860724,
+      "loss": 0.0195,
+      "step": 1530
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.0033414452336728573,
+      "learning_rate": 0.00014637883008356548,
+      "loss": 0.026,
+      "step": 1540
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.005320267751812935,
+      "learning_rate": 0.00014603064066852367,
+      "loss": 0.005,
+      "step": 1550
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.010981897823512554,
+      "learning_rate": 0.0001456824512534819,
+      "loss": 0.0015,
+      "step": 1560
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 2.1311142444610596,
+      "learning_rate": 0.00014533426183844012,
+      "loss": 0.0041,
+      "step": 1570
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 4.274787425994873,
+      "learning_rate": 0.00014498607242339836,
+      "loss": 0.0626,
+      "step": 1580
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 3.3070502281188965,
+      "learning_rate": 0.00014463788300835654,
+      "loss": 0.0131,
+      "step": 1590
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 6.9698100090026855,
+      "learning_rate": 0.00014428969359331475,
+      "loss": 0.0478,
+      "step": 1600
+    },
+    {
+      "epoch": 1.11,
+      "eval_accuracy": 0.9583333333333334,
+      "eval_loss": 0.14253316819667816,
+      "eval_runtime": 36.1253,
+      "eval_samples_per_second": 19.266,
+      "eval_steps_per_second": 2.408,
+      "step": 1600
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 5.856411457061768,
+      "learning_rate": 0.000143941504178273,
+      "loss": 0.1343,
+      "step": 1610
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.012498567812144756,
+      "learning_rate": 0.0001435933147632312,
+      "loss": 0.0011,
+      "step": 1620
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.861133575439453,
+      "learning_rate": 0.00014324512534818942,
+      "loss": 0.0418,
+      "step": 1630
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.04692409932613373,
+      "learning_rate": 0.00014289693593314763,
+      "loss": 0.0013,
+      "step": 1640
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.024860292673110962,
+      "learning_rate": 0.00014254874651810587,
+      "loss": 0.0646,
+      "step": 1650
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.05027357116341591,
+      "learning_rate": 0.00014220055710306408,
+      "loss": 0.0231,
+      "step": 1660
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.010157397948205471,
+      "learning_rate": 0.0001418523676880223,
+      "loss": 0.0695,
+      "step": 1670
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 1.822034239768982,
+      "learning_rate": 0.0001415041782729805,
+      "loss": 0.0329,
+      "step": 1680
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.07127121835947037,
+      "learning_rate": 0.00014115598885793872,
+      "loss": 0.0189,
+      "step": 1690
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.02665788121521473,
+      "learning_rate": 0.00014080779944289695,
+      "loss": 0.0107,
+      "step": 1700
+    },
+    {
+      "epoch": 1.18,
+      "eval_accuracy": 0.9683908045977011,
+      "eval_loss": 0.11352017521858215,
+      "eval_runtime": 36.5799,
+      "eval_samples_per_second": 19.027,
+      "eval_steps_per_second": 2.378,
+      "step": 1700
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.025212422013282776,
+      "learning_rate": 0.00014045961002785517,
+      "loss": 0.0591,
+      "step": 1710
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.021063433960080147,
+      "learning_rate": 0.00014011142061281338,
+      "loss": 0.0031,
+      "step": 1720
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.005293009802699089,
+      "learning_rate": 0.0001397632311977716,
+      "loss": 0.0099,
+      "step": 1730
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.031639717519283295,
+      "learning_rate": 0.00013941504178272983,
+      "loss": 0.0018,
+      "step": 1740
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.44839808344841003,
+      "learning_rate": 0.00013906685236768804,
+      "loss": 0.0218,
+      "step": 1750
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.039643511176109314,
+      "learning_rate": 0.00013871866295264623,
+      "loss": 0.0462,
+      "step": 1760
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.003339031245559454,
+      "learning_rate": 0.00013837047353760446,
+      "loss": 0.0008,
+      "step": 1770
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.009900330565869808,
+      "learning_rate": 0.00013802228412256268,
+      "loss": 0.0399,
+      "step": 1780
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 4.178069591522217,
+      "learning_rate": 0.00013767409470752092,
+      "loss": 0.067,
+      "step": 1790
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.026575196534395218,
+      "learning_rate": 0.0001373259052924791,
+      "loss": 0.0021,
+      "step": 1800
+    },
+    {
+      "epoch": 1.25,
+      "eval_accuracy": 0.9597701149425287,
+      "eval_loss": 0.14276058971881866,
+      "eval_runtime": 35.9178,
+      "eval_samples_per_second": 19.378,
+      "eval_steps_per_second": 2.422,
+      "step": 1800
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.005829383619129658,
+      "learning_rate": 0.00013697771587743734,
+      "loss": 0.0068,
+      "step": 1810
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.019658759236335754,
+      "learning_rate": 0.00013662952646239555,
+      "loss": 0.0008,
+      "step": 1820
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 5.781700611114502,
+      "learning_rate": 0.00013628133704735376,
+      "loss": 0.0246,
+      "step": 1830
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.789889395236969,
+      "learning_rate": 0.00013593314763231198,
+      "loss": 0.009,
+      "step": 1840
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.04666072502732277,
+      "learning_rate": 0.0001355849582172702,
+      "loss": 0.0059,
+      "step": 1850
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.002111963462084532,
+      "learning_rate": 0.00013523676880222843,
+      "loss": 0.0107,
+      "step": 1860
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.010087203234434128,
+      "learning_rate": 0.00013488857938718664,
+      "loss": 0.0467,
+      "step": 1870
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.08458948135376,
+      "learning_rate": 0.00013454038997214485,
+      "loss": 0.0124,
+      "step": 1880
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 6.12792444229126,
+      "learning_rate": 0.00013419220055710306,
+      "loss": 0.0593,
+      "step": 1890
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.0023462555836886168,
+      "learning_rate": 0.00013384401114206127,
+      "loss": 0.036,
+      "step": 1900
+    },
+    {
+      "epoch": 1.32,
+      "eval_accuracy": 0.9583333333333334,
+      "eval_loss": 0.18506751954555511,
+      "eval_runtime": 36.1885,
+      "eval_samples_per_second": 19.233,
+      "eval_steps_per_second": 2.404,
+      "step": 1900
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.010145030915737152,
+      "learning_rate": 0.0001334958217270195,
+      "loss": 0.0411,
+      "step": 1910
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.005521192215383053,
+      "learning_rate": 0.00013314763231197772,
+      "loss": 0.0095,
+      "step": 1920
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.008944313041865826,
+      "learning_rate": 0.00013279944289693594,
+      "loss": 0.0061,
+      "step": 1930
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.036939240992069244,
+      "learning_rate": 0.00013245125348189415,
+      "loss": 0.0295,
+      "step": 1940
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.004508269485086203,
+      "learning_rate": 0.0001321030640668524,
+      "loss": 0.015,
+      "step": 1950
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.0245398860424757,
+      "learning_rate": 0.0001317548746518106,
+      "loss": 0.0122,
+      "step": 1960
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.008592234924435616,
+      "learning_rate": 0.0001314066852367688,
+      "loss": 0.0519,
+      "step": 1970
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.026266993954777718,
+      "learning_rate": 0.00013105849582172702,
+      "loss": 0.047,
+      "step": 1980
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.8703653216362,
+      "learning_rate": 0.00013071030640668524,
+      "loss": 0.0858,
+      "step": 1990
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.16321466863155365,
+      "learning_rate": 0.00013036211699164347,
+      "loss": 0.0733,
+      "step": 2000
+    },
+    {
+      "epoch": 1.39,
+      "eval_accuracy": 0.9583333333333334,
+      "eval_loss": 0.18010464310646057,
+      "eval_runtime": 35.975,
+      "eval_samples_per_second": 19.347,
+      "eval_steps_per_second": 2.418,
+      "step": 2000
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.02395610883831978,
+      "learning_rate": 0.00013001392757660166,
+      "loss": 0.0481,
+      "step": 2010
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 2.885974645614624,
+      "learning_rate": 0.0001296657381615599,
+      "loss": 0.0926,
+      "step": 2020
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.23348748683929443,
+      "learning_rate": 0.0001293175487465181,
+      "loss": 0.0128,
+      "step": 2030
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.011263038031756878,
+      "learning_rate": 0.00012896935933147635,
+      "loss": 0.0595,
+      "step": 2040
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.003669577185064554,
+      "learning_rate": 0.00012862116991643453,
+      "loss": 0.0271,
+      "step": 2050
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.061473917216062546,
+      "learning_rate": 0.00012827298050139275,
+      "loss": 0.0705,
+      "step": 2060
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.21758386492729187,
+      "learning_rate": 0.00012792479108635098,
+      "loss": 0.0289,
+      "step": 2070
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.921506404876709,
+      "learning_rate": 0.0001275766016713092,
+      "loss": 0.0441,
+      "step": 2080
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.011570905335247517,
+      "learning_rate": 0.00012722841225626744,
+      "loss": 0.0016,
+      "step": 2090
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 5.713520050048828,
+      "learning_rate": 0.00012688022284122562,
+      "loss": 0.0549,
+      "step": 2100
+    },
+    {
+      "epoch": 1.46,
+      "eval_accuracy": 0.9597701149425287,
+      "eval_loss": 0.19172786176204681,
+      "eval_runtime": 35.7108,
+      "eval_samples_per_second": 19.49,
+      "eval_steps_per_second": 2.436,
+      "step": 2100
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.07785508036613464,
+      "learning_rate": 0.00012653203342618386,
+      "loss": 0.0217,
+      "step": 2110
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.004833151586353779,
+      "learning_rate": 0.00012618384401114207,
+      "loss": 0.0013,
+      "step": 2120
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.050749145448207855,
+      "learning_rate": 0.00012583565459610028,
+      "loss": 0.0231,
+      "step": 2130
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.002657771809026599,
+      "learning_rate": 0.0001254874651810585,
+      "loss": 0.0006,
+      "step": 2140
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.0026397837791591883,
+      "learning_rate": 0.0001251392757660167,
+      "loss": 0.0128,
+      "step": 2150
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.0028812838718295097,
+      "learning_rate": 0.00012479108635097495,
+      "loss": 0.0008,
+      "step": 2160
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 7.40750789642334,
+      "learning_rate": 0.00012444289693593316,
+      "loss": 0.0542,
+      "step": 2170
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.023035351186990738,
+      "learning_rate": 0.00012409470752089137,
+      "loss": 0.0618,
+      "step": 2180
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 3.932640552520752,
+      "learning_rate": 0.00012374651810584958,
+      "loss": 0.0346,
+      "step": 2190
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.043167661875486374,
+      "learning_rate": 0.00012339832869080782,
+      "loss": 0.0442,
+      "step": 2200
+    },
+    {
+      "epoch": 1.53,
+      "eval_accuracy": 0.9655172413793104,
+      "eval_loss": 0.15375293791294098,
+      "eval_runtime": 36.1114,
+      "eval_samples_per_second": 19.274,
+      "eval_steps_per_second": 2.409,
+      "step": 2200
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.030572544783353806,
+      "learning_rate": 0.00012305013927576603,
+      "loss": 0.0301,
+      "step": 2210
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 2.4444499015808105,
+      "learning_rate": 0.00012270194986072422,
+      "loss": 0.0648,
+      "step": 2220
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.003966104239225388,
+      "learning_rate": 0.00012235376044568246,
+      "loss": 0.0046,
+      "step": 2230
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 6.563431739807129,
+      "learning_rate": 0.00012200557103064068,
+      "loss": 0.071,
+      "step": 2240
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.04225797951221466,
+      "learning_rate": 0.0001216573816155989,
+      "loss": 0.0171,
+      "step": 2250
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.1912388801574707,
+      "learning_rate": 0.00012130919220055709,
+      "loss": 0.08,
+      "step": 2260
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.0018349624006077647,
+      "learning_rate": 0.00012096100278551532,
+      "loss": 0.0048,
+      "step": 2270
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.0031464290805161,
+      "learning_rate": 0.00012061281337047354,
+      "loss": 0.0044,
+      "step": 2280
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.003731086850166321,
+      "learning_rate": 0.00012026462395543177,
+      "loss": 0.0253,
+      "step": 2290
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 7.218816757202148,
+      "learning_rate": 0.00011991643454039,
+      "loss": 0.0196,
+      "step": 2300
+    },
+    {
+      "epoch": 1.6,
+      "eval_accuracy": 0.9698275862068966,
+      "eval_loss": 0.14107204973697662,
+      "eval_runtime": 36.5366,
+      "eval_samples_per_second": 19.049,
+      "eval_steps_per_second": 2.381,
+      "step": 2300
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 1.3289762735366821,
+      "learning_rate": 0.00011956824512534819,
+      "loss": 0.0037,
+      "step": 2310
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 2.8097541332244873,
+      "learning_rate": 0.0001192200557103064,
+      "loss": 0.0184,
+      "step": 2320
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 1.9819353818893433,
+      "learning_rate": 0.00011887186629526463,
+      "loss": 0.0692,
+      "step": 2330
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 3.6161446571350098,
+      "learning_rate": 0.00011855849582172702,
+      "loss": 0.1015,
+      "step": 2340
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.16102848947048187,
+      "learning_rate": 0.00011821030640668524,
+      "loss": 0.0155,
+      "step": 2350
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.002351599046960473,
+      "learning_rate": 0.00011786211699164347,
+      "loss": 0.0617,
+      "step": 2360
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.1253902018070221,
+      "learning_rate": 0.00011751392757660167,
+      "loss": 0.0134,
+      "step": 2370
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.6419570446014404,
+      "learning_rate": 0.00011716573816155989,
+      "loss": 0.0171,
+      "step": 2380
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.0014169831993058324,
+      "learning_rate": 0.00011681754874651812,
+      "loss": 0.0059,
+      "step": 2390
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.004712065681815147,
+      "learning_rate": 0.00011646935933147633,
+      "loss": 0.0809,
+      "step": 2400
+    },
+    {
+      "epoch": 1.67,
+      "eval_accuracy": 0.9540229885057471,
+      "eval_loss": 0.18621940910816193,
+      "eval_runtime": 35.9547,
+      "eval_samples_per_second": 19.358,
+      "eval_steps_per_second": 2.42,
+      "step": 2400
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.059306591749191284,
+      "learning_rate": 0.00011612116991643454,
+      "loss": 0.0824,
+      "step": 2410
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.07501501590013504,
+      "learning_rate": 0.00011577298050139275,
+      "loss": 0.0294,
+      "step": 2420
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.5017964839935303,
+      "learning_rate": 0.00011542479108635098,
+      "loss": 0.062,
+      "step": 2430
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 1.136009931564331,
+      "learning_rate": 0.0001150766016713092,
+      "loss": 0.0084,
+      "step": 2440
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.11244765669107437,
+      "learning_rate": 0.00011472841225626743,
+      "loss": 0.0784,
+      "step": 2450
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.004342706874012947,
+      "learning_rate": 0.00011438022284122563,
+      "loss": 0.0065,
+      "step": 2460
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.06110939756035805,
+      "learning_rate": 0.00011403203342618385,
+      "loss": 0.0014,
+      "step": 2470
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.06438643485307693,
+      "learning_rate": 0.00011368384401114207,
+      "loss": 0.0036,
+      "step": 2480
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.048323530703783035,
+      "learning_rate": 0.00011333565459610029,
+      "loss": 0.0059,
+      "step": 2490
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.016341280192136765,
+      "learning_rate": 0.00011298746518105849,
+      "loss": 0.0004,
+      "step": 2500
+    },
+    {
+      "epoch": 1.74,
+      "eval_accuracy": 0.9698275862068966,
+      "eval_loss": 0.1324520707130432,
+      "eval_runtime": 35.782,
+      "eval_samples_per_second": 19.451,
+      "eval_steps_per_second": 2.431,
+      "step": 2500
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.6630181670188904,
+      "learning_rate": 0.00011263927576601671,
+      "loss": 0.0017,
+      "step": 2510
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.002616755198687315,
+      "learning_rate": 0.00011229108635097494,
+      "loss": 0.0008,
+      "step": 2520
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.006078706588596106,
+      "learning_rate": 0.00011194289693593317,
+      "loss": 0.0005,
+      "step": 2530
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 1.398128867149353,
+      "learning_rate": 0.00011159470752089136,
+      "loss": 0.0052,
+      "step": 2540
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.0013051325222477317,
+      "learning_rate": 0.00011124651810584959,
+      "loss": 0.0601,
+      "step": 2550
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.10400926321744919,
+      "learning_rate": 0.0001108983286908078,
+      "loss": 0.003,
+      "step": 2560
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.0016845627687871456,
+      "learning_rate": 0.00011055013927576603,
+      "loss": 0.0799,
+      "step": 2570
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 8.756908416748047,
+      "learning_rate": 0.00011020194986072423,
+      "loss": 0.0414,
+      "step": 2580
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.06296553462743759,
+      "learning_rate": 0.00010985376044568245,
+      "loss": 0.0085,
+      "step": 2590
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.002325402805581689,
+      "learning_rate": 0.00010950557103064068,
+      "loss": 0.0404,
+      "step": 2600
+    },
+    {
+      "epoch": 1.81,
+      "eval_accuracy": 0.9712643678160919,
+      "eval_loss": 0.12463492900133133,
+      "eval_runtime": 36.0715,
+      "eval_samples_per_second": 19.295,
+      "eval_steps_per_second": 2.412,
+      "step": 2600
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 2.0505995750427246,
+      "learning_rate": 0.0001091573816155989,
+      "loss": 0.0131,
+      "step": 2610
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.005136621650308371,
+      "learning_rate": 0.0001088091922005571,
+      "loss": 0.0086,
+      "step": 2620
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.0078201275318861,
+      "learning_rate": 0.00010846100278551533,
+      "loss": 0.0017,
+      "step": 2630
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 5.730374813079834,
+      "learning_rate": 0.00010811281337047354,
+      "loss": 0.0106,
+      "step": 2640
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.0022723209112882614,
+      "learning_rate": 0.00010776462395543176,
+      "loss": 0.0104,
+      "step": 2650
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.0025097934994846582,
+      "learning_rate": 0.00010741643454038999,
+      "loss": 0.0248,
+      "step": 2660
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 7.8781256675720215,
+      "learning_rate": 0.00010706824512534819,
+      "loss": 0.0314,
+      "step": 2670
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 6.506507396697998,
+      "learning_rate": 0.00010672005571030641,
+      "loss": 0.022,
+      "step": 2680
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.18162275850772858,
+      "learning_rate": 0.00010637186629526464,
+      "loss": 0.0193,
+      "step": 2690
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.22412802278995514,
+      "learning_rate": 0.00010602367688022285,
+      "loss": 0.0691,
+      "step": 2700
+    },
+    {
+      "epoch": 1.88,
+      "eval_accuracy": 0.9597701149425287,
+      "eval_loss": 0.19613413512706757,
+      "eval_runtime": 35.9377,
+      "eval_samples_per_second": 19.367,
+      "eval_steps_per_second": 2.421,
+      "step": 2700
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.00789021048694849,
+      "learning_rate": 0.00010567548746518106,
+      "loss": 0.1014,
+      "step": 2710
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.0036270658019930124,
+      "learning_rate": 0.00010532729805013927,
+      "loss": 0.0066,
+      "step": 2720
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.0017741814954206347,
+      "learning_rate": 0.0001049791086350975,
+      "loss": 0.0017,
+      "step": 2730
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.007776726968586445,
+      "learning_rate": 0.00010463091922005572,
+      "loss": 0.0006,
+      "step": 2740
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.0036039133556187153,
+      "learning_rate": 0.00010428272980501392,
+      "loss": 0.0069,
+      "step": 2750
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.020857617259025574,
+      "learning_rate": 0.00010393454038997215,
+      "loss": 0.0084,
+      "step": 2760
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.00476573221385479,
+      "learning_rate": 0.00010358635097493037,
+      "loss": 0.0089,
+      "step": 2770
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.007087323348969221,
+      "learning_rate": 0.00010323816155988859,
+      "loss": 0.003,
+      "step": 2780
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.12047570943832397,
+      "learning_rate": 0.0001028899721448468,
+      "loss": 0.0353,
+      "step": 2790
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 8.084424018859863,
+      "learning_rate": 0.00010254178272980501,
+      "loss": 0.0088,
+      "step": 2800
+    },
+    {
+      "epoch": 1.95,
+      "eval_accuracy": 0.9683908045977011,
+      "eval_loss": 0.18410901725292206,
+      "eval_runtime": 35.8036,
+      "eval_samples_per_second": 19.439,
+      "eval_steps_per_second": 2.43,
+      "step": 2800
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.0028939053881913424,
+      "learning_rate": 0.00010219359331476323,
+      "loss": 0.0852,
+      "step": 2810
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.015490409918129444,
+      "learning_rate": 0.00010184540389972146,
+      "loss": 0.0008,
+      "step": 2820
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.21892137825489044,
+      "learning_rate": 0.00010149721448467969,
+      "loss": 0.0007,
+      "step": 2830
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.0013207359006628394,
+      "learning_rate": 0.00010114902506963788,
+      "loss": 0.003,
+      "step": 2840
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.013448349200189114,
+      "learning_rate": 0.00010080083565459611,
+      "loss": 0.0066,
+      "step": 2850
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.00878733117133379,
+      "learning_rate": 0.00010045264623955432,
+      "loss": 0.0496,
+      "step": 2860
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 13.995903968811035,
+      "learning_rate": 0.00010010445682451255,
+      "loss": 0.0989,
+      "step": 2870
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.14911749958992004,
+      "learning_rate": 9.975626740947076e-05,
+      "loss": 0.0027,
+      "step": 2880
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.003861737437546253,
+      "learning_rate": 9.940807799442897e-05,
+      "loss": 0.0322,
+      "step": 2890
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.005346905440092087,
+      "learning_rate": 9.90598885793872e-05,
+      "loss": 0.0029,
+      "step": 2900
+    },
+    {
+      "epoch": 2.02,
+      "eval_accuracy": 0.9813218390804598,
+      "eval_loss": 0.10571122169494629,
+      "eval_runtime": 36.3232,
+      "eval_samples_per_second": 19.161,
+      "eval_steps_per_second": 2.395,
+      "step": 2900
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.02600006014108658,
+      "learning_rate": 9.871169916434541e-05,
+      "loss": 0.0008,
+      "step": 2910
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.0034855674020946026,
+      "learning_rate": 9.836350974930363e-05,
+      "loss": 0.0017,
+      "step": 2920
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.012216300703585148,
+      "learning_rate": 9.801532033426185e-05,
+      "loss": 0.0003,
+      "step": 2930
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.0010572533356025815,
+      "learning_rate": 9.766713091922006e-05,
+      "loss": 0.0004,
+      "step": 2940
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.000851622608024627,
+      "learning_rate": 9.731894150417827e-05,
+      "loss": 0.0031,
+      "step": 2950
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.24186454713344574,
+      "learning_rate": 9.69707520891365e-05,
+      "loss": 0.0003,
+      "step": 2960
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.029137974604964256,
+      "learning_rate": 9.66225626740947e-05,
+      "loss": 0.0003,
+      "step": 2970
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.0058034202083945274,
+      "learning_rate": 9.627437325905293e-05,
+      "loss": 0.0005,
+      "step": 2980
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.0009422635193914175,
+      "learning_rate": 9.592618384401114e-05,
+      "loss": 0.0002,
+      "step": 2990
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.003287246450781822,
+      "learning_rate": 9.557799442896937e-05,
+      "loss": 0.0005,
+      "step": 3000
+    },
+    {
+      "epoch": 2.09,
+      "eval_accuracy": 0.9741379310344828,
+      "eval_loss": 0.11307180672883987,
+      "eval_runtime": 35.848,
+      "eval_samples_per_second": 19.415,
+      "eval_steps_per_second": 2.427,
+      "step": 3000
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.0038968026638031006,
+      "learning_rate": 9.522980501392758e-05,
+      "loss": 0.0007,
+      "step": 3010
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.003602625336498022,
+      "learning_rate": 9.488161559888579e-05,
+      "loss": 0.0003,
+      "step": 3020
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.0007726295152679086,
+      "learning_rate": 9.453342618384402e-05,
+      "loss": 0.0016,
+      "step": 3030
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.0014457539655268192,
+      "learning_rate": 9.418523676880223e-05,
+      "loss": 0.0361,
+      "step": 3040
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.001288870582357049,
+      "learning_rate": 9.383704735376046e-05,
+      "loss": 0.0002,
+      "step": 3050
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.0008597253472544253,
+      "learning_rate": 9.348885793871867e-05,
+      "loss": 0.0003,
+      "step": 3060
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.0012300481321290135,
+      "learning_rate": 9.314066852367689e-05,
+      "loss": 0.0002,
+      "step": 3070
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.0006355264922603965,
+      "learning_rate": 9.27924791086351e-05,
+      "loss": 0.0001,
+      "step": 3080
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.005312491673976183,
+      "learning_rate": 9.244428969359332e-05,
+      "loss": 0.0002,
+      "step": 3090
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.0009683069074526429,
+      "learning_rate": 9.209610027855153e-05,
+      "loss": 0.0001,
+      "step": 3100
+    },
+    {
+      "epoch": 2.16,
+      "eval_accuracy": 0.9813218390804598,
+      "eval_loss": 0.08920785784721375,
+      "eval_runtime": 35.6113,
+      "eval_samples_per_second": 19.544,
+      "eval_steps_per_second": 2.443,
+      "step": 3100
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.0020216393750160933,
+      "learning_rate": 9.174791086350975e-05,
+      "loss": 0.0052,
+      "step": 3110
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.5278111696243286,
+      "learning_rate": 9.139972144846797e-05,
+      "loss": 0.0033,
+      "step": 3120
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.0029855158645659685,
+      "learning_rate": 9.105153203342619e-05,
+      "loss": 0.0539,
+      "step": 3130
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.013236282393336296,
+      "learning_rate": 9.07033426183844e-05,
+      "loss": 0.0003,
+      "step": 3140
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.052337899804115295,
+      "learning_rate": 9.035515320334263e-05,
+      "loss": 0.001,
+      "step": 3150
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.0012746230931952596,
+      "learning_rate": 9.000696378830084e-05,
+      "loss": 0.0434,
+      "step": 3160
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.07627440243959427,
+      "learning_rate": 8.965877437325905e-05,
+      "loss": 0.0004,
+      "step": 3170
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.003083089366555214,
+      "learning_rate": 8.931058495821726e-05,
+      "loss": 0.0002,
+      "step": 3180
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.0020660432055592537,
+      "learning_rate": 8.896239554317549e-05,
+      "loss": 0.0003,
+      "step": 3190
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.04302741959691048,
+      "learning_rate": 8.861420612813372e-05,
+      "loss": 0.0002,
+      "step": 3200
+    },
+    {
+      "epoch": 2.23,
+      "eval_accuracy": 0.9827586206896551,
+      "eval_loss": 0.07567945867776871,
+      "eval_runtime": 35.3934,
+      "eval_samples_per_second": 19.665,
+      "eval_steps_per_second": 2.458,
+      "step": 3200
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.0009345992002636194,
+      "learning_rate": 8.826601671309193e-05,
+      "loss": 0.0441,
+      "step": 3210
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.002595352241769433,
+      "learning_rate": 8.791782729805015e-05,
+      "loss": 0.0002,
+      "step": 3220
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.002003292553126812,
+      "learning_rate": 8.756963788300836e-05,
+      "loss": 0.0007,
+      "step": 3230
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.0009823666187003255,
+      "learning_rate": 8.722144846796658e-05,
+      "loss": 0.0002,
+      "step": 3240
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.0008656879072077572,
+      "learning_rate": 8.687325905292479e-05,
+      "loss": 0.0004,
+      "step": 3250
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.0006070456001907587,
+      "learning_rate": 8.652506963788301e-05,
+      "loss": 0.0004,
+      "step": 3260
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.0006921704043634236,
+      "learning_rate": 8.617688022284123e-05,
+      "loss": 0.0196,
+      "step": 3270
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.0018148680683225393,
+      "learning_rate": 8.582869080779945e-05,
+      "loss": 0.0001,
+      "step": 3280
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.010168994776904583,
+      "learning_rate": 8.548050139275766e-05,
+      "loss": 0.0005,
+      "step": 3290
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.0014962274581193924,
+      "learning_rate": 8.513231197771589e-05,
+      "loss": 0.0186,
+      "step": 3300
+    },
+    {
+      "epoch": 2.3,
+      "eval_accuracy": 0.978448275862069,
+      "eval_loss": 0.07941145449876785,
+      "eval_runtime": 35.5001,
+      "eval_samples_per_second": 19.606,
+      "eval_steps_per_second": 2.451,
+      "step": 3300
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.0014951284974813461,
+      "learning_rate": 8.47841225626741e-05,
+      "loss": 0.0001,
+      "step": 3310
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.0012492133537307382,
+      "learning_rate": 8.443593314763231e-05,
+      "loss": 0.017,
+      "step": 3320
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.001966628013178706,
+      "learning_rate": 8.408774373259052e-05,
+      "loss": 0.0003,
+      "step": 3330
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.003503795713186264,
+      "learning_rate": 8.373955431754875e-05,
+      "loss": 0.0019,
+      "step": 3340
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.0009698419598862529,
+      "learning_rate": 8.339136490250696e-05,
+      "loss": 0.0002,
+      "step": 3350
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.003532392904162407,
+      "learning_rate": 8.304317548746519e-05,
+      "loss": 0.061,
+      "step": 3360
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.0015911379596218467,
+      "learning_rate": 8.26949860724234e-05,
+      "loss": 0.0002,
+      "step": 3370
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.003713688813149929,
+      "learning_rate": 8.234679665738162e-05,
+      "loss": 0.0003,
+      "step": 3380
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.0017646539490669966,
+      "learning_rate": 8.199860724233984e-05,
+      "loss": 0.0535,
+      "step": 3390
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.0015788556775078177,
+      "learning_rate": 8.165041782729805e-05,
+      "loss": 0.0127,
+      "step": 3400
+    },
+    {
+      "epoch": 2.37,
+      "eval_accuracy": 0.9770114942528736,
+      "eval_loss": 0.109991654753685,
+      "eval_runtime": 35.5883,
+      "eval_samples_per_second": 19.557,
+      "eval_steps_per_second": 2.445,
+      "step": 3400
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.0013490252895280719,
+      "learning_rate": 8.130222841225627e-05,
+      "loss": 0.0002,
+      "step": 3410
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.0011618455173447728,
+      "learning_rate": 8.095403899721449e-05,
+      "loss": 0.0041,
+      "step": 3420
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.0008495352230966091,
+      "learning_rate": 8.060584958217271e-05,
+      "loss": 0.0002,
+      "step": 3430
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.002711158012971282,
+      "learning_rate": 8.025766016713092e-05,
+      "loss": 0.0346,
+      "step": 3440
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.023644153028726578,
+      "learning_rate": 7.990947075208915e-05,
+      "loss": 0.005,
+      "step": 3450
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.006983724422752857,
+      "learning_rate": 7.956128133704736e-05,
+      "loss": 0.0424,
+      "step": 3460
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.02986639365553856,
+      "learning_rate": 7.921309192200557e-05,
+      "loss": 0.0318,
+      "step": 3470
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 4.136352062225342,
+      "learning_rate": 7.886490250696378e-05,
+      "loss": 0.0032,
+      "step": 3480
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.015517113730311394,
+      "learning_rate": 7.851671309192201e-05,
+      "loss": 0.0305,
+      "step": 3490
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.0017729534301906824,
+      "learning_rate": 7.816852367688022e-05,
+      "loss": 0.0048,
+      "step": 3500
+    },
+    {
+      "epoch": 2.44,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.13860931992530823,
+      "eval_runtime": 36.2444,
+      "eval_samples_per_second": 19.203,
+      "eval_steps_per_second": 2.4,
+      "step": 3500
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.2275744378566742,
+      "learning_rate": 7.782033426183845e-05,
+      "loss": 0.0207,
+      "step": 3510
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.0016194531926885247,
+      "learning_rate": 7.747214484679666e-05,
+      "loss": 0.0454,
+      "step": 3520
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.00831185095012188,
+      "learning_rate": 7.712395543175488e-05,
+      "loss": 0.0002,
+      "step": 3530
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.0064757768996059895,
+      "learning_rate": 7.67757660167131e-05,
+      "loss": 0.0015,
+      "step": 3540
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.19444404542446136,
+      "learning_rate": 7.642757660167131e-05,
+      "loss": 0.0038,
+      "step": 3550
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.0022411602549254894,
+      "learning_rate": 7.607938718662952e-05,
+      "loss": 0.0003,
+      "step": 3560
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.0066094715148210526,
+      "learning_rate": 7.573119777158775e-05,
+      "loss": 0.0307,
+      "step": 3570
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.00163698208052665,
+      "learning_rate": 7.538300835654597e-05,
+      "loss": 0.0053,
+      "step": 3580
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.00275797420181334,
+      "learning_rate": 7.503481894150418e-05,
+      "loss": 0.0009,
+      "step": 3590
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.014838581904768944,
+      "learning_rate": 7.468662952646241e-05,
+      "loss": 0.0048,
+      "step": 3600
+    },
+    {
+      "epoch": 2.51,
+      "eval_accuracy": 0.9899425287356322,
+      "eval_loss": 0.06347225606441498,
+      "eval_runtime": 35.6377,
+      "eval_samples_per_second": 19.53,
+      "eval_steps_per_second": 2.441,
+      "step": 3600
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.013059583492577076,
+      "learning_rate": 7.433844011142062e-05,
+      "loss": 0.0004,
+      "step": 3610
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.02046467363834381,
+      "learning_rate": 7.399025069637883e-05,
+      "loss": 0.004,
+      "step": 3620
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.2419157475233078,
+      "learning_rate": 7.364206128133704e-05,
+      "loss": 0.0059,
+      "step": 3630
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.001298550865612924,
+      "learning_rate": 7.329387186629527e-05,
+      "loss": 0.0003,
+      "step": 3640
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.006013019476085901,
+      "learning_rate": 7.294568245125348e-05,
+      "loss": 0.0004,
+      "step": 3650
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.07667834311723709,
+      "learning_rate": 7.259749303621171e-05,
+      "loss": 0.0004,
+      "step": 3660
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.0010899449698626995,
+      "learning_rate": 7.224930362116992e-05,
+      "loss": 0.0001,
+      "step": 3670
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.0033650537952780724,
+      "learning_rate": 7.190111420612814e-05,
+      "loss": 0.0005,
+      "step": 3680
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.0012150221737101674,
+      "learning_rate": 7.155292479108636e-05,
+      "loss": 0.0001,
+      "step": 3690
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.0012052224483340979,
+      "learning_rate": 7.120473537604457e-05,
+      "loss": 0.001,
+      "step": 3700
+    },
+    {
+      "epoch": 2.58,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.09972245991230011,
+      "eval_runtime": 36.1164,
+      "eval_samples_per_second": 19.271,
+      "eval_steps_per_second": 2.409,
+      "step": 3700
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.0028800666332244873,
+      "learning_rate": 7.085654596100278e-05,
+      "loss": 0.0002,
+      "step": 3710
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.0007534552132710814,
+      "learning_rate": 7.0508356545961e-05,
+      "loss": 0.0371,
+      "step": 3720
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.02767573855817318,
+      "learning_rate": 7.016016713091922e-05,
+      "loss": 0.0002,
+      "step": 3730
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.0017805310199037194,
+      "learning_rate": 6.981197771587744e-05,
+      "loss": 0.0003,
+      "step": 3740
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.01443151943385601,
+      "learning_rate": 6.946378830083565e-05,
+      "loss": 0.0002,
+      "step": 3750
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.00520982826128602,
+      "learning_rate": 6.911559888579388e-05,
+      "loss": 0.02,
+      "step": 3760
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.00234880018979311,
+      "learning_rate": 6.876740947075209e-05,
+      "loss": 0.0253,
+      "step": 3770
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.0004435311129782349,
+      "learning_rate": 6.84192200557103e-05,
+      "loss": 0.0001,
+      "step": 3780
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.0015614174772053957,
+      "learning_rate": 6.807103064066853e-05,
+      "loss": 0.0265,
+      "step": 3790
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.0008992806542664766,
+      "learning_rate": 6.772284122562674e-05,
+      "loss": 0.0005,
+      "step": 3800
+    },
+    {
+      "epoch": 2.65,
+      "eval_accuracy": 0.9755747126436781,
+      "eval_loss": 0.11192366480827332,
+      "eval_runtime": 35.6523,
+      "eval_samples_per_second": 19.522,
+      "eval_steps_per_second": 2.44,
+      "step": 3800
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.001526621519587934,
+      "learning_rate": 6.737465181058497e-05,
+      "loss": 0.0001,
+      "step": 3810
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.0012564824428409338,
+      "learning_rate": 6.702646239554318e-05,
+      "loss": 0.0184,
+      "step": 3820
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.010799840092658997,
+      "learning_rate": 6.66782729805014e-05,
+      "loss": 0.0002,
+      "step": 3830
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.09518575668334961,
+      "learning_rate": 6.633008356545962e-05,
+      "loss": 0.0233,
+      "step": 3840
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.007948394864797592,
+      "learning_rate": 6.598189415041783e-05,
+      "loss": 0.0538,
+      "step": 3850
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.0013590805465355515,
+      "learning_rate": 6.563370473537604e-05,
+      "loss": 0.012,
+      "step": 3860
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.004664690233767033,
+      "learning_rate": 6.528551532033427e-05,
+      "loss": 0.0022,
+      "step": 3870
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.0020023442339152098,
+      "learning_rate": 6.493732590529248e-05,
+      "loss": 0.0004,
+      "step": 3880
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.002935326425358653,
+      "learning_rate": 6.45891364902507e-05,
+      "loss": 0.0001,
+      "step": 3890
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.007063873112201691,
+      "learning_rate": 6.424094707520891e-05,
+      "loss": 0.0006,
+      "step": 3900
+    },
+    {
+      "epoch": 2.72,
+      "eval_accuracy": 0.9712643678160919,
+      "eval_loss": 0.12915293872356415,
+      "eval_runtime": 35.4193,
+      "eval_samples_per_second": 19.65,
+      "eval_steps_per_second": 2.456,
+      "step": 3900
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.028898164629936218,
+      "learning_rate": 6.389275766016714e-05,
+      "loss": 0.0004,
+      "step": 3910
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.016497179865837097,
+      "learning_rate": 6.354456824512535e-05,
+      "loss": 0.0018,
+      "step": 3920
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.00926460325717926,
+      "learning_rate": 6.319637883008356e-05,
+      "loss": 0.0001,
+      "step": 3930
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.0021442470606416464,
+      "learning_rate": 6.284818941504178e-05,
+      "loss": 0.0021,
+      "step": 3940
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.00117313745431602,
+      "learning_rate": 6.25e-05,
+      "loss": 0.0167,
+      "step": 3950
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.001675875741057098,
+      "learning_rate": 6.215181058495823e-05,
+      "loss": 0.0002,
+      "step": 3960
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 4.56508731842041,
+      "learning_rate": 6.180362116991644e-05,
+      "loss": 0.0458,
+      "step": 3970
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.00593574857339263,
+      "learning_rate": 6.145543175487466e-05,
+      "loss": 0.0001,
+      "step": 3980
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 2.249992847442627,
+      "learning_rate": 6.110724233983288e-05,
+      "loss": 0.0032,
+      "step": 3990
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.0021550431847572327,
+      "learning_rate": 6.0759052924791095e-05,
+      "loss": 0.0003,
+      "step": 4000
+    },
+    {
+      "epoch": 2.79,
+      "eval_accuracy": 0.9770114942528736,
+      "eval_loss": 0.1185581386089325,
+      "eval_runtime": 35.5879,
+      "eval_samples_per_second": 19.557,
+      "eval_steps_per_second": 2.445,
+      "step": 4000
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.003284281585365534,
+      "learning_rate": 6.0410863509749307e-05,
+      "loss": 0.0051,
+      "step": 4010
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.0009602207574062049,
+      "learning_rate": 6.0062674094707525e-05,
+      "loss": 0.0004,
+      "step": 4020
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.0013085228856652975,
+      "learning_rate": 5.971448467966574e-05,
+      "loss": 0.0044,
+      "step": 4030
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.0013867886736989021,
+      "learning_rate": 5.936629526462396e-05,
+      "loss": 0.0003,
+      "step": 4040
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.0016945902025327086,
+      "learning_rate": 5.9018105849582175e-05,
+      "loss": 0.0005,
+      "step": 4050
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.0058686388656497,
+      "learning_rate": 5.866991643454039e-05,
+      "loss": 0.0003,
+      "step": 4060
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.0007193565252237022,
+      "learning_rate": 5.8321727019498605e-05,
+      "loss": 0.0008,
+      "step": 4070
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.002980075776576996,
+      "learning_rate": 5.797353760445683e-05,
+      "loss": 0.0001,
+      "step": 4080
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.0009382622665725648,
+      "learning_rate": 5.762534818941504e-05,
+      "loss": 0.0005,
+      "step": 4090
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.0006443078746087849,
+      "learning_rate": 5.727715877437326e-05,
+      "loss": 0.0137,
+      "step": 4100
+    },
+    {
+      "epoch": 2.86,
+      "eval_accuracy": 0.9770114942528736,
+      "eval_loss": 0.0968838483095169,
+      "eval_runtime": 36.4506,
+      "eval_samples_per_second": 19.094,
+      "eval_steps_per_second": 2.387,
+      "step": 4100
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.02481170743703842,
+      "learning_rate": 5.692896935933147e-05,
+      "loss": 0.0007,
+      "step": 4110
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.0029068405274301767,
+      "learning_rate": 5.65807799442897e-05,
+      "loss": 0.0003,
+      "step": 4120
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 1.1100047826766968,
+      "learning_rate": 5.623259052924791e-05,
+      "loss": 0.0014,
+      "step": 4130
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.0005633348482660949,
+      "learning_rate": 5.588440111420613e-05,
+      "loss": 0.003,
+      "step": 4140
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.12589029967784882,
+      "learning_rate": 5.5536211699164355e-05,
+      "loss": 0.0003,
+      "step": 4150
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.0005603305180557072,
+      "learning_rate": 5.5188022284122566e-05,
+      "loss": 0.0005,
+      "step": 4160
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.0006461963639594615,
+      "learning_rate": 5.4839832869080785e-05,
+      "loss": 0.0001,
+      "step": 4170
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.0012363989371806383,
+      "learning_rate": 5.4491643454039e-05,
+      "loss": 0.0004,
+      "step": 4180
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.001515595125965774,
+      "learning_rate": 5.414345403899722e-05,
+      "loss": 0.0005,
+      "step": 4190
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.0007989232544787228,
+      "learning_rate": 5.3795264623955434e-05,
+      "loss": 0.0001,
+      "step": 4200
+    },
+    {
+      "epoch": 2.92,
+      "eval_accuracy": 0.9841954022988506,
+      "eval_loss": 0.073846735060215,
+      "eval_runtime": 36.351,
+      "eval_samples_per_second": 19.147,
+      "eval_steps_per_second": 2.393,
+      "step": 4200
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.0004774934204760939,
+      "learning_rate": 5.344707520891365e-05,
+      "loss": 0.0001,
+      "step": 4210
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.0007342536700889468,
+      "learning_rate": 5.3098885793871865e-05,
+      "loss": 0.0001,
+      "step": 4220
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.6081188917160034,
+      "learning_rate": 5.275069637883009e-05,
+      "loss": 0.0006,
+      "step": 4230
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.0013682940043509007,
+      "learning_rate": 5.24025069637883e-05,
+      "loss": 0.0001,
+      "step": 4240
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.001264879247173667,
+      "learning_rate": 5.205431754874652e-05,
+      "loss": 0.0001,
+      "step": 4250
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.0004952333983965218,
+      "learning_rate": 5.170612813370473e-05,
+      "loss": 0.0002,
+      "step": 4260
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.0011145187309011817,
+      "learning_rate": 5.135793871866296e-05,
+      "loss": 0.0002,
+      "step": 4270
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.0011305679799988866,
+      "learning_rate": 5.100974930362117e-05,
+      "loss": 0.0073,
+      "step": 4280
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.000979065429419279,
+      "learning_rate": 5.066155988857939e-05,
+      "loss": 0.0006,
+      "step": 4290
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.0009217975893989205,
+      "learning_rate": 5.03133704735376e-05,
+      "loss": 0.0001,
+      "step": 4300
+    },
+    {
+      "epoch": 2.99,
+      "eval_accuracy": 0.9827586206896551,
+      "eval_loss": 0.1235894039273262,
+      "eval_runtime": 35.7354,
+      "eval_samples_per_second": 19.476,
+      "eval_steps_per_second": 2.435,
+      "step": 4300
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.00048277797759510577,
+      "learning_rate": 4.9965181058495826e-05,
+      "loss": 0.0001,
+      "step": 4310
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.0013075217138975859,
+      "learning_rate": 4.9616991643454045e-05,
+      "loss": 0.0001,
+      "step": 4320
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.0005105934687890112,
+      "learning_rate": 4.926880222841226e-05,
+      "loss": 0.0001,
+      "step": 4330
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.013046924956142902,
+      "learning_rate": 4.8920612813370476e-05,
+      "loss": 0.0001,
+      "step": 4340
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.0009469912620261312,
+      "learning_rate": 4.8572423398328694e-05,
+      "loss": 0.0045,
+      "step": 4350
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.003529737703502178,
+      "learning_rate": 4.8224233983286906e-05,
+      "loss": 0.0001,
+      "step": 4360
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.0012883865274488926,
+      "learning_rate": 4.7876044568245125e-05,
+      "loss": 0.0001,
+      "step": 4370
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.0005208416259847581,
+      "learning_rate": 4.7527855153203343e-05,
+      "loss": 0.0001,
+      "step": 4380
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.004692881368100643,
+      "learning_rate": 4.717966573816156e-05,
+      "loss": 0.0063,
+      "step": 4390
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.043299201875925064,
+      "learning_rate": 4.6831476323119774e-05,
+      "loss": 0.0001,
+      "step": 4400
+    },
+    {
+      "epoch": 3.06,
+      "eval_accuracy": 0.985632183908046,
+      "eval_loss": 0.09321270138025284,
+      "eval_runtime": 35.0423,
+      "eval_samples_per_second": 19.862,
+      "eval_steps_per_second": 2.483,
+      "step": 4400
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.0009556158329360187,
+      "learning_rate": 4.648328690807799e-05,
+      "loss": 0.0001,
+      "step": 4410
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 4.4719648361206055,
+      "learning_rate": 4.613509749303621e-05,
+      "loss": 0.0047,
+      "step": 4420
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.0004217967507429421,
+      "learning_rate": 4.582172701949861e-05,
+      "loss": 0.0193,
+      "step": 4430
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.011219095438718796,
+      "learning_rate": 4.547353760445683e-05,
+      "loss": 0.0175,
+      "step": 4440
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.0016105608083307743,
+      "learning_rate": 4.5125348189415044e-05,
+      "loss": 0.0439,
+      "step": 4450
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.0010759801371023059,
+      "learning_rate": 4.477715877437326e-05,
+      "loss": 0.0001,
+      "step": 4460
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.0011237241560593247,
+      "learning_rate": 4.442896935933148e-05,
+      "loss": 0.0001,
+      "step": 4470
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.0017227695789188147,
+      "learning_rate": 4.40807799442897e-05,
+      "loss": 0.0001,
+      "step": 4480
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.0004417230957187712,
+      "learning_rate": 4.373259052924791e-05,
+      "loss": 0.0001,
+      "step": 4490
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.0016022748313844204,
+      "learning_rate": 4.338440111420613e-05,
+      "loss": 0.0001,
+      "step": 4500
+    },
+    {
+      "epoch": 3.13,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.09920793026685715,
+      "eval_runtime": 35.9448,
+      "eval_samples_per_second": 19.363,
+      "eval_steps_per_second": 2.42,
+      "step": 4500
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.0005612379754893482,
+      "learning_rate": 4.303621169916435e-05,
+      "loss": 0.0001,
+      "step": 4510
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0011157976696267724,
+      "learning_rate": 4.268802228412256e-05,
+      "loss": 0.0001,
+      "step": 4520
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0010938127525150776,
+      "learning_rate": 4.233983286908078e-05,
+      "loss": 0.0001,
+      "step": 4530
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.0006874927203170955,
+      "learning_rate": 4.1991643454039e-05,
+      "loss": 0.0001,
+      "step": 4540
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 0.002226391341537237,
+      "learning_rate": 4.164345403899722e-05,
+      "loss": 0.0013,
+      "step": 4550
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.0004934744210913777,
+      "learning_rate": 4.129526462395543e-05,
+      "loss": 0.0002,
+      "step": 4560
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.00039926779572851956,
+      "learning_rate": 4.094707520891365e-05,
+      "loss": 0.0001,
+      "step": 4570
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 0.0009613159345462918,
+      "learning_rate": 4.0598885793871866e-05,
+      "loss": 0.0033,
+      "step": 4580
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.0009614640148356557,
+      "learning_rate": 4.0250696378830085e-05,
+      "loss": 0.0005,
+      "step": 4590
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.0007666044984944165,
+      "learning_rate": 3.9902506963788303e-05,
+      "loss": 0.0001,
+      "step": 4600
+    },
+    {
+      "epoch": 3.2,
+      "eval_accuracy": 0.9827586206896551,
+      "eval_loss": 0.09604837745428085,
+      "eval_runtime": 35.3895,
+      "eval_samples_per_second": 19.667,
+      "eval_steps_per_second": 2.458,
+      "step": 4600
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 0.6162603497505188,
+      "learning_rate": 3.955431754874652e-05,
+      "loss": 0.0003,
+      "step": 4610
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.0006573472637683153,
+      "learning_rate": 3.920612813370474e-05,
+      "loss": 0.0001,
+      "step": 4620
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.0006694823387078941,
+      "learning_rate": 3.885793871866296e-05,
+      "loss": 0.0001,
+      "step": 4630
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 0.0006889757933095098,
+      "learning_rate": 3.850974930362117e-05,
+      "loss": 0.0001,
+      "step": 4640
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 0.0007866056403145194,
+      "learning_rate": 3.816155988857939e-05,
+      "loss": 0.0005,
+      "step": 4650
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 0.0006213219021447003,
+      "learning_rate": 3.781337047353761e-05,
+      "loss": 0.0001,
+      "step": 4660
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 0.0006275447667576373,
+      "learning_rate": 3.746518105849583e-05,
+      "loss": 0.0001,
+      "step": 4670
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 0.000448089704150334,
+      "learning_rate": 3.711699164345404e-05,
+      "loss": 0.0002,
+      "step": 4680
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.0015304009430110455,
+      "learning_rate": 3.676880222841226e-05,
+      "loss": 0.0001,
+      "step": 4690
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.06754755973815918,
+      "learning_rate": 3.642061281337048e-05,
+      "loss": 0.0001,
+      "step": 4700
+    },
+    {
+      "epoch": 3.27,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.11231694370508194,
+      "eval_runtime": 36.4603,
+      "eval_samples_per_second": 19.089,
+      "eval_steps_per_second": 2.386,
+      "step": 4700
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 0.001035011257044971,
+      "learning_rate": 3.607242339832869e-05,
+      "loss": 0.0001,
+      "step": 4710
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 0.00041941230301745236,
+      "learning_rate": 3.572423398328691e-05,
+      "loss": 0.0001,
+      "step": 4720
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 0.0006384252919815481,
+      "learning_rate": 3.5376044568245126e-05,
+      "loss": 0.0001,
+      "step": 4730
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 0.0006129765533842146,
+      "learning_rate": 3.5027855153203345e-05,
+      "loss": 0.0001,
+      "step": 4740
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 0.0009090271196328104,
+      "learning_rate": 3.4679665738161556e-05,
+      "loss": 0.0001,
+      "step": 4750
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 0.0004404740175232291,
+      "learning_rate": 3.4331476323119775e-05,
+      "loss": 0.0001,
+      "step": 4760
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 0.0005690119578503072,
+      "learning_rate": 3.3983286908077994e-05,
+      "loss": 0.0001,
+      "step": 4770
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.0005757323815487325,
+      "learning_rate": 3.363509749303621e-05,
+      "loss": 0.0001,
+      "step": 4780
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 0.0005803864914923906,
+      "learning_rate": 3.328690807799443e-05,
+      "loss": 0.0001,
+      "step": 4790
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 0.0007648465107195079,
+      "learning_rate": 3.293871866295265e-05,
+      "loss": 0.0001,
+      "step": 4800
+    },
+    {
+      "epoch": 3.34,
+      "eval_accuracy": 0.9813218390804598,
+      "eval_loss": 0.11066480726003647,
+      "eval_runtime": 35.9909,
+      "eval_samples_per_second": 19.338,
+      "eval_steps_per_second": 2.417,
+      "step": 4800
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 0.00033829055610112846,
+      "learning_rate": 3.259052924791087e-05,
+      "loss": 0.0001,
+      "step": 4810
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 0.0030458769761025906,
+      "learning_rate": 3.224233983286909e-05,
+      "loss": 0.0003,
+      "step": 4820
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 0.0004884973168373108,
+      "learning_rate": 3.18941504178273e-05,
+      "loss": 0.0001,
+      "step": 4830
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 0.0005915808724239469,
+      "learning_rate": 3.154596100278552e-05,
+      "loss": 0.0,
+      "step": 4840
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 0.0005569536588154733,
+      "learning_rate": 3.1197771587743737e-05,
+      "loss": 0.0001,
+      "step": 4850
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 0.00037319512921385467,
+      "learning_rate": 3.0849582172701955e-05,
+      "loss": 0.0,
+      "step": 4860
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 0.0006390858325175941,
+      "learning_rate": 3.050139275766017e-05,
+      "loss": 0.0002,
+      "step": 4870
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 0.010524235665798187,
+      "learning_rate": 3.0153203342618386e-05,
+      "loss": 0.0001,
+      "step": 4880
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 0.0003795890952460468,
+      "learning_rate": 2.98050139275766e-05,
+      "loss": 0.0,
+      "step": 4890
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 0.0016755288233980536,
+      "learning_rate": 2.945682451253482e-05,
+      "loss": 0.0029,
+      "step": 4900
+    },
+    {
+      "epoch": 3.41,
+      "eval_accuracy": 0.9841954022988506,
+      "eval_loss": 0.10405226796865463,
+      "eval_runtime": 35.9093,
+      "eval_samples_per_second": 19.382,
+      "eval_steps_per_second": 2.423,
+      "step": 4900
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 0.0005780430510640144,
+      "learning_rate": 2.9108635097493035e-05,
+      "loss": 0.0,
+      "step": 4910
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 0.0005223274347372353,
+      "learning_rate": 2.8760445682451254e-05,
+      "loss": 0.0001,
+      "step": 4920
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 0.0003509995876811445,
+      "learning_rate": 2.841225626740947e-05,
+      "loss": 0.0,
+      "step": 4930
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 0.0009933110559359193,
+      "learning_rate": 2.8064066852367688e-05,
+      "loss": 0.0001,
+      "step": 4940
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 0.0005258971359580755,
+      "learning_rate": 2.7715877437325903e-05,
+      "loss": 0.0001,
+      "step": 4950
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 0.0003142243658658117,
+      "learning_rate": 2.736768802228412e-05,
+      "loss": 0.0008,
+      "step": 4960
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 0.0010210133623331785,
+      "learning_rate": 2.7019498607242337e-05,
+      "loss": 0.0001,
+      "step": 4970
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 0.0005668088560923934,
+      "learning_rate": 2.6671309192200562e-05,
+      "loss": 0.0031,
+      "step": 4980
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 0.0004998073563911021,
+      "learning_rate": 2.6323119777158778e-05,
+      "loss": 0.0,
+      "step": 4990
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 0.0008013307233341038,
+      "learning_rate": 2.5974930362116996e-05,
+      "loss": 0.0001,
+      "step": 5000
+    },
+    {
+      "epoch": 3.48,
+      "eval_accuracy": 0.9827586206896551,
+      "eval_loss": 0.10744819790124893,
+      "eval_runtime": 35.6882,
+      "eval_samples_per_second": 19.502,
+      "eval_steps_per_second": 2.438,
+      "step": 5000
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 0.0004506228433456272,
+      "learning_rate": 2.562674094707521e-05,
+      "loss": 0.0,
+      "step": 5010
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 0.000541824207175523,
+      "learning_rate": 2.527855153203343e-05,
+      "loss": 0.0248,
+      "step": 5020
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 0.0005250126123428345,
+      "learning_rate": 2.4930362116991646e-05,
+      "loss": 0.0,
+      "step": 5030
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 0.0003384416049811989,
+      "learning_rate": 2.4582172701949864e-05,
+      "loss": 0.0005,
+      "step": 5040
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 0.004039893392473459,
+      "learning_rate": 2.423398328690808e-05,
+      "loss": 0.0,
+      "step": 5050
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 0.0003602771903388202,
+      "learning_rate": 2.3885793871866298e-05,
+      "loss": 0.0001,
+      "step": 5060
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 0.000374146387912333,
+      "learning_rate": 2.3537604456824514e-05,
+      "loss": 0.0,
+      "step": 5070
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 0.00026975912624038756,
+      "learning_rate": 2.318941504178273e-05,
+      "loss": 0.0001,
+      "step": 5080
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 0.0005988151533529162,
+      "learning_rate": 2.2841225626740948e-05,
+      "loss": 0.0001,
+      "step": 5090
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 0.0007265285239554942,
+      "learning_rate": 2.2493036211699163e-05,
+      "loss": 0.0001,
+      "step": 5100
+    },
+    {
+      "epoch": 3.55,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.11111408472061157,
+      "eval_runtime": 35.643,
+      "eval_samples_per_second": 19.527,
+      "eval_steps_per_second": 2.441,
+      "step": 5100
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 0.0004558263754006475,
+      "learning_rate": 2.214484679665738e-05,
+      "loss": 0.0,
+      "step": 5110
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 0.000497504835948348,
+      "learning_rate": 2.17966573816156e-05,
+      "loss": 0.0,
+      "step": 5120
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 0.017960894852876663,
+      "learning_rate": 2.144846796657382e-05,
+      "loss": 0.0004,
+      "step": 5130
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 0.0007236815872602165,
+      "learning_rate": 2.1100278551532034e-05,
+      "loss": 0.0001,
+      "step": 5140
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 0.001991676865145564,
+      "learning_rate": 2.0752089136490253e-05,
+      "loss": 0.0001,
+      "step": 5150
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 0.0005619129515253007,
+      "learning_rate": 2.0403899721448468e-05,
+      "loss": 0.0001,
+      "step": 5160
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 0.0004067553672939539,
+      "learning_rate": 2.0055710306406687e-05,
+      "loss": 0.0001,
+      "step": 5170
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 0.06427491456270218,
+      "learning_rate": 1.9707520891364902e-05,
+      "loss": 0.0085,
+      "step": 5180
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 0.0005257455050013959,
+      "learning_rate": 1.935933147632312e-05,
+      "loss": 0.0001,
+      "step": 5190
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 0.004010786768049002,
+      "learning_rate": 1.9011142061281336e-05,
+      "loss": 0.0001,
+      "step": 5200
+    },
+    {
+      "epoch": 3.62,
+      "eval_accuracy": 0.978448275862069,
+      "eval_loss": 0.10880725830793381,
+      "eval_runtime": 35.7466,
+      "eval_samples_per_second": 19.47,
+      "eval_steps_per_second": 2.434,
+      "step": 5200
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 0.0003503792395349592,
+      "learning_rate": 1.8662952646239558e-05,
+      "loss": 0.0001,
+      "step": 5210
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 0.0003166968817822635,
+      "learning_rate": 1.8314763231197773e-05,
+      "loss": 0.0002,
+      "step": 5220
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 0.00039611879037693143,
+      "learning_rate": 1.7966573816155992e-05,
+      "loss": 0.0001,
+      "step": 5230
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 0.0004587644652929157,
+      "learning_rate": 1.7618384401114207e-05,
+      "loss": 0.0002,
+      "step": 5240
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 0.000461429386632517,
+      "learning_rate": 1.7270194986072426e-05,
+      "loss": 0.0002,
+      "step": 5250
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 0.8542178869247437,
+      "learning_rate": 1.692200557103064e-05,
+      "loss": 0.0008,
+      "step": 5260
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 0.00043836917029693723,
+      "learning_rate": 1.6573816155988857e-05,
+      "loss": 0.0,
+      "step": 5270
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 0.001168058137409389,
+      "learning_rate": 1.6225626740947075e-05,
+      "loss": 0.0004,
+      "step": 5280
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 0.0005670097307302058,
+      "learning_rate": 1.587743732590529e-05,
+      "loss": 0.0001,
+      "step": 5290
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 0.020862262696027756,
+      "learning_rate": 1.552924791086351e-05,
+      "loss": 0.0001,
+      "step": 5300
+    },
+    {
+      "epoch": 3.69,
+      "eval_accuracy": 0.9813218390804598,
+      "eval_loss": 0.09361202269792557,
+      "eval_runtime": 35.5593,
+      "eval_samples_per_second": 19.573,
+      "eval_steps_per_second": 2.447,
+      "step": 5300
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.00047057002666406333,
+      "learning_rate": 1.518105849582173e-05,
+      "loss": 0.0102,
+      "step": 5310
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.0008214833214879036,
+      "learning_rate": 1.4832869080779947e-05,
+      "loss": 0.0001,
+      "step": 5320
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 0.007291738875210285,
+      "learning_rate": 1.4484679665738164e-05,
+      "loss": 0.0,
+      "step": 5330
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 0.000499303569085896,
+      "learning_rate": 1.413649025069638e-05,
+      "loss": 0.0,
+      "step": 5340
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 0.000553811842110008,
+      "learning_rate": 1.3788300835654596e-05,
+      "loss": 0.0,
+      "step": 5350
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 0.0004532486782409251,
+      "learning_rate": 1.3440111420612813e-05,
+      "loss": 0.0,
+      "step": 5360
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 0.00035664939787238836,
+      "learning_rate": 1.309192200557103e-05,
+      "loss": 0.0001,
+      "step": 5370
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 0.0002885190478991717,
+      "learning_rate": 1.2743732590529247e-05,
+      "loss": 0.0,
+      "step": 5380
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 0.0007975110784173012,
+      "learning_rate": 1.2395543175487466e-05,
+      "loss": 0.0001,
+      "step": 5390
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 0.0006704545812681317,
+      "learning_rate": 1.2047353760445683e-05,
+      "loss": 0.0001,
+      "step": 5400
+    },
+    {
+      "epoch": 3.76,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.09153711050748825,
+      "eval_runtime": 35.5171,
+      "eval_samples_per_second": 19.596,
+      "eval_steps_per_second": 2.45,
+      "step": 5400
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 0.0008886194555088878,
+      "learning_rate": 1.16991643454039e-05,
+      "loss": 0.0,
+      "step": 5410
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 0.051304157823324203,
+      "learning_rate": 1.1350974930362116e-05,
+      "loss": 0.0001,
+      "step": 5420
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 0.00032604343141429126,
+      "learning_rate": 1.1002785515320335e-05,
+      "loss": 0.0001,
+      "step": 5430
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 0.0004452233260963112,
+      "learning_rate": 1.0654596100278552e-05,
+      "loss": 0.0001,
+      "step": 5440
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 0.001050635357387364,
+      "learning_rate": 1.0306406685236769e-05,
+      "loss": 0.0,
+      "step": 5450
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 0.0005059978575445712,
+      "learning_rate": 9.958217270194986e-06,
+      "loss": 0.0,
+      "step": 5460
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 0.00035790822585113347,
+      "learning_rate": 9.610027855153205e-06,
+      "loss": 0.0001,
+      "step": 5470
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 0.0065688504837453365,
+      "learning_rate": 9.261838440111422e-06,
+      "loss": 0.0001,
+      "step": 5480
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 0.01191110722720623,
+      "learning_rate": 8.913649025069639e-06,
+      "loss": 0.0004,
+      "step": 5490
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 0.0005021314136683941,
+      "learning_rate": 8.565459610027856e-06,
+      "loss": 0.0001,
+      "step": 5500
+    },
+    {
+      "epoch": 3.83,
+      "eval_accuracy": 0.9798850574712644,
+      "eval_loss": 0.08972268551588058,
+      "eval_runtime": 35.479,
+      "eval_samples_per_second": 19.617,
+      "eval_steps_per_second": 2.452,
+      "step": 5500
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 0.0006523649790324271,
+      "learning_rate": 8.217270194986073e-06,
+      "loss": 0.0001,
+      "step": 5510
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 0.0007299577700905502,
+      "learning_rate": 7.869080779944291e-06,
+      "loss": 0.0001,
+      "step": 5520
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 0.0002770805440377444,
+      "learning_rate": 7.5208913649025075e-06,
+      "loss": 0.0413,
+      "step": 5530
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 0.002696505980566144,
+      "learning_rate": 7.1727019498607245e-06,
+      "loss": 0.0001,
+      "step": 5540
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 0.002345402492210269,
+      "learning_rate": 6.8245125348189415e-06,
+      "loss": 0.0,
+      "step": 5550
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 0.001016572816297412,
+      "learning_rate": 6.4763231197771585e-06,
+      "loss": 0.0001,
+      "step": 5560
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 0.001638455200009048,
+      "learning_rate": 6.128133704735376e-06,
+      "loss": 0.0001,
+      "step": 5570
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 0.0015767073491588235,
+      "learning_rate": 5.779944289693594e-06,
+      "loss": 0.0001,
+      "step": 5580
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 0.000313255179207772,
+      "learning_rate": 5.43175487465181e-06,
+      "loss": 0.0,
+      "step": 5590
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 0.0012458935379981995,
+      "learning_rate": 5.083565459610028e-06,
+      "loss": 0.0001,
+      "step": 5600
+    },
+    {
+      "epoch": 3.9,
+      "eval_accuracy": 0.9770114942528736,
+      "eval_loss": 0.08747635781764984,
+      "eval_runtime": 35.3235,
+      "eval_samples_per_second": 19.704,
+      "eval_steps_per_second": 2.463,
+      "step": 5600
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 0.0006569348042830825,
+      "learning_rate": 4.735376044568245e-06,
+      "loss": 0.0001,
+      "step": 5610
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 0.000606011541094631,
+      "learning_rate": 4.387186629526462e-06,
+      "loss": 0.0001,
+      "step": 5620
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.318962574005127,
+      "learning_rate": 4.03899721448468e-06,
+      "loss": 0.0031,
+      "step": 5630
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 0.00032933251350186765,
+      "learning_rate": 3.690807799442897e-06,
+      "loss": 0.0,
+      "step": 5640
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 0.0018586514052003622,
+      "learning_rate": 3.3426183844011147e-06,
+      "loss": 0.0005,
+      "step": 5650
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 0.0009259448852390051,
+      "learning_rate": 2.9944289693593313e-06,
+      "loss": 0.0001,
+      "step": 5660
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 0.0009393952204845846,
+      "learning_rate": 2.6462395543175487e-06,
+      "loss": 0.0001,
+      "step": 5670
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 0.00043099027243442833,
+      "learning_rate": 2.298050139275766e-06,
+      "loss": 0.0001,
+      "step": 5680
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 0.00032449286663904786,
+      "learning_rate": 1.9498607242339835e-06,
+      "loss": 0.0,
+      "step": 5690
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 0.0003699425724335015,
+      "learning_rate": 1.6016713091922007e-06,
+      "loss": 0.0,
+      "step": 5700
+    },
+    {
+      "epoch": 3.97,
+      "eval_accuracy": 0.978448275862069,
+      "eval_loss": 0.08557858318090439,
+      "eval_runtime": 35.5296,
+      "eval_samples_per_second": 19.589,
+      "eval_steps_per_second": 2.449,
+      "step": 5700
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 0.0005763766821473837,
+      "learning_rate": 1.253481894150418e-06,
+      "loss": 0.0001,
+      "step": 5710
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 0.0004730022919829935,
+      "learning_rate": 9.052924791086352e-07,
+      "loss": 0.0001,
+      "step": 5720
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 0.0002761252399068326,
+      "learning_rate": 5.571030640668524e-07,
+      "loss": 0.0001,
+      "step": 5730
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.002425793092697859,
+      "learning_rate": 2.0891364902506967e-07,
+      "loss": 0.0001,
+      "step": 5740
+    },
+    {
+      "epoch": 4.0,
+      "step": 5744,
+      "total_flos": 2.099744383597491e+19,
+      "train_loss": 0.053249289760298635,
+      "train_runtime": 10114.0647,
+      "train_samples_per_second": 9.084,
+      "train_steps_per_second": 0.568
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5744,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "total_flos": 2.099744383597491e+19,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}