diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.746105968952179,
-  "best_model_checkpoint": "./vit-base-hate-meme/checkpoint-1064",
+  "best_metric": 0.6965588331222534,
+  "best_model_checkpoint": "./vit-base-hate-meme/checkpoint-532",
   "epoch": 8.0,
   "eval_steps": 500,
   "global_step": 4256,
@@ -10,3059 +10,3051 @@
   "log_history": [
     {
       "epoch": 0.02,
-      "grad_norm": 1.893226146697998,
-      "learning_rate": 3.6e-06,
-      "loss": 0.6079,
+      "grad_norm": 6.126795768737793,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.7608,
       "step": 10
     },
     {
       "epoch": 0.04,
-      "grad_norm": 1.8366069793701172,
-      "learning_rate": 7.2e-06,
-      "loss": 0.6672,
+      "grad_norm": 5.251371383666992,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.7016,
       "step": 20
     },
     {
       "epoch": 0.06,
-      "grad_norm": 3.005919933319092,
-      "learning_rate": 1.1200000000000001e-05,
-      "loss": 0.6246,
+      "grad_norm": 5.532265663146973,
+      "learning_rate": 1.2e-05,
+      "loss": 0.6429,
       "step": 30
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.47935283184051514,
-      "learning_rate": 1.52e-05,
-      "loss": 0.6929,
+      "grad_norm": 5.4727888107299805,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.7306,
       "step": 40
     },
     {
       "epoch": 0.09,
-      "grad_norm": 3.6089649200439453,
-      "learning_rate": 1.9200000000000003e-05,
-      "loss": 0.6635,
+      "grad_norm": 5.2745771408081055,
+      "learning_rate": 2e-05,
+      "loss": 0.7079,
       "step": 50
     },
     {
       "epoch": 0.11,
-      "grad_norm": 1.0585061311721802,
-      "learning_rate": 2.32e-05,
-      "loss": 0.626,
+      "grad_norm": 4.616114616394043,
+      "learning_rate": 2.36e-05,
+      "loss": 0.671,
       "step": 60
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.16031716763973236,
-      "learning_rate": 2.7200000000000004e-05,
-      "loss": 0.6391,
+      "grad_norm": 4.487009048461914,
+      "learning_rate": 2.7600000000000003e-05,
+      "loss": 0.6954,
       "step": 70
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.742310106754303,
-      "learning_rate": 3.12e-05,
-      "loss": 0.6518,
+      "grad_norm": 6.081264495849609,
+      "learning_rate": 3.16e-05,
+      "loss": 0.6968,
       "step": 80
     },
     {
       "epoch": 0.17,
-      "grad_norm": 1.0254610776901245,
-      "learning_rate": 3.52e-05,
-      "loss": 0.6408,
+      "grad_norm": 4.151589393615723,
+      "learning_rate": 3.56e-05,
+      "loss": 0.6523,
       "step": 90
     },
     {
       "epoch": 0.19,
-      "grad_norm": 1.074569821357727,
-      "learning_rate": 3.9200000000000004e-05,
-      "loss": 0.6446,
+      "grad_norm": 4.241616249084473,
+      "learning_rate": 3.960000000000001e-05,
+      "loss": 0.6785,
       "step": 100
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.3864551782608032,
-      "learning_rate": 4.32e-05,
-      "loss": 0.6591,
+      "grad_norm": 4.035406112670898,
+      "learning_rate": 4.36e-05,
+      "loss": 0.6779,
       "step": 110
     },
     {
       "epoch": 0.23,
-      "grad_norm": 1.1516258716583252,
-      "learning_rate": 4.72e-05,
-      "loss": 0.6288,
+      "grad_norm": 4.655460357666016,
+      "learning_rate": 4.76e-05,
+      "loss": 0.6418,
       "step": 120
     },
     {
       "epoch": 0.24,
-      "grad_norm": 1.2458890676498413,
-      "learning_rate": 5.1200000000000004e-05,
-      "loss": 0.6325,
+      "grad_norm": 3.683321952819824,
+      "learning_rate": 5.16e-05,
+      "loss": 0.6415,
       "step": 130
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.7807619571685791,
-      "learning_rate": 5.520000000000001e-05,
-      "loss": 0.662,
+      "grad_norm": 5.034379959106445,
+      "learning_rate": 5.560000000000001e-05,
+      "loss": 0.7168,
       "step": 140
     },
     {
       "epoch": 0.28,
-      "grad_norm": 4.871969699859619,
-      "learning_rate": 5.88e-05,
-      "loss": 0.6877,
+      "grad_norm": 7.606465816497803,
+      "learning_rate": 5.96e-05,
+      "loss": 0.7146,
       "step": 150
     },
     {
       "epoch": 0.3,
-      "grad_norm": 2.10189151763916,
-      "learning_rate": 6.280000000000001e-05,
-      "loss": 0.6344,
+      "grad_norm": 3.5599539279937744,
+      "learning_rate": 6.36e-05,
+      "loss": 0.6495,
       "step": 160
     },
     {
       "epoch": 0.32,
-      "grad_norm": 5.75629186630249,
-      "learning_rate": 6.680000000000001e-05,
-      "loss": 0.6487,
+      "grad_norm": 8.060068130493164,
+      "learning_rate": 6.76e-05,
+      "loss": 0.6523,
       "step": 170
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.9313517212867737,
-      "learning_rate": 7.08e-05,
-      "loss": 0.5723,
+      "grad_norm": 2.229842185974121,
+      "learning_rate": 7.16e-05,
+      "loss": 0.6252,
       "step": 180
     },
     {
       "epoch": 0.36,
-      "grad_norm": 3.9287912845611572,
-      "learning_rate": 7.48e-05,
-      "loss": 0.6724,
+      "grad_norm": 2.0872910022735596,
+      "learning_rate": 7.560000000000001e-05,
+      "loss": 0.6729,
       "step": 190
     },
     {
       "epoch": 0.38,
-      "grad_norm": 4.517734527587891,
-      "learning_rate": 7.88e-05,
-      "loss": 0.647,
+      "grad_norm": 5.176510334014893,
+      "learning_rate": 7.960000000000001e-05,
+      "loss": 0.6706,
       "step": 200
     },
     {
       "epoch": 0.39,
-      "grad_norm": 1.1182656288146973,
-      "learning_rate": 8.28e-05,
-      "loss": 0.6423,
+      "grad_norm": 3.0015692710876465,
+      "learning_rate": 8.36e-05,
+      "loss": 0.6279,
       "step": 210
     },
     {
       "epoch": 0.41,
-      "grad_norm": 3.048739194869995,
-      "learning_rate": 8.680000000000001e-05,
-      "loss": 0.5778,
+      "grad_norm": 3.563908576965332,
+      "learning_rate": 8.76e-05,
+      "loss": 0.5934,
       "step": 220
     },
     {
       "epoch": 0.43,
-      "grad_norm": 1.1923900842666626,
-      "learning_rate": 9.080000000000001e-05,
-      "loss": 0.6203,
+      "grad_norm": 4.134763240814209,
+      "learning_rate": 9.16e-05,
+      "loss": 0.6436,
       "step": 230
     },
     {
       "epoch": 0.45,
-      "grad_norm": 2.969618082046509,
-      "learning_rate": 9.48e-05,
-      "loss": 0.6514,
+      "grad_norm": 6.163305759429932,
+      "learning_rate": 9.56e-05,
+      "loss": 0.6519,
       "step": 240
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.6900493502616882,
-      "learning_rate": 9.88e-05,
-      "loss": 0.6581,
+      "grad_norm": 4.374207019805908,
+      "learning_rate": 9.960000000000001e-05,
+      "loss": 0.6637,
       "step": 250
     },
     {
       "epoch": 0.49,
-      "grad_norm": 2.1176040172576904,
-      "learning_rate": 0.0001028,
-      "loss": 0.6104,
+      "grad_norm": 4.880618572235107,
+      "learning_rate": 0.00010360000000000001,
+      "loss": 0.6486,
       "step": 260
     },
     {
       "epoch": 0.51,
-      "grad_norm": 1.029479742050171,
-      "learning_rate": 0.00010680000000000001,
-      "loss": 0.5896,
+      "grad_norm": 1.537030577659607,
+      "learning_rate": 0.00010760000000000001,
+      "loss": 0.622,
       "step": 270
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.8468737602233887,
-      "learning_rate": 0.00011080000000000001,
-      "loss": 0.6067,
+      "grad_norm": 3.5295188426971436,
+      "learning_rate": 0.00011160000000000002,
+      "loss": 0.6234,
       "step": 280
     },
     {
       "epoch": 0.55,
-      "grad_norm": 1.5066865682601929,
-      "learning_rate": 0.0001148,
-      "loss": 0.6,
+      "grad_norm": 1.5816266536712646,
+      "learning_rate": 0.00011559999999999999,
+      "loss": 0.6321,
       "step": 290
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.6331378817558289,
-      "learning_rate": 0.0001188,
-      "loss": 0.6437,
+      "grad_norm": 1.2675076723098755,
+      "learning_rate": 0.0001192,
+      "loss": 0.7066,
       "step": 300
     },
     {
       "epoch": 0.58,
-      "grad_norm": 2.0694966316223145,
-      "learning_rate": 0.0001228,
-      "loss": 0.6304,
+      "grad_norm": 2.379913568496704,
+      "learning_rate": 0.0001232,
+      "loss": 0.6625,
       "step": 310
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.4095279574394226,
-      "learning_rate": 0.00012680000000000002,
-      "loss": 0.5867,
+      "grad_norm": 2.8097121715545654,
+      "learning_rate": 0.0001272,
+      "loss": 0.6018,
       "step": 320
     },
     {
       "epoch": 0.62,
-      "grad_norm": 3.9158384799957275,
-      "learning_rate": 0.0001308,
-      "loss": 0.6397,
+      "grad_norm": 6.209559917449951,
+      "learning_rate": 0.00013120000000000002,
+      "loss": 0.6946,
       "step": 330
     },
     {
       "epoch": 0.64,
-      "grad_norm": 1.578360915184021,
-      "learning_rate": 0.00013480000000000002,
-      "loss": 0.6616,
+      "grad_norm": 4.973850727081299,
+      "learning_rate": 0.0001352,
+      "loss": 0.6724,
       "step": 340
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.3877038359642029,
-      "learning_rate": 0.00013879999999999999,
-      "loss": 0.6161,
+      "grad_norm": 1.5640188455581665,
+      "learning_rate": 0.0001392,
+      "loss": 0.6227,
       "step": 350
     },
     {
       "epoch": 0.68,
-      "grad_norm": 1.0857969522476196,
-      "learning_rate": 0.0001428,
-      "loss": 0.6281,
+      "grad_norm": 3.425607442855835,
+      "learning_rate": 0.0001432,
+      "loss": 0.6515,
       "step": 360
     },
     {
       "epoch": 0.7,
-      "grad_norm": 4.571174621582031,
-      "learning_rate": 0.00014680000000000002,
-      "loss": 0.6338,
+      "grad_norm": 2.7312209606170654,
+      "learning_rate": 0.0001472,
+      "loss": 0.6477,
       "step": 370
     },
     {
       "epoch": 0.71,
-      "grad_norm": 1.5768461227416992,
-      "learning_rate": 0.0001508,
-      "loss": 0.6602,
+      "grad_norm": 2.8867738246917725,
+      "learning_rate": 0.00015120000000000002,
+      "loss": 0.6713,
       "step": 380
     },
     {
       "epoch": 0.73,
-      "grad_norm": 1.675762414932251,
-      "learning_rate": 0.00015480000000000002,
-      "loss": 0.6649,
+      "grad_norm": 0.647611141204834,
+      "learning_rate": 0.0001552,
+      "loss": 0.6554,
       "step": 390
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.6675221920013428,
-      "learning_rate": 0.0001588,
-      "loss": 0.6342,
+      "grad_norm": 1.5883524417877197,
+      "learning_rate": 0.00015920000000000002,
+      "loss": 0.6394,
       "step": 400
     },
     {
       "epoch": 0.77,
-      "grad_norm": 3.4393932819366455,
-      "learning_rate": 0.0001628,
-      "loss": 0.6678,
+      "grad_norm": 3.988485097885132,
+      "learning_rate": 0.0001632,
+      "loss": 0.6706,
       "step": 410
     },
     {
       "epoch": 0.79,
-      "grad_norm": 4.380407810211182,
-      "learning_rate": 0.0001668,
-      "loss": 0.6496,
+      "grad_norm": 7.253389358520508,
+      "learning_rate": 0.0001672,
+      "loss": 0.6658,
       "step": 420
     },
     {
       "epoch": 0.81,
-      "grad_norm": 1.5338869094848633,
-      "learning_rate": 0.0001708,
-      "loss": 0.6416,
+      "grad_norm": 0.7725067138671875,
+      "learning_rate": 0.00017120000000000001,
+      "loss": 0.7349,
       "step": 430
     },
     {
       "epoch": 0.83,
-      "grad_norm": 4.133732318878174,
-      "learning_rate": 0.00017480000000000002,
-      "loss": 0.6142,
+      "grad_norm": 3.858313798904419,
+      "learning_rate": 0.0001752,
+      "loss": 0.5985,
       "step": 440
     },
     {
       "epoch": 0.85,
-      "grad_norm": 1.7674825191497803,
-      "learning_rate": 0.0001788,
-      "loss": 0.5776,
+      "grad_norm": 3.0623714923858643,
+      "learning_rate": 0.00017920000000000002,
+      "loss": 0.5857,
       "step": 450
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.7282730937004089,
-      "learning_rate": 0.00018280000000000003,
-      "loss": 0.6721,
+      "grad_norm": 2.89859938621521,
+      "learning_rate": 0.0001832,
+      "loss": 0.7285,
       "step": 460
     },
     {
       "epoch": 0.88,
-      "grad_norm": 3.9304282665252686,
-      "learning_rate": 0.00018680000000000001,
-      "loss": 0.6354,
+      "grad_norm": 5.508333683013916,
+      "learning_rate": 0.00018720000000000002,
+      "loss": 0.643,
       "step": 470
     },
     {
       "epoch": 0.9,
-      "grad_norm": 2.655442476272583,
-      "learning_rate": 0.0001908,
-      "loss": 0.6835,
+      "grad_norm": 1.188262939453125,
+      "learning_rate": 0.0001912,
+      "loss": 0.664,
       "step": 480
     },
     {
       "epoch": 0.92,
-      "grad_norm": 3.454685926437378,
-      "learning_rate": 0.0001948,
-      "loss": 0.6148,
+      "grad_norm": 2.729386329650879,
+      "learning_rate": 0.0001952,
+      "loss": 0.622,
       "step": 490
     },
     {
       "epoch": 0.94,
-      "grad_norm": 2.250034809112549,
-      "learning_rate": 0.0001988,
-      "loss": 0.6026,
+      "grad_norm": 1.0896118879318237,
+      "learning_rate": 0.00019920000000000002,
+      "loss": 0.5781,
       "step": 500
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.33308330178260803,
-      "learning_rate": 0.0001996272630457934,
-      "loss": 0.6638,
+      "grad_norm": 1.856227159500122,
+      "learning_rate": 0.0001995740149094782,
+      "loss": 0.6579,
       "step": 510
     },
     {
       "epoch": 0.98,
-      "grad_norm": 2.148402214050293,
-      "learning_rate": 0.00019909478168264112,
-      "loss": 0.6184,
+      "grad_norm": 0.9613101482391357,
+      "learning_rate": 0.0001990415335463259,
+      "loss": 0.6003,
       "step": 520
     },
     {
       "epoch": 1.0,
-      "grad_norm": 4.5404582023620605,
-      "learning_rate": 0.00019856230031948884,
-      "loss": 0.6483,
+      "grad_norm": 2.3655054569244385,
+      "learning_rate": 0.0001985090521831736,
+      "loss": 0.6407,
       "step": 530
     },
     {
       "epoch": 1.0,
-      "eval_accuracy": 0.5,
-      "eval_loss": 0.7933283448219299,
-      "eval_runtime": 5.2926,
-      "eval_samples_per_second": 94.472,
-      "eval_steps_per_second": 3.023,
+      "eval_loss": 0.6965588331222534,
+      "eval_runtime": 5.5301,
+      "eval_samples_per_second": 90.415,
+      "eval_steps_per_second": 2.893,
       "step": 532
     },
     {
       "epoch": 1.02,
-      "grad_norm": 0.8757955431938171,
-      "learning_rate": 0.00019802981895633654,
-      "loss": 0.56,
+      "grad_norm": 4.0246710777282715,
+      "learning_rate": 0.0001979765708200213,
+      "loss": 0.6144,
       "step": 540
     },
     {
       "epoch": 1.03,
-      "grad_norm": 2.5476624965667725,
-      "learning_rate": 0.00019749733759318424,
-      "loss": 0.7584,
+      "grad_norm": 0.397265762090683,
+      "learning_rate": 0.000197444089456869,
+      "loss": 0.7252,
       "step": 550
     },
     {
       "epoch": 1.05,
-      "grad_norm": 4.550638675689697,
-      "learning_rate": 0.00019696485623003196,
-      "loss": 0.6235,
+      "grad_norm": 6.538491249084473,
+      "learning_rate": 0.00019691160809371674,
+      "loss": 0.6467,
       "step": 560
     },
     {
       "epoch": 1.07,
-      "grad_norm": 5.46172571182251,
-      "learning_rate": 0.00019643237486687966,
-      "loss": 0.6473,
+      "grad_norm": 5.099808692932129,
+      "learning_rate": 0.00019637912673056444,
+      "loss": 0.6668,
       "step": 570
     },
     {
       "epoch": 1.09,
-      "grad_norm": 2.1486427783966064,
-      "learning_rate": 0.00019589989350372736,
-      "loss": 0.6259,
+      "grad_norm": 2.9582018852233887,
+      "learning_rate": 0.00019584664536741216,
+      "loss": 0.6395,
       "step": 580
     },
     {
       "epoch": 1.11,
-      "grad_norm": 4.4022955894470215,
-      "learning_rate": 0.0001953674121405751,
-      "loss": 0.6031,
+      "grad_norm": 6.390412330627441,
+      "learning_rate": 0.00019531416400425986,
+      "loss": 0.6392,
       "step": 590
     },
     {
       "epoch": 1.13,
-      "grad_norm": 1.3872004747390747,
-      "learning_rate": 0.0001948349307774228,
-      "loss": 0.5789,
+      "grad_norm": 0.7637470960617065,
+      "learning_rate": 0.00019478168264110756,
+      "loss": 0.6227,
       "step": 600
     },
     {
       "epoch": 1.15,
-      "grad_norm": 1.193093180656433,
-      "learning_rate": 0.0001943024494142705,
-      "loss": 0.6409,
+      "grad_norm": 3.5703885555267334,
+      "learning_rate": 0.00019424920127795528,
+      "loss": 0.6981,
       "step": 610
     },
     {
       "epoch": 1.17,
-      "grad_norm": 4.4846415519714355,
-      "learning_rate": 0.0001937699680511182,
-      "loss": 0.6433,
+      "grad_norm": 0.8668873310089111,
+      "learning_rate": 0.000193716719914803,
+      "loss": 0.6175,
       "step": 620
     },
     {
       "epoch": 1.18,
-      "grad_norm": 1.3275984525680542,
-      "learning_rate": 0.00019323748668796594,
-      "loss": 0.6853,
+      "grad_norm": 2.3757476806640625,
+      "learning_rate": 0.0001931842385516507,
+      "loss": 0.6463,
       "step": 630
     },
     {
       "epoch": 1.2,
-      "grad_norm": 0.5516121983528137,
-      "learning_rate": 0.00019270500532481363,
-      "loss": 0.6577,
+      "grad_norm": 0.4950419068336487,
+      "learning_rate": 0.0001926517571884984,
+      "loss": 0.6656,
       "step": 640
     },
     {
       "epoch": 1.22,
-      "grad_norm": 0.6763603091239929,
-      "learning_rate": 0.00019217252396166133,
-      "loss": 0.6181,
+      "grad_norm": 0.4995131492614746,
+      "learning_rate": 0.00019211927582534613,
+      "loss": 0.6087,
       "step": 650
     },
     {
       "epoch": 1.24,
-      "grad_norm": 0.7651398181915283,
-      "learning_rate": 0.00019164004259850906,
-      "loss": 0.6698,
+      "grad_norm": 4.226231098175049,
+      "learning_rate": 0.00019158679446219383,
+      "loss": 0.6768,
       "step": 660
     },
     {
       "epoch": 1.26,
-      "grad_norm": 6.038695335388184,
-      "learning_rate": 0.00019110756123535678,
-      "loss": 0.6425,
+      "grad_norm": 4.077276706695557,
+      "learning_rate": 0.00019105431309904153,
+      "loss": 0.6551,
       "step": 670
     },
     {
       "epoch": 1.28,
-      "grad_norm": 1.9138271808624268,
-      "learning_rate": 0.00019057507987220448,
-      "loss": 0.6428,
+      "grad_norm": 0.6081713438034058,
+      "learning_rate": 0.00019052183173588926,
+      "loss": 0.6334,
       "step": 680
     },
     {
       "epoch": 1.3,
-      "grad_norm": 2.1620535850524902,
-      "learning_rate": 0.00019004259850905218,
-      "loss": 0.6643,
+      "grad_norm": 1.841720461845398,
+      "learning_rate": 0.00018998935037273698,
+      "loss": 0.6541,
       "step": 690
     },
     {
       "epoch": 1.32,
-      "grad_norm": 0.9087329506874084,
-      "learning_rate": 0.0001895101171458999,
-      "loss": 0.6042,
+      "grad_norm": 0.833160400390625,
+      "learning_rate": 0.00018945686900958468,
+      "loss": 0.5732,
       "step": 700
     },
     {
       "epoch": 1.33,
-      "grad_norm": 3.8417811393737793,
-      "learning_rate": 0.0001889776357827476,
-      "loss": 0.7502,
+      "grad_norm": 1.2434287071228027,
+      "learning_rate": 0.00018892438764643238,
+      "loss": 0.7005,
       "step": 710
     },
     {
       "epoch": 1.35,
-      "grad_norm": 3.6006288528442383,
-      "learning_rate": 0.0001884451544195953,
-      "loss": 0.6141,
+      "grad_norm": 1.5411890745162964,
+      "learning_rate": 0.0001883919062832801,
+      "loss": 0.604,
       "step": 720
     },
     {
       "epoch": 1.37,
-      "grad_norm": 1.9083914756774902,
-      "learning_rate": 0.00018791267305644303,
-      "loss": 0.6221,
+      "grad_norm": 0.5615927577018738,
+      "learning_rate": 0.0001878594249201278,
+      "loss": 0.6982,
       "step": 730
     },
     {
       "epoch": 1.39,
-      "grad_norm": 2.8606581687927246,
-      "learning_rate": 0.00018738019169329076,
-      "loss": 0.593,
+      "grad_norm": 1.1369315385818481,
+      "learning_rate": 0.0001873269435569755,
+      "loss": 0.6451,
       "step": 740
     },
     {
       "epoch": 1.41,
-      "grad_norm": 1.3666915893554688,
-      "learning_rate": 0.00018684771033013845,
-      "loss": 0.6307,
+      "grad_norm": 0.7013240456581116,
+      "learning_rate": 0.00018679446219382323,
+      "loss": 0.6697,
       "step": 750
     },
     {
       "epoch": 1.43,
-      "grad_norm": 0.7459832429885864,
-      "learning_rate": 0.00018631522896698615,
-      "loss": 0.6774,
+      "grad_norm": 0.7162796854972839,
+      "learning_rate": 0.00018626198083067095,
+      "loss": 0.6793,
       "step": 760
     },
     {
       "epoch": 1.45,
-      "grad_norm": 0.8331255912780762,
-      "learning_rate": 0.00018578274760383388,
-      "loss": 0.6637,
+      "grad_norm": 0.7037357091903687,
+      "learning_rate": 0.00018572949946751865,
+      "loss": 0.6546,
       "step": 770
     },
     {
       "epoch": 1.47,
-      "grad_norm": 4.606655120849609,
-      "learning_rate": 0.00018525026624068158,
-      "loss": 0.5936,
+      "grad_norm": 3.4446752071380615,
+      "learning_rate": 0.00018519701810436635,
+      "loss": 0.5589,
       "step": 780
     },
     {
       "epoch": 1.48,
-      "grad_norm": 0.44328516721725464,
-      "learning_rate": 0.00018471778487752928,
-      "loss": 0.6676,
+      "grad_norm": 2.896850109100342,
+      "learning_rate": 0.00018466453674121408,
+      "loss": 0.6368,
       "step": 790
     },
     {
       "epoch": 1.5,
-      "grad_norm": 4.314159393310547,
-      "learning_rate": 0.00018418530351437703,
-      "loss": 0.7248,
+      "grad_norm": 0.8148677945137024,
+      "learning_rate": 0.00018413205537806177,
+      "loss": 0.6817,
       "step": 800
     },
     {
       "epoch": 1.52,
-      "grad_norm": 1.6179046630859375,
-      "learning_rate": 0.00018365282215122473,
-      "loss": 0.673,
+      "grad_norm": 1.7753351926803589,
+      "learning_rate": 0.00018359957401490947,
+      "loss": 0.6491,
       "step": 810
     },
     {
       "epoch": 1.54,
-      "grad_norm": 4.103793144226074,
-      "learning_rate": 0.00018312034078807243,
-      "loss": 0.6637,
+      "grad_norm": 1.8540890216827393,
+      "learning_rate": 0.0001830670926517572,
+      "loss": 0.6578,
       "step": 820
     },
     {
       "epoch": 1.56,
-      "grad_norm": 0.7133169770240784,
-      "learning_rate": 0.00018258785942492015,
-      "loss": 0.6193,
+      "grad_norm": 1.7155886888504028,
+      "learning_rate": 0.00018253461128860492,
+      "loss": 0.6473,
       "step": 830
     },
     {
       "epoch": 1.58,
-      "grad_norm": 1.252029299736023,
-      "learning_rate": 0.00018205537806176785,
-      "loss": 0.594,
+      "grad_norm": 0.892581582069397,
+      "learning_rate": 0.00018200212992545262,
+      "loss": 0.6021,
       "step": 840
     },
     {
       "epoch": 1.6,
-      "grad_norm": 0.786902904510498,
-      "learning_rate": 0.00018152289669861555,
-      "loss": 0.5598,
+      "grad_norm": 0.7859013080596924,
+      "learning_rate": 0.00018146964856230032,
+      "loss": 0.6064,
       "step": 850
     },
     {
       "epoch": 1.62,
-      "grad_norm": 5.060324668884277,
-      "learning_rate": 0.00018099041533546325,
-      "loss": 0.6586,
+      "grad_norm": 6.423700332641602,
+      "learning_rate": 0.00018093716719914805,
+      "loss": 0.612,
       "step": 860
     },
     {
       "epoch": 1.64,
-      "grad_norm": 2.4381039142608643,
-      "learning_rate": 0.000180457933972311,
-      "loss": 0.6487,
+      "grad_norm": 1.622660517692566,
+      "learning_rate": 0.00018040468583599575,
+      "loss": 0.6688,
       "step": 870
     },
     {
       "epoch": 1.65,
-      "grad_norm": 2.3903331756591797,
-      "learning_rate": 0.0001799254526091587,
-      "loss": 0.624,
+      "grad_norm": 1.050557255744934,
+      "learning_rate": 0.00017987220447284344,
+      "loss": 0.6549,
       "step": 880
     },
     {
       "epoch": 1.67,
-      "grad_norm": 1.2366740703582764,
-      "learning_rate": 0.0001793929712460064,
-      "loss": 0.5997,
+      "grad_norm": 3.1418724060058594,
+      "learning_rate": 0.00017933972310969117,
+      "loss": 0.6104,
       "step": 890
     },
     {
       "epoch": 1.69,
-      "grad_norm": 3.3629353046417236,
-      "learning_rate": 0.00017886048988285412,
-      "loss": 0.5978,
+      "grad_norm": 3.0087502002716064,
+      "learning_rate": 0.0001788072417465389,
+      "loss": 0.6451,
       "step": 900
     },
     {
       "epoch": 1.71,
-      "grad_norm": 2.4150805473327637,
-      "learning_rate": 0.00017832800851970182,
-      "loss": 0.6342,
+      "grad_norm": 3.000000476837158,
+      "learning_rate": 0.0001782747603833866,
+      "loss": 0.6247,
       "step": 910
     },
     {
       "epoch": 1.73,
-      "grad_norm": 0.2921897768974304,
-      "learning_rate": 0.00017779552715654952,
-      "loss": 0.655,
+      "grad_norm": 1.215047001838684,
+      "learning_rate": 0.0001777422790202343,
+      "loss": 0.6418,
       "step": 920
     },
     {
       "epoch": 1.75,
-      "grad_norm": 2.8717799186706543,
-      "learning_rate": 0.00017726304579339722,
-      "loss": 0.6383,
+      "grad_norm": 2.694948434829712,
+      "learning_rate": 0.00017720979765708202,
+      "loss": 0.657,
       "step": 930
     },
     {
       "epoch": 1.77,
-      "grad_norm": 1.2297135591506958,
-      "learning_rate": 0.00017673056443024497,
-      "loss": 0.6154,
+      "grad_norm": 3.017652988433838,
+      "learning_rate": 0.00017667731629392972,
+      "loss": 0.6534,
       "step": 940
     },
     {
       "epoch": 1.79,
-      "grad_norm": 2.8782236576080322,
-      "learning_rate": 0.00017619808306709267,
-      "loss": 0.6469,
+      "grad_norm": 3.342782497406006,
+      "learning_rate": 0.00017614483493077742,
+      "loss": 0.66,
       "step": 950
     },
     {
       "epoch": 1.8,
-      "grad_norm": 3.708266019821167,
-      "learning_rate": 0.00017566560170394037,
-      "loss": 0.6462,
+      "grad_norm": 3.4681971073150635,
+      "learning_rate": 0.00017561235356762514,
+      "loss": 0.6681,
       "step": 960
     },
     {
       "epoch": 1.82,
-      "grad_norm": 0.30842864513397217,
-      "learning_rate": 0.0001751331203407881,
-      "loss": 0.5934,
+      "grad_norm": 0.638871431350708,
+      "learning_rate": 0.00017507987220447287,
+      "loss": 0.6062,
       "step": 970
     },
     {
       "epoch": 1.84,
-      "grad_norm": 0.9341667294502258,
-      "learning_rate": 0.0001746006389776358,
-      "loss": 0.5741,
+      "grad_norm": 0.615951657295227,
+      "learning_rate": 0.00017454739084132057,
+      "loss": 0.5745,
       "step": 980
     },
     {
       "epoch": 1.86,
-      "grad_norm": 1.7768970727920532,
-      "learning_rate": 0.0001740681576144835,
-      "loss": 0.6028,
+      "grad_norm": 2.9999630451202393,
+      "learning_rate": 0.00017401490947816826,
+      "loss": 0.6137,
       "step": 990
     },
     {
       "epoch": 1.88,
-      "grad_norm": 4.887693881988525,
-      "learning_rate": 0.0001735356762513312,
-      "loss": 0.6171,
+      "grad_norm": 3.212564468383789,
+      "learning_rate": 0.000173482428115016,
+      "loss": 0.5914,
       "step": 1000
     },
     {
       "epoch": 1.9,
-      "grad_norm": 2.1910202503204346,
-      "learning_rate": 0.00017300319488817894,
-      "loss": 0.6039,
+      "grad_norm": 2.063244104385376,
+      "learning_rate": 0.0001729499467518637,
+      "loss": 0.5904,
       "step": 1010
     },
     {
       "epoch": 1.92,
-      "grad_norm": 5.127533912658691,
-      "learning_rate": 0.00017247071352502664,
-      "loss": 0.6404,
+      "grad_norm": 4.356505393981934,
+      "learning_rate": 0.00017241746538871139,
+      "loss": 0.6904,
       "step": 1020
     },
     {
       "epoch": 1.94,
-      "grad_norm": 4.148878574371338,
-      "learning_rate": 0.00017193823216187434,
-      "loss": 0.6367,
+      "grad_norm": 2.85433292388916,
+      "learning_rate": 0.0001718849840255591,
+      "loss": 0.6621,
       "step": 1030
     },
     {
       "epoch": 1.95,
-      "grad_norm": 0.9742037057876587,
-      "learning_rate": 0.00017140575079872207,
-      "loss": 0.6539,
+      "grad_norm": 0.8630966544151306,
+      "learning_rate": 0.00017135250266240684,
+      "loss": 0.6591,
       "step": 1040
     },
     {
       "epoch": 1.97,
-      "grad_norm": 7.3390913009643555,
-      "learning_rate": 0.00017087326943556976,
-      "loss": 0.6369,
+      "grad_norm": 7.148132801055908,
+      "learning_rate": 0.00017082002129925454,
+      "loss": 0.6562,
       "step": 1050
     },
     {
       "epoch": 1.99,
-      "grad_norm": 0.8897514343261719,
-      "learning_rate": 0.00017034078807241746,
-      "loss": 0.6203,
+      "grad_norm": 0.8249239921569824,
+      "learning_rate": 0.00017028753993610223,
+      "loss": 0.5847,
       "step": 1060
     },
     {
       "epoch": 2.0,
-      "eval_accuracy": 0.498,
-      "eval_loss": 0.746105968952179,
-      "eval_runtime": 5.0077,
-      "eval_samples_per_second": 99.846,
-      "eval_steps_per_second": 3.195,
+      "eval_loss": 0.7979968190193176,
+      "eval_runtime": 4.8835,
+      "eval_samples_per_second": 102.387,
+      "eval_steps_per_second": 3.276,
       "step": 1064
     },
     {
       "epoch": 2.01,
-      "grad_norm": 5.566906452178955,
-      "learning_rate": 0.0001698083067092652,
-      "loss": 0.7104,
+      "grad_norm": 0.5970304012298584,
+      "learning_rate": 0.00016975505857294996,
+      "loss": 0.6398,
       "step": 1070
     },
     {
       "epoch": 2.03,
-      "grad_norm": 4.016073703765869,
-      "learning_rate": 0.00016927582534611291,
-      "loss": 0.6763,
+      "grad_norm": 2.641237258911133,
+      "learning_rate": 0.00016922257720979766,
+      "loss": 0.631,
       "step": 1080
     },
     {
       "epoch": 2.05,
-      "grad_norm": 0.48662686347961426,
-      "learning_rate": 0.0001687433439829606,
-      "loss": 0.6283,
+      "grad_norm": 0.5415599942207336,
+      "learning_rate": 0.00016869009584664536,
+      "loss": 0.6337,
       "step": 1090
     },
     {
       "epoch": 2.07,
-      "grad_norm": 3.80304217338562,
-      "learning_rate": 0.0001682108626198083,
-      "loss": 0.5875,
+      "grad_norm": 5.68765115737915,
+      "learning_rate": 0.00016815761448349308,
+      "loss": 0.6154,
       "step": 1100
     },
     {
       "epoch": 2.09,
-      "grad_norm": 3.6678249835968018,
-      "learning_rate": 0.00016767838125665604,
-      "loss": 0.6383,
+      "grad_norm": 3.6129837036132812,
+      "learning_rate": 0.0001676251331203408,
+      "loss": 0.68,
       "step": 1110
     },
     {
       "epoch": 2.11,
-      "grad_norm": 0.7282447814941406,
-      "learning_rate": 0.00016714589989350374,
-      "loss": 0.5975,
+      "grad_norm": 0.3377436399459839,
+      "learning_rate": 0.0001670926517571885,
+      "loss": 0.6461,
       "step": 1120
     },
     {
       "epoch": 2.12,
-      "grad_norm": 5.8518829345703125,
-      "learning_rate": 0.00016661341853035143,
-      "loss": 0.6223,
+      "grad_norm": 5.473901271820068,
+      "learning_rate": 0.0001665601703940362,
+      "loss": 0.6727,
       "step": 1130
     },
     {
       "epoch": 2.14,
-      "grad_norm": 0.6626885533332825,
-      "learning_rate": 0.00016608093716719916,
-      "loss": 0.6059,
+      "grad_norm": 0.664064884185791,
+      "learning_rate": 0.00016602768903088393,
+      "loss": 0.5921,
       "step": 1140
     },
     {
       "epoch": 2.16,
-      "grad_norm": 2.4838600158691406,
-      "learning_rate": 0.00016554845580404689,
-      "loss": 0.6218,
+      "grad_norm": 1.5953819751739502,
+      "learning_rate": 0.00016549520766773163,
+      "loss": 0.6052,
       "step": 1150
     },
     {
       "epoch": 2.18,
-      "grad_norm": 3.6144859790802,
-      "learning_rate": 0.00016501597444089458,
-      "loss": 0.601,
+      "grad_norm": 2.9201242923736572,
+      "learning_rate": 0.00016496272630457933,
+      "loss": 0.6278,
       "step": 1160
     },
     {
       "epoch": 2.2,
-      "grad_norm": 0.9720486998558044,
-      "learning_rate": 0.00016448349307774228,
-      "loss": 0.5471,
+      "grad_norm": 2.1254889965057373,
+      "learning_rate": 0.00016443024494142705,
+      "loss": 0.5732,
       "step": 1170
     },
     {
       "epoch": 2.22,
-      "grad_norm": 5.14749002456665,
-      "learning_rate": 0.00016395101171459,
-      "loss": 0.6341,
+      "grad_norm": 4.827600002288818,
+      "learning_rate": 0.00016389776357827478,
+      "loss": 0.5687,
       "step": 1180
     },
     {
       "epoch": 2.24,
-      "grad_norm": 4.361828804016113,
-      "learning_rate": 0.0001634185303514377,
-      "loss": 0.5915,
+      "grad_norm": 4.18930721282959,
+      "learning_rate": 0.00016336528221512248,
+      "loss": 0.5923,
       "step": 1190
     },
     {
       "epoch": 2.26,
-      "grad_norm": 1.6119446754455566,
-      "learning_rate": 0.0001628860489882854,
-      "loss": 0.658,
+      "grad_norm": 3.059593439102173,
+      "learning_rate": 0.00016283280085197018,
+      "loss": 0.6697,
       "step": 1200
     },
     {
       "epoch": 2.27,
-      "grad_norm": 0.6508371233940125,
-      "learning_rate": 0.00016235356762513313,
-      "loss": 0.5626,
+      "grad_norm": 0.7229830622673035,
+      "learning_rate": 0.0001623003194888179,
+      "loss": 0.5956,
       "step": 1210
     },
     {
       "epoch": 2.29,
-      "grad_norm": 0.7702106237411499,
-      "learning_rate": 0.00016182108626198086,
-      "loss": 0.6118,
+      "grad_norm": 0.41750237345695496,
+      "learning_rate": 0.0001617678381256656,
+      "loss": 0.6543,
       "step": 1220
     },
     {
       "epoch": 2.31,
-      "grad_norm": 0.45979470014572144,
-      "learning_rate": 0.00016128860489882855,
-      "loss": 0.6097,
+      "grad_norm": 1.0170073509216309,
+      "learning_rate": 0.0001612353567625133,
+      "loss": 0.6222,
       "step": 1230
     },
     {
       "epoch": 2.33,
-      "grad_norm": 1.6143121719360352,
-      "learning_rate": 0.00016075612353567625,
-      "loss": 0.6675,
+      "grad_norm": 1.94419527053833,
+      "learning_rate": 0.00016070287539936103,
+      "loss": 0.6379,
       "step": 1240
     },
     {
       "epoch": 2.35,
-      "grad_norm": 2.4905166625976562,
-      "learning_rate": 0.00016022364217252398,
-      "loss": 0.5858,
+      "grad_norm": 2.2258355617523193,
+      "learning_rate": 0.00016017039403620875,
+      "loss": 0.5887,
       "step": 1250
     },
     {
       "epoch": 2.37,
-      "grad_norm": 0.7425104379653931,
-      "learning_rate": 0.00015969116080937168,
-      "loss": 0.5901,
+      "grad_norm": 1.861475944519043,
+      "learning_rate": 0.00015963791267305645,
+      "loss": 0.5848,
       "step": 1260
     },
     {
       "epoch": 2.39,
-      "grad_norm": 1.7030450105667114,
-      "learning_rate": 0.00015915867944621938,
-      "loss": 0.5834,
+      "grad_norm": 1.9490504264831543,
+      "learning_rate": 0.00015910543130990418,
+      "loss": 0.6283,
       "step": 1270
     },
     {
       "epoch": 2.41,
-      "grad_norm": 3.350250720977783,
-      "learning_rate": 0.0001586261980830671,
-      "loss": 0.554,
+      "grad_norm": 0.9739388823509216,
+      "learning_rate": 0.00015857294994675187,
+      "loss": 0.6027,
       "step": 1280
     },
     {
       "epoch": 2.42,
-      "grad_norm": 2.625653028488159,
-      "learning_rate": 0.00015809371671991483,
-      "loss": 0.5784,
+      "grad_norm": 1.012650966644287,
+      "learning_rate": 0.00015804046858359957,
+      "loss": 0.5685,
       "step": 1290
     },
     {
       "epoch": 2.44,
-      "grad_norm": 1.9641324281692505,
-      "learning_rate": 0.00015756123535676253,
-      "loss": 0.6203,
+      "grad_norm": 0.5264813303947449,
+      "learning_rate": 0.00015750798722044727,
+      "loss": 0.619,
       "step": 1300
     },
     {
       "epoch": 2.46,
-      "grad_norm": 1.912189245223999,
-      "learning_rate": 0.00015702875399361022,
-      "loss": 0.6344,
+      "grad_norm": 1.5455741882324219,
+      "learning_rate": 0.00015697550585729502,
+      "loss": 0.6441,
       "step": 1310
     },
     {
       "epoch": 2.48,
-      "grad_norm": 2.581179141998291,
-      "learning_rate": 0.00015649627263045795,
-      "loss": 0.6706,
+      "grad_norm": 1.331289529800415,
+      "learning_rate": 0.00015644302449414272,
+      "loss": 0.6583,
       "step": 1320
     },
     {
       "epoch": 2.5,
-      "grad_norm": 0.7273750901222229,
-      "learning_rate": 0.00015596379126730565,
-      "loss": 0.6587,
+      "grad_norm": 1.2682970762252808,
+      "learning_rate": 0.00015591054313099042,
+      "loss": 0.6573,
       "step": 1330
     },
     {
       "epoch": 2.52,
-      "grad_norm": 0.5495343804359436,
-      "learning_rate": 0.00015543130990415335,
-      "loss": 0.6005,
+      "grad_norm": 2.107792377471924,
+      "learning_rate": 0.00015537806176783815,
+      "loss": 0.6044,
       "step": 1340
     },
     {
       "epoch": 2.54,
-      "grad_norm": 2.7823123931884766,
-      "learning_rate": 0.00015489882854100107,
-      "loss": 0.5477,
+      "grad_norm": 0.5763486623764038,
+      "learning_rate": 0.00015484558040468585,
+      "loss": 0.5277,
       "step": 1350
     },
     {
       "epoch": 2.56,
-      "grad_norm": 2.560255527496338,
-      "learning_rate": 0.0001543663471778488,
-      "loss": 0.6153,
+      "grad_norm": 5.478113651275635,
+      "learning_rate": 0.00015431309904153354,
+      "loss": 0.6821,
       "step": 1360
     },
     {
       "epoch": 2.58,
-      "grad_norm": 4.055157661437988,
-      "learning_rate": 0.0001538338658146965,
-      "loss": 0.6093,
+      "grad_norm": 3.25028395652771,
+      "learning_rate": 0.00015378061767838124,
+      "loss": 0.6411,
       "step": 1370
     },
     {
       "epoch": 2.59,
-      "grad_norm": 1.9480986595153809,
-      "learning_rate": 0.0001533013844515442,
-      "loss": 0.653,
+      "grad_norm": 1.4305381774902344,
+      "learning_rate": 0.000153248136315229,
+      "loss": 0.6273,
       "step": 1380
     },
     {
       "epoch": 2.61,
-      "grad_norm": 0.6847740411758423,
-      "learning_rate": 0.00015276890308839192,
-      "loss": 0.5842,
+      "grad_norm": 1.6666404008865356,
+      "learning_rate": 0.0001527156549520767,
+      "loss": 0.6636,
       "step": 1390
     },
     {
       "epoch": 2.63,
-      "grad_norm": 5.935375690460205,
-      "learning_rate": 0.00015223642172523962,
-      "loss": 0.5934,
+      "grad_norm": 4.375661373138428,
+      "learning_rate": 0.0001521831735889244,
+      "loss": 0.5989,
       "step": 1400
     },
     {
       "epoch": 2.65,
-      "grad_norm": 2.9190566539764404,
-      "learning_rate": 0.00015170394036208732,
-      "loss": 0.5746,
+      "grad_norm": 3.827533483505249,
+      "learning_rate": 0.00015165069222577212,
+      "loss": 0.5814,
       "step": 1410
     },
     {
       "epoch": 2.67,
-      "grad_norm": 1.8768565654754639,
-      "learning_rate": 0.00015117145899893504,
-      "loss": 0.5792,
+      "grad_norm": 2.938608169555664,
+      "learning_rate": 0.00015111821086261982,
+      "loss": 0.5575,
       "step": 1420
     },
     {
       "epoch": 2.69,
-      "grad_norm": 0.8437117338180542,
-      "learning_rate": 0.00015063897763578277,
-      "loss": 0.6118,
+      "grad_norm": 2.966648817062378,
+      "learning_rate": 0.00015058572949946752,
+      "loss": 0.623,
       "step": 1430
     },
     {
       "epoch": 2.71,
-      "grad_norm": 2.211174488067627,
-      "learning_rate": 0.00015010649627263047,
-      "loss": 0.5693,
+      "grad_norm": 3.7637674808502197,
+      "learning_rate": 0.00015005324813631521,
+      "loss": 0.5792,
       "step": 1440
     },
     {
       "epoch": 2.73,
-      "grad_norm": 4.7668609619140625,
-      "learning_rate": 0.00014957401490947817,
-      "loss": 0.6926,
+      "grad_norm": 2.175443172454834,
+      "learning_rate": 0.00014952076677316297,
+      "loss": 0.6161,
       "step": 1450
     },
     {
       "epoch": 2.74,
-      "grad_norm": 1.8787531852722168,
-      "learning_rate": 0.0001490415335463259,
-      "loss": 0.5893,
+      "grad_norm": 0.6871302723884583,
+      "learning_rate": 0.00014898828541001067,
+      "loss": 0.6145,
       "step": 1460
     },
     {
       "epoch": 2.76,
-      "grad_norm": 3.1121323108673096,
-      "learning_rate": 0.0001485090521831736,
-      "loss": 0.6001,
+      "grad_norm": 2.2254390716552734,
+      "learning_rate": 0.00014845580404685836,
+      "loss": 0.5978,
       "step": 1470
     },
     {
       "epoch": 2.78,
-      "grad_norm": 1.5263334512710571,
-      "learning_rate": 0.0001479765708200213,
-      "loss": 0.6538,
+      "grad_norm": 2.5566136837005615,
+      "learning_rate": 0.0001479233226837061,
+      "loss": 0.6676,
       "step": 1480
     },
     {
       "epoch": 2.8,
-      "grad_norm": 1.4740121364593506,
-      "learning_rate": 0.00014744408945686902,
-      "loss": 0.633,
+      "grad_norm": 1.8439075946807861,
+      "learning_rate": 0.0001473908413205538,
+      "loss": 0.6386,
       "step": 1490
     },
     {
       "epoch": 2.82,
-      "grad_norm": 2.5917043685913086,
-      "learning_rate": 0.00014691160809371674,
-      "loss": 0.619,
+      "grad_norm": 2.6792383193969727,
+      "learning_rate": 0.0001468583599574015,
+      "loss": 0.6562,
       "step": 1500
     },
     {
       "epoch": 2.84,
-      "grad_norm": 1.5224074125289917,
-      "learning_rate": 0.00014637912673056444,
-      "loss": 0.5376,
+      "grad_norm": 0.6229275465011597,
+      "learning_rate": 0.0001463258785942492,
+      "loss": 0.5914,
       "step": 1510
     },
     {
       "epoch": 2.86,
-      "grad_norm": 1.7697654962539673,
-      "learning_rate": 0.00014584664536741214,
-      "loss": 0.5838,
+      "grad_norm": 5.081364631652832,
+      "learning_rate": 0.00014579339723109694,
+      "loss": 0.5784,
       "step": 1520
     },
     {
       "epoch": 2.88,
-      "grad_norm": 4.307724475860596,
-      "learning_rate": 0.00014531416400425986,
-      "loss": 0.5861,
+      "grad_norm": 4.535241603851318,
+      "learning_rate": 0.00014526091586794464,
+      "loss": 0.5823,
       "step": 1530
     },
     {
       "epoch": 2.89,
-      "grad_norm": 2.4162418842315674,
-      "learning_rate": 0.00014478168264110756,
-      "loss": 0.7074,
+      "grad_norm": 0.9447025060653687,
+      "learning_rate": 0.00014472843450479234,
+      "loss": 0.6573,
       "step": 1540
     },
     {
       "epoch": 2.91,
-      "grad_norm": 3.3249616622924805,
-      "learning_rate": 0.00014424920127795526,
-      "loss": 0.6251,
+      "grad_norm": 1.4323358535766602,
+      "learning_rate": 0.00014419595314164006,
+      "loss": 0.5481,
       "step": 1550
     },
     {
       "epoch": 2.93,
-      "grad_norm": 3.486440420150757,
-      "learning_rate": 0.000143716719914803,
-      "loss": 0.6949,
+      "grad_norm": 2.6219279766082764,
+      "learning_rate": 0.00014366347177848776,
+      "loss": 0.6331,
       "step": 1560
     },
     {
       "epoch": 2.95,
-      "grad_norm": 1.9070874452590942,
-      "learning_rate": 0.0001431842385516507,
-      "loss": 0.6384,
+      "grad_norm": 0.7964478135108948,
+      "learning_rate": 0.00014313099041533546,
+      "loss": 0.6071,
       "step": 1570
     },
     {
       "epoch": 2.97,
-      "grad_norm": 0.9392743706703186,
-      "learning_rate": 0.0001426517571884984,
-      "loss": 0.6292,
+      "grad_norm": 1.5989696979522705,
+      "learning_rate": 0.00014259850905218318,
+      "loss": 0.621,
       "step": 1580
     },
     {
       "epoch": 2.99,
-      "grad_norm": 1.008888840675354,
-      "learning_rate": 0.0001421192758253461,
-      "loss": 0.5955,
+      "grad_norm": 2.2981741428375244,
+      "learning_rate": 0.0001420660276890309,
+      "loss": 0.6144,
       "step": 1590
     },
     {
       "epoch": 3.0,
-      "eval_accuracy": 0.518,
-      "eval_loss": 0.7538835406303406,
-      "eval_runtime": 5.5893,
-      "eval_samples_per_second": 89.456,
-      "eval_steps_per_second": 2.863,
+      "eval_loss": 0.7764111161231995,
+      "eval_runtime": 4.9574,
+      "eval_samples_per_second": 100.859,
+      "eval_steps_per_second": 3.227,
       "step": 1596
     },
     {
       "epoch": 3.01,
-      "grad_norm": 1.014275074005127,
-      "learning_rate": 0.00014158679446219384,
-      "loss": 0.6254,
+      "grad_norm": 1.12291419506073,
+      "learning_rate": 0.0001415335463258786,
+      "loss": 0.6111,
       "step": 1600
     },
     {
       "epoch": 3.03,
-      "grad_norm": 3.639158248901367,
-      "learning_rate": 0.00014105431309904153,
-      "loss": 0.5718,
+      "grad_norm": 3.5445947647094727,
+      "learning_rate": 0.0001410010649627263,
+      "loss": 0.5686,
       "step": 1610
     },
     {
       "epoch": 3.05,
-      "grad_norm": 3.8647546768188477,
-      "learning_rate": 0.00014052183173588923,
-      "loss": 0.5798,
+      "grad_norm": 5.601056098937988,
+      "learning_rate": 0.00014046858359957403,
+      "loss": 0.5344,
       "step": 1620
     },
     {
       "epoch": 3.06,
-      "grad_norm": 2.522571086883545,
-      "learning_rate": 0.00013998935037273696,
-      "loss": 0.596,
+      "grad_norm": 2.273033380508423,
+      "learning_rate": 0.00013993610223642173,
+      "loss": 0.5662,
       "step": 1630
     },
     {
       "epoch": 3.08,
-      "grad_norm": 0.9586620330810547,
-      "learning_rate": 0.00013945686900958468,
-      "loss": 0.546,
+      "grad_norm": 0.932608962059021,
+      "learning_rate": 0.00013940362087326943,
+      "loss": 0.5788,
       "step": 1640
     },
     {
       "epoch": 3.1,
-      "grad_norm": 1.2226234674453735,
-      "learning_rate": 0.00013892438764643238,
-      "loss": 0.5209,
+      "grad_norm": 1.4313303232192993,
+      "learning_rate": 0.00013887113951011716,
+      "loss": 0.6199,
       "step": 1650
     },
     {
       "epoch": 3.12,
-      "grad_norm": 1.6508853435516357,
-      "learning_rate": 0.00013839190628328008,
-      "loss": 0.5763,
+      "grad_norm": 1.161855936050415,
+      "learning_rate": 0.00013833865814696488,
+      "loss": 0.5824,
       "step": 1660
     },
     {
       "epoch": 3.14,
-      "grad_norm": 15.09642505645752,
-      "learning_rate": 0.0001378594249201278,
-      "loss": 0.5614,
+      "grad_norm": 3.128478527069092,
+      "learning_rate": 0.00013780617678381258,
+      "loss": 0.5774,
       "step": 1670
     },
     {
       "epoch": 3.16,
-      "grad_norm": 2.6017839908599854,
-      "learning_rate": 0.0001373269435569755,
-      "loss": 0.6123,
+      "grad_norm": 2.3241376876831055,
+      "learning_rate": 0.00013727369542066028,
+      "loss": 0.6157,
       "step": 1680
     },
     {
       "epoch": 3.18,
-      "grad_norm": 2.848982572555542,
-      "learning_rate": 0.0001367944621938232,
-      "loss": 0.5747,
+      "grad_norm": 3.1155593395233154,
+      "learning_rate": 0.000136741214057508,
+      "loss": 0.6216,
       "step": 1690
     },
     {
       "epoch": 3.2,
-      "grad_norm": 2.5202741622924805,
-      "learning_rate": 0.00013626198083067093,
-      "loss": 0.5505,
+      "grad_norm": 1.408307433128357,
+      "learning_rate": 0.0001362087326943557,
+      "loss": 0.6149,
       "step": 1700
     },
     {
       "epoch": 3.21,
-      "grad_norm": 1.2070810794830322,
-      "learning_rate": 0.00013572949946751866,
-      "loss": 0.4952,
+      "grad_norm": 1.0534541606903076,
+      "learning_rate": 0.0001356762513312034,
+      "loss": 0.579,
       "step": 1710
     },
     {
       "epoch": 3.23,
-      "grad_norm": 1.999449610710144,
-      "learning_rate": 0.00013519701810436635,
-      "loss": 0.4874,
+      "grad_norm": 1.5528515577316284,
+      "learning_rate": 0.00013514376996805113,
+      "loss": 0.5386,
       "step": 1720
     },
     {
       "epoch": 3.25,
-      "grad_norm": 1.2801100015640259,
-      "learning_rate": 0.00013466453674121405,
-      "loss": 0.568,
+      "grad_norm": 2.1062605381011963,
+      "learning_rate": 0.00013461128860489885,
+      "loss": 0.6308,
       "step": 1730
     },
     {
       "epoch": 3.27,
-      "grad_norm": 1.8408716917037964,
-      "learning_rate": 0.00013413205537806178,
-      "loss": 0.5894,
+      "grad_norm": 2.3048410415649414,
+      "learning_rate": 0.00013407880724174655,
+      "loss": 0.6255,
       "step": 1740
     },
     {
       "epoch": 3.29,
-      "grad_norm": 2.0327749252319336,
-      "learning_rate": 0.00013359957401490948,
-      "loss": 0.5749,
+      "grad_norm": 3.4321556091308594,
+      "learning_rate": 0.00013354632587859425,
+      "loss": 0.6106,
       "step": 1750
     },
     {
       "epoch": 3.31,
-      "grad_norm": 1.4758570194244385,
-      "learning_rate": 0.00013306709265175718,
-      "loss": 0.5308,
+      "grad_norm": 1.6664164066314697,
+      "learning_rate": 0.00013301384451544198,
+      "loss": 0.5444,
       "step": 1760
     },
     {
       "epoch": 3.33,
-      "grad_norm": 1.3630428314208984,
-      "learning_rate": 0.00013253461128860493,
-      "loss": 0.5475,
+      "grad_norm": 2.9253435134887695,
+      "learning_rate": 0.00013248136315228967,
+      "loss": 0.5808,
       "step": 1770
     },
     {
       "epoch": 3.35,
-      "grad_norm": 0.8149157166481018,
-      "learning_rate": 0.00013200212992545263,
-      "loss": 0.5295,
+      "grad_norm": 2.220623731613159,
+      "learning_rate": 0.00013194888178913737,
+      "loss": 0.5715,
       "step": 1780
     },
     {
       "epoch": 3.36,
-      "grad_norm": 5.804959297180176,
-      "learning_rate": 0.00013146964856230033,
-      "loss": 0.5317,
+      "grad_norm": 1.8884316682815552,
+      "learning_rate": 0.0001314164004259851,
+      "loss": 0.5269,
       "step": 1790
     },
     {
       "epoch": 3.38,
-      "grad_norm": 1.0680649280548096,
-      "learning_rate": 0.00013093716719914802,
-      "loss": 0.5561,
+      "grad_norm": 1.592698335647583,
+      "learning_rate": 0.00013088391906283282,
+      "loss": 0.5806,
       "step": 1800
     },
     {
       "epoch": 3.4,
-      "grad_norm": 2.4818127155303955,
-      "learning_rate": 0.00013040468583599575,
-      "loss": 0.6085,
+      "grad_norm": 1.9080153703689575,
+      "learning_rate": 0.00013035143769968052,
+      "loss": 0.6371,
       "step": 1810
     },
     {
       "epoch": 3.42,
-      "grad_norm": 1.4538164138793945,
-      "learning_rate": 0.00012987220447284345,
-      "loss": 0.5614,
+      "grad_norm": 2.2922542095184326,
+      "learning_rate": 0.00012981895633652822,
+      "loss": 0.6221,
       "step": 1820
     },
     {
       "epoch": 3.44,
-      "grad_norm": 2.130237102508545,
-      "learning_rate": 0.00012933972310969115,
-      "loss": 0.5255,
+      "grad_norm": 1.9848778247833252,
+      "learning_rate": 0.00012928647497337595,
+      "loss": 0.5616,
       "step": 1830
     },
     {
       "epoch": 3.46,
-      "grad_norm": 4.228801250457764,
-      "learning_rate": 0.0001288072417465389,
-      "loss": 0.592,
+      "grad_norm": 3.399679660797119,
+      "learning_rate": 0.00012875399361022365,
+      "loss": 0.6284,
       "step": 1840
     },
     {
       "epoch": 3.48,
-      "grad_norm": 1.1142246723175049,
-      "learning_rate": 0.0001282747603833866,
-      "loss": 0.5262,
+      "grad_norm": 1.6504648923873901,
+      "learning_rate": 0.00012822151224707134,
+      "loss": 0.5542,
       "step": 1850
     },
     {
       "epoch": 3.5,
-      "grad_norm": 2.6029281616210938,
-      "learning_rate": 0.0001277422790202343,
-      "loss": 0.541,
+      "grad_norm": 3.0508906841278076,
+      "learning_rate": 0.00012768903088391907,
+      "loss": 0.5928,
       "step": 1860
     },
     {
       "epoch": 3.52,
-      "grad_norm": 3.902634382247925,
-      "learning_rate": 0.000127209797657082,
-      "loss": 0.5725,
+      "grad_norm": 3.353853464126587,
+      "learning_rate": 0.0001271565495207668,
+      "loss": 0.5858,
       "step": 1870
     },
     {
       "epoch": 3.53,
-      "grad_norm": 5.952896595001221,
-      "learning_rate": 0.00012667731629392972,
-      "loss": 0.597,
+      "grad_norm": 3.4700353145599365,
+      "learning_rate": 0.0001266240681576145,
+      "loss": 0.6719,
       "step": 1880
     },
     {
       "epoch": 3.55,
-      "grad_norm": 1.5133850574493408,
-      "learning_rate": 0.00012614483493077742,
-      "loss": 0.4974,
+      "grad_norm": 0.9324420094490051,
+      "learning_rate": 0.0001260915867944622,
+      "loss": 0.5396,
       "step": 1890
     },
     {
       "epoch": 3.57,
-      "grad_norm": 1.7601635456085205,
-      "learning_rate": 0.00012561235356762512,
-      "loss": 0.5825,
+      "grad_norm": 1.1515653133392334,
+      "learning_rate": 0.00012555910543130992,
+      "loss": 0.6344,
       "step": 1900
     },
     {
       "epoch": 3.59,
-      "grad_norm": 2.166303873062134,
-      "learning_rate": 0.00012507987220447287,
-      "loss": 0.4884,
+      "grad_norm": 1.1183910369873047,
+      "learning_rate": 0.00012502662406815762,
+      "loss": 0.5346,
       "step": 1910
     },
     {
       "epoch": 3.61,
-      "grad_norm": 1.2939203977584839,
-      "learning_rate": 0.00012454739084132057,
-      "loss": 0.5576,
+      "grad_norm": 0.988416850566864,
+      "learning_rate": 0.00012449414270500531,
+      "loss": 0.611,
       "step": 1920
     },
     {
       "epoch": 3.63,
-      "grad_norm": 3.2266693115234375,
-      "learning_rate": 0.00012401490947816827,
-      "loss": 0.5598,
+      "grad_norm": 2.163034439086914,
+      "learning_rate": 0.00012396166134185304,
+      "loss": 0.5806,
       "step": 1930
     },
     {
       "epoch": 3.65,
-      "grad_norm": 1.2390942573547363,
-      "learning_rate": 0.00012348242811501597,
-      "loss": 0.507,
+      "grad_norm": 0.785860538482666,
+      "learning_rate": 0.00012342917997870077,
+      "loss": 0.5344,
       "step": 1940
     },
     {
       "epoch": 3.67,
-      "grad_norm": 1.1552023887634277,
-      "learning_rate": 0.0001229499467518637,
-      "loss": 0.5781,
+      "grad_norm": 1.1014069318771362,
+      "learning_rate": 0.00012289669861554846,
+      "loss": 0.5961,
       "step": 1950
     },
     {
       "epoch": 3.68,
-      "grad_norm": 1.372052788734436,
-      "learning_rate": 0.0001224174653887114,
-      "loss": 0.5315,
+      "grad_norm": 1.5329209566116333,
+      "learning_rate": 0.00012236421725239616,
+      "loss": 0.5685,
       "step": 1960
     },
     {
       "epoch": 3.7,
-      "grad_norm": 2.0448787212371826,
-      "learning_rate": 0.0001218849840255591,
-      "loss": 0.5402,
+      "grad_norm": 2.057729721069336,
+      "learning_rate": 0.00012183173588924388,
+      "loss": 0.6254,
       "step": 1970
     },
     {
       "epoch": 3.72,
-      "grad_norm": 1.5772067308425903,
-      "learning_rate": 0.00012135250266240683,
-      "loss": 0.518,
+      "grad_norm": 1.8107123374938965,
+      "learning_rate": 0.00012129925452609159,
+      "loss": 0.5715,
       "step": 1980
     },
     {
       "epoch": 3.74,
-      "grad_norm": 2.118274450302124,
-      "learning_rate": 0.00012082002129925454,
-      "loss": 0.5661,
+      "grad_norm": 2.0086817741394043,
+      "learning_rate": 0.0001207667731629393,
+      "loss": 0.5606,
       "step": 1990
     },
     {
       "epoch": 3.76,
-      "grad_norm": 2.9228522777557373,
-      "learning_rate": 0.00012028753993610224,
-      "loss": 0.5419,
+      "grad_norm": 3.7626357078552246,
+      "learning_rate": 0.000120234291799787,
+      "loss": 0.5334,
       "step": 2000
     },
     {
       "epoch": 3.78,
-      "grad_norm": 3.01971697807312,
-      "learning_rate": 0.00011975505857294995,
-      "loss": 0.5731,
+      "grad_norm": 3.1028711795806885,
+      "learning_rate": 0.00011970181043663474,
+      "loss": 0.6453,
       "step": 2010
     },
     {
       "epoch": 3.8,
-      "grad_norm": 1.2747687101364136,
-      "learning_rate": 0.00011922257720979766,
-      "loss": 0.5621,
+      "grad_norm": 1.0698882341384888,
+      "learning_rate": 0.00011916932907348244,
+      "loss": 0.5851,
       "step": 2020
     },
     {
       "epoch": 3.82,
-      "grad_norm": 2.164107084274292,
-      "learning_rate": 0.00011869009584664536,
-      "loss": 0.5818,
+      "grad_norm": 2.4677796363830566,
+      "learning_rate": 0.00011863684771033015,
+      "loss": 0.6008,
       "step": 2030
     },
     {
       "epoch": 3.83,
-      "grad_norm": 1.2156935930252075,
-      "learning_rate": 0.00011815761448349307,
-      "loss": 0.561,
+      "grad_norm": 1.6755846738815308,
+      "learning_rate": 0.00011810436634717785,
+      "loss": 0.5903,
       "step": 2040
     },
     {
       "epoch": 3.85,
-      "grad_norm": 1.5158003568649292,
-      "learning_rate": 0.0001176251331203408,
-      "loss": 0.5255,
+      "grad_norm": 1.8782827854156494,
+      "learning_rate": 0.00011757188498402556,
+      "loss": 0.6018,
       "step": 2050
     },
     {
       "epoch": 3.87,
-      "grad_norm": 1.795130729675293,
-      "learning_rate": 0.00011709265175718851,
-      "loss": 0.5451,
+      "grad_norm": 1.1513298749923706,
+      "learning_rate": 0.00011703940362087327,
+      "loss": 0.5819,
       "step": 2060
     },
     {
       "epoch": 3.89,
-      "grad_norm": 2.8883111476898193,
-      "learning_rate": 0.00011656017039403621,
-      "loss": 0.539,
+      "grad_norm": 2.3282463550567627,
+      "learning_rate": 0.000116506922257721,
+      "loss": 0.5476,
       "step": 2070
     },
     {
       "epoch": 3.91,
-      "grad_norm": 2.8420214653015137,
-      "learning_rate": 0.00011602768903088392,
-      "loss": 0.5655,
+      "grad_norm": 2.159614324569702,
+      "learning_rate": 0.00011597444089456871,
+      "loss": 0.5934,
       "step": 2080
     },
     {
       "epoch": 3.93,
-      "grad_norm": 1.114376425743103,
-      "learning_rate": 0.00011549520766773163,
-      "loss": 0.4676,
+      "grad_norm": 1.6344852447509766,
+      "learning_rate": 0.00011544195953141641,
+      "loss": 0.4688,
       "step": 2090
     },
     {
       "epoch": 3.95,
-      "grad_norm": 1.6993155479431152,
-      "learning_rate": 0.00011496272630457933,
-      "loss": 0.5308,
+      "grad_norm": 1.4373761415481567,
+      "learning_rate": 0.00011490947816826412,
+      "loss": 0.5604,
       "step": 2100
     },
     {
       "epoch": 3.97,
-      "grad_norm": 1.809404969215393,
-      "learning_rate": 0.00011443024494142705,
-      "loss": 0.4535,
+      "grad_norm": 2.425197124481201,
+      "learning_rate": 0.00011437699680511182,
+      "loss": 0.4934,
       "step": 2110
     },
     {
       "epoch": 3.98,
-      "grad_norm": 3.3252310752868652,
-      "learning_rate": 0.00011389776357827477,
-      "loss": 0.4796,
+      "grad_norm": 3.7428033351898193,
+      "learning_rate": 0.00011384451544195953,
+      "loss": 0.5307,
       "step": 2120
     },
     {
       "epoch": 4.0,
-      "eval_accuracy": 0.536,
-      "eval_loss": 0.9386767745018005,
-      "eval_runtime": 5.6156,
-      "eval_samples_per_second": 89.037,
-      "eval_steps_per_second": 2.849,
+      "eval_loss": 0.9912688136100769,
+      "eval_runtime": 5.4585,
+      "eval_samples_per_second": 91.6,
+      "eval_steps_per_second": 2.931,
       "step": 2128
     },
     {
       "epoch": 4.0,
-      "grad_norm": 1.6969894170761108,
-      "learning_rate": 0.00011336528221512248,
-      "loss": 0.5216,
+      "grad_norm": 1.4787501096725464,
+      "learning_rate": 0.00011331203407880724,
+      "loss": 0.5672,
       "step": 2130
     },
     {
       "epoch": 4.02,
-      "grad_norm": 3.2442750930786133,
-      "learning_rate": 0.00011283280085197018,
-      "loss": 0.4074,
+      "grad_norm": 3.0657687187194824,
+      "learning_rate": 0.00011277955271565497,
+      "loss": 0.4967,
       "step": 2140
     },
     {
       "epoch": 4.04,
-      "grad_norm": 2.2770748138427734,
-      "learning_rate": 0.0001123003194888179,
-      "loss": 0.3409,
+      "grad_norm": 1.8531616926193237,
+      "learning_rate": 0.00011224707135250268,
+      "loss": 0.4264,
       "step": 2150
     },
     {
       "epoch": 4.06,
-      "grad_norm": 2.7340738773345947,
-      "learning_rate": 0.0001117678381256656,
-      "loss": 0.4351,
+      "grad_norm": 2.5244009494781494,
+      "learning_rate": 0.00011171458998935038,
+      "loss": 0.5213,
       "step": 2160
     },
     {
       "epoch": 4.08,
-      "grad_norm": 2.154231071472168,
-      "learning_rate": 0.0001112353567625133,
-      "loss": 0.4451,
+      "grad_norm": 1.9924266338348389,
+      "learning_rate": 0.00011118210862619809,
+      "loss": 0.5376,
       "step": 2170
     },
     {
       "epoch": 4.1,
-      "grad_norm": 2.5511295795440674,
-      "learning_rate": 0.00011070287539936102,
-      "loss": 0.4674,
+      "grad_norm": 1.3312138319015503,
+      "learning_rate": 0.0001106496272630458,
+      "loss": 0.5506,
       "step": 2180
     },
     {
       "epoch": 4.12,
-      "grad_norm": 1.8929212093353271,
-      "learning_rate": 0.00011017039403620874,
-      "loss": 0.4184,
+      "grad_norm": 0.9579734206199646,
+      "learning_rate": 0.0001101171458998935,
+      "loss": 0.4378,
       "step": 2190
     },
     {
       "epoch": 4.14,
-      "grad_norm": 1.6195924282073975,
-      "learning_rate": 0.00010963791267305645,
-      "loss": 0.3456,
+      "grad_norm": 1.8804982900619507,
+      "learning_rate": 0.00010958466453674121,
+      "loss": 0.4032,
       "step": 2200
     },
     {
       "epoch": 4.15,
-      "grad_norm": 2.4593372344970703,
-      "learning_rate": 0.00010910543130990417,
-      "loss": 0.3398,
+      "grad_norm": 1.5957534313201904,
+      "learning_rate": 0.00010905218317358894,
+      "loss": 0.5157,
       "step": 2210
     },
     {
       "epoch": 4.17,
-      "grad_norm": 4.4898457527160645,
-      "learning_rate": 0.00010857294994675187,
-      "loss": 0.3741,
+      "grad_norm": 3.0028767585754395,
+      "learning_rate": 0.00010851970181043665,
+      "loss": 0.5172,
       "step": 2220
     },
     {
       "epoch": 4.19,
-      "grad_norm": 3.513038158416748,
-      "learning_rate": 0.00010804046858359958,
-      "loss": 0.5035,
+      "grad_norm": 2.0464072227478027,
+      "learning_rate": 0.00010798722044728435,
+      "loss": 0.5099,
       "step": 2230
     },
     {
       "epoch": 4.21,
-      "grad_norm": 2.14664626121521,
-      "learning_rate": 0.00010750798722044728,
-      "loss": 0.3854,
+      "grad_norm": 2.517566204071045,
+      "learning_rate": 0.00010745473908413206,
+      "loss": 0.5021,
       "step": 2240
     },
     {
       "epoch": 4.23,
-      "grad_norm": 1.8807299137115479,
-      "learning_rate": 0.00010697550585729502,
-      "loss": 0.3503,
+      "grad_norm": 1.1620187759399414,
+      "learning_rate": 0.00010692225772097977,
+      "loss": 0.4711,
       "step": 2250
     },
     {
       "epoch": 4.25,
-      "grad_norm": 3.9768781661987305,
-      "learning_rate": 0.00010644302449414271,
-      "loss": 0.4615,
+      "grad_norm": 1.9912981986999512,
+      "learning_rate": 0.00010638977635782747,
+      "loss": 0.5725,
       "step": 2260
     },
     {
       "epoch": 4.27,
-      "grad_norm": 4.187206745147705,
-      "learning_rate": 0.00010591054313099043,
-      "loss": 0.3798,
+      "grad_norm": 0.9746074080467224,
+      "learning_rate": 0.00010585729499467518,
+      "loss": 0.5196,
       "step": 2270
     },
     {
       "epoch": 4.29,
-      "grad_norm": 5.973431587219238,
-      "learning_rate": 0.00010537806176783814,
-      "loss": 0.4394,
+      "grad_norm": 3.125993490219116,
+      "learning_rate": 0.00010532481363152291,
+      "loss": 0.55,
       "step": 2280
     },
     {
       "epoch": 4.3,
-      "grad_norm": 2.4600276947021484,
-      "learning_rate": 0.00010484558040468584,
-      "loss": 0.4936,
+      "grad_norm": 1.6319783926010132,
+      "learning_rate": 0.00010479233226837062,
+      "loss": 0.5553,
       "step": 2290
     },
     {
       "epoch": 4.32,
-      "grad_norm": 2.6891305446624756,
-      "learning_rate": 0.00010431309904153355,
-      "loss": 0.4825,
+      "grad_norm": 2.2027761936187744,
+      "learning_rate": 0.00010425985090521832,
+      "loss": 0.5072,
       "step": 2300
     },
     {
       "epoch": 4.34,
-      "grad_norm": 3.481198787689209,
-      "learning_rate": 0.00010378061767838125,
-      "loss": 0.5209,
+      "grad_norm": 3.756495475769043,
+      "learning_rate": 0.00010372736954206603,
+      "loss": 0.5979,
       "step": 2310
     },
     {
       "epoch": 4.36,
-      "grad_norm": 1.2965519428253174,
-      "learning_rate": 0.0001033013844515442,
-      "loss": 0.4865,
+      "grad_norm": 0.9569351673126221,
+      "learning_rate": 0.00010319488817891375,
+      "loss": 0.5269,
       "step": 2320
     },
     {
       "epoch": 4.38,
-      "grad_norm": 1.5025973320007324,
-      "learning_rate": 0.00010276890308839191,
-      "loss": 0.4439,
+      "grad_norm": 1.8240845203399658,
+      "learning_rate": 0.00010266240681576144,
+      "loss": 0.5554,
       "step": 2330
     },
     {
       "epoch": 4.4,
-      "grad_norm": 3.154419422149658,
-      "learning_rate": 0.00010223642172523961,
-      "loss": 0.4129,
+      "grad_norm": 1.4558720588684082,
+      "learning_rate": 0.00010212992545260916,
+      "loss": 0.4523,
       "step": 2340
     },
     {
       "epoch": 4.42,
-      "grad_norm": 3.1447813510894775,
-      "learning_rate": 0.00010170394036208732,
-      "loss": 0.4973,
+      "grad_norm": 1.966886281967163,
+      "learning_rate": 0.00010159744408945688,
+      "loss": 0.5397,
       "step": 2350
     },
     {
       "epoch": 4.44,
-      "grad_norm": 2.3173677921295166,
-      "learning_rate": 0.00010117145899893504,
-      "loss": 0.4538,
+      "grad_norm": 2.034513473510742,
+      "learning_rate": 0.0001010649627263046,
+      "loss": 0.4826,
       "step": 2360
     },
     {
       "epoch": 4.45,
-      "grad_norm": 1.7222325801849365,
-      "learning_rate": 0.00010063897763578276,
-      "loss": 0.4326,
+      "grad_norm": 2.524667263031006,
+      "learning_rate": 0.00010053248136315229,
+      "loss": 0.4602,
       "step": 2370
     },
     {
       "epoch": 4.47,
-      "grad_norm": 1.9806593656539917,
-      "learning_rate": 0.00010010649627263047,
-      "loss": 0.4568,
+      "grad_norm": 1.973412036895752,
+      "learning_rate": 0.00010005324813631522,
+      "loss": 0.5769,
       "step": 2380
     },
     {
       "epoch": 4.49,
-      "grad_norm": 1.6622909307479858,
-      "learning_rate": 9.957401490947817e-05,
-      "loss": 0.4504,
+      "grad_norm": 2.0014846324920654,
+      "learning_rate": 9.952076677316294e-05,
+      "loss": 0.464,
       "step": 2390
     },
     {
       "epoch": 4.51,
-      "grad_norm": 1.593268871307373,
-      "learning_rate": 9.904153354632588e-05,
-      "loss": 0.3772,
+      "grad_norm": 1.190698266029358,
+      "learning_rate": 9.898828541001066e-05,
+      "loss": 0.4745,
       "step": 2400
     },
     {
       "epoch": 4.53,
-      "grad_norm": 2.6867918968200684,
-      "learning_rate": 9.850905218317358e-05,
-      "loss": 0.4217,
+      "grad_norm": 1.8334747552871704,
+      "learning_rate": 9.845580404685837e-05,
+      "loss": 0.4763,
       "step": 2410
     },
     {
       "epoch": 4.55,
-      "grad_norm": 2.0145435333251953,
-      "learning_rate": 9.797657082002131e-05,
-      "loss": 0.3476,
+      "grad_norm": 1.5026592016220093,
+      "learning_rate": 9.792332268370608e-05,
+      "loss": 0.4714,
       "step": 2420
     },
     {
       "epoch": 4.57,
-      "grad_norm": 2.0304441452026367,
-      "learning_rate": 9.744408945686902e-05,
-      "loss": 0.3252,
+      "grad_norm": 3.0638644695281982,
+      "learning_rate": 9.739084132055378e-05,
+      "loss": 0.3997,
       "step": 2430
     },
     {
       "epoch": 4.59,
-      "grad_norm": 4.424630165100098,
-      "learning_rate": 9.691160809371672e-05,
-      "loss": 0.5274,
+      "grad_norm": 3.050144672393799,
+      "learning_rate": 9.68583599574015e-05,
+      "loss": 0.4471,
       "step": 2440
     },
     {
       "epoch": 4.61,
-      "grad_norm": 1.801672101020813,
-      "learning_rate": 9.637912673056444e-05,
-      "loss": 0.4023,
+      "grad_norm": 2.4229140281677246,
+      "learning_rate": 9.63258785942492e-05,
+      "loss": 0.4742,
       "step": 2450
     },
     {
       "epoch": 4.62,
-      "grad_norm": 2.912360906600952,
-      "learning_rate": 9.584664536741214e-05,
-      "loss": 0.3901,
+      "grad_norm": 2.2217772006988525,
+      "learning_rate": 9.579339723109692e-05,
+      "loss": 0.5514,
       "step": 2460
     },
     {
       "epoch": 4.64,
-      "grad_norm": 2.676485538482666,
-      "learning_rate": 9.531416400425986e-05,
-      "loss": 0.4589,
+      "grad_norm": 2.5604331493377686,
+      "learning_rate": 9.526091586794463e-05,
+      "loss": 0.538,
       "step": 2470
     },
     {
       "epoch": 4.66,
-      "grad_norm": 2.1935267448425293,
-      "learning_rate": 9.478168264110757e-05,
-      "loss": 0.5248,
+      "grad_norm": 1.479228138923645,
+      "learning_rate": 9.472843450479234e-05,
+      "loss": 0.5559,
       "step": 2480
     },
     {
       "epoch": 4.68,
-      "grad_norm": 2.977781295776367,
-      "learning_rate": 9.424920127795528e-05,
-      "loss": 0.4278,
+      "grad_norm": 1.5853182077407837,
+      "learning_rate": 9.419595314164005e-05,
+      "loss": 0.4786,
       "step": 2490
     },
     {
       "epoch": 4.7,
-      "grad_norm": 4.009756088256836,
-      "learning_rate": 9.371671991480299e-05,
-      "loss": 0.4132,
+      "grad_norm": 4.546787261962891,
+      "learning_rate": 9.366347177848775e-05,
+      "loss": 0.5031,
       "step": 2500
     },
     {
       "epoch": 4.72,
-      "grad_norm": 3.5155155658721924,
-      "learning_rate": 9.318423855165069e-05,
-      "loss": 0.3934,
+      "grad_norm": 2.7307677268981934,
+      "learning_rate": 9.313099041533548e-05,
+      "loss": 0.4968,
       "step": 2510
     },
     {
       "epoch": 4.74,
-      "grad_norm": 2.687786817550659,
-      "learning_rate": 9.265175718849842e-05,
-      "loss": 0.3679,
+      "grad_norm": 2.9583988189697266,
+      "learning_rate": 9.259850905218317e-05,
+      "loss": 0.5033,
       "step": 2520
     },
     {
       "epoch": 4.76,
-      "grad_norm": 2.9000587463378906,
-      "learning_rate": 9.211927582534611e-05,
-      "loss": 0.483,
+      "grad_norm": 2.163015604019165,
+      "learning_rate": 9.206602768903089e-05,
+      "loss": 0.533,
       "step": 2530
     },
     {
       "epoch": 4.77,
-      "grad_norm": 2.172851085662842,
-      "learning_rate": 9.158679446219383e-05,
-      "loss": 0.3782,
+      "grad_norm": 2.1034317016601562,
+      "learning_rate": 9.15335463258786e-05,
+      "loss": 0.4756,
       "step": 2540
     },
     {
       "epoch": 4.79,
-      "grad_norm": 4.356163024902344,
-      "learning_rate": 9.105431309904154e-05,
-      "loss": 0.4852,
+      "grad_norm": 3.1241562366485596,
+      "learning_rate": 9.100106496272631e-05,
+      "loss": 0.5204,
       "step": 2550
     },
     {
       "epoch": 4.81,
-      "grad_norm": 3.4460835456848145,
-      "learning_rate": 9.052183173588925e-05,
-      "loss": 0.4019,
+      "grad_norm": 1.9084091186523438,
+      "learning_rate": 9.046858359957402e-05,
+      "loss": 0.4767,
       "step": 2560
     },
     {
       "epoch": 4.83,
-      "grad_norm": 3.1064889430999756,
-      "learning_rate": 8.998935037273696e-05,
-      "loss": 0.397,
+      "grad_norm": 2.3724541664123535,
+      "learning_rate": 8.993610223642172e-05,
+      "loss": 0.4941,
       "step": 2570
     },
     {
       "epoch": 4.85,
-      "grad_norm": 2.628739833831787,
-      "learning_rate": 8.945686900958466e-05,
-      "loss": 0.385,
+      "grad_norm": 1.9276816844940186,
+      "learning_rate": 8.940362087326945e-05,
+      "loss": 0.5011,
       "step": 2580
     },
     {
       "epoch": 4.87,
-      "grad_norm": 1.598198413848877,
-      "learning_rate": 8.892438764643239e-05,
-      "loss": 0.2814,
+      "grad_norm": 3.0274977684020996,
+      "learning_rate": 8.887113951011715e-05,
+      "loss": 0.4222,
       "step": 2590
     },
     {
       "epoch": 4.89,
-      "grad_norm": 4.517331123352051,
-      "learning_rate": 8.839190628328009e-05,
-      "loss": 0.3773,
+      "grad_norm": 3.4712843894958496,
+      "learning_rate": 8.833865814696486e-05,
+      "loss": 0.4536,
       "step": 2600
     },
     {
       "epoch": 4.91,
-      "grad_norm": 3.0325510501861572,
-      "learning_rate": 8.78594249201278e-05,
-      "loss": 0.3999,
+      "grad_norm": 1.3534293174743652,
+      "learning_rate": 8.780617678381257e-05,
+      "loss": 0.4915,
       "step": 2610
     },
     {
       "epoch": 4.92,
-      "grad_norm": 2.6737074851989746,
-      "learning_rate": 8.732694355697551e-05,
-      "loss": 0.4072,
+      "grad_norm": 2.5444459915161133,
+      "learning_rate": 8.727369542066028e-05,
+      "loss": 0.5122,
       "step": 2620
     },
     {
       "epoch": 4.94,
-      "grad_norm": 2.7080023288726807,
-      "learning_rate": 8.679446219382322e-05,
-      "loss": 0.4635,
+      "grad_norm": 2.352358102798462,
+      "learning_rate": 8.6741214057508e-05,
+      "loss": 0.5088,
       "step": 2630
     },
     {
       "epoch": 4.96,
-      "grad_norm": 2.1088662147521973,
-      "learning_rate": 8.626198083067093e-05,
-      "loss": 0.4483,
+      "grad_norm": 2.807650327682495,
+      "learning_rate": 8.620873269435569e-05,
+      "loss": 0.5527,
       "step": 2640
     },
     {
       "epoch": 4.98,
-      "grad_norm": 1.194525957107544,
-      "learning_rate": 8.572949946751863e-05,
-      "loss": 0.4549,
+      "grad_norm": 1.5027140378952026,
+      "learning_rate": 8.567625133120342e-05,
+      "loss": 0.5642,
       "step": 2650
     },
     {
       "epoch": 5.0,
-      "grad_norm": 12.147368431091309,
-      "learning_rate": 8.519701810436636e-05,
-      "loss": 0.425,
+      "grad_norm": 9.080309867858887,
+      "learning_rate": 8.514376996805112e-05,
+      "loss": 0.5545,
       "step": 2660
     },
     {
       "epoch": 5.0,
-      "eval_accuracy": 0.538,
-      "eval_loss": 0.9824263453483582,
-      "eval_runtime": 5.1851,
-      "eval_samples_per_second": 96.431,
-      "eval_steps_per_second": 3.086,
+      "eval_loss": 0.9512230157852173,
+      "eval_runtime": 5.1199,
+      "eval_samples_per_second": 97.657,
+      "eval_steps_per_second": 3.125,
       "step": 2660
     },
     {
       "epoch": 5.02,
-      "grad_norm": 1.088959813117981,
-      "learning_rate": 8.466453674121406e-05,
-      "loss": 0.295,
+      "grad_norm": 1.3162051439285278,
+      "learning_rate": 8.461128860489883e-05,
+      "loss": 0.405,
       "step": 2670
     },
     {
       "epoch": 5.04,
-      "grad_norm": 1.6955100297927856,
-      "learning_rate": 8.413205537806177e-05,
-      "loss": 0.2041,
+      "grad_norm": 2.260577440261841,
+      "learning_rate": 8.407880724174654e-05,
+      "loss": 0.3254,
       "step": 2680
     },
     {
       "epoch": 5.06,
-      "grad_norm": 2.099268913269043,
-      "learning_rate": 8.359957401490948e-05,
-      "loss": 0.3164,
+      "grad_norm": 2.4929864406585693,
+      "learning_rate": 8.354632587859425e-05,
+      "loss": 0.4028,
       "step": 2690
     },
     {
       "epoch": 5.08,
-      "grad_norm": 3.751316547393799,
-      "learning_rate": 8.30670926517572e-05,
-      "loss": 0.2167,
+      "grad_norm": 5.688662052154541,
+      "learning_rate": 8.301384451544197e-05,
+      "loss": 0.2906,
       "step": 2700
     },
     {
       "epoch": 5.09,
-      "grad_norm": 0.7041321396827698,
-      "learning_rate": 8.25346112886049e-05,
-      "loss": 0.3246,
+      "grad_norm": 1.3222452402114868,
+      "learning_rate": 8.248136315228966e-05,
+      "loss": 0.3261,
       "step": 2710
     },
     {
       "epoch": 5.11,
-      "grad_norm": 1.6721878051757812,
-      "learning_rate": 8.20021299254526e-05,
-      "loss": 0.2365,
+      "grad_norm": 1.3454262018203735,
+      "learning_rate": 8.194888178913739e-05,
+      "loss": 0.3013,
       "step": 2720
     },
     {
       "epoch": 5.13,
-      "grad_norm": 3.7711753845214844,
-      "learning_rate": 8.146964856230033e-05,
-      "loss": 0.22,
+      "grad_norm": 3.683896064758301,
+      "learning_rate": 8.141640042598509e-05,
+      "loss": 0.3953,
       "step": 2730
     },
     {
       "epoch": 5.15,
-      "grad_norm": 5.828824520111084,
-      "learning_rate": 8.093716719914803e-05,
-      "loss": 0.3559,
+      "grad_norm": 5.352532863616943,
+      "learning_rate": 8.08839190628328e-05,
+      "loss": 0.4251,
       "step": 2740
     },
     {
       "epoch": 5.17,
-      "grad_norm": 2.093899726867676,
-      "learning_rate": 8.040468583599574e-05,
-      "loss": 0.2061,
+      "grad_norm": 2.004856586456299,
+      "learning_rate": 8.035143769968051e-05,
+      "loss": 0.3757,
       "step": 2750
     },
     {
       "epoch": 5.19,
-      "grad_norm": 4.466729164123535,
-      "learning_rate": 7.987220447284345e-05,
-      "loss": 0.3012,
+      "grad_norm": 3.9098947048187256,
+      "learning_rate": 7.981895633652823e-05,
+      "loss": 0.3575,
       "step": 2760
     },
     {
       "epoch": 5.21,
-      "grad_norm": 2.4526288509368896,
-      "learning_rate": 7.933972310969116e-05,
-      "loss": 0.2209,
+      "grad_norm": 3.314652919769287,
+      "learning_rate": 7.928647497337594e-05,
+      "loss": 0.3759,
       "step": 2770
     },
     {
       "epoch": 5.23,
-      "grad_norm": 4.241008281707764,
-      "learning_rate": 7.880724174653888e-05,
-      "loss": 0.2233,
+      "grad_norm": 5.735342979431152,
+      "learning_rate": 7.875399361022364e-05,
+      "loss": 0.346,
       "step": 2780
     },
     {
       "epoch": 5.24,
-      "grad_norm": 2.828065872192383,
-      "learning_rate": 7.827476038338658e-05,
-      "loss": 0.1914,
+      "grad_norm": 1.5605205297470093,
+      "learning_rate": 7.822151224707136e-05,
+      "loss": 0.2185,
       "step": 2790
     },
     {
       "epoch": 5.26,
-      "grad_norm": 4.489712715148926,
-      "learning_rate": 7.77422790202343e-05,
-      "loss": 0.2209,
+      "grad_norm": 4.125953197479248,
+      "learning_rate": 7.768903088391907e-05,
+      "loss": 0.4666,
       "step": 2800
     },
     {
       "epoch": 5.28,
-      "grad_norm": 7.091925621032715,
-      "learning_rate": 7.7209797657082e-05,
-      "loss": 0.2372,
+      "grad_norm": 3.755160331726074,
+      "learning_rate": 7.715654952076677e-05,
+      "loss": 0.4115,
       "step": 2810
     },
     {
       "epoch": 5.3,
-      "grad_norm": 3.1179749965667725,
-      "learning_rate": 7.667731629392971e-05,
-      "loss": 0.3655,
+      "grad_norm": 3.822338104248047,
+      "learning_rate": 7.66240681576145e-05,
+      "loss": 0.3338,
       "step": 2820
     },
     {
       "epoch": 5.32,
-      "grad_norm": 2.333217144012451,
-      "learning_rate": 7.614483493077744e-05,
-      "loss": 0.2942,
+      "grad_norm": 7.3369269371032715,
+      "learning_rate": 7.60915867944622e-05,
+      "loss": 0.4178,
       "step": 2830
     },
     {
       "epoch": 5.34,
-      "grad_norm": 3.904360771179199,
-      "learning_rate": 7.561235356762514e-05,
-      "loss": 0.2113,
+      "grad_norm": 3.8311543464660645,
+      "learning_rate": 7.555910543130991e-05,
+      "loss": 0.3084,
       "step": 2840
     },
     {
       "epoch": 5.36,
-      "grad_norm": 3.5696709156036377,
-      "learning_rate": 7.507987220447285e-05,
-      "loss": 0.1951,
+      "grad_norm": 4.502466678619385,
+      "learning_rate": 7.502662406815761e-05,
+      "loss": 0.2962,
       "step": 2850
     },
     {
       "epoch": 5.38,
-      "grad_norm": 2.9150383472442627,
-      "learning_rate": 7.454739084132055e-05,
-      "loss": 0.2142,
+      "grad_norm": 3.489075183868408,
+      "learning_rate": 7.449414270500533e-05,
+      "loss": 0.3073,
       "step": 2860
     },
     {
       "epoch": 5.39,
-      "grad_norm": 3.8467230796813965,
-      "learning_rate": 7.401490947816827e-05,
-      "loss": 0.2315,
+      "grad_norm": 2.344810962677002,
+      "learning_rate": 7.396166134185304e-05,
+      "loss": 0.3059,
       "step": 2870
     },
     {
       "epoch": 5.41,
-      "grad_norm": 2.5664279460906982,
-      "learning_rate": 7.348242811501597e-05,
-      "loss": 0.3081,
+      "grad_norm": 4.37916374206543,
+      "learning_rate": 7.342917997870074e-05,
+      "loss": 0.3749,
       "step": 2880
     },
     {
       "epoch": 5.43,
-      "grad_norm": 2.859978199005127,
-      "learning_rate": 7.294994675186368e-05,
-      "loss": 0.2426,
+      "grad_norm": 2.2266342639923096,
+      "learning_rate": 7.289669861554847e-05,
+      "loss": 0.2954,
       "step": 2890
     },
     {
       "epoch": 5.45,
-      "grad_norm": 2.69465970993042,
-      "learning_rate": 7.241746538871141e-05,
-      "loss": 0.3018,
+      "grad_norm": 2.1551311016082764,
+      "learning_rate": 7.236421725239617e-05,
+      "loss": 0.3377,
       "step": 2900
     },
     {
       "epoch": 5.47,
-      "grad_norm": 3.34781551361084,
-      "learning_rate": 7.188498402555911e-05,
-      "loss": 0.1438,
+      "grad_norm": 5.243711948394775,
+      "learning_rate": 7.183173588924388e-05,
+      "loss": 0.2986,
       "step": 2910
     },
     {
       "epoch": 5.49,
-      "grad_norm": 4.8066182136535645,
-      "learning_rate": 7.135250266240682e-05,
-      "loss": 0.224,
+      "grad_norm": 3.0937271118164062,
+      "learning_rate": 7.129925452609159e-05,
+      "loss": 0.2355,
       "step": 2920
     },
     {
       "epoch": 5.51,
-      "grad_norm": 0.3521786630153656,
-      "learning_rate": 7.082002129925452e-05,
-      "loss": 0.2222,
+      "grad_norm": 1.6880041360855103,
+      "learning_rate": 7.07667731629393e-05,
+      "loss": 0.2808,
       "step": 2930
     },
     {
       "epoch": 5.53,
-      "grad_norm": 2.7981643676757812,
-      "learning_rate": 7.028753993610224e-05,
-      "loss": 0.3419,
+      "grad_norm": 2.438985586166382,
+      "learning_rate": 7.023429179978702e-05,
+      "loss": 0.4147,
       "step": 2940
     },
     {
       "epoch": 5.55,
-      "grad_norm": 4.798647403717041,
-      "learning_rate": 6.975505857294996e-05,
-      "loss": 0.2757,
+      "grad_norm": 2.1790943145751953,
+      "learning_rate": 6.970181043663471e-05,
+      "loss": 0.3619,
       "step": 2950
     },
     {
       "epoch": 5.56,
-      "grad_norm": 4.175738334655762,
-      "learning_rate": 6.922257720979765e-05,
-      "loss": 0.2388,
+      "grad_norm": 4.307677745819092,
+      "learning_rate": 6.916932907348244e-05,
+      "loss": 0.3657,
       "step": 2960
     },
     {
       "epoch": 5.58,
-      "grad_norm": 3.5461745262145996,
-      "learning_rate": 6.869009584664538e-05,
-      "loss": 0.1961,
+      "grad_norm": 4.3886566162109375,
+      "learning_rate": 6.863684771033014e-05,
+      "loss": 0.3333,
       "step": 2970
     },
     {
       "epoch": 5.6,
-      "grad_norm": 2.714567184448242,
-      "learning_rate": 6.815761448349308e-05,
-      "loss": 0.2225,
+      "grad_norm": 2.579944610595703,
+      "learning_rate": 6.810436634717785e-05,
+      "loss": 0.4132,
       "step": 2980
     },
     {
       "epoch": 5.62,
-      "grad_norm": 4.761994361877441,
-      "learning_rate": 6.762513312034079e-05,
-      "loss": 0.23,
+      "grad_norm": 2.7568283081054688,
+      "learning_rate": 6.757188498402556e-05,
+      "loss": 0.2668,
       "step": 2990
     },
     {
       "epoch": 5.64,
-      "grad_norm": 2.0303542613983154,
-      "learning_rate": 6.70926517571885e-05,
-      "loss": 0.1996,
+      "grad_norm": 1.8786532878875732,
+      "learning_rate": 6.703940362087328e-05,
+      "loss": 0.3285,
       "step": 3000
     },
     {
       "epoch": 5.66,
-      "grad_norm": 3.595022678375244,
-      "learning_rate": 6.656017039403621e-05,
-      "loss": 0.2089,
+      "grad_norm": 1.1324492692947388,
+      "learning_rate": 6.650692225772099e-05,
+      "loss": 0.3481,
       "step": 3010
     },
     {
       "epoch": 5.68,
-      "grad_norm": 4.19135856628418,
-      "learning_rate": 6.602768903088393e-05,
-      "loss": 0.2199,
+      "grad_norm": 1.0097510814666748,
+      "learning_rate": 6.597444089456869e-05,
+      "loss": 0.2596,
       "step": 3020
     },
     {
       "epoch": 5.7,
-      "grad_norm": 6.599343299865723,
-      "learning_rate": 6.549520766773163e-05,
-      "loss": 0.2658,
+      "grad_norm": 5.024123191833496,
+      "learning_rate": 6.544195953141641e-05,
+      "loss": 0.2748,
       "step": 3030
     },
     {
       "epoch": 5.71,
-      "grad_norm": 4.5252685546875,
-      "learning_rate": 6.496272630457935e-05,
-      "loss": 0.1459,
+      "grad_norm": 4.416189193725586,
+      "learning_rate": 6.490947816826411e-05,
+      "loss": 0.3929,
       "step": 3040
     },
     {
       "epoch": 5.73,
-      "grad_norm": 2.3499250411987305,
-      "learning_rate": 6.443024494142705e-05,
-      "loss": 0.2447,
+      "grad_norm": 2.678957462310791,
+      "learning_rate": 6.437699680511182e-05,
+      "loss": 0.2433,
       "step": 3050
     },
     {
       "epoch": 5.75,
-      "grad_norm": 4.795304298400879,
-      "learning_rate": 6.389776357827476e-05,
-      "loss": 0.1987,
+      "grad_norm": 2.553661346435547,
+      "learning_rate": 6.384451544195953e-05,
+      "loss": 0.2908,
       "step": 3060
     },
     {
       "epoch": 5.77,
-      "grad_norm": 2.328916311264038,
-      "learning_rate": 6.336528221512247e-05,
-      "loss": 0.2369,
+      "grad_norm": 2.256429433822632,
+      "learning_rate": 6.331203407880725e-05,
+      "loss": 0.2666,
       "step": 3070
     },
     {
       "epoch": 5.79,
-      "grad_norm": 5.857803821563721,
-      "learning_rate": 6.283280085197019e-05,
-      "loss": 0.2668,
+      "grad_norm": 5.489305019378662,
+      "learning_rate": 6.277955271565496e-05,
+      "loss": 0.3884,
       "step": 3080
     },
     {
       "epoch": 5.81,
-      "grad_norm": 6.459561347961426,
-      "learning_rate": 6.23003194888179e-05,
-      "loss": 0.2438,
+      "grad_norm": 4.188615322113037,
+      "learning_rate": 6.224707135250266e-05,
+      "loss": 0.3637,
       "step": 3090
     },
     {
       "epoch": 5.83,
-      "grad_norm": 3.3628289699554443,
-      "learning_rate": 6.17678381256656e-05,
-      "loss": 0.2694,
+      "grad_norm": 3.3800997734069824,
+      "learning_rate": 6.171458998935038e-05,
+      "loss": 0.2935,
       "step": 3100
     },
     {
       "epoch": 5.85,
-      "grad_norm": 3.036752223968506,
-      "learning_rate": 6.123535676251332e-05,
-      "loss": 0.3594,
+      "grad_norm": 1.9177793264389038,
+      "learning_rate": 6.118210862619808e-05,
+      "loss": 0.3361,
       "step": 3110
     },
     {
       "epoch": 5.86,
-      "grad_norm": 1.0607565641403198,
-      "learning_rate": 6.070287539936103e-05,
-      "loss": 0.1804,
+      "grad_norm": 1.3074955940246582,
+      "learning_rate": 6.0649627263045794e-05,
+      "loss": 0.2567,
       "step": 3120
     },
     {
       "epoch": 5.88,
-      "grad_norm": 3.0828421115875244,
-      "learning_rate": 6.017039403620873e-05,
-      "loss": 0.2408,
+      "grad_norm": 4.866466045379639,
+      "learning_rate": 6.01171458998935e-05,
+      "loss": 0.3597,
       "step": 3130
     },
     {
       "epoch": 5.9,
-      "grad_norm": 6.958425045013428,
-      "learning_rate": 5.963791267305645e-05,
-      "loss": 0.2278,
+      "grad_norm": 2.4640395641326904,
+      "learning_rate": 5.958466453674122e-05,
+      "loss": 0.2919,
       "step": 3140
     },
     {
       "epoch": 5.92,
-      "grad_norm": 6.2445526123046875,
-      "learning_rate": 5.910543130990416e-05,
-      "loss": 0.1687,
+      "grad_norm": 4.659738540649414,
+      "learning_rate": 5.9052183173588923e-05,
+      "loss": 0.3253,
       "step": 3150
     },
     {
       "epoch": 5.94,
-      "grad_norm": 4.447784900665283,
-      "learning_rate": 5.857294994675186e-05,
-      "loss": 0.2019,
+      "grad_norm": 3.3256399631500244,
+      "learning_rate": 5.8519701810436636e-05,
+      "loss": 0.3053,
       "step": 3160
     },
     {
       "epoch": 5.96,
-      "grad_norm": 2.6770880222320557,
-      "learning_rate": 5.8040468583599575e-05,
-      "loss": 0.2857,
+      "grad_norm": 1.7987494468688965,
+      "learning_rate": 5.7987220447284354e-05,
+      "loss": 0.3421,
       "step": 3170
     },
     {
       "epoch": 5.98,
-      "grad_norm": 2.762964963912964,
-      "learning_rate": 5.750798722044729e-05,
-      "loss": 0.2092,
+      "grad_norm": 3.481025218963623,
+      "learning_rate": 5.745473908413206e-05,
+      "loss": 0.2801,
       "step": 3180
     },
     {
       "epoch": 6.0,
-      "grad_norm": 4.038713455200195,
-      "learning_rate": 5.6975505857295e-05,
-      "loss": 0.2869,
+      "grad_norm": 3.000406265258789,
+      "learning_rate": 5.6922257720979765e-05,
+      "loss": 0.3243,
       "step": 3190
     },
     {
       "epoch": 6.0,
-      "eval_accuracy": 0.542,
-      "eval_loss": 1.6185880899429321,
-      "eval_runtime": 5.2709,
-      "eval_samples_per_second": 94.861,
-      "eval_steps_per_second": 3.036,
+      "eval_loss": 1.507144808769226,
+      "eval_runtime": 4.9299,
+      "eval_samples_per_second": 101.421,
+      "eval_steps_per_second": 3.245,
       "step": 3192
     },
     {
       "epoch": 6.02,
-      "grad_norm": 4.724943161010742,
-      "learning_rate": 5.6443024494142705e-05,
-      "loss": 0.1294,
+      "grad_norm": 1.909064531326294,
+      "learning_rate": 5.6389776357827484e-05,
+      "loss": 0.195,
       "step": 3200
     },
     {
       "epoch": 6.03,
-      "grad_norm": 0.55290687084198,
-      "learning_rate": 5.5910543130990424e-05,
-      "loss": 0.0765,
+      "grad_norm": 1.058613896369934,
+      "learning_rate": 5.585729499467519e-05,
+      "loss": 0.1155,
       "step": 3210
     },
     {
       "epoch": 6.05,
-      "grad_norm": 2.6764791011810303,
-      "learning_rate": 5.537806176783813e-05,
-      "loss": 0.0919,
+      "grad_norm": 3.8664138317108154,
+      "learning_rate": 5.53248136315229e-05,
+      "loss": 0.2217,
       "step": 3220
     },
     {
       "epoch": 6.07,
-      "grad_norm": 0.35876473784446716,
-      "learning_rate": 5.4845580404685834e-05,
-      "loss": 0.1147,
+      "grad_norm": 0.6019624471664429,
+      "learning_rate": 5.479233226837061e-05,
+      "loss": 0.1283,
       "step": 3230
     },
     {
       "epoch": 6.09,
-      "grad_norm": 7.607916355133057,
-      "learning_rate": 5.4313099041533546e-05,
-      "loss": 0.1175,
+      "grad_norm": 4.904506683349609,
+      "learning_rate": 5.4259850905218326e-05,
+      "loss": 0.1153,
       "step": 3240
     },
     {
       "epoch": 6.11,
-      "grad_norm": 0.5861133337020874,
-      "learning_rate": 5.3780617678381265e-05,
-      "loss": 0.041,
+      "grad_norm": 2.7440154552459717,
+      "learning_rate": 5.372736954206603e-05,
+      "loss": 0.0852,
       "step": 3250
     },
     {
       "epoch": 6.13,
-      "grad_norm": 7.388729095458984,
-      "learning_rate": 5.324813631522897e-05,
-      "loss": 0.0932,
+      "grad_norm": 1.7608120441436768,
+      "learning_rate": 5.3194888178913736e-05,
+      "loss": 0.1631,
       "step": 3260
     },
     {
       "epoch": 6.15,
-      "grad_norm": 0.1238517016172409,
-      "learning_rate": 5.2715654952076676e-05,
-      "loss": 0.1329,
+      "grad_norm": 0.817927360534668,
+      "learning_rate": 5.2662406815761455e-05,
+      "loss": 0.0899,
       "step": 3270
     },
     {
       "epoch": 6.17,
-      "grad_norm": 2.567272424697876,
-      "learning_rate": 5.2183173588924395e-05,
-      "loss": 0.0433,
+      "grad_norm": 0.4968046545982361,
+      "learning_rate": 5.212992545260916e-05,
+      "loss": 0.1774,
       "step": 3280
     },
     {
       "epoch": 6.18,
-      "grad_norm": 2.4135849475860596,
-      "learning_rate": 5.16506922257721e-05,
-      "loss": 0.068,
+      "grad_norm": 3.789479970932007,
+      "learning_rate": 5.159744408945687e-05,
+      "loss": 0.1012,
       "step": 3290
     },
     {
       "epoch": 6.2,
-      "grad_norm": 0.1513521671295166,
-      "learning_rate": 5.1118210862619806e-05,
-      "loss": 0.0358,
+      "grad_norm": 2.593512535095215,
+      "learning_rate": 5.106496272630458e-05,
+      "loss": 0.0886,
       "step": 3300
     },
     {
       "epoch": 6.22,
-      "grad_norm": 0.09611086547374725,
-      "learning_rate": 5.058572949946752e-05,
-      "loss": 0.1276,
+      "grad_norm": 1.9879395961761475,
+      "learning_rate": 5.05324813631523e-05,
+      "loss": 0.0865,
       "step": 3310
     },
     {
       "epoch": 6.24,
-      "grad_norm": 0.34191715717315674,
-      "learning_rate": 5.0053248136315237e-05,
-      "loss": 0.0796,
+      "grad_norm": 0.262031614780426,
+      "learning_rate": 5e-05,
+      "loss": 0.1385,
       "step": 3320
     },
     {
       "epoch": 6.26,
-      "grad_norm": 0.03613116592168808,
-      "learning_rate": 4.952076677316294e-05,
-      "loss": 0.0595,
+      "grad_norm": 0.5335056781768799,
+      "learning_rate": 4.9467518636847715e-05,
+      "loss": 0.12,
       "step": 3330
     },
     {
       "epoch": 6.28,
-      "grad_norm": 9.316259384155273,
-      "learning_rate": 4.8988285410010654e-05,
-      "loss": 0.0588,
+      "grad_norm": 3.7105541229248047,
+      "learning_rate": 4.893503727369542e-05,
+      "loss": 0.2251,
       "step": 3340
     },
     {
       "epoch": 6.3,
-      "grad_norm": 9.937971115112305,
-      "learning_rate": 4.845580404685836e-05,
-      "loss": 0.1649,
+      "grad_norm": 5.374883651733398,
+      "learning_rate": 4.840255591054313e-05,
+      "loss": 0.154,
       "step": 3350
     },
     {
       "epoch": 6.32,
-      "grad_norm": 8.708675384521484,
-      "learning_rate": 4.792332268370607e-05,
-      "loss": 0.2051,
+      "grad_norm": 11.730541229248047,
+      "learning_rate": 4.7870074547390844e-05,
+      "loss": 0.2578,
       "step": 3360
     },
     {
       "epoch": 6.33,
-      "grad_norm": 1.0412534475326538,
-      "learning_rate": 4.7390841320553784e-05,
-      "loss": 0.117,
+      "grad_norm": 2.778977632522583,
+      "learning_rate": 4.7337593184238556e-05,
+      "loss": 0.1043,
       "step": 3370
     },
     {
       "epoch": 6.35,
-      "grad_norm": 0.2932884395122528,
-      "learning_rate": 4.6858359957401496e-05,
-      "loss": 0.1319,
+      "grad_norm": 1.2093865871429443,
+      "learning_rate": 4.680511182108626e-05,
+      "loss": 0.1851,
       "step": 3380
     },
     {
       "epoch": 6.37,
-      "grad_norm": 9.108260154724121,
-      "learning_rate": 4.632587859424921e-05,
-      "loss": 0.0514,
+      "grad_norm": 0.44236135482788086,
+      "learning_rate": 4.6272630457933974e-05,
+      "loss": 0.193,
       "step": 3390
     },
     {
       "epoch": 6.39,
-      "grad_norm": 5.959402084350586,
-      "learning_rate": 4.579339723109691e-05,
-      "loss": 0.0938,
+      "grad_norm": 5.98400354385376,
+      "learning_rate": 4.5740149094781686e-05,
+      "loss": 0.1475,
       "step": 3400
     },
     {
       "epoch": 6.41,
-      "grad_norm": 2.6586625576019287,
-      "learning_rate": 4.5260915867944625e-05,
-      "loss": 0.094,
+      "grad_norm": 9.247739791870117,
+      "learning_rate": 4.520766773162939e-05,
+      "loss": 0.1972,
       "step": 3410
     },
     {
       "epoch": 6.43,
-      "grad_norm": 0.6427993178367615,
-      "learning_rate": 4.472843450479233e-05,
-      "loss": 0.0934,
+      "grad_norm": 0.7436373829841614,
+      "learning_rate": 4.46751863684771e-05,
+      "loss": 0.141,
       "step": 3420
     },
     {
       "epoch": 6.45,
-      "grad_norm": 4.278239727020264,
-      "learning_rate": 4.419595314164004e-05,
-      "loss": 0.1062,
+      "grad_norm": 2.57186222076416,
+      "learning_rate": 4.4142705005324815e-05,
+      "loss": 0.1439,
       "step": 3430
     },
     {
       "epoch": 6.47,
-      "grad_norm": 12.555079460144043,
-      "learning_rate": 4.3663471778487755e-05,
-      "loss": 0.0382,
+      "grad_norm": 6.6432600021362305,
+      "learning_rate": 4.361022364217253e-05,
+      "loss": 0.1118,
       "step": 3440
     },
     {
       "epoch": 6.48,
-      "grad_norm": 7.286933422088623,
-      "learning_rate": 4.313099041533547e-05,
-      "loss": 0.07,
+      "grad_norm": 5.326605796813965,
+      "learning_rate": 4.307774227902024e-05,
+      "loss": 0.1989,
       "step": 3450
     },
     {
       "epoch": 6.5,
-      "grad_norm": 0.022136323153972626,
-      "learning_rate": 4.259850905218318e-05,
-      "loss": 0.0973,
+      "grad_norm": 4.456277370452881,
+      "learning_rate": 4.2545260915867945e-05,
+      "loss": 0.1541,
       "step": 3460
     },
     {
       "epoch": 6.52,
-      "grad_norm": 13.952143669128418,
-      "learning_rate": 4.2066027689030885e-05,
-      "loss": 0.0956,
+      "grad_norm": 0.17808414995670319,
+      "learning_rate": 4.201277955271566e-05,
+      "loss": 0.0874,
       "step": 3470
     },
     {
       "epoch": 6.54,
-      "grad_norm": 5.638768672943115,
-      "learning_rate": 4.15335463258786e-05,
-      "loss": 0.0594,
+      "grad_norm": 14.033349990844727,
+      "learning_rate": 4.148029818956336e-05,
+      "loss": 0.1112,
       "step": 3480
     },
     {
       "epoch": 6.56,
-      "grad_norm": 0.06895623356103897,
-      "learning_rate": 4.10010649627263e-05,
-      "loss": 0.1302,
+      "grad_norm": 7.038208484649658,
+      "learning_rate": 4.094781682641108e-05,
+      "loss": 0.2453,
       "step": 3490
     },
     {
       "epoch": 6.58,
-      "grad_norm": 0.5818095207214355,
-      "learning_rate": 4.0468583599574014e-05,
-      "loss": 0.1488,
+      "grad_norm": 1.385833501815796,
+      "learning_rate": 4.041533546325879e-05,
+      "loss": 0.134,
       "step": 3500
     },
     {
       "epoch": 6.6,
-      "grad_norm": 0.13191522657871246,
-      "learning_rate": 3.9989350372736954e-05,
-      "loss": 0.0878,
+      "grad_norm": 3.095097303390503,
+      "learning_rate": 3.98828541001065e-05,
+      "loss": 0.1328,
       "step": 3510
     },
     {
       "epoch": 6.62,
-      "grad_norm": 18.15912628173828,
-      "learning_rate": 3.9456869009584666e-05,
-      "loss": 0.1311,
+      "grad_norm": 0.5865158438682556,
+      "learning_rate": 3.935037273695421e-05,
+      "loss": 0.1189,
       "step": 3520
     },
     {
       "epoch": 6.64,
-      "grad_norm": 13.288291931152344,
-      "learning_rate": 3.892438764643238e-05,
-      "loss": 0.2676,
+      "grad_norm": 6.275442600250244,
+      "learning_rate": 3.8817891373801916e-05,
+      "loss": 0.2252,
       "step": 3530
     },
     {
       "epoch": 6.65,
-      "grad_norm": 2.8228518962860107,
-      "learning_rate": 3.839190628328009e-05,
-      "loss": 0.0771,
+      "grad_norm": 4.78378438949585,
+      "learning_rate": 3.828541001064963e-05,
+      "loss": 0.2046,
       "step": 3540
     },
     {
       "epoch": 6.67,
-      "grad_norm": 0.11967118829488754,
-      "learning_rate": 3.7859424920127795e-05,
-      "loss": 0.047,
+      "grad_norm": 0.8807236552238464,
+      "learning_rate": 3.775292864749734e-05,
+      "loss": 0.0828,
       "step": 3550
     },
     {
       "epoch": 6.69,
-      "grad_norm": 0.17968201637268066,
-      "learning_rate": 3.732694355697551e-05,
-      "loss": 0.0551,
+      "grad_norm": 0.18408401310443878,
+      "learning_rate": 3.722044728434505e-05,
+      "loss": 0.0516,
       "step": 3560
     },
     {
       "epoch": 6.71,
-      "grad_norm": 3.643249273300171,
-      "learning_rate": 3.679446219382322e-05,
-      "loss": 0.0916,
+      "grad_norm": 8.150348663330078,
+      "learning_rate": 3.668796592119276e-05,
+      "loss": 0.2113,
       "step": 3570
     },
     {
       "epoch": 6.73,
-      "grad_norm": 1.9196984767913818,
-      "learning_rate": 3.6261980830670925e-05,
-      "loss": 0.0497,
+      "grad_norm": 4.728875637054443,
+      "learning_rate": 3.615548455804047e-05,
+      "loss": 0.1763,
       "step": 3580
     },
     {
       "epoch": 6.75,
-      "grad_norm": 14.608271598815918,
-      "learning_rate": 3.572949946751864e-05,
-      "loss": 0.2575,
+      "grad_norm": 8.548792839050293,
+      "learning_rate": 3.562300319488818e-05,
+      "loss": 0.1872,
       "step": 3590
     },
     {
       "epoch": 6.77,
-      "grad_norm": 0.6130942106246948,
-      "learning_rate": 3.519701810436635e-05,
-      "loss": 0.1304,
+      "grad_norm": 0.4689349830150604,
+      "learning_rate": 3.509052183173589e-05,
+      "loss": 0.0679,
       "step": 3600
     },
     {
       "epoch": 6.79,
-      "grad_norm": 0.11332878470420837,
-      "learning_rate": 3.466453674121406e-05,
-      "loss": 0.0753,
+      "grad_norm": 1.130393624305725,
+      "learning_rate": 3.45580404685836e-05,
+      "loss": 0.15,
       "step": 3610
     },
     {
       "epoch": 6.8,
-      "grad_norm": 0.6603028178215027,
-      "learning_rate": 3.413205537806177e-05,
-      "loss": 0.0852,
+      "grad_norm": 1.4335063695907593,
+      "learning_rate": 3.402555910543131e-05,
+      "loss": 0.1507,
       "step": 3620
     },
     {
       "epoch": 6.82,
-      "grad_norm": 4.780795097351074,
-      "learning_rate": 3.359957401490948e-05,
-      "loss": 0.0702,
+      "grad_norm": 1.1329718828201294,
+      "learning_rate": 3.3493077742279024e-05,
+      "loss": 0.1569,
       "step": 3630
     },
     {
       "epoch": 6.84,
-      "grad_norm": 2.858586549758911,
-      "learning_rate": 3.306709265175719e-05,
-      "loss": 0.1232,
+      "grad_norm": 0.7500413656234741,
+      "learning_rate": 3.2960596379126736e-05,
+      "loss": 0.1514,
       "step": 3640
     },
     {
       "epoch": 6.86,
-      "grad_norm": 4.370979309082031,
-      "learning_rate": 3.2534611288604896e-05,
-      "loss": 0.1168,
+      "grad_norm": 0.21835221350193024,
+      "learning_rate": 3.242811501597444e-05,
+      "loss": 0.1324,
       "step": 3650
     },
     {
       "epoch": 6.88,
-      "grad_norm": 0.04837312549352646,
-      "learning_rate": 3.2002129925452615e-05,
-      "loss": 0.0296,
+      "grad_norm": 1.765376329421997,
+      "learning_rate": 3.1895633652822154e-05,
+      "loss": 0.0542,
       "step": 3660
     },
     {
       "epoch": 6.9,
-      "grad_norm": 0.31512418389320374,
-      "learning_rate": 3.146964856230032e-05,
-      "loss": 0.1051,
+      "grad_norm": 1.007416844367981,
+      "learning_rate": 3.136315228966986e-05,
+      "loss": 0.1132,
       "step": 3670
     },
     {
       "epoch": 6.92,
-      "grad_norm": 3.6252450942993164,
-      "learning_rate": 3.093716719914803e-05,
-      "loss": 0.1216,
+      "grad_norm": 0.4407300353050232,
+      "learning_rate": 3.083067092651757e-05,
+      "loss": 0.1133,
       "step": 3680
     },
     {
       "epoch": 6.94,
-      "grad_norm": 0.5809699892997742,
-      "learning_rate": 3.0404685835995745e-05,
-      "loss": 0.0501,
+      "grad_norm": 3.9314990043640137,
+      "learning_rate": 3.029818956336528e-05,
+      "loss": 0.153,
       "step": 3690
     },
     {
       "epoch": 6.95,
-      "grad_norm": 0.15533213317394257,
-      "learning_rate": 2.987220447284345e-05,
-      "loss": 0.0621,
+      "grad_norm": 9.002704620361328,
+      "learning_rate": 2.9765708200212995e-05,
+      "loss": 0.1732,
       "step": 3700
     },
     {
       "epoch": 6.97,
-      "grad_norm": 0.04270392283797264,
-      "learning_rate": 2.9339723109691166e-05,
-      "loss": 0.0353,
+      "grad_norm": 8.162276268005371,
+      "learning_rate": 2.9233226837060707e-05,
+      "loss": 0.1315,
       "step": 3710
     },
     {
       "epoch": 6.99,
-      "grad_norm": 0.01716708578169346,
-      "learning_rate": 2.880724174653887e-05,
-      "loss": 0.0337,
+      "grad_norm": 0.3330124318599701,
+      "learning_rate": 2.8700745473908413e-05,
+      "loss": 0.1371,
       "step": 3720
     },
     {
       "epoch": 7.0,
-      "eval_accuracy": 0.538,
-      "eval_loss": 2.5670464038848877,
-      "eval_runtime": 5.0132,
-      "eval_samples_per_second": 99.737,
-      "eval_steps_per_second": 3.192,
+      "eval_loss": 2.0203089714050293,
+      "eval_runtime": 4.9645,
+      "eval_samples_per_second": 100.715,
+      "eval_steps_per_second": 3.223,
       "step": 3724
     },
     {
       "epoch": 7.01,
-      "grad_norm": 0.029269874095916748,
-      "learning_rate": 2.8274760383386583e-05,
-      "loss": 0.0997,
+      "grad_norm": 1.0700572729110718,
+      "learning_rate": 2.8168264110756125e-05,
+      "loss": 0.0909,
       "step": 3730
     },
     {
       "epoch": 7.03,
-      "grad_norm": 0.11788934469223022,
-      "learning_rate": 2.7742279020234292e-05,
-      "loss": 0.0082,
+      "grad_norm": 0.4397372007369995,
+      "learning_rate": 2.7635782747603834e-05,
+      "loss": 0.0137,
       "step": 3740
     },
     {
       "epoch": 7.05,
-      "grad_norm": 0.01990850828588009,
-      "learning_rate": 2.7209797657082004e-05,
-      "loss": 0.0251,
+      "grad_norm": 6.55562686920166,
+      "learning_rate": 2.7103301384451546e-05,
+      "loss": 0.0651,
       "step": 3750
     },
     {
       "epoch": 7.07,
-      "grad_norm": 0.38145899772644043,
-      "learning_rate": 2.6677316293929716e-05,
-      "loss": 0.013,
+      "grad_norm": 0.3805689215660095,
+      "learning_rate": 2.6570820021299255e-05,
+      "loss": 0.0863,
       "step": 3760
     },
     {
       "epoch": 7.09,
-      "grad_norm": 2.5921037197113037,
-      "learning_rate": 2.6144834930777425e-05,
-      "loss": 0.0601,
+      "grad_norm": 3.0128049850463867,
+      "learning_rate": 2.6038338658146967e-05,
+      "loss": 0.0868,
       "step": 3770
     },
     {
       "epoch": 7.11,
-      "grad_norm": 0.03015085868537426,
-      "learning_rate": 2.5612353567625137e-05,
-      "loss": 0.0365,
+      "grad_norm": 2.7534263134002686,
+      "learning_rate": 2.550585729499468e-05,
+      "loss": 0.0183,
       "step": 3780
     },
     {
       "epoch": 7.12,
-      "grad_norm": 9.706396102905273,
-      "learning_rate": 2.5079872204472842e-05,
-      "loss": 0.0722,
+      "grad_norm": 0.03968283161520958,
+      "learning_rate": 2.4973375931842384e-05,
+      "loss": 0.0045,
       "step": 3790
     },
     {
       "epoch": 7.14,
-      "grad_norm": 0.02435947395861149,
-      "learning_rate": 2.4547390841320554e-05,
-      "loss": 0.0082,
+      "grad_norm": 11.932687759399414,
+      "learning_rate": 2.44408945686901e-05,
+      "loss": 0.0418,
       "step": 3800
     },
     {
       "epoch": 7.16,
-      "grad_norm": 0.030334001407027245,
-      "learning_rate": 2.4014909478168267e-05,
-      "loss": 0.011,
+      "grad_norm": 0.20918692648410797,
+      "learning_rate": 2.390841320553781e-05,
+      "loss": 0.0048,
       "step": 3810
     },
     {
       "epoch": 7.18,
-      "grad_norm": 3.0113189220428467,
-      "learning_rate": 2.3482428115015975e-05,
-      "loss": 0.0485,
+      "grad_norm": 0.48621994256973267,
+      "learning_rate": 2.3375931842385517e-05,
+      "loss": 0.0285,
       "step": 3820
     },
     {
       "epoch": 7.2,
-      "grad_norm": 0.021948257461190224,
-      "learning_rate": 2.2949946751863684e-05,
-      "loss": 0.0384,
+      "grad_norm": 0.008626734837889671,
+      "learning_rate": 2.284345047923323e-05,
+      "loss": 0.0382,
       "step": 3830
     },
     {
       "epoch": 7.22,
-      "grad_norm": 0.02243644930422306,
-      "learning_rate": 2.2417465388711396e-05,
-      "loss": 0.0103,
+      "grad_norm": 12.129301071166992,
+      "learning_rate": 2.2310969116080938e-05,
+      "loss": 0.1381,
       "step": 3840
     },
     {
       "epoch": 7.24,
-      "grad_norm": 0.03432391956448555,
-      "learning_rate": 2.188498402555911e-05,
-      "loss": 0.0436,
+      "grad_norm": 3.3187191486358643,
+      "learning_rate": 2.1778487752928647e-05,
+      "loss": 0.034,
       "step": 3850
     },
     {
       "epoch": 7.26,
-      "grad_norm": 0.05325201898813248,
-      "learning_rate": 2.1352502662406817e-05,
-      "loss": 0.0519,
+      "grad_norm": 19.58869171142578,
+      "learning_rate": 2.124600638977636e-05,
+      "loss": 0.078,
       "step": 3860
     },
     {
       "epoch": 7.27,
-      "grad_norm": 0.01565093919634819,
-      "learning_rate": 2.0820021299254526e-05,
-      "loss": 0.0225,
+      "grad_norm": 0.011677253991365433,
+      "learning_rate": 2.071352502662407e-05,
+      "loss": 0.0278,
       "step": 3870
     },
     {
       "epoch": 7.29,
-      "grad_norm": 0.14941458404064178,
-      "learning_rate": 2.0287539936102238e-05,
-      "loss": 0.0014,
+      "grad_norm": 0.0037781130522489548,
+      "learning_rate": 2.018104366347178e-05,
+      "loss": 0.019,
       "step": 3880
     },
     {
       "epoch": 7.31,
-      "grad_norm": 0.012375161051750183,
-      "learning_rate": 1.9755058572949947e-05,
-      "loss": 0.0155,
+      "grad_norm": 0.012461444362998009,
+      "learning_rate": 1.964856230031949e-05,
+      "loss": 0.022,
       "step": 3890
     },
     {
       "epoch": 7.33,
-      "grad_norm": 0.004966201260685921,
-      "learning_rate": 1.9222577209797655e-05,
-      "loss": 0.0131,
+      "grad_norm": 0.13421136140823364,
+      "learning_rate": 1.91160809371672e-05,
+      "loss": 0.0591,
       "step": 3900
     },
     {
       "epoch": 7.35,
-      "grad_norm": 0.013580994680523872,
-      "learning_rate": 1.869009584664537e-05,
-      "loss": 0.0054,
+      "grad_norm": 0.015879683196544647,
+      "learning_rate": 1.858359957401491e-05,
+      "loss": 0.0587,
       "step": 3910
     },
     {
       "epoch": 7.37,
-      "grad_norm": 0.006246180739253759,
-      "learning_rate": 1.815761448349308e-05,
-      "loss": 0.0953,
+      "grad_norm": 0.006961719132959843,
+      "learning_rate": 1.805111821086262e-05,
+      "loss": 0.0549,
       "step": 3920
     },
     {
       "epoch": 7.39,
-      "grad_norm": 0.05573183670639992,
-      "learning_rate": 1.762513312034079e-05,
-      "loss": 0.0454,
+      "grad_norm": 0.0046995761804282665,
+      "learning_rate": 1.7518636847710333e-05,
+      "loss": 0.0096,
       "step": 3930
     },
     {
       "epoch": 7.41,
-      "grad_norm": 0.023609144613146782,
-      "learning_rate": 1.70926517571885e-05,
-      "loss": 0.0313,
+      "grad_norm": 3.304826021194458,
+      "learning_rate": 1.6986155484558042e-05,
+      "loss": 0.083,
       "step": 3940
     },
     {
       "epoch": 7.42,
-      "grad_norm": 0.01226514670997858,
-      "learning_rate": 1.656017039403621e-05,
-      "loss": 0.0448,
+      "grad_norm": 0.3136584758758545,
+      "learning_rate": 1.645367412140575e-05,
+      "loss": 0.0191,
       "step": 3950
     },
     {
       "epoch": 7.44,
-      "grad_norm": 20.82164764404297,
-      "learning_rate": 1.6027689030883918e-05,
-      "loss": 0.1159,
+      "grad_norm": 2.098071575164795,
+      "learning_rate": 1.5921192758253463e-05,
+      "loss": 0.1322,
       "step": 3960
     },
     {
       "epoch": 7.46,
-      "grad_norm": 0.018167337402701378,
-      "learning_rate": 1.549520766773163e-05,
-      "loss": 0.0377,
+      "grad_norm": 0.0669000893831253,
+      "learning_rate": 1.5388711395101172e-05,
+      "loss": 0.0662,
       "step": 3970
     },
     {
       "epoch": 7.48,
-      "grad_norm": 0.031441397964954376,
-      "learning_rate": 1.4962726304579342e-05,
-      "loss": 0.0299,
+      "grad_norm": 0.024517908692359924,
+      "learning_rate": 1.485623003194888e-05,
+      "loss": 0.0431,
       "step": 3980
     },
     {
       "epoch": 7.5,
-      "grad_norm": 0.2981891632080078,
-      "learning_rate": 1.4430244941427051e-05,
-      "loss": 0.0301,
+      "grad_norm": 0.01538030430674553,
+      "learning_rate": 1.4323748668796594e-05,
+      "loss": 0.0201,
       "step": 3990
     },
     {
       "epoch": 7.52,
-      "grad_norm": 0.052271194756031036,
-      "learning_rate": 1.3897763578274761e-05,
-      "loss": 0.0014,
+      "grad_norm": 0.019790129736065865,
+      "learning_rate": 1.3791267305644303e-05,
+      "loss": 0.0768,
       "step": 4000
     },
     {
       "epoch": 7.54,
-      "grad_norm": 0.07910721004009247,
-      "learning_rate": 1.3365282215122472e-05,
-      "loss": 0.0029,
+      "grad_norm": 0.6820014119148254,
+      "learning_rate": 1.3258785942492014e-05,
+      "loss": 0.0222,
       "step": 4010
     },
     {
       "epoch": 7.56,
-      "grad_norm": 1.7630151510238647,
-      "learning_rate": 1.283280085197018e-05,
-      "loss": 0.0421,
+      "grad_norm": 0.027402225881814957,
+      "learning_rate": 1.2726304579339724e-05,
+      "loss": 0.0191,
       "step": 4020
     },
     {
       "epoch": 7.58,
-      "grad_norm": 0.013812140561640263,
-      "learning_rate": 1.2300319488817893e-05,
-      "loss": 0.0221,
+      "grad_norm": 0.012008791789412498,
+      "learning_rate": 1.2193823216187434e-05,
+      "loss": 0.0137,
       "step": 4030
     },
     {
       "epoch": 7.59,
-      "grad_norm": 0.013484804891049862,
-      "learning_rate": 1.1767838125665601e-05,
-      "loss": 0.0011,
+      "grad_norm": 10.289400100708008,
+      "learning_rate": 1.1661341853035145e-05,
+      "loss": 0.0738,
       "step": 4040
     },
     {
       "epoch": 7.61,
-      "grad_norm": 0.031141534447669983,
-      "learning_rate": 1.1235356762513314e-05,
-      "loss": 0.0963,
+      "grad_norm": 0.019787069410085678,
+      "learning_rate": 1.1128860489882854e-05,
+      "loss": 0.0201,
       "step": 4050
     },
     {
       "epoch": 7.63,
-      "grad_norm": 0.1125926822423935,
-      "learning_rate": 1.0702875399361024e-05,
-      "loss": 0.0206,
+      "grad_norm": 3.301687717437744,
+      "learning_rate": 1.0596379126730564e-05,
+      "loss": 0.0393,
       "step": 4060
     },
     {
       "epoch": 7.65,
-      "grad_norm": 1.0204988718032837,
-      "learning_rate": 1.0170394036208733e-05,
-      "loss": 0.0019,
+      "grad_norm": 1.3089189529418945,
+      "learning_rate": 1.0063897763578276e-05,
+      "loss": 0.031,
       "step": 4070
     },
     {
       "epoch": 7.67,
-      "grad_norm": 0.020539429038763046,
-      "learning_rate": 9.637912673056443e-06,
-      "loss": 0.0474,
+      "grad_norm": 0.03497765213251114,
+      "learning_rate": 9.531416400425985e-06,
+      "loss": 0.159,
       "step": 4080
     },
     {
       "epoch": 7.69,
-      "grad_norm": 0.02552681416273117,
-      "learning_rate": 9.105431309904154e-06,
-      "loss": 0.0263,
+      "grad_norm": 0.23914293944835663,
+      "learning_rate": 8.998935037273695e-06,
+      "loss": 0.0757,
       "step": 4090
     },
     {
       "epoch": 7.71,
-      "grad_norm": 0.0711381658911705,
-      "learning_rate": 8.572949946751864e-06,
-      "loss": 0.0225,
+      "grad_norm": 0.011268123984336853,
+      "learning_rate": 8.466453674121406e-06,
+      "loss": 0.0051,
       "step": 4100
     },
     {
       "epoch": 7.73,
-      "grad_norm": 0.031245287507772446,
-      "learning_rate": 8.040468583599574e-06,
-      "loss": 0.0012,
+      "grad_norm": 0.021614540368318558,
+      "learning_rate": 7.933972310969116e-06,
+      "loss": 0.0606,
       "step": 4110
     },
     {
       "epoch": 7.74,
-      "grad_norm": 0.014275019988417625,
-      "learning_rate": 7.507987220447285e-06,
-      "loss": 0.0434,
+      "grad_norm": 2.5865726470947266,
+      "learning_rate": 7.4014909478168266e-06,
+      "loss": 0.0776,
       "step": 4120
     },
     {
       "epoch": 7.76,
-      "grad_norm": 0.060072433203458786,
-      "learning_rate": 6.975505857294995e-06,
-      "loss": 0.0846,
+      "grad_norm": 0.5956721305847168,
+      "learning_rate": 6.869009584664538e-06,
+      "loss": 0.0222,
       "step": 4130
     },
     {
       "epoch": 7.78,
-      "grad_norm": 2.3613195419311523,
-      "learning_rate": 6.443024494142705e-06,
-      "loss": 0.0444,
+      "grad_norm": 5.844602108001709,
+      "learning_rate": 6.336528221512247e-06,
+      "loss": 0.0293,
       "step": 4140
     },
     {
       "epoch": 7.8,
-      "grad_norm": 0.03658520057797432,
-      "learning_rate": 5.910543130990415e-06,
-      "loss": 0.0022,
+      "grad_norm": 0.03582174703478813,
+      "learning_rate": 5.804046858359958e-06,
+      "loss": 0.032,
       "step": 4150
     },
     {
       "epoch": 7.82,
-      "grad_norm": 0.14258913695812225,
-      "learning_rate": 5.378061767838126e-06,
-      "loss": 0.0537,
+      "grad_norm": 4.13277006149292,
+      "learning_rate": 5.2715654952076674e-06,
+      "loss": 0.0635,
       "step": 4160
     },
     {
       "epoch": 7.84,
-      "grad_norm": 0.017876233905553818,
-      "learning_rate": 4.845580404685836e-06,
-      "loss": 0.0654,
+      "grad_norm": 0.02260272204875946,
+      "learning_rate": 4.739084132055379e-06,
+      "loss": 0.0024,
       "step": 4170
     },
     {
       "epoch": 7.86,
-      "grad_norm": 0.039430875331163406,
-      "learning_rate": 4.3130990415335465e-06,
-      "loss": 0.0241,
+      "grad_norm": 0.026146868243813515,
+      "learning_rate": 4.206602768903089e-06,
+      "loss": 0.0522,
       "step": 4180
     },
     {
       "epoch": 7.88,
-      "grad_norm": 0.07222764194011688,
-      "learning_rate": 3.780617678381257e-06,
-      "loss": 0.0018,
+      "grad_norm": 0.3275425434112549,
+      "learning_rate": 3.6741214057507987e-06,
+      "loss": 0.0492,
       "step": 4190
     },
     {
       "epoch": 7.89,
-      "grad_norm": 0.2034127116203308,
-      "learning_rate": 3.2481363152289674e-06,
-      "loss": 0.0322,
+      "grad_norm": 0.07726357877254486,
+      "learning_rate": 3.141640042598509e-06,
+      "loss": 0.0205,
       "step": 4200
     },
     {
       "epoch": 7.91,
-      "grad_norm": 32.39055633544922,
-      "learning_rate": 2.7156549520766774e-06,
-      "loss": 0.0278,
+      "grad_norm": 0.6571511626243591,
+      "learning_rate": 2.6091586794462196e-06,
+      "loss": 0.0593,
       "step": 4210
     },
     {
       "epoch": 7.93,
-      "grad_norm": 0.02537735551595688,
-      "learning_rate": 2.183173588924388e-06,
-      "loss": 0.0016,
+      "grad_norm": 0.18334336578845978,
+      "learning_rate": 2.0766773162939296e-06,
+      "loss": 0.0052,
       "step": 4220
     },
     {
       "epoch": 7.95,
-      "grad_norm": 5.338490009307861,
-      "learning_rate": 1.650692225772098e-06,
-      "loss": 0.0305,
+      "grad_norm": 0.056655995547771454,
+      "learning_rate": 1.54419595314164e-06,
+      "loss": 0.0362,
       "step": 4230
     },
     {
       "epoch": 7.97,
-      "grad_norm": 0.04093256965279579,
-      "learning_rate": 1.1182108626198083e-06,
-      "loss": 0.0439,
+      "grad_norm": 0.3170097768306732,
+      "learning_rate": 1.0117145899893504e-06,
+      "loss": 0.0056,
       "step": 4240
     },
     {
       "epoch": 7.99,
-      "grad_norm": 0.01651330478489399,
-      "learning_rate": 5.857294994675187e-07,
-      "loss": 0.0166,
+      "grad_norm": 6.882564544677734,
+      "learning_rate": 4.792332268370607e-07,
+      "loss": 0.0571,
       "step": 4250
     },
     {
       "epoch": 8.0,
-      "eval_accuracy": 0.538,
-      "eval_loss": 2.806075096130371,
-      "eval_runtime": 5.1896,
-      "eval_samples_per_second": 96.347,
-      "eval_steps_per_second": 3.083,
+      "eval_loss": 2.9498751163482666,
+      "eval_runtime": 5.0819,
+      "eval_samples_per_second": 98.389,
+      "eval_steps_per_second": 3.148,
       "step": 4256
     },
     {
       "epoch": 8.0,
       "step": 4256,
       "total_flos": 5.269455293792256e+18,
-      "train_loss": 0.40350076897398923,
-      "train_runtime": 1386.2467,
-      "train_samples_per_second": 49.053,
-      "train_steps_per_second": 3.07
+      "train_loss": 0.4405817036390921,
+      "train_runtime": 1400.9279,
+      "train_samples_per_second": 48.539,
+      "train_steps_per_second": 3.038
     }
   ],
   "logging_steps": 10,