diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.6442861557006836,
-  "best_model_checkpoint": "Action_model/checkpoint-600",
+  "best_metric": 0.9484250545501709,
+  "best_model_checkpoint": "Action_model/checkpoint-300",
   "epoch": 10.0,
   "eval_steps": 100,
   "global_step": 6280,
@@ -10,2768 +10,2768 @@
   "log_history": [
     {
       "epoch": 0.03,
-      "grad_norm": 1.9006062746047974,
+      "grad_norm": 6.281271457672119,
       "learning_rate": 9.968152866242038e-05,
-      "loss": 2.1282,
+      "loss": 0.2185,
       "step": 20
     },
     {
       "epoch": 0.06,
-      "grad_norm": 2.270326614379883,
+      "grad_norm": 0.12211454659700394,
       "learning_rate": 9.936305732484077e-05,
-      "loss": 1.9389,
+      "loss": 0.1594,
       "step": 40
     },
     {
       "epoch": 0.1,
-      "grad_norm": 2.4455759525299072,
+      "grad_norm": 6.89764404296875,
       "learning_rate": 9.904458598726115e-05,
-      "loss": 1.7096,
+      "loss": 0.2992,
       "step": 60
     },
     {
       "epoch": 0.13,
-      "grad_norm": 2.7374861240386963,
+      "grad_norm": 7.044023513793945,
       "learning_rate": 9.872611464968153e-05,
-      "loss": 1.4553,
+      "loss": 0.2345,
       "step": 80
     },
     {
       "epoch": 0.16,
-      "grad_norm": 2.2748217582702637,
+      "grad_norm": 0.09122150391340256,
       "learning_rate": 9.840764331210192e-05,
-      "loss": 1.3291,
+      "loss": 0.1624,
       "step": 100
     },
     {
       "epoch": 0.16,
-      "eval_accuracy": 0.7457142857142857,
-      "eval_loss": 1.1432547569274902,
-      "eval_runtime": 24.2055,
-      "eval_samples_per_second": 43.379,
-      "eval_steps_per_second": 5.453,
+      "eval_accuracy": 0.7638095238095238,
+      "eval_loss": 1.0534248352050781,
+      "eval_runtime": 15.7781,
+      "eval_samples_per_second": 66.548,
+      "eval_steps_per_second": 8.366,
       "step": 100
     },
     {
       "epoch": 0.19,
-      "grad_norm": 6.182733535766602,
+      "grad_norm": 4.277255058288574,
       "learning_rate": 9.80891719745223e-05,
-      "loss": 1.2402,
+      "loss": 0.2162,
       "step": 120
     },
     {
       "epoch": 0.22,
-      "grad_norm": 3.0395169258117676,
+      "grad_norm": 8.193276405334473,
       "learning_rate": 9.777070063694268e-05,
-      "loss": 1.1275,
+      "loss": 0.3221,
       "step": 140
     },
     {
       "epoch": 0.25,
-      "grad_norm": 3.663072347640991,
+      "grad_norm": 5.246768474578857,
       "learning_rate": 9.745222929936307e-05,
-      "loss": 1.0061,
+      "loss": 0.223,
       "step": 160
     },
     {
       "epoch": 0.29,
-      "grad_norm": 2.044201135635376,
+      "grad_norm": 3.064364194869995,
       "learning_rate": 9.713375796178345e-05,
-      "loss": 1.0052,
+      "loss": 0.2204,
       "step": 180
     },
     {
       "epoch": 0.32,
-      "grad_norm": 3.233858585357666,
+      "grad_norm": 14.690643310546875,
       "learning_rate": 9.681528662420382e-05,
-      "loss": 0.9409,
+      "loss": 0.2926,
       "step": 200
     },
     {
       "epoch": 0.32,
-      "eval_accuracy": 0.7685714285714286,
-      "eval_loss": 0.8621137738227844,
-      "eval_runtime": 14.6062,
-      "eval_samples_per_second": 71.887,
-      "eval_steps_per_second": 9.037,
+      "eval_accuracy": 0.6866666666666666,
+      "eval_loss": 1.3483974933624268,
+      "eval_runtime": 14.3127,
+      "eval_samples_per_second": 73.361,
+      "eval_steps_per_second": 9.223,
       "step": 200
     },
     {
       "epoch": 0.35,
-      "grad_norm": 3.4736602306365967,
+      "grad_norm": 6.638768196105957,
       "learning_rate": 9.649681528662421e-05,
-      "loss": 0.8728,
+      "loss": 0.3432,
       "step": 220
     },
     {
       "epoch": 0.38,
-      "grad_norm": 4.214045524597168,
+      "grad_norm": 3.001636266708374,
       "learning_rate": 9.617834394904459e-05,
-      "loss": 0.865,
+      "loss": 0.3027,
       "step": 240
     },
     {
       "epoch": 0.41,
-      "grad_norm": 4.457535743713379,
+      "grad_norm": 1.0896390676498413,
       "learning_rate": 9.585987261146497e-05,
-      "loss": 0.8243,
+      "loss": 0.2674,
       "step": 260
     },
     {
       "epoch": 0.45,
-      "grad_norm": 5.426912307739258,
+      "grad_norm": 7.582240581512451,
       "learning_rate": 9.554140127388536e-05,
-      "loss": 0.8089,
+      "loss": 0.2708,
       "step": 280
     },
     {
       "epoch": 0.48,
-      "grad_norm": 3.6690785884857178,
+      "grad_norm": 0.8179775476455688,
       "learning_rate": 9.522292993630574e-05,
-      "loss": 0.7815,
+      "loss": 0.159,
       "step": 300
     },
     {
       "epoch": 0.48,
-      "eval_accuracy": 0.7676190476190476,
-      "eval_loss": 0.7599405646324158,
-      "eval_runtime": 14.7234,
-      "eval_samples_per_second": 71.315,
-      "eval_steps_per_second": 8.965,
+      "eval_accuracy": 0.7723809523809524,
+      "eval_loss": 0.9484250545501709,
+      "eval_runtime": 14.2308,
+      "eval_samples_per_second": 73.784,
+      "eval_steps_per_second": 9.276,
       "step": 300
     },
     {
       "epoch": 0.51,
-      "grad_norm": 3.9266645908355713,
+      "grad_norm": 8.498051643371582,
       "learning_rate": 9.490445859872612e-05,
-      "loss": 0.856,
+      "loss": 0.2362,
       "step": 320
     },
     {
       "epoch": 0.54,
-      "grad_norm": 4.882758617401123,
+      "grad_norm": 7.327170372009277,
       "learning_rate": 9.458598726114651e-05,
-      "loss": 0.7119,
+      "loss": 0.2343,
       "step": 340
     },
     {
       "epoch": 0.57,
-      "grad_norm": 4.982381343841553,
+      "grad_norm": 10.750676155090332,
       "learning_rate": 9.426751592356689e-05,
-      "loss": 0.7165,
+      "loss": 0.2969,
       "step": 360
     },
     {
       "epoch": 0.61,
-      "grad_norm": 9.61684799194336,
+      "grad_norm": 4.64677619934082,
       "learning_rate": 9.394904458598726e-05,
-      "loss": 0.7334,
+      "loss": 0.275,
       "step": 380
     },
     {
       "epoch": 0.64,
-      "grad_norm": 4.234207630157471,
-      "learning_rate": 9.364649681528663e-05,
-      "loss": 0.6803,
+      "grad_norm": 1.3999114036560059,
+      "learning_rate": 9.363057324840766e-05,
+      "loss": 0.2145,
       "step": 400
     },
     {
       "epoch": 0.64,
-      "eval_accuracy": 0.7209523809523809,
-      "eval_loss": 0.88614422082901,
-      "eval_runtime": 14.4486,
-      "eval_samples_per_second": 72.671,
-      "eval_steps_per_second": 9.136,
+      "eval_accuracy": 0.7476190476190476,
+      "eval_loss": 1.0013974905014038,
+      "eval_runtime": 14.3316,
+      "eval_samples_per_second": 73.265,
+      "eval_steps_per_second": 9.21,
       "step": 400
     },
     {
       "epoch": 0.67,
-      "grad_norm": 4.2712836265563965,
-      "learning_rate": 9.332802547770702e-05,
-      "loss": 0.7381,
+      "grad_norm": 3.5272552967071533,
+      "learning_rate": 9.331210191082803e-05,
+      "loss": 0.1733,
       "step": 420
     },
     {
       "epoch": 0.7,
-      "grad_norm": 4.896230220794678,
-      "learning_rate": 9.30095541401274e-05,
-      "loss": 0.6985,
+      "grad_norm": 7.54526948928833,
+      "learning_rate": 9.299363057324841e-05,
+      "loss": 0.2729,
       "step": 440
     },
     {
       "epoch": 0.73,
-      "grad_norm": 6.688229560852051,
-      "learning_rate": 9.269108280254777e-05,
-      "loss": 0.6238,
+      "grad_norm": 5.566946506500244,
+      "learning_rate": 9.26751592356688e-05,
+      "loss": 0.3801,
       "step": 460
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.8742395043373108,
-      "learning_rate": 9.237261146496817e-05,
-      "loss": 0.5231,
+      "grad_norm": 5.137397289276123,
+      "learning_rate": 9.235668789808918e-05,
+      "loss": 0.2219,
       "step": 480
     },
     {
       "epoch": 0.8,
-      "grad_norm": 7.481076240539551,
-      "learning_rate": 9.205414012738854e-05,
-      "loss": 0.6277,
+      "grad_norm": 6.159224033355713,
+      "learning_rate": 9.203821656050956e-05,
+      "loss": 0.1889,
       "step": 500
     },
     {
       "epoch": 0.8,
-      "eval_accuracy": 0.7304761904761905,
-      "eval_loss": 0.8189947009086609,
-      "eval_runtime": 14.388,
-      "eval_samples_per_second": 72.977,
-      "eval_steps_per_second": 9.174,
+      "eval_accuracy": 0.7457142857142857,
+      "eval_loss": 1.0321154594421387,
+      "eval_runtime": 14.339,
+      "eval_samples_per_second": 73.227,
+      "eval_steps_per_second": 9.206,
       "step": 500
     },
     {
       "epoch": 0.83,
-      "grad_norm": 4.198249340057373,
-      "learning_rate": 9.173566878980892e-05,
-      "loss": 0.5832,
+      "grad_norm": 0.11031434684991837,
+      "learning_rate": 9.171974522292994e-05,
+      "loss": 0.2699,
       "step": 520
     },
     {
       "epoch": 0.86,
-      "grad_norm": 4.3164286613464355,
-      "learning_rate": 9.141719745222931e-05,
-      "loss": 0.6607,
+      "grad_norm": 7.686229705810547,
+      "learning_rate": 9.140127388535033e-05,
+      "loss": 0.3293,
       "step": 540
     },
     {
       "epoch": 0.89,
-      "grad_norm": 4.7557759284973145,
-      "learning_rate": 9.109872611464969e-05,
-      "loss": 0.5325,
+      "grad_norm": 6.5040717124938965,
+      "learning_rate": 9.10828025477707e-05,
+      "loss": 0.2556,
       "step": 560
     },
     {
       "epoch": 0.92,
-      "grad_norm": 3.6213605403900146,
-      "learning_rate": 9.078025477707007e-05,
-      "loss": 0.6377,
+      "grad_norm": 6.538149833679199,
+      "learning_rate": 9.076433121019108e-05,
+      "loss": 0.3262,
       "step": 580
     },
     {
       "epoch": 0.96,
-      "grad_norm": 3.4348907470703125,
-      "learning_rate": 9.046178343949046e-05,
-      "loss": 0.5569,
+      "grad_norm": 5.0560526847839355,
+      "learning_rate": 9.044585987261147e-05,
+      "loss": 0.3064,
       "step": 600
     },
     {
       "epoch": 0.96,
-      "eval_accuracy": 0.799047619047619,
-      "eval_loss": 0.6442861557006836,
-      "eval_runtime": 14.5781,
-      "eval_samples_per_second": 72.026,
-      "eval_steps_per_second": 9.055,
+      "eval_accuracy": 0.7314285714285714,
+      "eval_loss": 1.0795009136199951,
+      "eval_runtime": 14.412,
+      "eval_samples_per_second": 72.856,
+      "eval_steps_per_second": 9.159,
       "step": 600
     },
     {
       "epoch": 0.99,
-      "grad_norm": 3.056196451187134,
-      "learning_rate": 9.014331210191084e-05,
-      "loss": 0.5262,
+      "grad_norm": 2.845521926879883,
+      "learning_rate": 9.012738853503185e-05,
+      "loss": 0.2473,
       "step": 620
     },
     {
       "epoch": 1.02,
-      "grad_norm": 5.837003707885742,
-      "learning_rate": 8.982484076433122e-05,
-      "loss": 0.5519,
+      "grad_norm": 2.4307384490966797,
+      "learning_rate": 8.980891719745223e-05,
+      "loss": 0.2096,
       "step": 640
     },
     {
       "epoch": 1.05,
-      "grad_norm": 3.373262643814087,
-      "learning_rate": 8.950636942675161e-05,
-      "loss": 0.5346,
+      "grad_norm": 0.9748784303665161,
+      "learning_rate": 8.949044585987262e-05,
+      "loss": 0.1928,
       "step": 660
     },
     {
       "epoch": 1.08,
-      "grad_norm": 23.955032348632812,
-      "learning_rate": 8.920382165605096e-05,
-      "loss": 0.5024,
+      "grad_norm": 7.010157585144043,
+      "learning_rate": 8.9171974522293e-05,
+      "loss": 0.2571,
       "step": 680
     },
     {
       "epoch": 1.11,
-      "grad_norm": 8.919329643249512,
-      "learning_rate": 8.888535031847133e-05,
-      "loss": 0.5029,
+      "grad_norm": 7.25006628036499,
+      "learning_rate": 8.885350318471338e-05,
+      "loss": 0.2195,
       "step": 700
     },
     {
       "epoch": 1.11,
-      "eval_accuracy": 0.7676190476190476,
-      "eval_loss": 0.74560546875,
-      "eval_runtime": 14.3938,
-      "eval_samples_per_second": 72.948,
-      "eval_steps_per_second": 9.171,
+      "eval_accuracy": 0.7628571428571429,
+      "eval_loss": 0.9885631799697876,
+      "eval_runtime": 14.3852,
+      "eval_samples_per_second": 72.992,
+      "eval_steps_per_second": 9.176,
       "step": 700
     },
     {
       "epoch": 1.15,
-      "grad_norm": 3.3955509662628174,
-      "learning_rate": 8.856687898089173e-05,
-      "loss": 0.5526,
+      "grad_norm": 6.138445854187012,
+      "learning_rate": 8.853503184713377e-05,
+      "loss": 0.2153,
       "step": 720
     },
     {
       "epoch": 1.18,
-      "grad_norm": 4.959121227264404,
-      "learning_rate": 8.82484076433121e-05,
-      "loss": 0.6102,
+      "grad_norm": 1.425584077835083,
+      "learning_rate": 8.821656050955415e-05,
+      "loss": 0.2032,
       "step": 740
     },
     {
       "epoch": 1.21,
-      "grad_norm": 4.158123970031738,
-      "learning_rate": 8.792993630573248e-05,
-      "loss": 0.5073,
+      "grad_norm": 13.131802558898926,
+      "learning_rate": 8.789808917197452e-05,
+      "loss": 0.2239,
       "step": 760
     },
     {
       "epoch": 1.24,
-      "grad_norm": 5.6492815017700195,
-      "learning_rate": 8.761146496815287e-05,
-      "loss": 0.5264,
+      "grad_norm": 9.984475135803223,
+      "learning_rate": 8.757961783439491e-05,
+      "loss": 0.2686,
       "step": 780
     },
     {
       "epoch": 1.27,
-      "grad_norm": 6.516519069671631,
-      "learning_rate": 8.729299363057325e-05,
-      "loss": 0.3852,
+      "grad_norm": 4.414975166320801,
+      "learning_rate": 8.726114649681529e-05,
+      "loss": 0.2982,
       "step": 800
     },
     {
       "epoch": 1.27,
-      "eval_accuracy": 0.7695238095238095,
-      "eval_loss": 0.7608746886253357,
-      "eval_runtime": 14.1377,
-      "eval_samples_per_second": 74.27,
-      "eval_steps_per_second": 9.337,
+      "eval_accuracy": 0.7590476190476191,
+      "eval_loss": 1.0292422771453857,
+      "eval_runtime": 14.0503,
+      "eval_samples_per_second": 74.732,
+      "eval_steps_per_second": 9.395,
       "step": 800
     },
     {
       "epoch": 1.31,
-      "grad_norm": 2.4917163848876953,
-      "learning_rate": 8.697452229299363e-05,
-      "loss": 0.3567,
+      "grad_norm": 3.8477492332458496,
+      "learning_rate": 8.694267515923567e-05,
+      "loss": 0.2279,
       "step": 820
     },
     {
       "epoch": 1.34,
-      "grad_norm": 6.754037380218506,
-      "learning_rate": 8.665605095541402e-05,
-      "loss": 0.3531,
+      "grad_norm": 1.814688801765442,
+      "learning_rate": 8.662420382165606e-05,
+      "loss": 0.1822,
       "step": 840
     },
     {
       "epoch": 1.37,
-      "grad_norm": 2.805572032928467,
-      "learning_rate": 8.63375796178344e-05,
-      "loss": 0.4395,
+      "grad_norm": 7.395686149597168,
+      "learning_rate": 8.630573248407644e-05,
+      "loss": 0.2657,
       "step": 860
     },
     {
       "epoch": 1.4,
-      "grad_norm": 6.810430526733398,
-      "learning_rate": 8.601910828025478e-05,
-      "loss": 0.4404,
+      "grad_norm": 0.7708375453948975,
+      "learning_rate": 8.598726114649682e-05,
+      "loss": 0.2323,
       "step": 880
     },
     {
       "epoch": 1.43,
-      "grad_norm": 3.569650888442993,
-      "learning_rate": 8.570063694267517e-05,
-      "loss": 0.4065,
+      "grad_norm": 0.4371287226676941,
+      "learning_rate": 8.566878980891721e-05,
+      "loss": 0.2477,
       "step": 900
     },
     {
       "epoch": 1.43,
-      "eval_accuracy": 0.7704761904761904,
-      "eval_loss": 0.7441606521606445,
-      "eval_runtime": 14.6102,
-      "eval_samples_per_second": 71.867,
-      "eval_steps_per_second": 9.035,
+      "eval_accuracy": 0.7247619047619047,
+      "eval_loss": 1.2390671968460083,
+      "eval_runtime": 14.2774,
+      "eval_samples_per_second": 73.543,
+      "eval_steps_per_second": 9.245,
       "step": 900
     },
     {
       "epoch": 1.46,
-      "grad_norm": 5.281124114990234,
-      "learning_rate": 8.538216560509554e-05,
-      "loss": 0.3558,
+      "grad_norm": 6.83594274520874,
+      "learning_rate": 8.535031847133759e-05,
+      "loss": 0.312,
       "step": 920
     },
     {
       "epoch": 1.5,
-      "grad_norm": 2.859487295150757,
-      "learning_rate": 8.506369426751592e-05,
-      "loss": 0.5225,
+      "grad_norm": 3.179515838623047,
+      "learning_rate": 8.503184713375796e-05,
+      "loss": 0.2738,
       "step": 940
     },
     {
       "epoch": 1.53,
-      "grad_norm": 1.9912675619125366,
-      "learning_rate": 8.474522292993631e-05,
-      "loss": 0.4057,
+      "grad_norm": 10.801413536071777,
+      "learning_rate": 8.471337579617836e-05,
+      "loss": 0.2332,
       "step": 960
     },
     {
       "epoch": 1.56,
-      "grad_norm": 2.6385486125946045,
-      "learning_rate": 8.442675159235669e-05,
-      "loss": 0.327,
+      "grad_norm": 0.5215848088264465,
+      "learning_rate": 8.439490445859873e-05,
+      "loss": 0.2813,
       "step": 980
     },
     {
       "epoch": 1.59,
-      "grad_norm": 1.4441651105880737,
-      "learning_rate": 8.410828025477707e-05,
-      "loss": 0.4319,
+      "grad_norm": 3.889249801635742,
+      "learning_rate": 8.407643312101911e-05,
+      "loss": 0.3076,
       "step": 1000
     },
     {
       "epoch": 1.59,
-      "eval_accuracy": 0.7542857142857143,
-      "eval_loss": 0.8239891529083252,
-      "eval_runtime": 14.7208,
-      "eval_samples_per_second": 71.328,
-      "eval_steps_per_second": 8.967,
+      "eval_accuracy": 0.7323809523809524,
+      "eval_loss": 1.1325620412826538,
+      "eval_runtime": 14.3244,
+      "eval_samples_per_second": 73.301,
+      "eval_steps_per_second": 9.215,
       "step": 1000
     },
     {
       "epoch": 1.62,
-      "grad_norm": 6.856269359588623,
-      "learning_rate": 8.378980891719746e-05,
-      "loss": 0.3086,
+      "grad_norm": 7.688146591186523,
+      "learning_rate": 8.37579617834395e-05,
+      "loss": 0.1596,
       "step": 1020
     },
     {
       "epoch": 1.66,
-      "grad_norm": 3.9138472080230713,
-      "learning_rate": 8.347133757961784e-05,
-      "loss": 0.434,
+      "grad_norm": 8.286615371704102,
+      "learning_rate": 8.343949044585988e-05,
+      "loss": 0.2221,
       "step": 1040
     },
     {
       "epoch": 1.69,
-      "grad_norm": 1.921924114227295,
-      "learning_rate": 8.315286624203822e-05,
-      "loss": 0.474,
+      "grad_norm": 6.622535705566406,
+      "learning_rate": 8.312101910828026e-05,
+      "loss": 0.2828,
       "step": 1060
     },
     {
       "epoch": 1.72,
-      "grad_norm": 2.406094789505005,
-      "learning_rate": 8.283439490445861e-05,
-      "loss": 0.4016,
+      "grad_norm": 5.936185359954834,
+      "learning_rate": 8.280254777070065e-05,
+      "loss": 0.2282,
       "step": 1080
     },
     {
       "epoch": 1.75,
-      "grad_norm": 3.4770796298980713,
-      "learning_rate": 8.251592356687899e-05,
-      "loss": 0.4167,
+      "grad_norm": 1.0975838899612427,
+      "learning_rate": 8.248407643312103e-05,
+      "loss": 0.1863,
       "step": 1100
     },
     {
       "epoch": 1.75,
-      "eval_accuracy": 0.7571428571428571,
-      "eval_loss": 0.8335052728652954,
-      "eval_runtime": 14.4944,
-      "eval_samples_per_second": 72.442,
-      "eval_steps_per_second": 9.107,
+      "eval_accuracy": 0.7047619047619048,
+      "eval_loss": 1.2596460580825806,
+      "eval_runtime": 14.2428,
+      "eval_samples_per_second": 73.721,
+      "eval_steps_per_second": 9.268,
       "step": 1100
     },
     {
       "epoch": 1.78,
-      "grad_norm": 3.806575298309326,
-      "learning_rate": 8.219745222929936e-05,
-      "loss": 0.3995,
+      "grad_norm": 2.0272772312164307,
+      "learning_rate": 8.21656050955414e-05,
+      "loss": 0.2084,
       "step": 1120
     },
     {
       "epoch": 1.82,
-      "grad_norm": 4.899787902832031,
-      "learning_rate": 8.187898089171974e-05,
-      "loss": 0.4137,
+      "grad_norm": 3.3026514053344727,
+      "learning_rate": 8.18471337579618e-05,
+      "loss": 0.2587,
       "step": 1140
     },
     {
       "epoch": 1.85,
-      "grad_norm": 3.540315866470337,
-      "learning_rate": 8.156050955414013e-05,
-      "loss": 0.3625,
+      "grad_norm": 1.8570455312728882,
+      "learning_rate": 8.152866242038217e-05,
+      "loss": 0.1259,
       "step": 1160
     },
     {
       "epoch": 1.88,
-      "grad_norm": 5.864512920379639,
-      "learning_rate": 8.124203821656051e-05,
-      "loss": 0.3156,
+      "grad_norm": 3.929997444152832,
+      "learning_rate": 8.121019108280255e-05,
+      "loss": 0.1459,
       "step": 1180
     },
     {
       "epoch": 1.91,
-      "grad_norm": 5.941263675689697,
-      "learning_rate": 8.092356687898089e-05,
-      "loss": 0.3778,
+      "grad_norm": 4.541895866394043,
+      "learning_rate": 8.089171974522294e-05,
+      "loss": 0.2577,
       "step": 1200
     },
     {
       "epoch": 1.91,
-      "eval_accuracy": 0.7666666666666667,
-      "eval_loss": 0.8210071325302124,
-      "eval_runtime": 14.6523,
-      "eval_samples_per_second": 71.661,
-      "eval_steps_per_second": 9.009,
+      "eval_accuracy": 0.7609523809523809,
+      "eval_loss": 1.0649347305297852,
+      "eval_runtime": 14.2644,
+      "eval_samples_per_second": 73.61,
+      "eval_steps_per_second": 9.254,
       "step": 1200
     },
     {
       "epoch": 1.94,
-      "grad_norm": 10.370551109313965,
-      "learning_rate": 8.060509554140128e-05,
-      "loss": 0.3759,
+      "grad_norm": 4.69288969039917,
+      "learning_rate": 8.057324840764332e-05,
+      "loss": 0.2256,
       "step": 1220
     },
     {
       "epoch": 1.97,
-      "grad_norm": 0.30926162004470825,
-      "learning_rate": 8.028662420382166e-05,
-      "loss": 0.3933,
+      "grad_norm": 2.169623851776123,
+      "learning_rate": 8.02547770700637e-05,
+      "loss": 0.1563,
       "step": 1240
     },
     {
       "epoch": 2.01,
-      "grad_norm": 3.72684907913208,
-      "learning_rate": 7.998407643312102e-05,
-      "loss": 0.4111,
+      "grad_norm": 9.649871826171875,
+      "learning_rate": 7.993630573248409e-05,
+      "loss": 0.184,
       "step": 1260
     },
     {
       "epoch": 2.04,
-      "grad_norm": 0.4711369276046753,
-      "learning_rate": 7.966560509554141e-05,
-      "loss": 0.3268,
+      "grad_norm": 10.572738647460938,
+      "learning_rate": 7.961783439490447e-05,
+      "loss": 0.1312,
       "step": 1280
     },
     {
       "epoch": 2.07,
-      "grad_norm": 2.9964778423309326,
-      "learning_rate": 7.934713375796179e-05,
-      "loss": 0.3818,
+      "grad_norm": 2.3869473934173584,
+      "learning_rate": 7.929936305732485e-05,
+      "loss": 0.1491,
       "step": 1300
     },
     {
       "epoch": 2.07,
-      "eval_accuracy": 0.7485714285714286,
-      "eval_loss": 0.8431442975997925,
-      "eval_runtime": 14.7089,
-      "eval_samples_per_second": 71.385,
-      "eval_steps_per_second": 8.974,
+      "eval_accuracy": 0.7561904761904762,
+      "eval_loss": 1.1044275760650635,
+      "eval_runtime": 14.2778,
+      "eval_samples_per_second": 73.541,
+      "eval_steps_per_second": 9.245,
       "step": 1300
     },
     {
       "epoch": 2.1,
-      "grad_norm": 4.478536128997803,
-      "learning_rate": 7.902866242038217e-05,
-      "loss": 0.4138,
+      "grad_norm": 3.4942755699157715,
+      "learning_rate": 7.898089171974524e-05,
+      "loss": 0.1304,
       "step": 1320
     },
     {
       "epoch": 2.13,
-      "grad_norm": 10.571870803833008,
-      "learning_rate": 7.871019108280256e-05,
-      "loss": 0.3231,
+      "grad_norm": 4.393723011016846,
+      "learning_rate": 7.866242038216561e-05,
+      "loss": 0.1951,
       "step": 1340
     },
     {
       "epoch": 2.17,
-      "grad_norm": 0.3552369773387909,
-      "learning_rate": 7.839171974522294e-05,
-      "loss": 0.2726,
+      "grad_norm": 2.8070526123046875,
+      "learning_rate": 7.834394904458599e-05,
+      "loss": 0.1785,
       "step": 1360
     },
     {
       "epoch": 2.2,
-      "grad_norm": 7.564175605773926,
-      "learning_rate": 7.807324840764331e-05,
-      "loss": 0.3614,
+      "grad_norm": 6.361399173736572,
+      "learning_rate": 7.802547770700638e-05,
+      "loss": 0.1891,
       "step": 1380
     },
     {
       "epoch": 2.23,
-      "grad_norm": 3.9198570251464844,
-      "learning_rate": 7.77547770700637e-05,
-      "loss": 0.3249,
+      "grad_norm": 0.7736614346504211,
+      "learning_rate": 7.770700636942676e-05,
+      "loss": 0.2635,
       "step": 1400
     },
     {
       "epoch": 2.23,
-      "eval_accuracy": 0.7276190476190476,
-      "eval_loss": 0.915591299533844,
-      "eval_runtime": 14.3099,
-      "eval_samples_per_second": 73.376,
-      "eval_steps_per_second": 9.224,
+      "eval_accuracy": 0.7447619047619047,
+      "eval_loss": 1.196480631828308,
+      "eval_runtime": 14.377,
+      "eval_samples_per_second": 73.034,
+      "eval_steps_per_second": 9.181,
       "step": 1400
     },
     {
       "epoch": 2.26,
-      "grad_norm": 3.698049306869507,
-      "learning_rate": 7.743630573248408e-05,
-      "loss": 0.3335,
+      "grad_norm": 3.0931506156921387,
+      "learning_rate": 7.738853503184714e-05,
+      "loss": 0.1778,
       "step": 1420
     },
     {
       "epoch": 2.29,
-      "grad_norm": 4.099252223968506,
-      "learning_rate": 7.711783439490446e-05,
-      "loss": 0.3994,
+      "grad_norm": 0.11563995480537415,
+      "learning_rate": 7.707006369426753e-05,
+      "loss": 0.1268,
       "step": 1440
     },
     {
       "epoch": 2.32,
-      "grad_norm": 3.5469987392425537,
-      "learning_rate": 7.679936305732485e-05,
-      "loss": 0.3912,
+      "grad_norm": 0.48315638303756714,
+      "learning_rate": 7.675159235668791e-05,
+      "loss": 0.2534,
       "step": 1460
     },
     {
       "epoch": 2.36,
-      "grad_norm": 0.5396450161933899,
-      "learning_rate": 7.648089171974523e-05,
-      "loss": 0.3337,
+      "grad_norm": 5.294414520263672,
+      "learning_rate": 7.643312101910829e-05,
+      "loss": 0.1391,
       "step": 1480
     },
     {
       "epoch": 2.39,
-      "grad_norm": 2.9178402423858643,
-      "learning_rate": 7.616242038216561e-05,
-      "loss": 0.2931,
+      "grad_norm": 0.42698147892951965,
+      "learning_rate": 7.611464968152868e-05,
+      "loss": 0.2597,
       "step": 1500
     },
     {
       "epoch": 2.39,
-      "eval_accuracy": 0.7371428571428571,
-      "eval_loss": 0.8948376178741455,
-      "eval_runtime": 14.4741,
-      "eval_samples_per_second": 72.543,
-      "eval_steps_per_second": 9.12,
+      "eval_accuracy": 0.7428571428571429,
+      "eval_loss": 1.2240893840789795,
+      "eval_runtime": 14.1539,
+      "eval_samples_per_second": 74.184,
+      "eval_steps_per_second": 9.326,
       "step": 1500
     },
     {
       "epoch": 2.42,
-      "grad_norm": 3.9901485443115234,
-      "learning_rate": 7.5843949044586e-05,
-      "loss": 0.3061,
+      "grad_norm": 8.944632530212402,
+      "learning_rate": 7.579617834394906e-05,
+      "loss": 0.2203,
       "step": 1520
     },
     {
       "epoch": 2.45,
-      "grad_norm": 5.539924621582031,
-      "learning_rate": 7.552547770700638e-05,
-      "loss": 0.354,
+      "grad_norm": 7.132127285003662,
+      "learning_rate": 7.547770700636943e-05,
+      "loss": 0.2442,
       "step": 1540
     },
     {
       "epoch": 2.48,
-      "grad_norm": 0.7245145440101624,
-      "learning_rate": 7.520700636942676e-05,
-      "loss": 0.3427,
+      "grad_norm": 5.285152912139893,
+      "learning_rate": 7.515923566878981e-05,
+      "loss": 0.174,
       "step": 1560
     },
     {
       "epoch": 2.52,
-      "grad_norm": 5.047821044921875,
-      "learning_rate": 7.488853503184715e-05,
-      "loss": 0.3581,
+      "grad_norm": 2.583677291870117,
+      "learning_rate": 7.484076433121019e-05,
+      "loss": 0.2125,
       "step": 1580
     },
     {
       "epoch": 2.55,
-      "grad_norm": 9.612607955932617,
-      "learning_rate": 7.457006369426752e-05,
-      "loss": 0.2808,
+      "grad_norm": 3.9505248069763184,
+      "learning_rate": 7.453821656050955e-05,
+      "loss": 0.2468,
       "step": 1600
     },
     {
       "epoch": 2.55,
-      "eval_accuracy": 0.7466666666666667,
-      "eval_loss": 0.9114470481872559,
-      "eval_runtime": 14.3031,
-      "eval_samples_per_second": 73.411,
-      "eval_steps_per_second": 9.229,
+      "eval_accuracy": 0.7390476190476191,
+      "eval_loss": 1.1452451944351196,
+      "eval_runtime": 14.2679,
+      "eval_samples_per_second": 73.592,
+      "eval_steps_per_second": 9.252,
       "step": 1600
     },
     {
       "epoch": 2.58,
-      "grad_norm": 2.8393077850341797,
-      "learning_rate": 7.42515923566879e-05,
-      "loss": 0.3319,
+      "grad_norm": 0.03219176083803177,
+      "learning_rate": 7.421974522292993e-05,
+      "loss": 0.1693,
       "step": 1620
     },
     {
       "epoch": 2.61,
-      "grad_norm": 2.83937668800354,
-      "learning_rate": 7.393312101910828e-05,
-      "loss": 0.3182,
+      "grad_norm": 0.1227966845035553,
+      "learning_rate": 7.390127388535032e-05,
+      "loss": 0.1948,
       "step": 1640
     },
     {
       "epoch": 2.64,
-      "grad_norm": 3.702991008758545,
-      "learning_rate": 7.361464968152867e-05,
-      "loss": 0.2819,
+      "grad_norm": 1.8267372846603394,
+      "learning_rate": 7.35828025477707e-05,
+      "loss": 0.1636,
       "step": 1660
     },
     {
       "epoch": 2.68,
-      "grad_norm": 3.6075496673583984,
-      "learning_rate": 7.329617834394905e-05,
-      "loss": 0.2595,
+      "grad_norm": 2.313892364501953,
+      "learning_rate": 7.326433121019108e-05,
+      "loss": 0.1555,
       "step": 1680
     },
     {
       "epoch": 2.71,
-      "grad_norm": 5.349125385284424,
-      "learning_rate": 7.297770700636943e-05,
-      "loss": 0.2767,
+      "grad_norm": 8.712140083312988,
+      "learning_rate": 7.294585987261147e-05,
+      "loss": 0.216,
       "step": 1700
     },
     {
       "epoch": 2.71,
-      "eval_accuracy": 0.7590476190476191,
-      "eval_loss": 0.8772222399711609,
-      "eval_runtime": 14.5323,
-      "eval_samples_per_second": 72.253,
-      "eval_steps_per_second": 9.083,
+      "eval_accuracy": 0.7276190476190476,
+      "eval_loss": 1.2419151067733765,
+      "eval_runtime": 14.3339,
+      "eval_samples_per_second": 73.253,
+      "eval_steps_per_second": 9.209,
       "step": 1700
     },
     {
       "epoch": 2.74,
-      "grad_norm": 5.251324653625488,
-      "learning_rate": 7.265923566878982e-05,
-      "loss": 0.281,
+      "grad_norm": 0.11808944493532181,
+      "learning_rate": 7.262738853503185e-05,
+      "loss": 0.168,
       "step": 1720
     },
     {
       "epoch": 2.77,
-      "grad_norm": 4.524991989135742,
-      "learning_rate": 7.23407643312102e-05,
-      "loss": 0.3164,
+      "grad_norm": 6.689340591430664,
+      "learning_rate": 7.230891719745222e-05,
+      "loss": 0.2301,
       "step": 1740
     },
     {
       "epoch": 2.8,
-      "grad_norm": 0.386074036359787,
-      "learning_rate": 7.202229299363057e-05,
-      "loss": 0.3269,
+      "grad_norm": 0.052781842648983,
+      "learning_rate": 7.199044585987262e-05,
+      "loss": 0.147,
       "step": 1760
     },
     {
       "epoch": 2.83,
-      "grad_norm": 6.747868061065674,
-      "learning_rate": 7.170382165605097e-05,
-      "loss": 0.2968,
+      "grad_norm": 0.02793816663324833,
+      "learning_rate": 7.1671974522293e-05,
+      "loss": 0.1428,
       "step": 1780
     },
     {
       "epoch": 2.87,
-      "grad_norm": 1.3398292064666748,
-      "learning_rate": 7.138535031847134e-05,
-      "loss": 0.3422,
+      "grad_norm": NaN,
+      "learning_rate": 7.136942675159236e-05,
+      "loss": 0.1971,
       "step": 1800
     },
     {
       "epoch": 2.87,
-      "eval_accuracy": 0.7676190476190476,
-      "eval_loss": 0.8072260022163391,
-      "eval_runtime": 14.6204,
-      "eval_samples_per_second": 71.817,
-      "eval_steps_per_second": 9.028,
+      "eval_accuracy": 0.7361904761904762,
+      "eval_loss": 1.1883373260498047,
+      "eval_runtime": 14.0347,
+      "eval_samples_per_second": 74.815,
+      "eval_steps_per_second": 9.405,
       "step": 1800
     },
     {
       "epoch": 2.9,
-      "grad_norm": 6.668325901031494,
-      "learning_rate": 7.106687898089172e-05,
-      "loss": 0.2825,
+      "grad_norm": 4.8182148933410645,
+      "learning_rate": 7.105095541401275e-05,
+      "loss": 0.1544,
       "step": 1820
     },
     {
       "epoch": 2.93,
-      "grad_norm": 0.14574064314365387,
-      "learning_rate": 7.074840764331211e-05,
-      "loss": 0.2408,
+      "grad_norm": 12.383788108825684,
+      "learning_rate": 7.073248407643313e-05,
+      "loss": 0.1332,
       "step": 1840
     },
     {
       "epoch": 2.96,
-      "grad_norm": 7.607251167297363,
-      "learning_rate": 7.042993630573249e-05,
-      "loss": 0.3295,
+      "grad_norm": 3.8350412845611572,
+      "learning_rate": 7.04140127388535e-05,
+      "loss": 0.2138,
       "step": 1860
     },
     {
       "epoch": 2.99,
-      "grad_norm": 0.10143504291772842,
-      "learning_rate": 7.011146496815287e-05,
-      "loss": 0.2425,
+      "grad_norm": 4.728395462036133,
+      "learning_rate": 7.00955414012739e-05,
+      "loss": 0.2154,
       "step": 1880
     },
     {
       "epoch": 3.03,
-      "grad_norm": 9.659887313842773,
-      "learning_rate": 6.979299363057326e-05,
-      "loss": 0.2441,
+      "grad_norm": 0.14295512437820435,
+      "learning_rate": 6.977707006369427e-05,
+      "loss": 0.2071,
       "step": 1900
     },
     {
       "epoch": 3.03,
-      "eval_accuracy": 0.7380952380952381,
-      "eval_loss": 0.9659485816955566,
-      "eval_runtime": 14.6043,
-      "eval_samples_per_second": 71.897,
-      "eval_steps_per_second": 9.038,
+      "eval_accuracy": 0.6952380952380952,
+      "eval_loss": 1.465928077697754,
+      "eval_runtime": 14.3172,
+      "eval_samples_per_second": 73.338,
+      "eval_steps_per_second": 9.22,
       "step": 1900
     },
     {
       "epoch": 3.06,
-      "grad_norm": 0.4748639762401581,
-      "learning_rate": 6.947452229299364e-05,
-      "loss": 0.2151,
+      "grad_norm": 3.1117920875549316,
+      "learning_rate": 6.945859872611465e-05,
+      "loss": 0.1382,
       "step": 1920
     },
     {
       "epoch": 3.09,
-      "grad_norm": 2.329375743865967,
-      "learning_rate": 6.915605095541401e-05,
-      "loss": 0.2014,
+      "grad_norm": 4.977789402008057,
+      "learning_rate": 6.914012738853504e-05,
+      "loss": 0.2122,
       "step": 1940
     },
     {
       "epoch": 3.12,
-      "grad_norm": 0.08076146245002747,
-      "learning_rate": 6.88375796178344e-05,
-      "loss": 0.3039,
+      "grad_norm": 1.647383451461792,
+      "learning_rate": 6.882165605095542e-05,
+      "loss": 0.1982,
       "step": 1960
     },
     {
       "epoch": 3.15,
-      "grad_norm": 2.8407723903656006,
-      "learning_rate": 6.851910828025478e-05,
-      "loss": 0.2391,
+      "grad_norm": 3.111419439315796,
+      "learning_rate": 6.85031847133758e-05,
+      "loss": 0.2113,
       "step": 1980
     },
     {
       "epoch": 3.18,
-      "grad_norm": 4.808467388153076,
-      "learning_rate": 6.820063694267516e-05,
-      "loss": 0.2438,
+      "grad_norm": 9.33786678314209,
+      "learning_rate": 6.818471337579619e-05,
+      "loss": 0.1535,
       "step": 2000
     },
     {
       "epoch": 3.18,
-      "eval_accuracy": 0.7923809523809524,
-      "eval_loss": 0.7905208468437195,
-      "eval_runtime": 14.5131,
-      "eval_samples_per_second": 72.349,
-      "eval_steps_per_second": 9.095,
+      "eval_accuracy": 0.7723809523809524,
+      "eval_loss": 1.0238806009292603,
+      "eval_runtime": 14.4564,
+      "eval_samples_per_second": 72.632,
+      "eval_steps_per_second": 9.131,
       "step": 2000
     },
     {
       "epoch": 3.22,
-      "grad_norm": 4.137459754943848,
-      "learning_rate": 6.788216560509555e-05,
-      "loss": 0.2438,
+      "grad_norm": 9.09780502319336,
+      "learning_rate": 6.786624203821657e-05,
+      "loss": 0.2061,
       "step": 2020
     },
     {
       "epoch": 3.25,
-      "grad_norm": 1.3424416780471802,
-      "learning_rate": 6.756369426751593e-05,
-      "loss": 0.3438,
+      "grad_norm": 4.574890613555908,
+      "learning_rate": 6.754777070063694e-05,
+      "loss": 0.2066,
       "step": 2040
     },
     {
       "epoch": 3.28,
-      "grad_norm": 9.731406211853027,
-      "learning_rate": 6.724522292993631e-05,
-      "loss": 0.2893,
+      "grad_norm": 10.347898483276367,
+      "learning_rate": 6.722929936305734e-05,
+      "loss": 0.216,
       "step": 2060
     },
     {
       "epoch": 3.31,
-      "grad_norm": 3.4848482608795166,
-      "learning_rate": 6.69267515923567e-05,
-      "loss": 0.2532,
+      "grad_norm": 0.9446568489074707,
+      "learning_rate": 6.691082802547771e-05,
+      "loss": 0.1014,
       "step": 2080
     },
     {
       "epoch": 3.34,
-      "grad_norm": 0.2547754943370819,
-      "learning_rate": 6.660828025477708e-05,
-      "loss": 0.3925,
+      "grad_norm": 2.516169786453247,
+      "learning_rate": 6.659235668789809e-05,
+      "loss": 0.1842,
       "step": 2100
     },
     {
       "epoch": 3.34,
-      "eval_accuracy": 0.7304761904761905,
-      "eval_loss": 0.9977201819419861,
-      "eval_runtime": 14.2446,
-      "eval_samples_per_second": 73.712,
-      "eval_steps_per_second": 9.267,
+      "eval_accuracy": 0.7390476190476191,
+      "eval_loss": 1.1967185735702515,
+      "eval_runtime": 14.3268,
+      "eval_samples_per_second": 73.289,
+      "eval_steps_per_second": 9.214,
       "step": 2100
     },
     {
       "epoch": 3.38,
-      "grad_norm": 0.40093258023262024,
-      "learning_rate": 6.628980891719746e-05,
-      "loss": 0.2749,
+      "grad_norm": 6.236209869384766,
+      "learning_rate": 6.627388535031848e-05,
+      "loss": 0.2118,
       "step": 2120
     },
     {
       "epoch": 3.41,
-      "grad_norm": 5.478763103485107,
-      "learning_rate": 6.597133757961785e-05,
-      "loss": 0.2362,
+      "grad_norm": 1.8366751670837402,
+      "learning_rate": 6.595541401273886e-05,
+      "loss": 0.1288,
       "step": 2140
     },
     {
       "epoch": 3.44,
-      "grad_norm": 6.652340888977051,
-      "learning_rate": 6.565286624203822e-05,
-      "loss": 0.1756,
+      "grad_norm": 2.917489528656006,
+      "learning_rate": 6.563694267515924e-05,
+      "loss": 0.1044,
       "step": 2160
     },
     {
       "epoch": 3.47,
-      "grad_norm": 0.4871610701084137,
-      "learning_rate": 6.53343949044586e-05,
-      "loss": 0.2548,
+      "grad_norm": 0.04419567063450813,
+      "learning_rate": 6.531847133757962e-05,
+      "loss": 0.1439,
       "step": 2180
     },
     {
       "epoch": 3.5,
-      "grad_norm": 0.6913163661956787,
-      "learning_rate": 6.5015923566879e-05,
-      "loss": 0.263,
+      "grad_norm": 6.143473148345947,
+      "learning_rate": 6.500000000000001e-05,
+      "loss": 0.2087,
       "step": 2200
     },
     {
       "epoch": 3.5,
-      "eval_accuracy": 0.7619047619047619,
-      "eval_loss": 0.8677366971969604,
-      "eval_runtime": 14.5691,
-      "eval_samples_per_second": 72.07,
-      "eval_steps_per_second": 9.06,
+      "eval_accuracy": 0.7466666666666667,
+      "eval_loss": 1.1403493881225586,
+      "eval_runtime": 14.4028,
+      "eval_samples_per_second": 72.903,
+      "eval_steps_per_second": 9.165,
       "step": 2200
     },
     {
       "epoch": 3.54,
-      "grad_norm": 4.722912788391113,
-      "learning_rate": 6.469745222929937e-05,
-      "loss": 0.2466,
+      "grad_norm": 6.739344120025635,
+      "learning_rate": 6.468152866242039e-05,
+      "loss": 0.1333,
       "step": 2220
     },
     {
       "epoch": 3.57,
-      "grad_norm": 2.7744407653808594,
-      "learning_rate": 6.437898089171975e-05,
-      "loss": 0.2491,
+      "grad_norm": 2.0759243965148926,
+      "learning_rate": 6.436305732484076e-05,
+      "loss": 0.1585,
       "step": 2240
     },
     {
       "epoch": 3.6,
-      "grad_norm": 2.703162431716919,
-      "learning_rate": 6.406050955414014e-05,
-      "loss": 0.1714,
+      "grad_norm": 0.363150954246521,
+      "learning_rate": 6.404458598726115e-05,
+      "loss": 0.1502,
       "step": 2260
     },
     {
       "epoch": 3.63,
-      "grad_norm": 6.555820941925049,
-      "learning_rate": 6.374203821656052e-05,
-      "loss": 0.2594,
+      "grad_norm": 4.845916271209717,
+      "learning_rate": 6.372611464968153e-05,
+      "loss": 0.1721,
       "step": 2280
     },
     {
       "epoch": 3.66,
-      "grad_norm": 1.9190115928649902,
-      "learning_rate": 6.34235668789809e-05,
-      "loss": 0.2585,
+      "grad_norm": 7.214461326599121,
+      "learning_rate": 6.340764331210191e-05,
+      "loss": 0.1658,
       "step": 2300
     },
     {
       "epoch": 3.66,
-      "eval_accuracy": 0.7390476190476191,
-      "eval_loss": 1.0279096364974976,
-      "eval_runtime": 14.5845,
-      "eval_samples_per_second": 71.994,
-      "eval_steps_per_second": 9.051,
+      "eval_accuracy": 0.7342857142857143,
+      "eval_loss": 1.2900978326797485,
+      "eval_runtime": 14.0814,
+      "eval_samples_per_second": 74.566,
+      "eval_steps_per_second": 9.374,
       "step": 2300
     },
     {
       "epoch": 3.69,
-      "grad_norm": 3.693655252456665,
-      "learning_rate": 6.310509554140129e-05,
-      "loss": 0.2109,
+      "grad_norm": 3.039654016494751,
+      "learning_rate": 6.30891719745223e-05,
+      "loss": 0.2372,
       "step": 2320
     },
     {
       "epoch": 3.73,
-      "grad_norm": 5.748462200164795,
-      "learning_rate": 6.278662420382167e-05,
-      "loss": 0.2292,
+      "grad_norm": 4.820199489593506,
+      "learning_rate": 6.277070063694268e-05,
+      "loss": 0.1637,
       "step": 2340
     },
     {
       "epoch": 3.76,
-      "grad_norm": 5.914999485015869,
-      "learning_rate": 6.246815286624203e-05,
-      "loss": 0.3045,
+      "grad_norm": 5.6595282554626465,
+      "learning_rate": 6.245222929936306e-05,
+      "loss": 0.1702,
       "step": 2360
     },
     {
       "epoch": 3.79,
-      "grad_norm": 4.841549873352051,
-      "learning_rate": 6.214968152866242e-05,
-      "loss": 0.2512,
+      "grad_norm": 6.2742180824279785,
+      "learning_rate": 6.213375796178345e-05,
+      "loss": 0.1126,
       "step": 2380
     },
     {
       "epoch": 3.82,
-      "grad_norm": 1.5212379693984985,
-      "learning_rate": 6.18312101910828e-05,
-      "loss": 0.2523,
+      "grad_norm": 2.786323308944702,
+      "learning_rate": 6.181528662420383e-05,
+      "loss": 0.1159,
       "step": 2400
     },
     {
       "epoch": 3.82,
-      "eval_accuracy": 0.7828571428571428,
-      "eval_loss": 0.8742047548294067,
-      "eval_runtime": 14.3756,
-      "eval_samples_per_second": 73.041,
-      "eval_steps_per_second": 9.182,
+      "eval_accuracy": 0.7438095238095238,
+      "eval_loss": 1.1825873851776123,
+      "eval_runtime": 14.2548,
+      "eval_samples_per_second": 73.659,
+      "eval_steps_per_second": 9.26,
       "step": 2400
     },
     {
       "epoch": 3.85,
-      "grad_norm": 5.37538480758667,
-      "learning_rate": 6.151273885350318e-05,
-      "loss": 0.2907,
+      "grad_norm": 2.5385348796844482,
+      "learning_rate": 6.14968152866242e-05,
+      "loss": 0.2796,
       "step": 2420
     },
     {
       "epoch": 3.89,
-      "grad_norm": 0.1197693943977356,
-      "learning_rate": 6.119426751592357e-05,
-      "loss": 0.166,
+      "grad_norm": 0.2509418725967407,
+      "learning_rate": 6.11783439490446e-05,
+      "loss": 0.1627,
       "step": 2440
     },
     {
       "epoch": 3.92,
-      "grad_norm": 0.6890983581542969,
-      "learning_rate": 6.0875796178343946e-05,
-      "loss": 0.3482,
+      "grad_norm": 11.263955116271973,
+      "learning_rate": 6.085987261146497e-05,
+      "loss": 0.1582,
       "step": 2460
     },
     {
       "epoch": 3.95,
-      "grad_norm": 6.912744045257568,
-      "learning_rate": 6.055732484076433e-05,
-      "loss": 0.262,
+      "grad_norm": 1.3140084743499756,
+      "learning_rate": 6.054140127388536e-05,
+      "loss": 0.1666,
       "step": 2480
     },
     {
       "epoch": 3.98,
-      "grad_norm": 5.4778733253479,
-      "learning_rate": 6.023885350318471e-05,
-      "loss": 0.2322,
+      "grad_norm": 12.360730171203613,
+      "learning_rate": 6.0222929936305736e-05,
+      "loss": 0.1498,
       "step": 2500
     },
     {
       "epoch": 3.98,
-      "eval_accuracy": 0.7790476190476191,
-      "eval_loss": 0.8817368745803833,
-      "eval_runtime": 14.4036,
-      "eval_samples_per_second": 72.899,
-      "eval_steps_per_second": 9.164,
+      "eval_accuracy": 0.741904761904762,
+      "eval_loss": 1.262666940689087,
+      "eval_runtime": 14.0899,
+      "eval_samples_per_second": 74.521,
+      "eval_steps_per_second": 9.368,
       "step": 2500
     },
     {
       "epoch": 4.01,
-      "grad_norm": 1.3964594602584839,
-      "learning_rate": 5.992038216560509e-05,
-      "loss": 0.1819,
+      "grad_norm": 5.474282264709473,
+      "learning_rate": 5.990445859872612e-05,
+      "loss": 0.1333,
       "step": 2520
     },
     {
       "epoch": 4.04,
-      "grad_norm": 0.07835082709789276,
-      "learning_rate": 5.960191082802548e-05,
-      "loss": 0.1875,
+      "grad_norm": 0.4006533920764923,
+      "learning_rate": 5.9585987261146505e-05,
+      "loss": 0.1433,
       "step": 2540
     },
     {
       "epoch": 4.08,
-      "grad_norm": 0.1384442001581192,
-      "learning_rate": 5.9283439490445855e-05,
-      "loss": 0.2671,
+      "grad_norm": 0.5337119102478027,
+      "learning_rate": 5.926751592356688e-05,
+      "loss": 0.1829,
       "step": 2560
     },
     {
       "epoch": 4.11,
-      "grad_norm": 1.913549780845642,
-      "learning_rate": 5.896496815286624e-05,
-      "loss": 0.21,
+      "grad_norm": 0.3745242655277252,
+      "learning_rate": 5.894904458598727e-05,
+      "loss": 0.1426,
       "step": 2580
     },
     {
       "epoch": 4.14,
-      "grad_norm": 1.3238245248794556,
-      "learning_rate": 5.8646496815286624e-05,
-      "loss": 0.1948,
+      "grad_norm": 1.1705845594406128,
+      "learning_rate": 5.863057324840765e-05,
+      "loss": 0.135,
       "step": 2600
     },
     {
       "epoch": 4.14,
-      "eval_accuracy": 0.7847619047619048,
-      "eval_loss": 0.8387181162834167,
-      "eval_runtime": 14.398,
-      "eval_samples_per_second": 72.927,
-      "eval_steps_per_second": 9.168,
+      "eval_accuracy": 0.76,
+      "eval_loss": 1.1382542848587036,
+      "eval_runtime": 14.0296,
+      "eval_samples_per_second": 74.842,
+      "eval_steps_per_second": 9.409,
       "step": 2600
     },
     {
       "epoch": 4.17,
-      "grad_norm": 4.234366416931152,
-      "learning_rate": 5.8328025477707e-05,
-      "loss": 0.1491,
+      "grad_norm": 2.097527027130127,
+      "learning_rate": 5.831210191082803e-05,
+      "loss": 0.1094,
       "step": 2620
     },
     {
       "epoch": 4.2,
-      "grad_norm": 2.7212390899658203,
-      "learning_rate": 5.8009554140127386e-05,
-      "loss": 0.1666,
+      "grad_norm": 0.44118282198905945,
+      "learning_rate": 5.7993630573248414e-05,
+      "loss": 0.2505,
       "step": 2640
     },
     {
       "epoch": 4.24,
-      "grad_norm": 0.8286333680152893,
-      "learning_rate": 5.769108280254777e-05,
-      "loss": 0.1682,
+      "grad_norm": 0.03466079756617546,
+      "learning_rate": 5.76751592356688e-05,
+      "loss": 0.0714,
       "step": 2660
     },
     {
       "epoch": 4.27,
-      "grad_norm": 11.05745792388916,
-      "learning_rate": 5.737261146496815e-05,
-      "loss": 0.2116,
+      "grad_norm": 3.324432373046875,
+      "learning_rate": 5.7356687898089176e-05,
+      "loss": 0.1761,
       "step": 2680
     },
     {
       "epoch": 4.3,
-      "grad_norm": 3.838848352432251,
-      "learning_rate": 5.705414012738853e-05,
-      "loss": 0.2318,
+      "grad_norm": 2.578779458999634,
+      "learning_rate": 5.703821656050956e-05,
+      "loss": 0.1492,
       "step": 2700
     },
     {
       "epoch": 4.3,
-      "eval_accuracy": 0.7123809523809523,
-      "eval_loss": 1.1542153358459473,
-      "eval_runtime": 14.4441,
-      "eval_samples_per_second": 72.694,
-      "eval_steps_per_second": 9.139,
+      "eval_accuracy": 0.7342857142857143,
+      "eval_loss": 1.231048583984375,
+      "eval_runtime": 14.1976,
+      "eval_samples_per_second": 73.956,
+      "eval_steps_per_second": 9.297,
       "step": 2700
     },
     {
       "epoch": 4.33,
-      "grad_norm": 7.583115100860596,
-      "learning_rate": 5.673566878980892e-05,
-      "loss": 0.1891,
+      "grad_norm": 2.340916156768799,
+      "learning_rate": 5.6719745222929946e-05,
+      "loss": 0.1776,
       "step": 2720
     },
     {
       "epoch": 4.36,
-      "grad_norm": 0.050061825662851334,
-      "learning_rate": 5.6417197452229296e-05,
-      "loss": 0.2521,
+      "grad_norm": 0.6476538181304932,
+      "learning_rate": 5.6401273885350323e-05,
+      "loss": 0.1338,
       "step": 2740
     },
     {
       "epoch": 4.39,
-      "grad_norm": 3.531385898590088,
-      "learning_rate": 5.609872611464968e-05,
-      "loss": 0.203,
+      "grad_norm": 1.5338772535324097,
+      "learning_rate": 5.608280254777071e-05,
+      "loss": 0.0915,
       "step": 2760
     },
     {
       "epoch": 4.43,
-      "grad_norm": 0.8870866894721985,
-      "learning_rate": 5.5780254777070065e-05,
-      "loss": 0.2052,
+      "grad_norm": 16.77225112915039,
+      "learning_rate": 5.576433121019109e-05,
+      "loss": 0.1523,
       "step": 2780
     },
     {
       "epoch": 4.46,
-      "grad_norm": 5.210803031921387,
-      "learning_rate": 5.546178343949044e-05,
-      "loss": 0.2184,
+      "grad_norm": 2.6961004734039307,
+      "learning_rate": 5.544585987261147e-05,
+      "loss": 0.0982,
       "step": 2800
     },
     {
       "epoch": 4.46,
-      "eval_accuracy": 0.7152380952380952,
-      "eval_loss": 1.1386553049087524,
-      "eval_runtime": 14.4274,
-      "eval_samples_per_second": 72.778,
-      "eval_steps_per_second": 9.149,
+      "eval_accuracy": 0.7104761904761905,
+      "eval_loss": 1.4144282341003418,
+      "eval_runtime": 14.0846,
+      "eval_samples_per_second": 74.549,
+      "eval_steps_per_second": 9.372,
       "step": 2800
     },
     {
       "epoch": 4.49,
-      "grad_norm": 2.6480255126953125,
-      "learning_rate": 5.514331210191083e-05,
-      "loss": 0.2484,
+      "grad_norm": 0.01971273683011532,
+      "learning_rate": 5.5127388535031855e-05,
+      "loss": 0.1875,
       "step": 2820
     },
     {
       "epoch": 4.52,
-      "grad_norm": 0.11620941758155823,
-      "learning_rate": 5.482484076433121e-05,
-      "loss": 0.1871,
+      "grad_norm": 0.44128119945526123,
+      "learning_rate": 5.480891719745224e-05,
+      "loss": 0.1622,
       "step": 2840
     },
     {
       "epoch": 4.55,
-      "grad_norm": 3.393413782119751,
-      "learning_rate": 5.450636942675159e-05,
-      "loss": 0.1796,
+      "grad_norm": 4.695611476898193,
+      "learning_rate": 5.449044585987262e-05,
+      "loss": 0.2536,
       "step": 2860
     },
     {
       "epoch": 4.59,
-      "grad_norm": 2.8002829551696777,
-      "learning_rate": 5.4187898089171974e-05,
-      "loss": 0.2563,
+      "grad_norm": 8.442815780639648,
+      "learning_rate": 5.4171974522293e-05,
+      "loss": 0.1699,
       "step": 2880
     },
     {
       "epoch": 4.62,
-      "grad_norm": 6.208673000335693,
-      "learning_rate": 5.386942675159236e-05,
-      "loss": 0.2484,
+      "grad_norm": 17.77415657043457,
+      "learning_rate": 5.3853503184713386e-05,
+      "loss": 0.1256,
       "step": 2900
     },
     {
       "epoch": 4.62,
-      "eval_accuracy": 0.7247619047619047,
-      "eval_loss": 1.0976409912109375,
-      "eval_runtime": 14.3178,
-      "eval_samples_per_second": 73.335,
-      "eval_steps_per_second": 9.219,
+      "eval_accuracy": 0.7171428571428572,
+      "eval_loss": 1.3512778282165527,
+      "eval_runtime": 14.2652,
+      "eval_samples_per_second": 73.606,
+      "eval_steps_per_second": 9.253,
       "step": 2900
     },
     {
       "epoch": 4.65,
-      "grad_norm": 2.2990214824676514,
-      "learning_rate": 5.3550955414012736e-05,
-      "loss": 0.1589,
+      "grad_norm": 17.13568115234375,
+      "learning_rate": 5.3535031847133764e-05,
+      "loss": 0.1093,
       "step": 2920
     },
     {
       "epoch": 4.68,
-      "grad_norm": 3.4065587520599365,
-      "learning_rate": 5.323248407643312e-05,
-      "loss": 0.1823,
+      "grad_norm": 1.0104303359985352,
+      "learning_rate": 5.321656050955415e-05,
+      "loss": 0.1486,
       "step": 2940
     },
     {
       "epoch": 4.71,
-      "grad_norm": 8.64983081817627,
-      "learning_rate": 5.2914012738853506e-05,
-      "loss": 0.214,
+      "grad_norm": 1.0769445896148682,
+      "learning_rate": 5.2898089171974527e-05,
+      "loss": 0.1452,
       "step": 2960
     },
     {
       "epoch": 4.75,
-      "grad_norm": 0.693291425704956,
-      "learning_rate": 5.2595541401273883e-05,
-      "loss": 0.2407,
+      "grad_norm": 3.9294614791870117,
+      "learning_rate": 5.257961783439491e-05,
+      "loss": 0.1679,
       "step": 2980
     },
     {
       "epoch": 4.78,
-      "grad_norm": 7.287389278411865,
-      "learning_rate": 5.227707006369427e-05,
-      "loss": 0.1575,
+      "grad_norm": 6.109922409057617,
+      "learning_rate": 5.2261146496815296e-05,
+      "loss": 0.1544,
       "step": 3000
     },
     {
       "epoch": 4.78,
-      "eval_accuracy": 0.7457142857142857,
-      "eval_loss": 1.0477701425552368,
-      "eval_runtime": 14.492,
-      "eval_samples_per_second": 72.454,
-      "eval_steps_per_second": 9.108,
+      "eval_accuracy": 0.7019047619047619,
+      "eval_loss": 1.4279906749725342,
+      "eval_runtime": 14.4347,
+      "eval_samples_per_second": 72.742,
+      "eval_steps_per_second": 9.145,
       "step": 3000
     },
     {
       "epoch": 4.81,
-      "grad_norm": 5.838592529296875,
-      "learning_rate": 5.1958598726114646e-05,
-      "loss": 0.2628,
+      "grad_norm": 1.6985957622528076,
+      "learning_rate": 5.1942675159235673e-05,
+      "loss": 0.1023,
       "step": 3020
     },
     {
       "epoch": 4.84,
-      "grad_norm": 5.59370756149292,
-      "learning_rate": 5.164012738853503e-05,
-      "loss": 0.1391,
+      "grad_norm": 12.082504272460938,
+      "learning_rate": 5.162420382165606e-05,
+      "loss": 0.1548,
       "step": 3040
     },
     {
       "epoch": 4.87,
-      "grad_norm": 1.1490521430969238,
-      "learning_rate": 5.1321656050955415e-05,
-      "loss": 0.2392,
+      "grad_norm": 3.0203235149383545,
+      "learning_rate": 5.130573248407644e-05,
+      "loss": 0.164,
       "step": 3060
     },
     {
       "epoch": 4.9,
-      "grad_norm": 4.223572731018066,
-      "learning_rate": 5.100318471337579e-05,
-      "loss": 0.2346,
+      "grad_norm": 1.4739816188812256,
+      "learning_rate": 5.098726114649682e-05,
+      "loss": 0.1553,
       "step": 3080
     },
     {
       "epoch": 4.94,
-      "grad_norm": 3.0692086219787598,
-      "learning_rate": 5.068471337579618e-05,
-      "loss": 0.2028,
+      "grad_norm": 0.4960005581378937,
+      "learning_rate": 5.0668789808917205e-05,
+      "loss": 0.0858,
       "step": 3100
     },
     {
       "epoch": 4.94,
-      "eval_accuracy": 0.7447619047619047,
-      "eval_loss": 1.0374425649642944,
-      "eval_runtime": 14.213,
-      "eval_samples_per_second": 73.876,
-      "eval_steps_per_second": 9.287,
+      "eval_accuracy": 0.7428571428571429,
+      "eval_loss": 1.2230653762817383,
+      "eval_runtime": 14.2515,
+      "eval_samples_per_second": 73.677,
+      "eval_steps_per_second": 9.262,
       "step": 3100
     },
     {
       "epoch": 4.97,
-      "grad_norm": 1.3413909673690796,
-      "learning_rate": 5.036624203821656e-05,
-      "loss": 0.1113,
+      "grad_norm": 0.42542117834091187,
+      "learning_rate": 5.035031847133759e-05,
+      "loss": 0.1885,
       "step": 3120
     },
     {
       "epoch": 5.0,
-      "grad_norm": 0.1451389044523239,
-      "learning_rate": 5.004777070063694e-05,
-      "loss": 0.1797,
+      "grad_norm": 0.01594359055161476,
+      "learning_rate": 5.003184713375797e-05,
+      "loss": 0.1823,
       "step": 3140
     },
     {
       "epoch": 5.03,
-      "grad_norm": 8.001447677612305,
-      "learning_rate": 4.9729299363057324e-05,
-      "loss": 0.1934,
+      "grad_norm": 5.340611934661865,
+      "learning_rate": 4.9713375796178345e-05,
+      "loss": 0.1836,
       "step": 3160
     },
     {
       "epoch": 5.06,
-      "grad_norm": 3.657029151916504,
-      "learning_rate": 4.941082802547771e-05,
-      "loss": 0.2077,
+      "grad_norm": 10.515073776245117,
+      "learning_rate": 4.939490445859873e-05,
+      "loss": 0.2215,
       "step": 3180
     },
     {
       "epoch": 5.1,
-      "grad_norm": 0.24746979773044586,
-      "learning_rate": 4.9092356687898087e-05,
-      "loss": 0.1173,
+      "grad_norm": 0.008586952462792397,
+      "learning_rate": 4.907643312101911e-05,
+      "loss": 0.1049,
       "step": 3200
     },
     {
       "epoch": 5.1,
-      "eval_accuracy": 0.7609523809523809,
-      "eval_loss": 1.0067189931869507,
-      "eval_runtime": 14.4189,
-      "eval_samples_per_second": 72.821,
-      "eval_steps_per_second": 9.155,
+      "eval_accuracy": 0.7352380952380952,
+      "eval_loss": 1.2775230407714844,
+      "eval_runtime": 14.3058,
+      "eval_samples_per_second": 73.397,
+      "eval_steps_per_second": 9.227,
       "step": 3200
     },
     {
       "epoch": 5.13,
-      "grad_norm": 1.1961082220077515,
-      "learning_rate": 4.877388535031847e-05,
-      "loss": 0.2267,
+      "grad_norm": 0.013670964166522026,
+      "learning_rate": 4.875796178343949e-05,
+      "loss": 0.1363,
       "step": 3220
     },
     {
       "epoch": 5.16,
-      "grad_norm": 6.003686904907227,
-      "learning_rate": 4.8455414012738856e-05,
-      "loss": 0.2543,
+      "grad_norm": 9.68531608581543,
+      "learning_rate": 4.8439490445859877e-05,
+      "loss": 0.1017,
       "step": 3240
     },
     {
       "epoch": 5.19,
-      "grad_norm": 16.887609481811523,
-      "learning_rate": 4.8136942675159233e-05,
-      "loss": 0.1849,
+      "grad_norm": 6.831617832183838,
+      "learning_rate": 4.8121019108280254e-05,
+      "loss": 0.1477,
       "step": 3260
     },
     {
       "epoch": 5.22,
-      "grad_norm": 1.4152742624282837,
-      "learning_rate": 4.781847133757962e-05,
-      "loss": 0.2119,
+      "grad_norm": 2.455923557281494,
+      "learning_rate": 4.780254777070064e-05,
+      "loss": 0.1324,
       "step": 3280
     },
     {
       "epoch": 5.25,
-      "grad_norm": 1.6081457138061523,
-      "learning_rate": 4.75e-05,
-      "loss": 0.1313,
+      "grad_norm": 4.937666416168213,
+      "learning_rate": 4.7484076433121024e-05,
+      "loss": 0.1361,
       "step": 3300
     },
     {
       "epoch": 5.25,
-      "eval_accuracy": 0.7266666666666667,
-      "eval_loss": 1.1970884799957275,
-      "eval_runtime": 14.5245,
-      "eval_samples_per_second": 72.292,
-      "eval_steps_per_second": 9.088,
+      "eval_accuracy": 0.7428571428571429,
+      "eval_loss": 1.2840485572814941,
+      "eval_runtime": 14.3537,
+      "eval_samples_per_second": 73.152,
+      "eval_steps_per_second": 9.196,
       "step": 3300
     },
     {
       "epoch": 5.29,
-      "grad_norm": 19.153751373291016,
-      "learning_rate": 4.718152866242038e-05,
-      "loss": 0.2472,
+      "grad_norm": 3.7385175228118896,
+      "learning_rate": 4.71656050955414e-05,
+      "loss": 0.1418,
       "step": 3320
     },
     {
       "epoch": 5.32,
-      "grad_norm": 4.610495567321777,
-      "learning_rate": 4.6863057324840765e-05,
-      "loss": 0.1872,
+      "grad_norm": 1.919258952140808,
+      "learning_rate": 4.6847133757961786e-05,
+      "loss": 0.1547,
       "step": 3340
     },
     {
       "epoch": 5.35,
-      "grad_norm": 1.8803948163986206,
-      "learning_rate": 4.654458598726115e-05,
-      "loss": 0.153,
+      "grad_norm": 0.17440159618854523,
+      "learning_rate": 4.652866242038217e-05,
+      "loss": 0.1155,
       "step": 3360
     },
     {
       "epoch": 5.38,
-      "grad_norm": 2.9183852672576904,
-      "learning_rate": 4.622611464968153e-05,
-      "loss": 0.1024,
+      "grad_norm": 10.752092361450195,
+      "learning_rate": 4.621019108280255e-05,
+      "loss": 0.1213,
       "step": 3380
     },
     {
       "epoch": 5.41,
-      "grad_norm": 11.132299423217773,
-      "learning_rate": 4.590764331210191e-05,
-      "loss": 0.2142,
+      "grad_norm": 0.030742734670639038,
+      "learning_rate": 4.589171974522293e-05,
+      "loss": 0.1505,
       "step": 3400
     },
     {
       "epoch": 5.41,
-      "eval_accuracy": 0.74,
-      "eval_loss": 1.1455038785934448,
-      "eval_runtime": 13.999,
-      "eval_samples_per_second": 75.006,
-      "eval_steps_per_second": 9.429,
+      "eval_accuracy": 0.7390476190476191,
+      "eval_loss": 1.3372679948806763,
+      "eval_runtime": 14.3909,
+      "eval_samples_per_second": 72.963,
+      "eval_steps_per_second": 9.172,
       "step": 3400
     },
     {
       "epoch": 5.45,
-      "grad_norm": 0.15127496421337128,
-      "learning_rate": 4.5589171974522296e-05,
-      "loss": 0.1674,
+      "grad_norm": 0.015463191084563732,
+      "learning_rate": 4.557324840764332e-05,
+      "loss": 0.0626,
       "step": 3420
     },
     {
       "epoch": 5.48,
-      "grad_norm": 4.990033149719238,
-      "learning_rate": 4.5270700636942674e-05,
-      "loss": 0.1906,
+      "grad_norm": 0.14180102944374084,
+      "learning_rate": 4.5254777070063695e-05,
+      "loss": 0.1208,
       "step": 3440
     },
     {
       "epoch": 5.51,
-      "grad_norm": 4.008519649505615,
-      "learning_rate": 4.495222929936306e-05,
-      "loss": 0.1497,
+      "grad_norm": 1.5291404724121094,
+      "learning_rate": 4.493630573248408e-05,
+      "loss": 0.1032,
       "step": 3460
     },
     {
       "epoch": 5.54,
-      "grad_norm": 1.5550432205200195,
-      "learning_rate": 4.4633757961783443e-05,
-      "loss": 0.1006,
+      "grad_norm": 3.196084976196289,
+      "learning_rate": 4.4617834394904464e-05,
+      "loss": 0.2026,
       "step": 3480
     },
     {
       "epoch": 5.57,
-      "grad_norm": 0.13081075251102448,
-      "learning_rate": 4.431528662420382e-05,
-      "loss": 0.1302,
+      "grad_norm": 6.2625837326049805,
+      "learning_rate": 4.429936305732484e-05,
+      "loss": 0.1244,
       "step": 3500
     },
     {
       "epoch": 5.57,
-      "eval_accuracy": 0.7628571428571429,
-      "eval_loss": 1.0319159030914307,
-      "eval_runtime": 14.3106,
-      "eval_samples_per_second": 73.372,
-      "eval_steps_per_second": 9.224,
+      "eval_accuracy": 0.7438095238095238,
+      "eval_loss": 1.2958623170852661,
+      "eval_runtime": 14.333,
+      "eval_samples_per_second": 73.258,
+      "eval_steps_per_second": 9.21,
       "step": 3500
     },
     {
       "epoch": 5.61,
-      "grad_norm": 4.138894081115723,
-      "learning_rate": 4.3996815286624206e-05,
-      "loss": 0.1988,
+      "grad_norm": 2.965757369995117,
+      "learning_rate": 4.398089171974523e-05,
+      "loss": 0.1397,
       "step": 3520
     },
     {
       "epoch": 5.64,
-      "grad_norm": 0.9299852848052979,
-      "learning_rate": 4.3678343949044584e-05,
-      "loss": 0.1264,
+      "grad_norm": 2.413609027862549,
+      "learning_rate": 4.366242038216561e-05,
+      "loss": 0.1064,
       "step": 3540
     },
     {
       "epoch": 5.67,
-      "grad_norm": 0.05426739901304245,
-      "learning_rate": 4.335987261146497e-05,
-      "loss": 0.184,
+      "grad_norm": 0.011031672358512878,
+      "learning_rate": 4.334394904458599e-05,
+      "loss": 0.0878,
       "step": 3560
     },
     {
       "epoch": 5.7,
-      "grad_norm": 0.8005551695823669,
-      "learning_rate": 4.304140127388535e-05,
-      "loss": 0.1535,
+      "grad_norm": 8.136690139770508,
+      "learning_rate": 4.3025477707006374e-05,
+      "loss": 0.1437,
       "step": 3580
     },
     {
       "epoch": 5.73,
-      "grad_norm": 6.299161911010742,
-      "learning_rate": 4.272292993630573e-05,
-      "loss": 0.2193,
+      "grad_norm": 9.638904571533203,
+      "learning_rate": 4.270700636942676e-05,
+      "loss": 0.1114,
       "step": 3600
     },
     {
       "epoch": 5.73,
-      "eval_accuracy": 0.7733333333333333,
-      "eval_loss": 0.9745522141456604,
-      "eval_runtime": 14.4492,
-      "eval_samples_per_second": 72.669,
-      "eval_steps_per_second": 9.135,
+      "eval_accuracy": 0.7380952380952381,
+      "eval_loss": 1.318123698234558,
+      "eval_runtime": 14.0757,
+      "eval_samples_per_second": 74.597,
+      "eval_steps_per_second": 9.378,
       "step": 3600
     },
     {
       "epoch": 5.76,
-      "grad_norm": 5.8440937995910645,
-      "learning_rate": 4.2404458598726115e-05,
-      "loss": 0.2346,
+      "grad_norm": 2.292039394378662,
+      "learning_rate": 4.2388535031847136e-05,
+      "loss": 0.1465,
       "step": 3620
     },
     {
       "epoch": 5.8,
-      "grad_norm": 0.3318318724632263,
-      "learning_rate": 4.20859872611465e-05,
-      "loss": 0.2378,
+      "grad_norm": 0.009739763103425503,
+      "learning_rate": 4.207006369426752e-05,
+      "loss": 0.1259,
       "step": 3640
     },
     {
       "epoch": 5.83,
-      "grad_norm": 6.423093318939209,
-      "learning_rate": 4.176751592356688e-05,
-      "loss": 0.2194,
+      "grad_norm": 0.010617711581289768,
+      "learning_rate": 4.1751592356687905e-05,
+      "loss": 0.1131,
       "step": 3660
     },
     {
       "epoch": 5.86,
-      "grad_norm": 1.1782708168029785,
-      "learning_rate": 4.144904458598726e-05,
-      "loss": 0.2442,
+      "grad_norm": 0.3622857332229614,
+      "learning_rate": 4.143312101910828e-05,
+      "loss": 0.1085,
       "step": 3680
     },
     {
       "epoch": 5.89,
-      "grad_norm": 4.644681930541992,
-      "learning_rate": 4.1130573248407647e-05,
-      "loss": 0.1778,
+      "grad_norm": 12.089427947998047,
+      "learning_rate": 4.111464968152867e-05,
+      "loss": 0.0851,
       "step": 3700
     },
     {
       "epoch": 5.89,
-      "eval_accuracy": 0.7552380952380953,
-      "eval_loss": 1.0206609964370728,
-      "eval_runtime": 14.406,
-      "eval_samples_per_second": 72.886,
-      "eval_steps_per_second": 9.163,
+      "eval_accuracy": 0.7457142857142857,
+      "eval_loss": 1.3287726640701294,
+      "eval_runtime": 14.124,
+      "eval_samples_per_second": 74.341,
+      "eval_steps_per_second": 9.346,
       "step": 3700
     },
     {
       "epoch": 5.92,
-      "grad_norm": 2.935955762863159,
-      "learning_rate": 4.0812101910828024e-05,
-      "loss": 0.2641,
+      "grad_norm": 6.460238933563232,
+      "learning_rate": 4.0796178343949045e-05,
+      "loss": 0.1665,
       "step": 3720
     },
     {
       "epoch": 5.96,
-      "grad_norm": 12.24996566772461,
-      "learning_rate": 4.049363057324841e-05,
-      "loss": 0.2185,
+      "grad_norm": 1.5777150392532349,
+      "learning_rate": 4.047770700636943e-05,
+      "loss": 0.0867,
       "step": 3740
     },
     {
       "epoch": 5.99,
-      "grad_norm": 4.923351287841797,
-      "learning_rate": 4.0175159235668793e-05,
-      "loss": 0.1169,
+      "grad_norm": 6.013555526733398,
+      "learning_rate": 4.0159235668789814e-05,
+      "loss": 0.1668,
       "step": 3760
     },
     {
       "epoch": 6.02,
-      "grad_norm": 3.076030731201172,
-      "learning_rate": 3.985668789808917e-05,
-      "loss": 0.1629,
+      "grad_norm": 0.11355218291282654,
+      "learning_rate": 3.984076433121019e-05,
+      "loss": 0.2062,
       "step": 3780
     },
     {
       "epoch": 6.05,
-      "grad_norm": 1.7281866073608398,
-      "learning_rate": 3.9538216560509556e-05,
-      "loss": 0.1003,
+      "grad_norm": 2.2892956733703613,
+      "learning_rate": 3.952229299363058e-05,
+      "loss": 0.0799,
       "step": 3800
     },
     {
       "epoch": 6.05,
-      "eval_accuracy": 0.7638095238095238,
-      "eval_loss": 1.0537958145141602,
-      "eval_runtime": 14.4611,
-      "eval_samples_per_second": 72.609,
-      "eval_steps_per_second": 9.128,
+      "eval_accuracy": 0.76,
+      "eval_loss": 1.1858717203140259,
+      "eval_runtime": 14.3784,
+      "eval_samples_per_second": 73.026,
+      "eval_steps_per_second": 9.18,
       "step": 3800
     },
     {
       "epoch": 6.08,
-      "grad_norm": 4.985991477966309,
-      "learning_rate": 3.921974522292994e-05,
-      "loss": 0.1808,
+      "grad_norm": 3.4917514324188232,
+      "learning_rate": 3.920382165605096e-05,
+      "loss": 0.1793,
       "step": 3820
     },
     {
       "epoch": 6.11,
-      "grad_norm": 0.1005251556634903,
-      "learning_rate": 3.890127388535032e-05,
-      "loss": 0.1514,
+      "grad_norm": 0.31258919835090637,
+      "learning_rate": 3.888535031847134e-05,
+      "loss": 0.1873,
       "step": 3840
     },
     {
       "epoch": 6.15,
-      "grad_norm": 4.247957706451416,
-      "learning_rate": 3.85828025477707e-05,
-      "loss": 0.1823,
+      "grad_norm": 3.590149164199829,
+      "learning_rate": 3.8566878980891724e-05,
+      "loss": 0.1306,
       "step": 3860
     },
     {
       "epoch": 6.18,
-      "grad_norm": 4.05401611328125,
-      "learning_rate": 3.8280254777070066e-05,
-      "loss": 0.1412,
+      "grad_norm": 0.22531455755233765,
+      "learning_rate": 3.824840764331211e-05,
+      "loss": 0.0829,
       "step": 3880
     },
     {
       "epoch": 6.21,
-      "grad_norm": 0.09717115759849548,
-      "learning_rate": 3.796178343949045e-05,
-      "loss": 0.1644,
+      "grad_norm": 4.326334476470947,
+      "learning_rate": 3.7929936305732486e-05,
+      "loss": 0.1331,
       "step": 3900
     },
     {
       "epoch": 6.21,
-      "eval_accuracy": 0.7304761904761905,
-      "eval_loss": 1.1832480430603027,
-      "eval_runtime": 14.3128,
-      "eval_samples_per_second": 73.361,
-      "eval_steps_per_second": 9.222,
+      "eval_accuracy": 0.7371428571428571,
+      "eval_loss": 1.2544463872909546,
+      "eval_runtime": 14.5697,
+      "eval_samples_per_second": 72.067,
+      "eval_steps_per_second": 9.06,
       "step": 3900
     },
     {
       "epoch": 6.24,
-      "grad_norm": 6.105861663818359,
-      "learning_rate": 3.7643312101910836e-05,
-      "loss": 0.2253,
+      "grad_norm": 2.9954144954681396,
+      "learning_rate": 3.761146496815287e-05,
+      "loss": 0.1022,
       "step": 3920
     },
     {
       "epoch": 6.27,
-      "grad_norm": 0.2512984573841095,
-      "learning_rate": 3.7324840764331207e-05,
-      "loss": 0.1698,
+      "grad_norm": 2.7083683013916016,
+      "learning_rate": 3.729299363057325e-05,
+      "loss": 0.168,
       "step": 3940
     },
     {
       "epoch": 6.31,
-      "grad_norm": 0.045377813279628754,
-      "learning_rate": 3.700636942675159e-05,
-      "loss": 0.1442,
+      "grad_norm": 7.03603458404541,
+      "learning_rate": 3.697452229299363e-05,
+      "loss": 0.1202,
       "step": 3960
     },
     {
       "epoch": 6.34,
-      "grad_norm": 8.247663497924805,
-      "learning_rate": 3.6687898089171976e-05,
-      "loss": 0.1449,
+      "grad_norm": 0.9397388100624084,
+      "learning_rate": 3.665605095541401e-05,
+      "loss": 0.0837,
       "step": 3980
     },
     {
       "epoch": 6.37,
-      "grad_norm": 10.677542686462402,
-      "learning_rate": 3.6369426751592353e-05,
-      "loss": 0.1843,
+      "grad_norm": 5.8838276863098145,
+      "learning_rate": 3.6337579617834395e-05,
+      "loss": 0.121,
       "step": 4000
     },
     {
       "epoch": 6.37,
-      "eval_accuracy": 0.7495238095238095,
-      "eval_loss": 1.081445336341858,
-      "eval_runtime": 14.074,
-      "eval_samples_per_second": 74.605,
-      "eval_steps_per_second": 9.379,
+      "eval_accuracy": 0.7533333333333333,
+      "eval_loss": 1.2185813188552856,
+      "eval_runtime": 14.2575,
+      "eval_samples_per_second": 73.645,
+      "eval_steps_per_second": 9.258,
       "step": 4000
     },
     {
       "epoch": 6.4,
-      "grad_norm": 1.8205310106277466,
-      "learning_rate": 3.605095541401274e-05,
-      "loss": 0.1621,
+      "grad_norm": 3.0600955486297607,
+      "learning_rate": 3.601910828025477e-05,
+      "loss": 0.1115,
       "step": 4020
     },
     {
       "epoch": 6.43,
-      "grad_norm": 4.124607086181641,
-      "learning_rate": 3.573248407643312e-05,
-      "loss": 0.1634,
+      "grad_norm": 2.251216173171997,
+      "learning_rate": 3.570063694267516e-05,
+      "loss": 0.1661,
       "step": 4040
     },
     {
       "epoch": 6.46,
-      "grad_norm": 0.3355584442615509,
-      "learning_rate": 3.54140127388535e-05,
-      "loss": 0.2083,
+      "grad_norm": 5.1470770835876465,
+      "learning_rate": 3.538216560509554e-05,
+      "loss": 0.1254,
       "step": 4060
     },
     {
       "epoch": 6.5,
-      "grad_norm": 0.12367378920316696,
-      "learning_rate": 3.5095541401273885e-05,
-      "loss": 0.148,
+      "grad_norm": 0.48159414529800415,
+      "learning_rate": 3.506369426751592e-05,
+      "loss": 0.1516,
       "step": 4080
     },
     {
       "epoch": 6.53,
-      "grad_norm": 0.021049553528428078,
-      "learning_rate": 3.477707006369427e-05,
-      "loss": 0.129,
+      "grad_norm": 3.215970754623413,
+      "learning_rate": 3.4745222929936305e-05,
+      "loss": 0.1276,
       "step": 4100
     },
     {
       "epoch": 6.53,
-      "eval_accuracy": 0.72,
-      "eval_loss": 1.247863531112671,
-      "eval_runtime": 14.2349,
-      "eval_samples_per_second": 73.762,
-      "eval_steps_per_second": 9.273,
+      "eval_accuracy": 0.7323809523809524,
+      "eval_loss": 1.2963875532150269,
+      "eval_runtime": 14.2554,
+      "eval_samples_per_second": 73.656,
+      "eval_steps_per_second": 9.26,
       "step": 4100
     },
     {
       "epoch": 6.56,
-      "grad_norm": 1.4821038246154785,
-      "learning_rate": 3.445859872611465e-05,
-      "loss": 0.1379,
+      "grad_norm": 0.053949933499097824,
+      "learning_rate": 3.442675159235669e-05,
+      "loss": 0.1387,
       "step": 4120
     },
     {
       "epoch": 6.59,
-      "grad_norm": 2.6189982891082764,
-      "learning_rate": 3.414012738853503e-05,
-      "loss": 0.1667,
+      "grad_norm": 3.1332573890686035,
+      "learning_rate": 3.410828025477707e-05,
+      "loss": 0.1468,
       "step": 4140
     },
     {
       "epoch": 6.62,
-      "grad_norm": 1.175626516342163,
-      "learning_rate": 3.3821656050955416e-05,
-      "loss": 0.2009,
+      "grad_norm": 1.160008192062378,
+      "learning_rate": 3.378980891719745e-05,
+      "loss": 0.1335,
       "step": 4160
     },
     {
       "epoch": 6.66,
-      "grad_norm": 5.35676908493042,
-      "learning_rate": 3.3503184713375794e-05,
-      "loss": 0.1821,
+      "grad_norm": 2.326484441757202,
+      "learning_rate": 3.3471337579617836e-05,
+      "loss": 0.0792,
       "step": 4180
     },
     {
       "epoch": 6.69,
-      "grad_norm": 2.7101659774780273,
-      "learning_rate": 3.318471337579618e-05,
-      "loss": 0.17,
+      "grad_norm": 8.954107284545898,
+      "learning_rate": 3.31687898089172e-05,
+      "loss": 0.1194,
       "step": 4200
     },
     {
       "epoch": 6.69,
-      "eval_accuracy": 0.741904761904762,
-      "eval_loss": 1.1575206518173218,
-      "eval_runtime": 14.4135,
-      "eval_samples_per_second": 72.848,
-      "eval_steps_per_second": 9.158,
+      "eval_accuracy": 0.7590476190476191,
+      "eval_loss": 1.1906794309616089,
+      "eval_runtime": 14.2444,
+      "eval_samples_per_second": 73.713,
+      "eval_steps_per_second": 9.267,
       "step": 4200
     },
     {
       "epoch": 6.72,
-      "grad_norm": 4.070112705230713,
-      "learning_rate": 3.286624203821656e-05,
-      "loss": 0.2064,
+      "grad_norm": 6.413561820983887,
+      "learning_rate": 3.2850318471337584e-05,
+      "loss": 0.1239,
       "step": 4220
     },
     {
       "epoch": 6.75,
-      "grad_norm": 0.4738242030143738,
-      "learning_rate": 3.254777070063694e-05,
-      "loss": 0.1836,
+      "grad_norm": 5.949230194091797,
+      "learning_rate": 3.253184713375796e-05,
+      "loss": 0.1326,
       "step": 4240
     },
     {
       "epoch": 6.78,
-      "grad_norm": 6.459200859069824,
-      "learning_rate": 3.2229299363057326e-05,
-      "loss": 0.1277,
+      "grad_norm": 7.957036972045898,
+      "learning_rate": 3.221337579617835e-05,
+      "loss": 0.0859,
       "step": 4260
     },
     {
       "epoch": 6.82,
-      "grad_norm": 4.038717746734619,
-      "learning_rate": 3.191082802547771e-05,
-      "loss": 0.1903,
+      "grad_norm": 0.8040112853050232,
+      "learning_rate": 3.189490445859873e-05,
+      "loss": 0.0546,
       "step": 4280
     },
     {
       "epoch": 6.85,
-      "grad_norm": 1.6985381841659546,
-      "learning_rate": 3.159235668789809e-05,
-      "loss": 0.2184,
+      "grad_norm": 12.735435485839844,
+      "learning_rate": 3.157643312101911e-05,
+      "loss": 0.1649,
       "step": 4300
     },
     {
       "epoch": 6.85,
-      "eval_accuracy": 0.7561904761904762,
-      "eval_loss": 1.0945791006088257,
-      "eval_runtime": 14.3853,
-      "eval_samples_per_second": 72.991,
-      "eval_steps_per_second": 9.176,
+      "eval_accuracy": 0.7104761904761905,
+      "eval_loss": 1.4679057598114014,
+      "eval_runtime": 14.4494,
+      "eval_samples_per_second": 72.667,
+      "eval_steps_per_second": 9.135,
       "step": 4300
     },
     {
       "epoch": 6.88,
-      "grad_norm": 1.8890398740768433,
-      "learning_rate": 3.127388535031847e-05,
-      "loss": 0.1224,
+      "grad_norm": 2.7187016010284424,
+      "learning_rate": 3.1257961783439494e-05,
+      "loss": 0.1065,
       "step": 4320
     },
     {
       "epoch": 6.91,
-      "grad_norm": 5.5975799560546875,
-      "learning_rate": 3.095541401273885e-05,
-      "loss": 0.1936,
+      "grad_norm": 6.976973056793213,
+      "learning_rate": 3.093949044585988e-05,
+      "loss": 0.1179,
       "step": 4340
     },
     {
       "epoch": 6.94,
-      "grad_norm": 0.805150032043457,
-      "learning_rate": 3.0636942675159235e-05,
-      "loss": 0.1103,
+      "grad_norm": 0.4394014775753021,
+      "learning_rate": 3.0621019108280256e-05,
+      "loss": 0.1598,
       "step": 4360
     },
     {
       "epoch": 6.97,
-      "grad_norm": 6.093379020690918,
-      "learning_rate": 3.031847133757962e-05,
-      "loss": 0.1666,
+      "grad_norm": 0.00932831410318613,
+      "learning_rate": 3.030254777070064e-05,
+      "loss": 0.1628,
       "step": 4380
     },
     {
       "epoch": 7.01,
-      "grad_norm": 0.4640253484249115,
-      "learning_rate": 3e-05,
-      "loss": 0.1506,
+      "grad_norm": 0.08716867864131927,
+      "learning_rate": 2.998407643312102e-05,
+      "loss": 0.0558,
       "step": 4400
     },
     {
       "epoch": 7.01,
-      "eval_accuracy": 0.7714285714285715,
-      "eval_loss": 1.0580005645751953,
-      "eval_runtime": 14.4045,
-      "eval_samples_per_second": 72.894,
-      "eval_steps_per_second": 9.164,
+      "eval_accuracy": 0.7533333333333333,
+      "eval_loss": 1.2028064727783203,
+      "eval_runtime": 14.2961,
+      "eval_samples_per_second": 73.446,
+      "eval_steps_per_second": 9.233,
       "step": 4400
     },
     {
       "epoch": 7.04,
-      "grad_norm": 0.06945107132196426,
-      "learning_rate": 2.9681528662420382e-05,
-      "loss": 0.1537,
+      "grad_norm": 0.5166774392127991,
+      "learning_rate": 2.9665605095541403e-05,
+      "loss": 0.091,
       "step": 4420
     },
     {
       "epoch": 7.07,
-      "grad_norm": 1.686673879623413,
-      "learning_rate": 2.9363057324840763e-05,
-      "loss": 0.1661,
+      "grad_norm": 10.801895141601562,
+      "learning_rate": 2.9347133757961787e-05,
+      "loss": 0.127,
       "step": 4440
     },
     {
       "epoch": 7.1,
-      "grad_norm": 3.4670166969299316,
-      "learning_rate": 2.9044585987261148e-05,
-      "loss": 0.1128,
+      "grad_norm": 2.0005249977111816,
+      "learning_rate": 2.902866242038217e-05,
+      "loss": 0.1329,
       "step": 4460
     },
     {
       "epoch": 7.13,
-      "grad_norm": 3.8089122772216797,
-      "learning_rate": 2.872611464968153e-05,
-      "loss": 0.1505,
+      "grad_norm": 0.037386998534202576,
+      "learning_rate": 2.871019108280255e-05,
+      "loss": 0.1703,
       "step": 4480
     },
     {
       "epoch": 7.17,
-      "grad_norm": 0.22587546706199646,
-      "learning_rate": 2.840764331210191e-05,
-      "loss": 0.1099,
+      "grad_norm": 0.06666604429483414,
+      "learning_rate": 2.8391719745222934e-05,
+      "loss": 0.0687,
       "step": 4500
     },
     {
       "epoch": 7.17,
-      "eval_accuracy": 0.7638095238095238,
-      "eval_loss": 1.0479028224945068,
-      "eval_runtime": 14.3476,
-      "eval_samples_per_second": 73.183,
-      "eval_steps_per_second": 9.2,
+      "eval_accuracy": 0.7380952380952381,
+      "eval_loss": 1.3242312669754028,
+      "eval_runtime": 14.4705,
+      "eval_samples_per_second": 72.561,
+      "eval_steps_per_second": 9.122,
       "step": 4500
     },
     {
       "epoch": 7.2,
-      "grad_norm": 0.4526523947715759,
-      "learning_rate": 2.8089171974522295e-05,
-      "loss": 0.0994,
+      "grad_norm": 0.09800776094198227,
+      "learning_rate": 2.8073248407643316e-05,
+      "loss": 0.1342,
       "step": 4520
     },
     {
       "epoch": 7.23,
-      "grad_norm": 2.4940195083618164,
-      "learning_rate": 2.7770700636942676e-05,
-      "loss": 0.1098,
+      "grad_norm": 0.050632625818252563,
+      "learning_rate": 2.7754777070063697e-05,
+      "loss": 0.0791,
       "step": 4540
     },
     {
       "epoch": 7.26,
-      "grad_norm": 1.7795358896255493,
-      "learning_rate": 2.7452229299363057e-05,
-      "loss": 0.1648,
+      "grad_norm": 5.7890777587890625,
+      "learning_rate": 2.743630573248408e-05,
+      "loss": 0.136,
       "step": 4560
     },
     {
       "epoch": 7.29,
-      "grad_norm": 8.93464469909668,
-      "learning_rate": 2.713375796178344e-05,
-      "loss": 0.151,
+      "grad_norm": 2.2278764247894287,
+      "learning_rate": 2.7117834394904462e-05,
+      "loss": 0.0671,
       "step": 4580
     },
     {
       "epoch": 7.32,
-      "grad_norm": 0.04857097566127777,
-      "learning_rate": 2.6815286624203823e-05,
-      "loss": 0.1226,
+      "grad_norm": 2.3305046558380127,
+      "learning_rate": 2.6799363057324844e-05,
+      "loss": 0.1419,
       "step": 4600
     },
     {
       "epoch": 7.32,
-      "eval_accuracy": 0.7495238095238095,
-      "eval_loss": 1.1306982040405273,
-      "eval_runtime": 14.3492,
-      "eval_samples_per_second": 73.175,
-      "eval_steps_per_second": 9.199,
+      "eval_accuracy": 0.76,
+      "eval_loss": 1.2328267097473145,
+      "eval_runtime": 14.3513,
+      "eval_samples_per_second": 73.164,
+      "eval_steps_per_second": 9.198,
       "step": 4600
     },
     {
       "epoch": 7.36,
-      "grad_norm": 1.8932238817214966,
-      "learning_rate": 2.6496815286624204e-05,
-      "loss": 0.1225,
+      "grad_norm": 0.22716645896434784,
+      "learning_rate": 2.6480891719745228e-05,
+      "loss": 0.0494,
       "step": 4620
     },
     {
       "epoch": 7.39,
-      "grad_norm": 10.580707550048828,
-      "learning_rate": 2.617834394904459e-05,
-      "loss": 0.1567,
+      "grad_norm": 0.10885431617498398,
+      "learning_rate": 2.616242038216561e-05,
+      "loss": 0.1191,
       "step": 4640
     },
     {
       "epoch": 7.42,
-      "grad_norm": 0.38718360662460327,
-      "learning_rate": 2.585987261146497e-05,
-      "loss": 0.1251,
+      "grad_norm": 0.01832442916929722,
+      "learning_rate": 2.584394904458599e-05,
+      "loss": 0.0763,
       "step": 4660
     },
     {
       "epoch": 7.45,
-      "grad_norm": 9.657524108886719,
-      "learning_rate": 2.554140127388535e-05,
-      "loss": 0.1076,
+      "grad_norm": 0.012117845937609673,
+      "learning_rate": 2.5525477707006372e-05,
+      "loss": 0.0664,
       "step": 4680
     },
     {
       "epoch": 7.48,
-      "grad_norm": 0.019253961741924286,
-      "learning_rate": 2.5222929936305732e-05,
-      "loss": 0.2122,
+      "grad_norm": 0.015309098176658154,
+      "learning_rate": 2.5207006369426756e-05,
+      "loss": 0.0901,
       "step": 4700
     },
     {
       "epoch": 7.48,
-      "eval_accuracy": 0.7285714285714285,
-      "eval_loss": 1.2837932109832764,
-      "eval_runtime": 14.1311,
-      "eval_samples_per_second": 74.304,
-      "eval_steps_per_second": 9.341,
+      "eval_accuracy": 0.7676190476190476,
+      "eval_loss": 1.186057209968567,
+      "eval_runtime": 14.3129,
+      "eval_samples_per_second": 73.361,
+      "eval_steps_per_second": 9.222,
       "step": 4700
     },
     {
       "epoch": 7.52,
-      "grad_norm": 0.8583599925041199,
-      "learning_rate": 2.4904458598726117e-05,
-      "loss": 0.1148,
+      "grad_norm": 4.8887224197387695,
+      "learning_rate": 2.4888535031847134e-05,
+      "loss": 0.09,
       "step": 4720
     },
     {
       "epoch": 7.55,
-      "grad_norm": 0.6162470579147339,
-      "learning_rate": 2.4585987261146498e-05,
-      "loss": 0.1698,
+      "grad_norm": 0.4825209677219391,
+      "learning_rate": 2.457006369426752e-05,
+      "loss": 0.0831,
       "step": 4740
     },
     {
       "epoch": 7.58,
-      "grad_norm": 0.9160044193267822,
-      "learning_rate": 2.426751592356688e-05,
-      "loss": 0.0871,
+      "grad_norm": 2.224686861038208,
+      "learning_rate": 2.42515923566879e-05,
+      "loss": 0.069,
       "step": 4760
     },
     {
       "epoch": 7.61,
-      "grad_norm": 0.59488445520401,
-      "learning_rate": 2.3949044585987263e-05,
-      "loss": 0.1455,
+      "grad_norm": 0.07285654544830322,
+      "learning_rate": 2.393312101910828e-05,
+      "loss": 0.1224,
       "step": 4780
     },
     {
       "epoch": 7.64,
-      "grad_norm": 18.75020980834961,
-      "learning_rate": 2.3630573248407645e-05,
-      "loss": 0.1565,
+      "grad_norm": 12.75794506072998,
+      "learning_rate": 2.3614649681528662e-05,
+      "loss": 0.1181,
       "step": 4800
     },
     {
       "epoch": 7.64,
-      "eval_accuracy": 0.7390476190476191,
-      "eval_loss": 1.2039564847946167,
-      "eval_runtime": 14.3722,
-      "eval_samples_per_second": 73.058,
-      "eval_steps_per_second": 9.184,
+      "eval_accuracy": 0.7352380952380952,
+      "eval_loss": 1.403079867362976,
+      "eval_runtime": 14.2734,
+      "eval_samples_per_second": 73.564,
+      "eval_steps_per_second": 9.248,
       "step": 4800
     },
     {
       "epoch": 7.68,
-      "grad_norm": 0.011890099383890629,
-      "learning_rate": 2.3312101910828026e-05,
-      "loss": 0.0881,
+      "grad_norm": 2.022967576980591,
+      "learning_rate": 2.3296178343949047e-05,
+      "loss": 0.1238,
       "step": 4820
     },
     {
       "epoch": 7.71,
-      "grad_norm": 2.5114028453826904,
-      "learning_rate": 2.299363057324841e-05,
-      "loss": 0.1385,
+      "grad_norm": 4.282898426055908,
+      "learning_rate": 2.2977707006369428e-05,
+      "loss": 0.101,
       "step": 4840
     },
     {
       "epoch": 7.74,
-      "grad_norm": 0.410961389541626,
-      "learning_rate": 2.267515923566879e-05,
-      "loss": 0.1422,
+      "grad_norm": 16.033994674682617,
+      "learning_rate": 2.265923566878981e-05,
+      "loss": 0.1177,
       "step": 4860
     },
     {
       "epoch": 7.77,
-      "grad_norm": 0.022722242400050163,
-      "learning_rate": 2.2356687898089173e-05,
-      "loss": 0.1176,
+      "grad_norm": 0.018929600715637207,
+      "learning_rate": 2.2340764331210194e-05,
+      "loss": 0.163,
       "step": 4880
     },
     {
       "epoch": 7.8,
-      "grad_norm": 0.026792345568537712,
-      "learning_rate": 2.2038216560509557e-05,
-      "loss": 0.151,
+      "grad_norm": 9.345541954040527,
+      "learning_rate": 2.2022292993630575e-05,
+      "loss": 0.1272,
       "step": 4900
     },
     {
       "epoch": 7.8,
-      "eval_accuracy": 0.7428571428571429,
-      "eval_loss": 1.2361027002334595,
-      "eval_runtime": 14.3298,
-      "eval_samples_per_second": 73.274,
-      "eval_steps_per_second": 9.212,
+      "eval_accuracy": 0.7438095238095238,
+      "eval_loss": 1.360774040222168,
+      "eval_runtime": 14.3415,
+      "eval_samples_per_second": 73.214,
+      "eval_steps_per_second": 9.204,
       "step": 4900
     },
     {
       "epoch": 7.83,
-      "grad_norm": 5.774230480194092,
-      "learning_rate": 2.171974522292994e-05,
-      "loss": 0.1663,
+      "grad_norm": 2.5780770778656006,
+      "learning_rate": 2.1703821656050956e-05,
+      "loss": 0.0676,
       "step": 4920
     },
     {
       "epoch": 7.87,
-      "grad_norm": 0.011655393987894058,
-      "learning_rate": 2.140127388535032e-05,
-      "loss": 0.1085,
+      "grad_norm": 3.4303979873657227,
+      "learning_rate": 2.138535031847134e-05,
+      "loss": 0.1564,
       "step": 4940
     },
     {
       "epoch": 7.9,
-      "grad_norm": 2.6619997024536133,
-      "learning_rate": 2.10828025477707e-05,
-      "loss": 0.115,
+      "grad_norm": 0.17826871573925018,
+      "learning_rate": 2.1066878980891722e-05,
+      "loss": 0.0475,
       "step": 4960
     },
     {
       "epoch": 7.93,
-      "grad_norm": 6.465090274810791,
-      "learning_rate": 2.0764331210191085e-05,
-      "loss": 0.1723,
+      "grad_norm": 2.283670663833618,
+      "learning_rate": 2.0748407643312103e-05,
+      "loss": 0.0894,
       "step": 4980
     },
     {
       "epoch": 7.96,
-      "grad_norm": 8.653176307678223,
-      "learning_rate": 2.0445859872611467e-05,
-      "loss": 0.0934,
+      "grad_norm": 0.17847611010074615,
+      "learning_rate": 2.0429936305732488e-05,
+      "loss": 0.0979,
       "step": 5000
     },
     {
       "epoch": 7.96,
-      "eval_accuracy": 0.7457142857142857,
-      "eval_loss": 1.198464274406433,
-      "eval_runtime": 14.2409,
-      "eval_samples_per_second": 73.731,
-      "eval_steps_per_second": 9.269,
+      "eval_accuracy": 0.7495238095238095,
+      "eval_loss": 1.3098150491714478,
+      "eval_runtime": 14.3921,
+      "eval_samples_per_second": 72.957,
+      "eval_steps_per_second": 9.172,
       "step": 5000
     },
     {
       "epoch": 7.99,
-      "grad_norm": 4.535229206085205,
-      "learning_rate": 2.0127388535031848e-05,
-      "loss": 0.1456,
+      "grad_norm": 0.007124764379113913,
+      "learning_rate": 2.011146496815287e-05,
+      "loss": 0.1294,
       "step": 5020
     },
     {
       "epoch": 8.03,
-      "grad_norm": 11.831160545349121,
-      "learning_rate": 1.9808917197452232e-05,
-      "loss": 0.1115,
+      "grad_norm": 0.2644352912902832,
+      "learning_rate": 1.979299363057325e-05,
+      "loss": 0.0774,
       "step": 5040
     },
     {
       "epoch": 8.06,
-      "grad_norm": 8.955262184143066,
-      "learning_rate": 1.9490445859872614e-05,
-      "loss": 0.1298,
+      "grad_norm": 0.16474761068820953,
+      "learning_rate": 1.947452229299363e-05,
+      "loss": 0.1332,
       "step": 5060
     },
     {
       "epoch": 8.09,
-      "grad_norm": 5.660279273986816,
-      "learning_rate": 1.9171974522292995e-05,
-      "loss": 0.1849,
+      "grad_norm": 0.009978950954973698,
+      "learning_rate": 1.9156050955414016e-05,
+      "loss": 0.1025,
       "step": 5080
     },
     {
       "epoch": 8.12,
-      "grad_norm": 6.7122087478637695,
-      "learning_rate": 1.885350318471338e-05,
-      "loss": 0.1374,
+      "grad_norm": 2.784860849380493,
+      "learning_rate": 1.8837579617834397e-05,
+      "loss": 0.0805,
       "step": 5100
     },
     {
       "epoch": 8.12,
-      "eval_accuracy": 0.7495238095238095,
-      "eval_loss": 1.136474370956421,
-      "eval_runtime": 14.3605,
-      "eval_samples_per_second": 73.117,
-      "eval_steps_per_second": 9.192,
+      "eval_accuracy": 0.7533333333333333,
+      "eval_loss": 1.2445318698883057,
+      "eval_runtime": 14.314,
+      "eval_samples_per_second": 73.355,
+      "eval_steps_per_second": 9.222,
       "step": 5100
     },
     {
       "epoch": 8.15,
-      "grad_norm": 0.83880215883255,
-      "learning_rate": 1.8535031847133757e-05,
-      "loss": 0.1406,
+      "grad_norm": 6.736931324005127,
+      "learning_rate": 1.8519108280254778e-05,
+      "loss": 0.0965,
       "step": 5120
     },
     {
       "epoch": 8.18,
-      "grad_norm": 7.110737323760986,
-      "learning_rate": 1.8216560509554138e-05,
-      "loss": 0.1445,
+      "grad_norm": 6.403193473815918,
+      "learning_rate": 1.820063694267516e-05,
+      "loss": 0.1262,
       "step": 5140
     },
     {
       "epoch": 8.22,
-      "grad_norm": 3.224792242050171,
-      "learning_rate": 1.7898089171974523e-05,
-      "loss": 0.1951,
+      "grad_norm": 0.31247055530548096,
+      "learning_rate": 1.788216560509554e-05,
+      "loss": 0.1447,
       "step": 5160
     },
     {
       "epoch": 8.25,
-      "grad_norm": 1.4524933099746704,
-      "learning_rate": 1.7579617834394904e-05,
-      "loss": 0.1013,
+      "grad_norm": 9.875935554504395,
+      "learning_rate": 1.7563694267515925e-05,
+      "loss": 0.0796,
       "step": 5180
     },
     {
       "epoch": 8.28,
-      "grad_norm": 5.544615268707275,
-      "learning_rate": 1.7261146496815285e-05,
-      "loss": 0.1799,
+      "grad_norm": 0.12406259775161743,
+      "learning_rate": 1.7245222929936306e-05,
+      "loss": 0.0354,
       "step": 5200
     },
     {
       "epoch": 8.28,
       "eval_accuracy": 0.758095238095238,
-      "eval_loss": 1.1371468305587769,
-      "eval_runtime": 14.4095,
-      "eval_samples_per_second": 72.869,
-      "eval_steps_per_second": 9.161,
+      "eval_loss": 1.2345435619354248,
+      "eval_runtime": 14.3786,
+      "eval_samples_per_second": 73.025,
+      "eval_steps_per_second": 9.18,
       "step": 5200
     },
     {
       "epoch": 8.31,
-      "grad_norm": 0.4378151297569275,
-      "learning_rate": 1.694267515923567e-05,
-      "loss": 0.156,
+      "grad_norm": 0.022050337865948677,
+      "learning_rate": 1.6926751592356687e-05,
+      "loss": 0.0951,
       "step": 5220
     },
     {
       "epoch": 8.34,
-      "grad_norm": 4.541534423828125,
-      "learning_rate": 1.662420382165605e-05,
-      "loss": 0.1158,
+      "grad_norm": 5.747735977172852,
+      "learning_rate": 1.660828025477707e-05,
+      "loss": 0.1071,
       "step": 5240
     },
     {
       "epoch": 8.38,
-      "grad_norm": 0.01598621904850006,
-      "learning_rate": 1.6305732484076432e-05,
-      "loss": 0.1636,
+      "grad_norm": 0.20012032985687256,
+      "learning_rate": 1.6289808917197453e-05,
+      "loss": 0.0581,
       "step": 5260
     },
     {
       "epoch": 8.41,
-      "grad_norm": 0.15167254209518433,
-      "learning_rate": 1.5987261146496817e-05,
+      "grad_norm": 20.06525993347168,
+      "learning_rate": 1.5971337579617834e-05,
       "loss": 0.0819,
       "step": 5280
     },
     {
       "epoch": 8.44,
-      "grad_norm": 2.8723442554473877,
-      "learning_rate": 1.5668789808917198e-05,
-      "loss": 0.1496,
+      "grad_norm": 0.022994108498096466,
+      "learning_rate": 1.5652866242038215e-05,
+      "loss": 0.0499,
       "step": 5300
     },
     {
       "epoch": 8.44,
-      "eval_accuracy": 0.7428571428571429,
-      "eval_loss": 1.1775193214416504,
-      "eval_runtime": 14.1642,
-      "eval_samples_per_second": 74.13,
-      "eval_steps_per_second": 9.319,
+      "eval_accuracy": 0.7571428571428571,
+      "eval_loss": 1.1775814294815063,
+      "eval_runtime": 14.4071,
+      "eval_samples_per_second": 72.88,
+      "eval_steps_per_second": 9.162,
       "step": 5300
     },
     {
       "epoch": 8.47,
-      "grad_norm": 5.163817405700684,
-      "learning_rate": 1.535031847133758e-05,
-      "loss": 0.126,
+      "grad_norm": 1.858337163925171,
+      "learning_rate": 1.53343949044586e-05,
+      "loss": 0.1257,
       "step": 5320
     },
     {
       "epoch": 8.5,
-      "grad_norm": 7.263587474822998,
-      "learning_rate": 1.5031847133757962e-05,
-      "loss": 0.1124,
+      "grad_norm": 0.0053113107569515705,
+      "learning_rate": 1.5015923566878981e-05,
+      "loss": 0.0348,
       "step": 5340
     },
     {
       "epoch": 8.54,
-      "grad_norm": 0.11626210063695908,
-      "learning_rate": 1.4713375796178345e-05,
-      "loss": 0.1274,
+      "grad_norm": 0.014819352887570858,
+      "learning_rate": 1.4697452229299364e-05,
+      "loss": 0.0669,
       "step": 5360
     },
     {
       "epoch": 8.57,
-      "grad_norm": 0.1953892856836319,
-      "learning_rate": 1.4394904458598726e-05,
-      "loss": 0.0721,
+      "grad_norm": 0.23732012510299683,
+      "learning_rate": 1.4378980891719745e-05,
+      "loss": 0.0379,
       "step": 5380
     },
     {
       "epoch": 8.6,
-      "grad_norm": 4.252742290496826,
-      "learning_rate": 1.4076433121019109e-05,
-      "loss": 0.0804,
+      "grad_norm": 3.4506566524505615,
+      "learning_rate": 1.4060509554140128e-05,
+      "loss": 0.1046,
       "step": 5400
     },
     {
       "epoch": 8.6,
-      "eval_accuracy": 0.7561904761904762,
-      "eval_loss": 1.1278163194656372,
-      "eval_runtime": 14.271,
-      "eval_samples_per_second": 73.576,
-      "eval_steps_per_second": 9.25,
+      "eval_accuracy": 0.76,
+      "eval_loss": 1.1939048767089844,
+      "eval_runtime": 14.1579,
+      "eval_samples_per_second": 74.164,
+      "eval_steps_per_second": 9.323,
       "step": 5400
     },
     {
       "epoch": 8.63,
-      "grad_norm": 0.10082034021615982,
-      "learning_rate": 1.375796178343949e-05,
-      "loss": 0.1376,
+      "grad_norm": 5.083927154541016,
+      "learning_rate": 1.374203821656051e-05,
+      "loss": 0.13,
       "step": 5420
     },
     {
       "epoch": 8.66,
-      "grad_norm": 0.05313138663768768,
-      "learning_rate": 1.3439490445859873e-05,
-      "loss": 0.0992,
+      "grad_norm": 0.006614590063691139,
+      "learning_rate": 1.3423566878980892e-05,
+      "loss": 0.102,
       "step": 5440
     },
     {
       "epoch": 8.69,
-      "grad_norm": 3.8297231197357178,
-      "learning_rate": 1.3121019108280256e-05,
-      "loss": 0.1511,
+      "grad_norm": 0.12619677186012268,
+      "learning_rate": 1.3105095541401275e-05,
+      "loss": 0.0753,
       "step": 5460
     },
     {
       "epoch": 8.73,
-      "grad_norm": 3.4488308429718018,
-      "learning_rate": 1.2802547770700637e-05,
-      "loss": 0.1603,
+      "grad_norm": 0.02025582641363144,
+      "learning_rate": 1.2786624203821656e-05,
+      "loss": 0.0855,
       "step": 5480
     },
     {
       "epoch": 8.76,
-      "grad_norm": 1.6856378316879272,
-      "learning_rate": 1.248407643312102e-05,
-      "loss": 0.12,
+      "grad_norm": 0.15310314297676086,
+      "learning_rate": 1.2468152866242039e-05,
+      "loss": 0.0912,
       "step": 5500
     },
     {
       "epoch": 8.76,
-      "eval_accuracy": 0.7533333333333333,
-      "eval_loss": 1.1209964752197266,
-      "eval_runtime": 14.1604,
-      "eval_samples_per_second": 74.15,
-      "eval_steps_per_second": 9.322,
+      "eval_accuracy": 0.7485714285714286,
+      "eval_loss": 1.2372716665267944,
+      "eval_runtime": 14.3245,
+      "eval_samples_per_second": 73.301,
+      "eval_steps_per_second": 9.215,
       "step": 5500
     },
     {
       "epoch": 8.79,
-      "grad_norm": 3.3723340034484863,
-      "learning_rate": 1.2165605095541403e-05,
-      "loss": 0.0811,
+      "grad_norm": 0.005766577553004026,
+      "learning_rate": 1.214968152866242e-05,
+      "loss": 0.0808,
       "step": 5520
     },
     {
       "epoch": 8.82,
-      "grad_norm": 4.700990676879883,
-      "learning_rate": 1.1847133757961784e-05,
-      "loss": 0.0977,
+      "grad_norm": 2.425058364868164,
+      "learning_rate": 1.1831210191082803e-05,
+      "loss": 0.1606,
       "step": 5540
     },
     {
       "epoch": 8.85,
-      "grad_norm": 2.095137596130371,
-      "learning_rate": 1.1528662420382167e-05,
-      "loss": 0.0514,
+      "grad_norm": 3.1397340297698975,
+      "learning_rate": 1.1512738853503186e-05,
+      "loss": 0.0553,
       "step": 5560
     },
     {
       "epoch": 8.89,
-      "grad_norm": 0.014027014374732971,
-      "learning_rate": 1.1210191082802548e-05,
-      "loss": 0.1034,
+      "grad_norm": 0.04639526456594467,
+      "learning_rate": 1.1194267515923567e-05,
+      "loss": 0.0602,
       "step": 5580
     },
     {
       "epoch": 8.92,
-      "grad_norm": 1.6214849948883057,
-      "learning_rate": 1.089171974522293e-05,
-      "loss": 0.099,
+      "grad_norm": 2.158754587173462,
+      "learning_rate": 1.087579617834395e-05,
+      "loss": 0.0589,
       "step": 5600
     },
     {
       "epoch": 8.92,
-      "eval_accuracy": 0.7485714285714286,
-      "eval_loss": 1.1295264959335327,
-      "eval_runtime": 14.482,
-      "eval_samples_per_second": 72.504,
-      "eval_steps_per_second": 9.115,
+      "eval_accuracy": 0.7552380952380953,
+      "eval_loss": 1.2165464162826538,
+      "eval_runtime": 14.3916,
+      "eval_samples_per_second": 72.959,
+      "eval_steps_per_second": 9.172,
       "step": 5600
     },
     {
       "epoch": 8.95,
-      "grad_norm": 1.0302766561508179,
-      "learning_rate": 1.0573248407643314e-05,
-      "loss": 0.0755,
+      "grad_norm": 0.23296405375003815,
+      "learning_rate": 1.0557324840764333e-05,
+      "loss": 0.1005,
       "step": 5620
     },
     {
       "epoch": 8.98,
-      "grad_norm": 3.192128896713257,
-      "learning_rate": 1.0254777070063695e-05,
-      "loss": 0.0791,
+      "grad_norm": 0.005250574089586735,
+      "learning_rate": 1.0238853503184714e-05,
+      "loss": 0.0531,
       "step": 5640
     },
     {
       "epoch": 9.01,
-      "grad_norm": 4.739717960357666,
-      "learning_rate": 9.936305732484078e-06,
-      "loss": 0.1346,
+      "grad_norm": 0.2323373258113861,
+      "learning_rate": 9.920382165605097e-06,
+      "loss": 0.0559,
       "step": 5660
     },
     {
       "epoch": 9.04,
-      "grad_norm": 0.05003926530480385,
-      "learning_rate": 9.617834394904459e-06,
-      "loss": 0.0694,
+      "grad_norm": 0.2171221822500229,
+      "learning_rate": 9.601910828025478e-06,
+      "loss": 0.0661,
       "step": 5680
     },
     {
       "epoch": 9.08,
-      "grad_norm": 1.395236611366272,
-      "learning_rate": 9.29936305732484e-06,
-      "loss": 0.1429,
+      "grad_norm": 0.14795228838920593,
+      "learning_rate": 9.28343949044586e-06,
+      "loss": 0.0829,
       "step": 5700
     },
     {
       "epoch": 9.08,
-      "eval_accuracy": 0.7390476190476191,
-      "eval_loss": 1.207868218421936,
-      "eval_runtime": 14.2527,
-      "eval_samples_per_second": 73.67,
-      "eval_steps_per_second": 9.261,
+      "eval_accuracy": 0.7504761904761905,
+      "eval_loss": 1.2683508396148682,
+      "eval_runtime": 14.4828,
+      "eval_samples_per_second": 72.5,
+      "eval_steps_per_second": 9.114,
       "step": 5700
     },
     {
       "epoch": 9.11,
-      "grad_norm": 4.653144836425781,
-      "learning_rate": 8.980891719745223e-06,
-      "loss": 0.1111,
+      "grad_norm": 3.622931480407715,
+      "learning_rate": 8.964968152866242e-06,
+      "loss": 0.0883,
       "step": 5720
     },
     {
       "epoch": 9.14,
-      "grad_norm": 12.179301261901855,
-      "learning_rate": 8.662420382165604e-06,
-      "loss": 0.0931,
+      "grad_norm": 0.058454740792512894,
+      "learning_rate": 8.646496815286623e-06,
+      "loss": 0.0853,
       "step": 5740
     },
     {
       "epoch": 9.17,
-      "grad_norm": 5.497566223144531,
-      "learning_rate": 8.343949044585987e-06,
-      "loss": 0.1509,
+      "grad_norm": 5.839081287384033,
+      "learning_rate": 8.328025477707006e-06,
+      "loss": 0.0751,
       "step": 5760
     },
     {
       "epoch": 9.2,
-      "grad_norm": 4.653054237365723,
-      "learning_rate": 8.02547770700637e-06,
-      "loss": 0.1349,
+      "grad_norm": 1.6214606761932373,
+      "learning_rate": 8.009554140127389e-06,
+      "loss": 0.0764,
       "step": 5780
     },
     {
       "epoch": 9.24,
-      "grad_norm": 4.78197717666626,
-      "learning_rate": 7.707006369426751e-06,
-      "loss": 0.0959,
+      "grad_norm": 7.4141926765441895,
+      "learning_rate": 7.69108280254777e-06,
+      "loss": 0.0897,
       "step": 5800
     },
     {
       "epoch": 9.24,
-      "eval_accuracy": 0.7476190476190476,
-      "eval_loss": 1.1450541019439697,
-      "eval_runtime": 14.2174,
-      "eval_samples_per_second": 73.853,
-      "eval_steps_per_second": 9.284,
+      "eval_accuracy": 0.7552380952380953,
+      "eval_loss": 1.2467057704925537,
+      "eval_runtime": 14.4317,
+      "eval_samples_per_second": 72.756,
+      "eval_steps_per_second": 9.147,
       "step": 5800
     },
     {
       "epoch": 9.27,
-      "grad_norm": 0.03227326646447182,
-      "learning_rate": 7.388535031847134e-06,
-      "loss": 0.135,
+      "grad_norm": 0.004910254385322332,
+      "learning_rate": 7.372611464968153e-06,
+      "loss": 0.1181,
       "step": 5820
     },
     {
       "epoch": 9.3,
-      "grad_norm": 0.03949011489748955,
-      "learning_rate": 7.070063694267516e-06,
-      "loss": 0.1352,
+      "grad_norm": 0.02709910087287426,
+      "learning_rate": 7.054140127388535e-06,
+      "loss": 0.0754,
       "step": 5840
     },
     {
       "epoch": 9.33,
-      "grad_norm": 0.12604393064975739,
-      "learning_rate": 6.751592356687898e-06,
-      "loss": 0.0613,
+      "grad_norm": 4.529952049255371,
+      "learning_rate": 6.735668789808917e-06,
+      "loss": 0.0882,
       "step": 5860
     },
     {
       "epoch": 9.36,
-      "grad_norm": 3.415341377258301,
-      "learning_rate": 6.433121019108281e-06,
-      "loss": 0.1164,
+      "grad_norm": 0.010457679629325867,
+      "learning_rate": 6.417197452229299e-06,
+      "loss": 0.0589,
       "step": 5880
     },
     {
       "epoch": 9.39,
-      "grad_norm": 0.011540273204445839,
-      "learning_rate": 6.114649681528663e-06,
-      "loss": 0.0263,
+      "grad_norm": 2.2108523845672607,
+      "learning_rate": 6.098726114649682e-06,
+      "loss": 0.1114,
       "step": 5900
     },
     {
       "epoch": 9.39,
-      "eval_accuracy": 0.7514285714285714,
-      "eval_loss": 1.117615818977356,
-      "eval_runtime": 14.3543,
-      "eval_samples_per_second": 73.149,
-      "eval_steps_per_second": 9.196,
+      "eval_accuracy": 0.7571428571428571,
+      "eval_loss": 1.2302595376968384,
+      "eval_runtime": 14.3407,
+      "eval_samples_per_second": 73.218,
+      "eval_steps_per_second": 9.205,
       "step": 5900
     },
     {
       "epoch": 9.43,
-      "grad_norm": 1.7297450304031372,
-      "learning_rate": 5.796178343949045e-06,
-      "loss": 0.1058,
+      "grad_norm": 0.09094621986150742,
+      "learning_rate": 5.780254777070064e-06,
+      "loss": 0.1128,
       "step": 5920
     },
     {
       "epoch": 9.46,
-      "grad_norm": 0.2807926833629608,
-      "learning_rate": 5.477707006369427e-06,
-      "loss": 0.1072,
+      "grad_norm": 8.727699279785156,
+      "learning_rate": 5.461783439490446e-06,
+      "loss": 0.0966,
       "step": 5940
     },
     {
       "epoch": 9.49,
-      "grad_norm": 4.0545244216918945,
-      "learning_rate": 5.15923566878981e-06,
-      "loss": 0.1031,
+      "grad_norm": 0.30840545892715454,
+      "learning_rate": 5.143312101910828e-06,
+      "loss": 0.0601,
       "step": 5960
     },
     {
       "epoch": 9.52,
-      "grad_norm": 7.815937042236328,
-      "learning_rate": 4.840764331210192e-06,
-      "loss": 0.0962,
+      "grad_norm": 0.003402425441890955,
+      "learning_rate": 4.824840764331211e-06,
+      "loss": 0.0563,
       "step": 5980
     },
     {
       "epoch": 9.55,
-      "grad_norm": 12.278979301452637,
-      "learning_rate": 4.522292993630573e-06,
-      "loss": 0.0936,
+      "grad_norm": 0.32906395196914673,
+      "learning_rate": 4.506369426751592e-06,
+      "loss": 0.0712,
       "step": 6000
     },
     {
       "epoch": 9.55,
-      "eval_accuracy": 0.7533333333333333,
-      "eval_loss": 1.1178562641143799,
-      "eval_runtime": 14.5366,
-      "eval_samples_per_second": 72.231,
-      "eval_steps_per_second": 9.081,
+      "eval_accuracy": 0.7638095238095238,
+      "eval_loss": 1.1997016668319702,
+      "eval_runtime": 14.373,
+      "eval_samples_per_second": 73.054,
+      "eval_steps_per_second": 9.184,
       "step": 6000
     },
     {
       "epoch": 9.59,
-      "grad_norm": 0.06204557418823242,
-      "learning_rate": 4.203821656050955e-06,
-      "loss": 0.1101,
+      "grad_norm": 2.1830148696899414,
+      "learning_rate": 4.187898089171974e-06,
+      "loss": 0.0718,
       "step": 6020
     },
     {
       "epoch": 9.62,
-      "grad_norm": 0.023558788001537323,
-      "learning_rate": 3.885350318471338e-06,
-      "loss": 0.0556,
+      "grad_norm": 0.03326411545276642,
+      "learning_rate": 3.869426751592356e-06,
+      "loss": 0.0736,
       "step": 6040
     },
     {
       "epoch": 9.65,
-      "grad_norm": 0.07364491373300552,
-      "learning_rate": 3.56687898089172e-06,
-      "loss": 0.1003,
+      "grad_norm": 0.06509624421596527,
+      "learning_rate": 3.5509554140127388e-06,
+      "loss": 0.0789,
       "step": 6060
     },
     {
       "epoch": 9.68,
-      "grad_norm": 2.476426839828491,
-      "learning_rate": 3.248407643312102e-06,
-      "loss": 0.1278,
+      "grad_norm": 7.084524154663086,
+      "learning_rate": 3.232484076433121e-06,
+      "loss": 0.1014,
       "step": 6080
     },
     {
       "epoch": 9.71,
-      "grad_norm": 0.04320213571190834,
-      "learning_rate": 2.9299363057324844e-06,
-      "loss": 0.1332,
+      "grad_norm": 8.154586791992188,
+      "learning_rate": 2.9140127388535032e-06,
+      "loss": 0.0621,
       "step": 6100
     },
     {
       "epoch": 9.71,
-      "eval_accuracy": 0.7485714285714286,
-      "eval_loss": 1.1386895179748535,
-      "eval_runtime": 14.088,
-      "eval_samples_per_second": 74.532,
-      "eval_steps_per_second": 9.37,
+      "eval_accuracy": 0.7628571428571429,
+      "eval_loss": 1.2093576192855835,
+      "eval_runtime": 14.431,
+      "eval_samples_per_second": 72.76,
+      "eval_steps_per_second": 9.147,
       "step": 6100
     },
     {
       "epoch": 9.75,
-      "grad_norm": 3.164142608642578,
-      "learning_rate": 2.6114649681528665e-06,
-      "loss": 0.1016,
+      "grad_norm": 7.626684665679932,
+      "learning_rate": 2.5955414012738857e-06,
+      "loss": 0.1057,
       "step": 6120
     },
     {
       "epoch": 9.78,
-      "grad_norm": 1.9194905757904053,
-      "learning_rate": 2.2929936305732485e-06,
-      "loss": 0.0907,
+      "grad_norm": 0.032657843083143234,
+      "learning_rate": 2.2770700636942673e-06,
+      "loss": 0.0633,
       "step": 6140
     },
     {
       "epoch": 9.81,
-      "grad_norm": 6.227757930755615,
-      "learning_rate": 1.9745222929936305e-06,
-      "loss": 0.1274,
+      "grad_norm": 8.062530517578125,
+      "learning_rate": 1.9585987261146497e-06,
+      "loss": 0.0749,
       "step": 6160
     },
     {
       "epoch": 9.84,
-      "grad_norm": 0.11668018996715546,
-      "learning_rate": 1.6560509554140127e-06,
-      "loss": 0.0701,
+      "grad_norm": 5.101759433746338,
+      "learning_rate": 1.640127388535032e-06,
+      "loss": 0.1027,
       "step": 6180
     },
     {
       "epoch": 9.87,
-      "grad_norm": 1.6369106769561768,
-      "learning_rate": 1.337579617834395e-06,
-      "loss": 0.071,
+      "grad_norm": 0.1060834750533104,
+      "learning_rate": 1.3216560509554142e-06,
+      "loss": 0.037,
       "step": 6200
     },
     {
       "epoch": 9.87,
-      "eval_accuracy": 0.7466666666666667,
-      "eval_loss": 1.1522855758666992,
-      "eval_runtime": 14.2788,
-      "eval_samples_per_second": 73.536,
-      "eval_steps_per_second": 9.244,
+      "eval_accuracy": 0.7619047619047619,
+      "eval_loss": 1.2147165536880493,
+      "eval_runtime": 14.2781,
+      "eval_samples_per_second": 73.539,
+      "eval_steps_per_second": 9.245,
       "step": 6200
     },
     {
       "epoch": 9.9,
-      "grad_norm": 0.5098825097084045,
-      "learning_rate": 1.019108280254777e-06,
-      "loss": 0.0679,
+      "grad_norm": 0.00528693338856101,
+      "learning_rate": 1.0031847133757962e-06,
+      "loss": 0.0861,
       "step": 6220
     },
     {
       "epoch": 9.94,
-      "grad_norm": 0.0343327559530735,
-      "learning_rate": 7.006369426751592e-07,
-      "loss": 0.1315,
+      "grad_norm": 6.1690802574157715,
+      "learning_rate": 6.847133757961784e-07,
+      "loss": 0.1415,
       "step": 6240
     },
     {
       "epoch": 9.97,
-      "grad_norm": 3.2321577072143555,
-      "learning_rate": 3.8216560509554143e-07,
-      "loss": 0.0521,
+      "grad_norm": 14.695515632629395,
+      "learning_rate": 3.6624203821656055e-07,
+      "loss": 0.1083,
       "step": 6260
     },
     {
       "epoch": 10.0,
-      "grad_norm": 0.04384122043848038,
-      "learning_rate": 6.369426751592356e-08,
-      "loss": 0.0769,
+      "grad_norm": 0.008544925600290298,
+      "learning_rate": 4.777070063694268e-08,
+      "loss": 0.0538,
       "step": 6280
     },
     {
       "epoch": 10.0,
       "step": 6280,
-      "total_flos": 7.783078535151575e+18,
-      "train_loss": 0.28272074579623097,
-      "train_runtime": 3631.8012,
-      "train_samples_per_second": 27.631,
-      "train_steps_per_second": 1.729
+      "total_flos": 7.791758260292837e+18,
+      "train_loss": 0.15229952573586422,
+      "train_runtime": 3498.5508,
+      "train_samples_per_second": 28.683,
+      "train_steps_per_second": 1.795
     }
   ],
   "logging_steps": 20,
@@ -2779,7 +2779,7 @@
   "num_input_tokens_seen": 0,
   "num_train_epochs": 10,
   "save_steps": 100,
-  "total_flos": 7.783078535151575e+18,
+  "total_flos": 7.791758260292837e+18,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null