diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,2632 +1,7840 @@
 {
-  "best_metric": 0.6788575409265064,
-  "best_model_checkpoint": "finetuned-FER2013/checkpoint-4040",
-  "epoch": 20.0,
+  "best_metric": 0.7133402995471961,
+  "best_model_checkpoint": "finetuned-FER2013/checkpoint-10706",
+  "epoch": 60.0,
   "eval_steps": 500,
-  "global_step": 4040,
+  "global_step": 12120,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.05,
-      "learning_rate": 1.2376237623762377e-07,
-      "loss": 2.073,
+      "learning_rate": 4.125412541254125e-08,
+      "loss": 2.0393,
       "step": 10
     },
     {
       "epoch": 0.1,
-      "learning_rate": 2.4752475247524754e-07,
-      "loss": 2.1018,
+      "learning_rate": 8.25082508250825e-08,
+      "loss": 2.0234,
       "step": 20
     },
     {
       "epoch": 0.15,
-      "learning_rate": 3.712871287128713e-07,
-      "loss": 2.0777,
+      "learning_rate": 1.2376237623762377e-07,
+      "loss": 2.0187,
       "step": 30
     },
     {
       "epoch": 0.2,
-      "learning_rate": 4.950495049504951e-07,
-      "loss": 2.0483,
+      "learning_rate": 1.65016501650165e-07,
+      "loss": 1.994,
       "step": 40
     },
     {
       "epoch": 0.25,
-      "learning_rate": 6.188118811881188e-07,
-      "loss": 2.0325,
+      "learning_rate": 2.0627062706270628e-07,
+      "loss": 1.9756,
       "step": 50
     },
     {
       "epoch": 0.3,
-      "learning_rate": 7.425742574257426e-07,
-      "loss": 1.9965,
+      "learning_rate": 2.4752475247524754e-07,
+      "loss": 2.0019,
       "step": 60
     },
     {
       "epoch": 0.35,
-      "learning_rate": 8.663366336633663e-07,
-      "loss": 1.9529,
+      "learning_rate": 2.887788778877888e-07,
+      "loss": 1.9724,
       "step": 70
     },
     {
       "epoch": 0.4,
-      "learning_rate": 9.900990099009902e-07,
-      "loss": 1.9238,
+      "learning_rate": 3.3003300330033e-07,
+      "loss": 1.9631,
       "step": 80
     },
     {
       "epoch": 0.45,
-      "learning_rate": 1.113861386138614e-06,
-      "loss": 1.9034,
+      "learning_rate": 3.712871287128713e-07,
+      "loss": 1.9461,
       "step": 90
     },
     {
       "epoch": 0.5,
-      "learning_rate": 1.2376237623762377e-06,
-      "loss": 1.8819,
+      "learning_rate": 4.1254125412541255e-07,
+      "loss": 1.9288,
       "step": 100
     },
     {
       "epoch": 0.54,
-      "learning_rate": 1.3613861386138616e-06,
-      "loss": 1.8242,
+      "learning_rate": 4.537953795379538e-07,
+      "loss": 1.925,
       "step": 110
     },
     {
       "epoch": 0.59,
-      "learning_rate": 1.4851485148514852e-06,
-      "loss": 1.8252,
+      "learning_rate": 4.950495049504951e-07,
+      "loss": 1.9099,
       "step": 120
     },
     {
       "epoch": 0.64,
-      "learning_rate": 1.6089108910891091e-06,
-      "loss": 1.7892,
+      "learning_rate": 5.363036303630363e-07,
+      "loss": 1.9031,
       "step": 130
     },
     {
       "epoch": 0.69,
-      "learning_rate": 1.7326732673267326e-06,
-      "loss": 1.7766,
+      "learning_rate": 5.775577557755776e-07,
+      "loss": 1.8606,
       "step": 140
     },
     {
       "epoch": 0.74,
-      "learning_rate": 1.8564356435643566e-06,
-      "loss": 1.7322,
+      "learning_rate": 6.188118811881188e-07,
+      "loss": 1.8663,
       "step": 150
     },
     {
       "epoch": 0.79,
-      "learning_rate": 1.9801980198019803e-06,
-      "loss": 1.6977,
+      "learning_rate": 6.6006600660066e-07,
+      "loss": 1.8232,
       "step": 160
     },
     {
       "epoch": 0.84,
-      "learning_rate": 2.103960396039604e-06,
-      "loss": 1.6751,
+      "learning_rate": 7.013201320132014e-07,
+      "loss": 1.8284,
       "step": 170
     },
     {
       "epoch": 0.89,
-      "learning_rate": 2.227722772277228e-06,
-      "loss": 1.6337,
+      "learning_rate": 7.425742574257426e-07,
+      "loss": 1.8139,
       "step": 180
     },
     {
       "epoch": 0.94,
-      "learning_rate": 2.3514851485148514e-06,
-      "loss": 1.5947,
+      "learning_rate": 7.838283828382839e-07,
+      "loss": 1.8134,
       "step": 190
     },
     {
       "epoch": 0.99,
-      "learning_rate": 2.4752475247524753e-06,
-      "loss": 1.5466,
+      "learning_rate": 8.250825082508251e-07,
+      "loss": 1.7999,
       "step": 200
     },
     {
       "epoch": 1.0,
-      "eval_accuracy": 0.4500174155346569,
-      "eval_loss": 1.502170443534851,
-      "eval_runtime": 38.4158,
-      "eval_samples_per_second": 74.735,
-      "eval_steps_per_second": 2.343,
+      "eval_accuracy": 0.2713340299547196,
+      "eval_loss": 1.7896360158920288,
+      "eval_runtime": 37.8591,
+      "eval_samples_per_second": 75.834,
+      "eval_steps_per_second": 2.377,
       "step": 202
     },
     {
       "epoch": 1.04,
-      "learning_rate": 2.5990099009900993e-06,
-      "loss": 1.5667,
+      "learning_rate": 8.663366336633663e-07,
+      "loss": 1.7895,
       "step": 210
     },
     {
       "epoch": 1.09,
-      "learning_rate": 2.7227722772277232e-06,
-      "loss": 1.551,
+      "learning_rate": 9.075907590759076e-07,
+      "loss": 1.7802,
       "step": 220
     },
     {
       "epoch": 1.14,
-      "learning_rate": 2.8465346534653464e-06,
-      "loss": 1.5325,
+      "learning_rate": 9.488448844884489e-07,
+      "loss": 1.7528,
       "step": 230
     },
     {
       "epoch": 1.19,
-      "learning_rate": 2.9702970297029703e-06,
-      "loss": 1.5274,
+      "learning_rate": 9.900990099009902e-07,
+      "loss": 1.751,
       "step": 240
     },
     {
       "epoch": 1.24,
-      "learning_rate": 3.0940594059405943e-06,
-      "loss": 1.4687,
+      "learning_rate": 1.0313531353135315e-06,
+      "loss": 1.722,
       "step": 250
     },
     {
       "epoch": 1.29,
-      "learning_rate": 3.2178217821782182e-06,
-      "loss": 1.4933,
+      "learning_rate": 1.0726072607260726e-06,
+      "loss": 1.7037,
       "step": 260
     },
     {
       "epoch": 1.34,
-      "learning_rate": 3.341584158415842e-06,
-      "loss": 1.4407,
+      "learning_rate": 1.113861386138614e-06,
+      "loss": 1.6982,
       "step": 270
     },
     {
       "epoch": 1.39,
-      "learning_rate": 3.4653465346534653e-06,
-      "loss": 1.4817,
+      "learning_rate": 1.1551155115511552e-06,
+      "loss": 1.7017,
       "step": 280
     },
     {
       "epoch": 1.44,
-      "learning_rate": 3.5891089108910892e-06,
-      "loss": 1.4594,
+      "learning_rate": 1.1963696369636966e-06,
+      "loss": 1.6585,
       "step": 290
     },
     {
       "epoch": 1.49,
-      "learning_rate": 3.712871287128713e-06,
-      "loss": 1.4494,
+      "learning_rate": 1.2376237623762377e-06,
+      "loss": 1.6565,
       "step": 300
     },
     {
       "epoch": 1.53,
-      "learning_rate": 3.836633663366337e-06,
-      "loss": 1.4094,
+      "learning_rate": 1.278877887788779e-06,
+      "loss": 1.6461,
       "step": 310
     },
     {
       "epoch": 1.58,
-      "learning_rate": 3.960396039603961e-06,
-      "loss": 1.4515,
+      "learning_rate": 1.32013201320132e-06,
+      "loss": 1.6283,
       "step": 320
     },
     {
       "epoch": 1.63,
-      "learning_rate": 4.084158415841584e-06,
-      "loss": 1.3378,
+      "learning_rate": 1.3613861386138616e-06,
+      "loss": 1.591,
       "step": 330
     },
     {
       "epoch": 1.68,
-      "learning_rate": 4.207920792079208e-06,
-      "loss": 1.3925,
+      "learning_rate": 1.4026402640264027e-06,
+      "loss": 1.5986,
       "step": 340
     },
     {
       "epoch": 1.73,
-      "learning_rate": 4.331683168316832e-06,
-      "loss": 1.3663,
+      "learning_rate": 1.443894389438944e-06,
+      "loss": 1.6095,
       "step": 350
     },
     {
       "epoch": 1.78,
-      "learning_rate": 4.455445544554456e-06,
-      "loss": 1.3544,
+      "learning_rate": 1.4851485148514852e-06,
+      "loss": 1.5793,
       "step": 360
     },
     {
       "epoch": 1.83,
-      "learning_rate": 4.57920792079208e-06,
-      "loss": 1.3758,
+      "learning_rate": 1.5264026402640265e-06,
+      "loss": 1.5953,
       "step": 370
     },
     {
       "epoch": 1.88,
-      "learning_rate": 4.702970297029703e-06,
-      "loss": 1.3395,
+      "learning_rate": 1.5676567656765678e-06,
+      "loss": 1.5272,
       "step": 380
     },
     {
       "epoch": 1.93,
-      "learning_rate": 4.826732673267327e-06,
-      "loss": 1.328,
+      "learning_rate": 1.6089108910891091e-06,
+      "loss": 1.5413,
       "step": 390
     },
     {
       "epoch": 1.98,
-      "learning_rate": 4.950495049504951e-06,
-      "loss": 1.3372,
+      "learning_rate": 1.6501650165016502e-06,
+      "loss": 1.5165,
       "step": 400
     },
     {
       "epoch": 2.0,
-      "eval_accuracy": 0.5632183908045977,
-      "eval_loss": 1.1727259159088135,
-      "eval_runtime": 38.6928,
-      "eval_samples_per_second": 74.2,
-      "eval_steps_per_second": 2.326,
+      "eval_accuracy": 0.47579240682688956,
+      "eval_loss": 1.4311269521713257,
+      "eval_runtime": 38.0229,
+      "eval_samples_per_second": 75.507,
+      "eval_steps_per_second": 2.367,
       "step": 404
     },
     {
       "epoch": 2.03,
-      "learning_rate": 4.991749174917492e-06,
-      "loss": 1.313,
+      "learning_rate": 1.6914191419141915e-06,
+      "loss": 1.4938,
       "step": 410
     },
     {
       "epoch": 2.08,
-      "learning_rate": 4.977997799779978e-06,
-      "loss": 1.3004,
+      "learning_rate": 1.7326732673267326e-06,
+      "loss": 1.4922,
       "step": 420
     },
     {
       "epoch": 2.13,
-      "learning_rate": 4.964246424642465e-06,
-      "loss": 1.2099,
+      "learning_rate": 1.7739273927392742e-06,
+      "loss": 1.4751,
       "step": 430
     },
     {
       "epoch": 2.18,
-      "learning_rate": 4.950495049504951e-06,
-      "loss": 1.311,
+      "learning_rate": 1.8151815181518153e-06,
+      "loss": 1.4986,
       "step": 440
     },
     {
       "epoch": 2.23,
-      "learning_rate": 4.936743674367437e-06,
-      "loss": 1.2735,
+      "learning_rate": 1.8564356435643566e-06,
+      "loss": 1.4757,
       "step": 450
     },
     {
       "epoch": 2.28,
-      "learning_rate": 4.922992299229923e-06,
-      "loss": 1.2598,
+      "learning_rate": 1.8976897689768977e-06,
+      "loss": 1.4634,
       "step": 460
     },
     {
       "epoch": 2.33,
-      "learning_rate": 4.90924092409241e-06,
-      "loss": 1.2941,
+      "learning_rate": 1.938943894389439e-06,
+      "loss": 1.4137,
       "step": 470
     },
     {
       "epoch": 2.38,
-      "learning_rate": 4.8954895489548956e-06,
-      "loss": 1.2459,
+      "learning_rate": 1.9801980198019803e-06,
+      "loss": 1.4223,
       "step": 480
     },
     {
       "epoch": 2.43,
-      "learning_rate": 4.881738173817382e-06,
-      "loss": 1.2597,
+      "learning_rate": 2.0214521452145217e-06,
+      "loss": 1.4068,
       "step": 490
     },
     {
       "epoch": 2.48,
-      "learning_rate": 4.867986798679869e-06,
-      "loss": 1.2726,
+      "learning_rate": 2.062706270627063e-06,
+      "loss": 1.4421,
       "step": 500
     },
     {
       "epoch": 2.52,
-      "learning_rate": 4.854235423542355e-06,
-      "loss": 1.3236,
+      "learning_rate": 2.103960396039604e-06,
+      "loss": 1.4113,
       "step": 510
     },
     {
       "epoch": 2.57,
-      "learning_rate": 4.840484048404841e-06,
-      "loss": 1.2716,
+      "learning_rate": 2.145214521452145e-06,
+      "loss": 1.4139,
       "step": 520
     },
     {
       "epoch": 2.62,
-      "learning_rate": 4.826732673267327e-06,
-      "loss": 1.2441,
+      "learning_rate": 2.186468646864687e-06,
+      "loss": 1.3938,
       "step": 530
     },
     {
       "epoch": 2.67,
-      "learning_rate": 4.812981298129814e-06,
-      "loss": 1.2843,
+      "learning_rate": 2.227722772277228e-06,
+      "loss": 1.3521,
       "step": 540
     },
     {
       "epoch": 2.72,
-      "learning_rate": 4.7992299229923e-06,
-      "loss": 1.1991,
+      "learning_rate": 2.268976897689769e-06,
+      "loss": 1.3233,
       "step": 550
     },
     {
       "epoch": 2.77,
-      "learning_rate": 4.785478547854786e-06,
-      "loss": 1.2845,
+      "learning_rate": 2.3102310231023105e-06,
+      "loss": 1.3656,
       "step": 560
     },
     {
       "epoch": 2.82,
-      "learning_rate": 4.771727172717272e-06,
-      "loss": 1.2553,
+      "learning_rate": 2.3514851485148514e-06,
+      "loss": 1.3232,
       "step": 570
     },
     {
       "epoch": 2.87,
-      "learning_rate": 4.757975797579759e-06,
-      "loss": 1.2146,
+      "learning_rate": 2.392739273927393e-06,
+      "loss": 1.3777,
       "step": 580
     },
     {
       "epoch": 2.92,
-      "learning_rate": 4.7442244224422445e-06,
-      "loss": 1.2662,
+      "learning_rate": 2.4339933993399344e-06,
+      "loss": 1.3559,
       "step": 590
     },
     {
       "epoch": 2.97,
-      "learning_rate": 4.730473047304731e-06,
-      "loss": 1.2372,
+      "learning_rate": 2.4752475247524753e-06,
+      "loss": 1.3381,
       "step": 600
     },
     {
       "epoch": 3.0,
-      "eval_accuracy": 0.6074538488331592,
-      "eval_loss": 1.0635906457901,
-      "eval_runtime": 38.5764,
-      "eval_samples_per_second": 74.424,
-      "eval_steps_per_second": 2.333,
+      "eval_accuracy": 0.5468477882270986,
+      "eval_loss": 1.206508755683899,
+      "eval_runtime": 38.9418,
+      "eval_samples_per_second": 73.725,
+      "eval_steps_per_second": 2.311,
       "step": 606
     },
     {
       "epoch": 3.02,
-      "learning_rate": 4.716721672167217e-06,
-      "loss": 1.2289,
+      "learning_rate": 2.516501650165017e-06,
+      "loss": 1.2892,
       "step": 610
     },
     {
       "epoch": 3.07,
-      "learning_rate": 4.702970297029703e-06,
-      "loss": 1.2239,
+      "learning_rate": 2.557755775577558e-06,
+      "loss": 1.359,
       "step": 620
     },
     {
       "epoch": 3.12,
-      "learning_rate": 4.689218921892189e-06,
-      "loss": 1.2212,
+      "learning_rate": 2.5990099009900993e-06,
+      "loss": 1.3016,
       "step": 630
     },
     {
       "epoch": 3.17,
-      "learning_rate": 4.675467546754676e-06,
-      "loss": 1.2033,
+      "learning_rate": 2.64026402640264e-06,
+      "loss": 1.3118,
       "step": 640
     },
     {
       "epoch": 3.22,
-      "learning_rate": 4.661716171617162e-06,
-      "loss": 1.2295,
+      "learning_rate": 2.681518151815182e-06,
+      "loss": 1.3381,
       "step": 650
     },
     {
       "epoch": 3.27,
-      "learning_rate": 4.6479647964796485e-06,
-      "loss": 1.2271,
+      "learning_rate": 2.7227722772277232e-06,
+      "loss": 1.3114,
       "step": 660
     },
     {
       "epoch": 3.32,
-      "learning_rate": 4.634213421342134e-06,
-      "loss": 1.2193,
+      "learning_rate": 2.764026402640264e-06,
+      "loss": 1.3113,
       "step": 670
     },
     {
       "epoch": 3.37,
-      "learning_rate": 4.620462046204621e-06,
-      "loss": 1.2324,
+      "learning_rate": 2.8052805280528055e-06,
+      "loss": 1.2894,
       "step": 680
     },
     {
       "epoch": 3.42,
-      "learning_rate": 4.606710671067108e-06,
-      "loss": 1.1368,
+      "learning_rate": 2.8465346534653464e-06,
+      "loss": 1.3256,
       "step": 690
     },
     {
       "epoch": 3.47,
-      "learning_rate": 4.592959295929593e-06,
-      "loss": 1.1451,
+      "learning_rate": 2.887788778877888e-06,
+      "loss": 1.3077,
       "step": 700
     },
     {
       "epoch": 3.51,
-      "learning_rate": 4.57920792079208e-06,
-      "loss": 1.1696,
+      "learning_rate": 2.9290429042904294e-06,
+      "loss": 1.2873,
       "step": 710
     },
     {
       "epoch": 3.56,
-      "learning_rate": 4.565456545654566e-06,
-      "loss": 1.2064,
+      "learning_rate": 2.9702970297029703e-06,
+      "loss": 1.3329,
       "step": 720
     },
     {
       "epoch": 3.61,
-      "learning_rate": 4.5517051705170525e-06,
-      "loss": 1.2161,
+      "learning_rate": 3.011551155115512e-06,
+      "loss": 1.2657,
       "step": 730
     },
     {
       "epoch": 3.66,
-      "learning_rate": 4.537953795379538e-06,
-      "loss": 1.1665,
+      "learning_rate": 3.052805280528053e-06,
+      "loss": 1.2467,
       "step": 740
     },
     {
       "epoch": 3.71,
-      "learning_rate": 4.524202420242025e-06,
-      "loss": 1.165,
+      "learning_rate": 3.0940594059405943e-06,
+      "loss": 1.3142,
       "step": 750
     },
     {
       "epoch": 3.76,
-      "learning_rate": 4.510451045104511e-06,
-      "loss": 1.232,
+      "learning_rate": 3.1353135313531356e-06,
+      "loss": 1.2681,
       "step": 760
     },
     {
       "epoch": 3.81,
-      "learning_rate": 4.4966996699669966e-06,
-      "loss": 1.2113,
+      "learning_rate": 3.1765676567656765e-06,
+      "loss": 1.2687,
       "step": 770
     },
     {
       "epoch": 3.86,
-      "learning_rate": 4.482948294829483e-06,
-      "loss": 1.1504,
+      "learning_rate": 3.2178217821782182e-06,
+      "loss": 1.2866,
       "step": 780
     },
     {
       "epoch": 3.91,
-      "learning_rate": 4.469196919691969e-06,
-      "loss": 1.1573,
+      "learning_rate": 3.259075907590759e-06,
+      "loss": 1.2432,
       "step": 790
     },
     {
       "epoch": 3.96,
-      "learning_rate": 4.455445544554456e-06,
-      "loss": 1.2096,
+      "learning_rate": 3.3003300330033004e-06,
+      "loss": 1.2629,
       "step": 800
     },
     {
       "epoch": 4.0,
-      "eval_accuracy": 0.6116335771508186,
-      "eval_loss": 1.019976019859314,
-      "eval_runtime": 38.619,
-      "eval_samples_per_second": 74.342,
-      "eval_steps_per_second": 2.33,
+      "eval_accuracy": 0.588645071403692,
+      "eval_loss": 1.105048656463623,
+      "eval_runtime": 44.3471,
+      "eval_samples_per_second": 64.739,
+      "eval_steps_per_second": 2.029,
       "step": 808
     },
     {
       "epoch": 4.01,
-      "learning_rate": 4.4416941694169415e-06,
-      "loss": 1.1387,
+      "learning_rate": 3.341584158415842e-06,
+      "loss": 1.2836,
       "step": 810
     },
     {
       "epoch": 4.06,
-      "learning_rate": 4.427942794279428e-06,
-      "loss": 1.1161,
+      "learning_rate": 3.382838283828383e-06,
+      "loss": 1.2383,
       "step": 820
     },
     {
       "epoch": 4.11,
-      "learning_rate": 4.414191419141915e-06,
-      "loss": 1.1883,
+      "learning_rate": 3.4240924092409244e-06,
+      "loss": 1.2334,
       "step": 830
     },
     {
       "epoch": 4.16,
-      "learning_rate": 4.400440044004401e-06,
-      "loss": 1.1215,
+      "learning_rate": 3.4653465346534653e-06,
+      "loss": 1.2634,
       "step": 840
     },
     {
       "epoch": 4.21,
-      "learning_rate": 4.386688668866887e-06,
-      "loss": 1.1195,
+      "learning_rate": 3.506600660066007e-06,
+      "loss": 1.275,
       "step": 850
     },
     {
       "epoch": 4.26,
-      "learning_rate": 4.372937293729374e-06,
-      "loss": 1.1991,
+      "learning_rate": 3.5478547854785484e-06,
+      "loss": 1.2509,
       "step": 860
     },
     {
       "epoch": 4.31,
-      "learning_rate": 4.35918591859186e-06,
-      "loss": 1.1645,
+      "learning_rate": 3.5891089108910892e-06,
+      "loss": 1.214,
       "step": 870
     },
     {
       "epoch": 4.36,
-      "learning_rate": 4.345434543454346e-06,
-      "loss": 1.1446,
+      "learning_rate": 3.6303630363036306e-06,
+      "loss": 1.2483,
       "step": 880
     },
     {
       "epoch": 4.41,
-      "learning_rate": 4.331683168316832e-06,
-      "loss": 1.1308,
+      "learning_rate": 3.6716171617161715e-06,
+      "loss": 1.2113,
       "step": 890
     },
     {
       "epoch": 4.46,
-      "learning_rate": 4.317931793179319e-06,
-      "loss": 1.1758,
+      "learning_rate": 3.712871287128713e-06,
+      "loss": 1.2304,
       "step": 900
     },
     {
       "epoch": 4.5,
-      "learning_rate": 4.304180418041805e-06,
-      "loss": 1.1254,
+      "learning_rate": 3.7541254125412545e-06,
+      "loss": 1.2507,
       "step": 910
     },
     {
       "epoch": 4.55,
-      "learning_rate": 4.29042904290429e-06,
-      "loss": 1.1217,
+      "learning_rate": 3.7953795379537954e-06,
+      "loss": 1.2655,
       "step": 920
     },
     {
       "epoch": 4.6,
-      "learning_rate": 4.276677667766777e-06,
-      "loss": 1.1633,
+      "learning_rate": 3.836633663366337e-06,
+      "loss": 1.2257,
       "step": 930
     },
     {
       "epoch": 4.65,
-      "learning_rate": 4.262926292629263e-06,
-      "loss": 1.1553,
+      "learning_rate": 3.877887788778878e-06,
+      "loss": 1.1953,
       "step": 940
     },
     {
       "epoch": 4.7,
-      "learning_rate": 4.2491749174917495e-06,
-      "loss": 1.1764,
+      "learning_rate": 3.91914191419142e-06,
+      "loss": 1.222,
       "step": 950
     },
     {
       "epoch": 4.75,
-      "learning_rate": 4.235423542354235e-06,
-      "loss": 1.1058,
+      "learning_rate": 3.960396039603961e-06,
+      "loss": 1.2276,
       "step": 960
     },
     {
       "epoch": 4.8,
-      "learning_rate": 4.221672167216722e-06,
-      "loss": 1.1587,
+      "learning_rate": 4.001650165016502e-06,
+      "loss": 1.2075,
       "step": 970
     },
     {
       "epoch": 4.85,
-      "learning_rate": 4.207920792079208e-06,
-      "loss": 1.1889,
+      "learning_rate": 4.042904290429043e-06,
+      "loss": 1.2429,
       "step": 980
     },
     {
       "epoch": 4.9,
-      "learning_rate": 4.194169416941694e-06,
-      "loss": 1.1176,
+      "learning_rate": 4.084158415841584e-06,
+      "loss": 1.1501,
       "step": 990
     },
     {
       "epoch": 4.95,
-      "learning_rate": 4.180418041804181e-06,
-      "loss": 1.1319,
+      "learning_rate": 4.125412541254126e-06,
+      "loss": 1.2582,
       "step": 1000
     },
     {
       "epoch": 5.0,
       "learning_rate": 4.166666666666667e-06,
-      "loss": 1.145,
+      "loss": 1.248,
       "step": 1010
     },
     {
       "epoch": 5.0,
-      "eval_accuracy": 0.6325322187391152,
-      "eval_loss": 0.9768889546394348,
-      "eval_runtime": 38.6786,
-      "eval_samples_per_second": 74.227,
-      "eval_steps_per_second": 2.327,
+      "eval_accuracy": 0.6053639846743295,
+      "eval_loss": 1.0424652099609375,
+      "eval_runtime": 37.658,
+      "eval_samples_per_second": 76.239,
+      "eval_steps_per_second": 2.39,
       "step": 1010
     },
     {
       "epoch": 5.05,
-      "learning_rate": 4.1529152915291535e-06,
-      "loss": 1.1321,
+      "learning_rate": 4.207920792079208e-06,
+      "loss": 1.1697,
       "step": 1020
     },
     {
       "epoch": 5.1,
-      "learning_rate": 4.139163916391639e-06,
-      "loss": 1.1268,
+      "learning_rate": 4.2491749174917495e-06,
+      "loss": 1.2253,
       "step": 1030
     },
     {
       "epoch": 5.15,
-      "learning_rate": 4.125412541254126e-06,
-      "loss": 1.1032,
+      "learning_rate": 4.29042904290429e-06,
+      "loss": 1.1684,
       "step": 1040
     },
     {
       "epoch": 5.2,
-      "learning_rate": 4.111661166116613e-06,
-      "loss": 1.1327,
+      "learning_rate": 4.331683168316832e-06,
+      "loss": 1.2072,
       "step": 1050
     },
     {
       "epoch": 5.25,
-      "learning_rate": 4.0979097909790984e-06,
-      "loss": 1.141,
+      "learning_rate": 4.372937293729374e-06,
+      "loss": 1.2264,
       "step": 1060
     },
     {
       "epoch": 5.3,
-      "learning_rate": 4.084158415841584e-06,
-      "loss": 1.1025,
+      "learning_rate": 4.414191419141915e-06,
+      "loss": 1.2196,
       "step": 1070
     },
     {
       "epoch": 5.35,
-      "learning_rate": 4.070407040704071e-06,
-      "loss": 1.1138,
+      "learning_rate": 4.455445544554456e-06,
+      "loss": 1.2138,
       "step": 1080
     },
     {
       "epoch": 5.4,
-      "learning_rate": 4.056655665566557e-06,
-      "loss": 1.122,
+      "learning_rate": 4.4966996699669966e-06,
+      "loss": 1.1952,
       "step": 1090
     },
     {
       "epoch": 5.45,
-      "learning_rate": 4.042904290429043e-06,
-      "loss": 1.0642,
+      "learning_rate": 4.537953795379538e-06,
+      "loss": 1.1586,
       "step": 1100
     },
     {
       "epoch": 5.5,
-      "learning_rate": 4.029152915291529e-06,
-      "loss": 1.1162,
+      "learning_rate": 4.57920792079208e-06,
+      "loss": 1.1803,
       "step": 1110
     },
     {
       "epoch": 5.54,
-      "learning_rate": 4.015401540154016e-06,
-      "loss": 1.0873,
+      "learning_rate": 4.620462046204621e-06,
+      "loss": 1.1986,
       "step": 1120
     },
     {
       "epoch": 5.59,
-      "learning_rate": 4.001650165016502e-06,
-      "loss": 1.1188,
+      "learning_rate": 4.661716171617162e-06,
+      "loss": 1.1891,
       "step": 1130
     },
     {
       "epoch": 5.64,
-      "learning_rate": 3.987898789878988e-06,
-      "loss": 1.0996,
+      "learning_rate": 4.702970297029703e-06,
+      "loss": 1.1526,
       "step": 1140
     },
     {
       "epoch": 5.69,
-      "learning_rate": 3.974147414741474e-06,
-      "loss": 1.1377,
+      "learning_rate": 4.7442244224422445e-06,
+      "loss": 1.1906,
       "step": 1150
     },
     {
       "epoch": 5.74,
-      "learning_rate": 3.960396039603961e-06,
-      "loss": 1.0857,
+      "learning_rate": 4.785478547854786e-06,
+      "loss": 1.1474,
       "step": 1160
     },
     {
       "epoch": 5.79,
-      "learning_rate": 3.9466446644664465e-06,
-      "loss": 1.0875,
+      "learning_rate": 4.826732673267327e-06,
+      "loss": 1.1113,
       "step": 1170
     },
     {
       "epoch": 5.84,
-      "learning_rate": 3.932893289328933e-06,
-      "loss": 1.0532,
+      "learning_rate": 4.867986798679869e-06,
+      "loss": 1.2198,
       "step": 1180
     },
     {
       "epoch": 5.89,
-      "learning_rate": 3.91914191419142e-06,
-      "loss": 1.1319,
+      "learning_rate": 4.90924092409241e-06,
+      "loss": 1.1695,
       "step": 1190
     },
     {
       "epoch": 5.94,
-      "learning_rate": 3.905390539053906e-06,
-      "loss": 1.1125,
+      "learning_rate": 4.950495049504951e-06,
+      "loss": 1.1517,
       "step": 1200
     },
     {
       "epoch": 5.99,
-      "learning_rate": 3.891639163916392e-06,
-      "loss": 1.1589,
+      "learning_rate": 4.991749174917492e-06,
+      "loss": 1.2007,
       "step": 1210
     },
     {
       "epoch": 6.0,
-      "eval_accuracy": 0.6405433646812957,
-      "eval_loss": 0.9515116214752197,
-      "eval_runtime": 38.8551,
-      "eval_samples_per_second": 73.89,
-      "eval_steps_per_second": 2.316,
+      "eval_accuracy": 0.6339254615116684,
+      "eval_loss": 0.9874390363693237,
+      "eval_runtime": 37.6179,
+      "eval_samples_per_second": 76.32,
+      "eval_steps_per_second": 2.392,
       "step": 1212
     },
     {
       "epoch": 6.04,
-      "learning_rate": 3.877887788778878e-06,
-      "loss": 1.1213,
+      "learning_rate": 4.996332966629997e-06,
+      "loss": 1.1902,
       "step": 1220
     },
     {
       "epoch": 6.09,
-      "learning_rate": 3.864136413641365e-06,
-      "loss": 1.0874,
+      "learning_rate": 4.991749174917492e-06,
+      "loss": 1.1337,
       "step": 1230
     },
     {
       "epoch": 6.14,
-      "learning_rate": 3.8503850385038505e-06,
-      "loss": 1.0901,
+      "learning_rate": 4.987165383204988e-06,
+      "loss": 1.1566,
       "step": 1240
     },
     {
       "epoch": 6.19,
-      "learning_rate": 3.836633663366337e-06,
-      "loss": 1.0979,
+      "learning_rate": 4.982581591492483e-06,
+      "loss": 1.1235,
       "step": 1250
     },
     {
       "epoch": 6.24,
-      "learning_rate": 3.822882288228823e-06,
-      "loss": 1.074,
+      "learning_rate": 4.977997799779978e-06,
+      "loss": 1.1312,
       "step": 1260
     },
     {
       "epoch": 6.29,
-      "learning_rate": 3.8091309130913096e-06,
-      "loss": 1.0936,
+      "learning_rate": 4.9734140080674735e-06,
+      "loss": 1.1486,
       "step": 1270
     },
     {
       "epoch": 6.34,
-      "learning_rate": 3.7953795379537954e-06,
-      "loss": 1.071,
+      "learning_rate": 4.968830216354969e-06,
+      "loss": 1.1042,
       "step": 1280
     },
     {
       "epoch": 6.39,
-      "learning_rate": 3.781628162816282e-06,
-      "loss": 1.1341,
+      "learning_rate": 4.964246424642465e-06,
+      "loss": 1.1302,
       "step": 1290
     },
     {
       "epoch": 6.44,
-      "learning_rate": 3.767876787678768e-06,
-      "loss": 1.0803,
+      "learning_rate": 4.95966263292996e-06,
+      "loss": 1.1386,
       "step": 1300
     },
     {
       "epoch": 6.49,
-      "learning_rate": 3.7541254125412545e-06,
-      "loss": 1.1026,
+      "learning_rate": 4.955078841217455e-06,
+      "loss": 1.1502,
       "step": 1310
     },
     {
       "epoch": 6.53,
-      "learning_rate": 3.7403740374037408e-06,
-      "loss": 1.0736,
+      "learning_rate": 4.950495049504951e-06,
+      "loss": 1.1429,
       "step": 1320
     },
     {
       "epoch": 6.58,
-      "learning_rate": 3.726622662266227e-06,
-      "loss": 1.1181,
+      "learning_rate": 4.945911257792446e-06,
+      "loss": 1.1652,
       "step": 1330
     },
     {
       "epoch": 6.63,
-      "learning_rate": 3.712871287128713e-06,
-      "loss": 1.0721,
+      "learning_rate": 4.941327466079941e-06,
+      "loss": 1.1129,
       "step": 1340
     },
     {
       "epoch": 6.68,
-      "learning_rate": 3.6991199119912e-06,
-      "loss": 1.0743,
+      "learning_rate": 4.936743674367437e-06,
+      "loss": 1.1433,
       "step": 1350
     },
     {
       "epoch": 6.73,
-      "learning_rate": 3.6853685368536857e-06,
-      "loss": 1.1077,
+      "learning_rate": 4.9321598826549326e-06,
+      "loss": 1.1434,
       "step": 1360
     },
     {
       "epoch": 6.78,
-      "learning_rate": 3.6716171617161715e-06,
-      "loss": 1.0815,
+      "learning_rate": 4.927576090942428e-06,
+      "loss": 1.1051,
       "step": 1370
     },
     {
       "epoch": 6.83,
-      "learning_rate": 3.657865786578658e-06,
-      "loss": 1.0834,
+      "learning_rate": 4.922992299229923e-06,
+      "loss": 1.1306,
       "step": 1380
     },
     {
       "epoch": 6.88,
-      "learning_rate": 3.6441144114411443e-06,
-      "loss": 1.1362,
+      "learning_rate": 4.918408507517418e-06,
+      "loss": 1.1473,
       "step": 1390
     },
     {
       "epoch": 6.93,
-      "learning_rate": 3.6303630363036306e-06,
-      "loss": 1.0713,
+      "learning_rate": 4.913824715804914e-06,
+      "loss": 1.1608,
       "step": 1400
     },
     {
       "epoch": 6.98,
-      "learning_rate": 3.616611661166117e-06,
-      "loss": 1.0752,
+      "learning_rate": 4.90924092409241e-06,
+      "loss": 1.083,
       "step": 1410
     },
     {
       "epoch": 7.0,
-      "eval_accuracy": 0.64576802507837,
-      "eval_loss": 0.9395281076431274,
-      "eval_runtime": 38.3561,
-      "eval_samples_per_second": 74.851,
-      "eval_steps_per_second": 2.346,
+      "eval_accuracy": 0.644723092998955,
+      "eval_loss": 0.9609948396682739,
+      "eval_runtime": 37.9401,
+      "eval_samples_per_second": 75.672,
+      "eval_steps_per_second": 2.372,
       "step": 1414
     },
     {
       "epoch": 7.03,
-      "learning_rate": 3.6028602860286034e-06,
-      "loss": 1.0504,
+      "learning_rate": 4.904657132379905e-06,
+      "loss": 1.1368,
       "step": 1420
     },
     {
       "epoch": 7.08,
-      "learning_rate": 3.5891089108910892e-06,
-      "loss": 1.1083,
+      "learning_rate": 4.9000733406674e-06,
+      "loss": 1.0952,
       "step": 1430
     },
     {
       "epoch": 7.13,
-      "learning_rate": 3.575357535753576e-06,
-      "loss": 1.0908,
+      "learning_rate": 4.8954895489548956e-06,
+      "loss": 1.1306,
       "step": 1440
     },
     {
       "epoch": 7.18,
-      "learning_rate": 3.5616061606160617e-06,
-      "loss": 1.1186,
+      "learning_rate": 4.890905757242391e-06,
+      "loss": 1.1101,
       "step": 1450
     },
     {
       "epoch": 7.23,
-      "learning_rate": 3.5478547854785484e-06,
-      "loss": 1.096,
+      "learning_rate": 4.886321965529887e-06,
+      "loss": 1.097,
       "step": 1460
     },
     {
       "epoch": 7.28,
-      "learning_rate": 3.534103410341034e-06,
-      "loss": 1.1121,
+      "learning_rate": 4.881738173817382e-06,
+      "loss": 1.1345,
       "step": 1470
     },
     {
       "epoch": 7.33,
-      "learning_rate": 3.520352035203521e-06,
-      "loss": 1.0535,
+      "learning_rate": 4.8771543821048775e-06,
+      "loss": 1.1167,
       "step": 1480
     },
     {
       "epoch": 7.38,
-      "learning_rate": 3.506600660066007e-06,
-      "loss": 1.0403,
+      "learning_rate": 4.872570590392373e-06,
+      "loss": 1.1038,
       "step": 1490
     },
     {
       "epoch": 7.43,
-      "learning_rate": 3.4928492849284933e-06,
-      "loss": 1.0904,
+      "learning_rate": 4.867986798679869e-06,
+      "loss": 1.1067,
       "step": 1500
     },
     {
       "epoch": 7.48,
-      "learning_rate": 3.4790979097909795e-06,
-      "loss": 1.0987,
+      "learning_rate": 4.863403006967364e-06,
+      "loss": 1.1384,
       "step": 1510
     },
     {
       "epoch": 7.52,
-      "learning_rate": 3.4653465346534653e-06,
-      "loss": 1.0786,
+      "learning_rate": 4.858819215254859e-06,
+      "loss": 1.0725,
       "step": 1520
     },
     {
       "epoch": 7.57,
-      "learning_rate": 3.451595159515952e-06,
-      "loss": 1.1133,
+      "learning_rate": 4.854235423542355e-06,
+      "loss": 1.1605,
       "step": 1530
     },
     {
       "epoch": 7.62,
-      "learning_rate": 3.4378437843784377e-06,
-      "loss": 1.0804,
+      "learning_rate": 4.84965163182985e-06,
+      "loss": 1.0845,
       "step": 1540
     },
     {
       "epoch": 7.67,
-      "learning_rate": 3.4240924092409244e-06,
-      "loss": 1.0237,
+      "learning_rate": 4.845067840117346e-06,
+      "loss": 1.1047,
       "step": 1550
     },
     {
       "epoch": 7.72,
-      "learning_rate": 3.4103410341034106e-06,
-      "loss": 1.0276,
+      "learning_rate": 4.840484048404841e-06,
+      "loss": 1.0937,
       "step": 1560
     },
     {
       "epoch": 7.77,
-      "learning_rate": 3.396589658965897e-06,
-      "loss": 1.0769,
+      "learning_rate": 4.835900256692337e-06,
+      "loss": 1.0797,
       "step": 1570
     },
     {
       "epoch": 7.82,
-      "learning_rate": 3.382838283828383e-06,
-      "loss": 1.0962,
+      "learning_rate": 4.831316464979832e-06,
+      "loss": 1.1292,
       "step": 1580
     },
     {
       "epoch": 7.87,
-      "learning_rate": 3.3690869086908697e-06,
-      "loss": 1.0586,
+      "learning_rate": 4.826732673267327e-06,
+      "loss": 1.1153,
       "step": 1590
     },
     {
       "epoch": 7.92,
-      "learning_rate": 3.3553355335533555e-06,
-      "loss": 1.0786,
+      "learning_rate": 4.822148881554822e-06,
+      "loss": 1.0888,
       "step": 1600
     },
     {
       "epoch": 7.97,
-      "learning_rate": 3.341584158415842e-06,
-      "loss": 1.0524,
+      "learning_rate": 4.8175650898423185e-06,
+      "loss": 1.1061,
       "step": 1610
     },
     {
       "epoch": 8.0,
-      "eval_accuracy": 0.64576802507837,
-      "eval_loss": 0.9330951571464539,
-      "eval_runtime": 38.8204,
-      "eval_samples_per_second": 73.956,
-      "eval_steps_per_second": 2.318,
+      "eval_accuracy": 0.6523859282479972,
+      "eval_loss": 0.9385119080543518,
+      "eval_runtime": 37.523,
+      "eval_samples_per_second": 76.513,
+      "eval_steps_per_second": 2.399,
       "step": 1616
     },
     {
       "epoch": 8.02,
-      "learning_rate": 3.327832783278328e-06,
-      "loss": 1.0425,
+      "learning_rate": 4.812981298129814e-06,
+      "loss": 1.1135,
       "step": 1620
     },
     {
       "epoch": 8.07,
-      "learning_rate": 3.3140814081408146e-06,
-      "loss": 1.0729,
+      "learning_rate": 4.808397506417309e-06,
+      "loss": 1.1188,
       "step": 1630
     },
     {
       "epoch": 8.12,
-      "learning_rate": 3.3003300330033004e-06,
-      "loss": 1.0748,
+      "learning_rate": 4.803813714704804e-06,
+      "loss": 1.0562,
       "step": 1640
     },
     {
       "epoch": 8.17,
-      "learning_rate": 3.286578657865787e-06,
-      "loss": 1.0766,
+      "learning_rate": 4.7992299229923e-06,
+      "loss": 1.0857,
       "step": 1650
     },
     {
       "epoch": 8.22,
-      "learning_rate": 3.272827282728273e-06,
-      "loss": 1.0779,
+      "learning_rate": 4.794646131279795e-06,
+      "loss": 1.0731,
       "step": 1660
     },
     {
       "epoch": 8.27,
-      "learning_rate": 3.259075907590759e-06,
-      "loss": 1.0591,
+      "learning_rate": 4.790062339567291e-06,
+      "loss": 1.0705,
       "step": 1670
     },
     {
       "epoch": 8.32,
-      "learning_rate": 3.2453245324532458e-06,
-      "loss": 1.0551,
+      "learning_rate": 4.785478547854786e-06,
+      "loss": 1.094,
       "step": 1680
     },
     {
       "epoch": 8.37,
-      "learning_rate": 3.2315731573157316e-06,
-      "loss": 1.031,
+      "learning_rate": 4.7808947561422815e-06,
+      "loss": 1.1015,
       "step": 1690
     },
     {
       "epoch": 8.42,
-      "learning_rate": 3.2178217821782182e-06,
-      "loss": 1.0348,
+      "learning_rate": 4.776310964429777e-06,
+      "loss": 1.0873,
       "step": 1700
     },
     {
       "epoch": 8.47,
-      "learning_rate": 3.204070407040704e-06,
-      "loss": 1.064,
+      "learning_rate": 4.771727172717272e-06,
+      "loss": 1.089,
       "step": 1710
     },
     {
       "epoch": 8.51,
-      "learning_rate": 3.1903190319031907e-06,
-      "loss": 1.0142,
+      "learning_rate": 4.767143381004767e-06,
+      "loss": 1.095,
       "step": 1720
     },
     {
       "epoch": 8.56,
-      "learning_rate": 3.1765676567656765e-06,
-      "loss": 1.0871,
+      "learning_rate": 4.7625595892922626e-06,
+      "loss": 1.0872,
       "step": 1730
     },
     {
       "epoch": 8.61,
-      "learning_rate": 3.162816281628163e-06,
-      "loss": 1.0586,
+      "learning_rate": 4.757975797579759e-06,
+      "loss": 1.0989,
       "step": 1740
     },
     {
       "epoch": 8.66,
-      "learning_rate": 3.1490649064906494e-06,
-      "loss": 1.05,
+      "learning_rate": 4.753392005867254e-06,
+      "loss": 1.1298,
       "step": 1750
     },
     {
       "epoch": 8.71,
-      "learning_rate": 3.1353135313531356e-06,
-      "loss": 1.0503,
+      "learning_rate": 4.748808214154749e-06,
+      "loss": 1.1369,
       "step": 1760
     },
     {
       "epoch": 8.76,
-      "learning_rate": 3.121562156215622e-06,
-      "loss": 1.0318,
+      "learning_rate": 4.7442244224422445e-06,
+      "loss": 1.0746,
       "step": 1770
     },
     {
       "epoch": 8.81,
-      "learning_rate": 3.1078107810781085e-06,
-      "loss": 1.087,
+      "learning_rate": 4.73964063072974e-06,
+      "loss": 1.0882,
       "step": 1780
     },
     {
       "epoch": 8.86,
-      "learning_rate": 3.0940594059405943e-06,
-      "loss": 1.0563,
+      "learning_rate": 4.735056839017235e-06,
+      "loss": 1.0684,
       "step": 1790
     },
     {
       "epoch": 8.91,
-      "learning_rate": 3.080308030803081e-06,
-      "loss": 1.0416,
+      "learning_rate": 4.730473047304731e-06,
+      "loss": 1.0504,
       "step": 1800
     },
     {
       "epoch": 8.96,
-      "learning_rate": 3.0665566556655667e-06,
-      "loss": 1.0829,
+      "learning_rate": 4.725889255592226e-06,
+      "loss": 1.0597,
       "step": 1810
     },
     {
       "epoch": 9.0,
-      "eval_accuracy": 0.6572622779519331,
-      "eval_loss": 0.9173280000686646,
-      "eval_runtime": 38.5578,
-      "eval_samples_per_second": 74.46,
-      "eval_steps_per_second": 2.334,
+      "eval_accuracy": 0.6516893068617207,
+      "eval_loss": 0.9154672026634216,
+      "eval_runtime": 37.7412,
+      "eval_samples_per_second": 76.071,
+      "eval_steps_per_second": 2.385,
       "step": 1818
     },
     {
       "epoch": 9.01,
-      "learning_rate": 3.052805280528053e-06,
-      "loss": 1.0785,
+      "learning_rate": 4.721305463879722e-06,
+      "loss": 1.0758,
       "step": 1820
     },
     {
       "epoch": 9.06,
-      "learning_rate": 3.039053905390539e-06,
-      "loss": 1.0458,
+      "learning_rate": 4.716721672167217e-06,
+      "loss": 1.0463,
       "step": 1830
     },
     {
       "epoch": 9.11,
-      "learning_rate": 3.0253025302530254e-06,
-      "loss": 1.0499,
+      "learning_rate": 4.712137880454712e-06,
+      "loss": 1.0305,
       "step": 1840
     },
     {
       "epoch": 9.16,
-      "learning_rate": 3.011551155115512e-06,
-      "loss": 1.0528,
+      "learning_rate": 4.7075540887422075e-06,
+      "loss": 1.0939,
       "step": 1850
     },
     {
       "epoch": 9.21,
-      "learning_rate": 2.997799779977998e-06,
-      "loss": 1.0891,
+      "learning_rate": 4.702970297029703e-06,
+      "loss": 1.0295,
       "step": 1860
     },
     {
       "epoch": 9.26,
-      "learning_rate": 2.9840484048404845e-06,
-      "loss": 1.0229,
+      "learning_rate": 4.698386505317199e-06,
+      "loss": 1.0499,
       "step": 1870
     },
     {
       "epoch": 9.31,
-      "learning_rate": 2.9702970297029703e-06,
-      "loss": 1.0544,
+      "learning_rate": 4.693802713604694e-06,
+      "loss": 1.1133,
       "step": 1880
     },
     {
       "epoch": 9.36,
-      "learning_rate": 2.956545654565457e-06,
-      "loss": 1.0305,
+      "learning_rate": 4.689218921892189e-06,
+      "loss": 1.0863,
       "step": 1890
     },
     {
       "epoch": 9.41,
-      "learning_rate": 2.9427942794279428e-06,
-      "loss": 1.036,
+      "learning_rate": 4.684635130179685e-06,
+      "loss": 1.1398,
       "step": 1900
     },
     {
       "epoch": 9.46,
-      "learning_rate": 2.9290429042904294e-06,
-      "loss": 1.0191,
+      "learning_rate": 4.68005133846718e-06,
+      "loss": 1.0313,
       "step": 1910
     },
     {
       "epoch": 9.5,
-      "learning_rate": 2.9152915291529156e-06,
-      "loss": 1.0612,
+      "learning_rate": 4.675467546754676e-06,
+      "loss": 1.0492,
       "step": 1920
     },
     {
       "epoch": 9.55,
-      "learning_rate": 2.901540154015402e-06,
-      "loss": 1.0258,
+      "learning_rate": 4.670883755042171e-06,
+      "loss": 1.016,
       "step": 1930
     },
     {
       "epoch": 9.6,
-      "learning_rate": 2.887788778877888e-06,
-      "loss": 0.9895,
+      "learning_rate": 4.666299963329667e-06,
+      "loss": 1.0541,
       "step": 1940
     },
     {
       "epoch": 9.65,
-      "learning_rate": 2.8740374037403747e-06,
-      "loss": 1.0716,
+      "learning_rate": 4.661716171617162e-06,
+      "loss": 1.0498,
       "step": 1950
     },
     {
       "epoch": 9.7,
-      "learning_rate": 2.8602860286028605e-06,
-      "loss": 1.0605,
+      "learning_rate": 4.657132379904657e-06,
+      "loss": 1.0077,
       "step": 1960
     },
     {
       "epoch": 9.75,
-      "learning_rate": 2.8465346534653464e-06,
-      "loss": 1.0462,
+      "learning_rate": 4.652548588192153e-06,
+      "loss": 1.0413,
       "step": 1970
     },
     {
       "epoch": 9.8,
-      "learning_rate": 2.832783278327833e-06,
-      "loss": 1.0582,
+      "learning_rate": 4.6479647964796485e-06,
+      "loss": 1.0875,
       "step": 1980
     },
     {
       "epoch": 9.85,
-      "learning_rate": 2.8190319031903192e-06,
-      "loss": 1.0484,
+      "learning_rate": 4.643381004767144e-06,
+      "loss": 1.1108,
       "step": 1990
     },
     {
       "epoch": 9.9,
-      "learning_rate": 2.8052805280528055e-06,
-      "loss": 0.9998,
+      "learning_rate": 4.638797213054639e-06,
+      "loss": 1.0434,
       "step": 2000
     },
     {
       "epoch": 9.95,
-      "learning_rate": 2.7915291529152917e-06,
-      "loss": 1.0437,
+      "learning_rate": 4.634213421342134e-06,
+      "loss": 1.0951,
       "step": 2010
     },
     {
       "epoch": 10.0,
-      "learning_rate": 2.7777777777777783e-06,
-      "loss": 1.0219,
+      "learning_rate": 4.62962962962963e-06,
+      "loss": 1.0511,
       "step": 2020
     },
     {
       "epoch": 10.0,
-      "eval_accuracy": 0.6597004528039011,
-      "eval_loss": 0.9113705158233643,
-      "eval_runtime": 38.5342,
-      "eval_samples_per_second": 74.505,
-      "eval_steps_per_second": 2.336,
+      "eval_accuracy": 0.6579588993382097,
+      "eval_loss": 0.9127991795539856,
+      "eval_runtime": 37.7505,
+      "eval_samples_per_second": 76.052,
+      "eval_steps_per_second": 2.384,
       "step": 2020
     },
     {
       "epoch": 10.05,
-      "learning_rate": 2.764026402640264e-06,
-      "loss": 1.0749,
+      "learning_rate": 4.625045837917126e-06,
+      "loss": 1.0362,
       "step": 2030
     },
     {
       "epoch": 10.1,
-      "learning_rate": 2.750275027502751e-06,
-      "loss": 1.0419,
+      "learning_rate": 4.620462046204621e-06,
+      "loss": 1.0655,
       "step": 2040
     },
     {
       "epoch": 10.15,
-      "learning_rate": 2.7365236523652366e-06,
-      "loss": 1.0407,
+      "learning_rate": 4.615878254492116e-06,
+      "loss": 1.0384,
       "step": 2050
     },
     {
       "epoch": 10.2,
-      "learning_rate": 2.7227722772277232e-06,
-      "loss": 0.9984,
+      "learning_rate": 4.6112944627796115e-06,
+      "loss": 1.0336,
       "step": 2060
     },
     {
       "epoch": 10.25,
-      "learning_rate": 2.709020902090209e-06,
-      "loss": 0.9575,
+      "learning_rate": 4.606710671067108e-06,
+      "loss": 1.0624,
       "step": 2070
     },
     {
       "epoch": 10.3,
-      "learning_rate": 2.6952695269526957e-06,
-      "loss": 1.021,
+      "learning_rate": 4.602126879354603e-06,
+      "loss": 1.068,
       "step": 2080
     },
     {
       "epoch": 10.35,
-      "learning_rate": 2.681518151815182e-06,
-      "loss": 1.0191,
+      "learning_rate": 4.597543087642098e-06,
+      "loss": 1.043,
       "step": 2090
     },
     {
       "epoch": 10.4,
-      "learning_rate": 2.667766776677668e-06,
-      "loss": 1.0502,
+      "learning_rate": 4.592959295929593e-06,
+      "loss": 1.0353,
       "step": 2100
     },
     {
       "epoch": 10.45,
-      "learning_rate": 2.6540154015401544e-06,
-      "loss": 1.0292,
+      "learning_rate": 4.588375504217089e-06,
+      "loss": 1.0328,
       "step": 2110
     },
     {
       "epoch": 10.5,
-      "learning_rate": 2.64026402640264e-06,
-      "loss": 0.9986,
+      "learning_rate": 4.583791712504585e-06,
+      "loss": 1.0516,
       "step": 2120
     },
     {
       "epoch": 10.54,
-      "learning_rate": 2.626512651265127e-06,
-      "loss": 0.9604,
+      "learning_rate": 4.57920792079208e-06,
+      "loss": 1.0249,
       "step": 2130
     },
     {
       "epoch": 10.59,
-      "learning_rate": 2.6127612761276126e-06,
-      "loss": 1.0127,
+      "learning_rate": 4.574624129079575e-06,
+      "loss": 1.0343,
       "step": 2140
     },
     {
       "epoch": 10.64,
-      "learning_rate": 2.5990099009900993e-06,
-      "loss": 1.048,
+      "learning_rate": 4.570040337367071e-06,
+      "loss": 1.0877,
       "step": 2150
     },
     {
       "epoch": 10.69,
-      "learning_rate": 2.585258525852585e-06,
-      "loss": 1.0475,
+      "learning_rate": 4.565456545654566e-06,
+      "loss": 1.0895,
       "step": 2160
     },
     {
       "epoch": 10.74,
-      "learning_rate": 2.5715071507150717e-06,
-      "loss": 1.0552,
+      "learning_rate": 4.560872753942061e-06,
+      "loss": 1.0344,
       "step": 2170
     },
     {
       "epoch": 10.79,
-      "learning_rate": 2.557755775577558e-06,
-      "loss": 1.0351,
+      "learning_rate": 4.556288962229556e-06,
+      "loss": 1.0582,
       "step": 2180
     },
     {
       "epoch": 10.84,
-      "learning_rate": 2.544004400440044e-06,
-      "loss": 1.0465,
+      "learning_rate": 4.5517051705170525e-06,
+      "loss": 1.0478,
       "step": 2190
     },
     {
       "epoch": 10.89,
-      "learning_rate": 2.5302530253025304e-06,
-      "loss": 1.0807,
+      "learning_rate": 4.547121378804548e-06,
+      "loss": 1.0696,
       "step": 2200
     },
     {
       "epoch": 10.94,
-      "learning_rate": 2.516501650165017e-06,
-      "loss": 1.0255,
+      "learning_rate": 4.542537587092043e-06,
+      "loss": 1.0541,
       "step": 2210
     },
     {
       "epoch": 10.99,
-      "learning_rate": 2.502750275027503e-06,
-      "loss": 0.9986,
+      "learning_rate": 4.537953795379538e-06,
+      "loss": 1.012,
       "step": 2220
     },
     {
       "epoch": 11.0,
-      "eval_accuracy": 0.6579588993382097,
-      "eval_loss": 0.9033769369125366,
-      "eval_runtime": 38.6552,
-      "eval_samples_per_second": 74.272,
-      "eval_steps_per_second": 2.328,
+      "eval_accuracy": 0.66597004528039,
+      "eval_loss": 0.9047526717185974,
+      "eval_runtime": 37.6785,
+      "eval_samples_per_second": 76.197,
+      "eval_steps_per_second": 2.389,
       "step": 2222
     },
     {
       "epoch": 11.04,
-      "learning_rate": 2.488998899889989e-06,
-      "loss": 1.0336,
+      "learning_rate": 4.533370003667034e-06,
+      "loss": 1.0867,
       "step": 2230
     },
     {
       "epoch": 11.09,
-      "learning_rate": 2.4752475247524753e-06,
-      "loss": 1.0238,
+      "learning_rate": 4.528786211954529e-06,
+      "loss": 1.0345,
       "step": 2240
     },
     {
       "epoch": 11.14,
-      "learning_rate": 2.4614961496149616e-06,
-      "loss": 1.04,
+      "learning_rate": 4.524202420242025e-06,
+      "loss": 1.0066,
       "step": 2250
     },
     {
       "epoch": 11.19,
-      "learning_rate": 2.4477447744774478e-06,
-      "loss": 1.0165,
+      "learning_rate": 4.51961862852952e-06,
+      "loss": 0.9869,
       "step": 2260
     },
     {
       "epoch": 11.24,
-      "learning_rate": 2.4339933993399344e-06,
-      "loss": 1.0155,
+      "learning_rate": 4.5150348368170155e-06,
+      "loss": 1.0201,
       "step": 2270
     },
     {
       "epoch": 11.29,
-      "learning_rate": 2.4202420242024207e-06,
-      "loss": 1.0273,
+      "learning_rate": 4.510451045104511e-06,
+      "loss": 0.9869,
       "step": 2280
     },
     {
       "epoch": 11.34,
-      "learning_rate": 2.406490649064907e-06,
-      "loss": 1.0048,
+      "learning_rate": 4.505867253392006e-06,
+      "loss": 0.9864,
       "step": 2290
     },
     {
       "epoch": 11.39,
-      "learning_rate": 2.392739273927393e-06,
-      "loss": 0.988,
+      "learning_rate": 4.501283461679501e-06,
+      "loss": 1.0796,
       "step": 2300
     },
     {
       "epoch": 11.44,
-      "learning_rate": 2.3789878987898793e-06,
-      "loss": 1.0388,
+      "learning_rate": 4.4966996699669966e-06,
+      "loss": 1.0156,
       "step": 2310
     },
     {
       "epoch": 11.49,
-      "learning_rate": 2.3652365236523656e-06,
-      "loss": 1.0389,
+      "learning_rate": 4.492115878254493e-06,
+      "loss": 1.0199,
       "step": 2320
     },
     {
       "epoch": 11.53,
-      "learning_rate": 2.3514851485148514e-06,
-      "loss": 1.0651,
+      "learning_rate": 4.487532086541988e-06,
+      "loss": 1.0733,
       "step": 2330
     },
     {
       "epoch": 11.58,
-      "learning_rate": 2.337733773377338e-06,
-      "loss": 1.0789,
+      "learning_rate": 4.482948294829483e-06,
+      "loss": 1.0598,
       "step": 2340
     },
     {
       "epoch": 11.63,
-      "learning_rate": 2.3239823982398242e-06,
-      "loss": 0.9802,
+      "learning_rate": 4.4783645031169785e-06,
+      "loss": 1.0724,
       "step": 2350
     },
     {
       "epoch": 11.68,
-      "learning_rate": 2.3102310231023105e-06,
-      "loss": 1.054,
+      "learning_rate": 4.473780711404474e-06,
+      "loss": 1.0272,
       "step": 2360
     },
     {
       "epoch": 11.73,
-      "learning_rate": 2.2964796479647967e-06,
-      "loss": 1.0351,
+      "learning_rate": 4.469196919691969e-06,
+      "loss": 1.0477,
       "step": 2370
     },
     {
       "epoch": 11.78,
-      "learning_rate": 2.282728272827283e-06,
-      "loss": 1.0146,
+      "learning_rate": 4.464613127979465e-06,
+      "loss": 1.0071,
       "step": 2380
     },
     {
       "epoch": 11.83,
-      "learning_rate": 2.268976897689769e-06,
-      "loss": 1.0293,
+      "learning_rate": 4.46002933626696e-06,
+      "loss": 1.0568,
       "step": 2390
     },
     {
       "epoch": 11.88,
-      "learning_rate": 2.2552255225522554e-06,
-      "loss": 0.9827,
+      "learning_rate": 4.455445544554456e-06,
+      "loss": 1.0426,
       "step": 2400
     },
     {
       "epoch": 11.93,
-      "learning_rate": 2.2414741474147416e-06,
-      "loss": 0.9807,
+      "learning_rate": 4.450861752841951e-06,
+      "loss": 1.0291,
       "step": 2410
     },
     {
       "epoch": 11.98,
-      "learning_rate": 2.227722772277228e-06,
-      "loss": 1.013,
+      "learning_rate": 4.446277961129446e-06,
+      "loss": 1.0479,
       "step": 2420
     },
     {
       "epoch": 12.0,
-      "eval_accuracy": 0.6656217345872518,
-      "eval_loss": 0.9003686904907227,
-      "eval_runtime": 38.455,
-      "eval_samples_per_second": 74.659,
-      "eval_steps_per_second": 2.34,
+      "eval_accuracy": 0.6729362591431557,
+      "eval_loss": 0.8820806741714478,
+      "eval_runtime": 37.6359,
+      "eval_samples_per_second": 76.283,
+      "eval_steps_per_second": 2.391,
       "step": 2424
     },
     {
       "epoch": 12.03,
-      "learning_rate": 2.213971397139714e-06,
-      "loss": 1.0289,
+      "learning_rate": 4.4416941694169415e-06,
+      "loss": 1.0517,
       "step": 2430
     },
     {
       "epoch": 12.08,
-      "learning_rate": 2.2002200220022003e-06,
-      "loss": 1.0303,
+      "learning_rate": 4.437110377704438e-06,
+      "loss": 0.9969,
       "step": 2440
     },
     {
       "epoch": 12.13,
-      "learning_rate": 2.186468646864687e-06,
-      "loss": 1.0304,
+      "learning_rate": 4.432526585991933e-06,
+      "loss": 1.0549,
       "step": 2450
     },
     {
       "epoch": 12.18,
-      "learning_rate": 2.172717271727173e-06,
-      "loss": 1.0318,
+      "learning_rate": 4.427942794279428e-06,
+      "loss": 0.9794,
       "step": 2460
     },
     {
       "epoch": 12.23,
-      "learning_rate": 2.1589658965896594e-06,
-      "loss": 1.0566,
+      "learning_rate": 4.423359002566923e-06,
+      "loss": 1.0239,
       "step": 2470
     },
     {
       "epoch": 12.28,
-      "learning_rate": 2.145214521452145e-06,
-      "loss": 1.0052,
+      "learning_rate": 4.418775210854419e-06,
+      "loss": 1.0048,
       "step": 2480
     },
     {
       "epoch": 12.33,
-      "learning_rate": 2.1314631463146314e-06,
-      "loss": 0.9798,
+      "learning_rate": 4.414191419141915e-06,
+      "loss": 1.051,
       "step": 2490
     },
     {
       "epoch": 12.38,
-      "learning_rate": 2.1177117711771177e-06,
-      "loss": 1.0139,
+      "learning_rate": 4.40960762742941e-06,
+      "loss": 1.0001,
       "step": 2500
     },
     {
       "epoch": 12.43,
-      "learning_rate": 2.103960396039604e-06,
-      "loss": 0.9932,
+      "learning_rate": 4.405023835716905e-06,
+      "loss": 0.9507,
       "step": 2510
     },
     {
       "epoch": 12.48,
-      "learning_rate": 2.0902090209020905e-06,
-      "loss": 1.0196,
+      "learning_rate": 4.400440044004401e-06,
+      "loss": 1.0807,
       "step": 2520
     },
     {
       "epoch": 12.52,
-      "learning_rate": 2.0764576457645768e-06,
-      "loss": 1.0261,
+      "learning_rate": 4.395856252291897e-06,
+      "loss": 1.0422,
       "step": 2530
     },
     {
       "epoch": 12.57,
-      "learning_rate": 2.062706270627063e-06,
-      "loss": 1.0209,
+      "learning_rate": 4.391272460579392e-06,
+      "loss": 1.0412,
       "step": 2540
     },
     {
       "epoch": 12.62,
-      "learning_rate": 2.0489548954895492e-06,
-      "loss": 0.9964,
+      "learning_rate": 4.386688668866887e-06,
+      "loss": 1.023,
       "step": 2550
     },
     {
       "epoch": 12.67,
-      "learning_rate": 2.0352035203520354e-06,
-      "loss": 1.0248,
+      "learning_rate": 4.3821048771543825e-06,
+      "loss": 1.0623,
       "step": 2560
     },
     {
       "epoch": 12.72,
-      "learning_rate": 2.0214521452145217e-06,
-      "loss": 0.9682,
+      "learning_rate": 4.377521085441878e-06,
+      "loss": 1.0074,
       "step": 2570
     },
     {
       "epoch": 12.77,
-      "learning_rate": 2.007700770077008e-06,
-      "loss": 1.0485,
+      "learning_rate": 4.372937293729374e-06,
+      "loss": 1.0022,
       "step": 2580
     },
     {
       "epoch": 12.82,
-      "learning_rate": 1.993949394939494e-06,
-      "loss": 0.996,
+      "learning_rate": 4.368353502016869e-06,
+      "loss": 0.9737,
       "step": 2590
     },
     {
       "epoch": 12.87,
-      "learning_rate": 1.9801980198019803e-06,
-      "loss": 1.0432,
+      "learning_rate": 4.363769710304364e-06,
+      "loss": 1.0175,
       "step": 2600
     },
     {
       "epoch": 12.92,
-      "learning_rate": 1.9664466446644666e-06,
-      "loss": 1.0554,
+      "learning_rate": 4.35918591859186e-06,
+      "loss": 1.0366,
       "step": 2610
     },
     {
       "epoch": 12.97,
-      "learning_rate": 1.952695269526953e-06,
-      "loss": 1.0133,
+      "learning_rate": 4.354602126879355e-06,
+      "loss": 0.9993,
       "step": 2620
     },
     {
       "epoch": 13.0,
-      "eval_accuracy": 0.6628352490421456,
-      "eval_loss": 0.8939908146858215,
-      "eval_runtime": 38.8578,
-      "eval_samples_per_second": 73.885,
-      "eval_steps_per_second": 2.316,
+      "eval_accuracy": 0.6746778126088471,
+      "eval_loss": 0.8769838809967041,
+      "eval_runtime": 37.7441,
+      "eval_samples_per_second": 76.065,
+      "eval_steps_per_second": 2.384,
       "step": 2626
     },
     {
       "epoch": 13.02,
-      "learning_rate": 1.938943894389439e-06,
-      "loss": 0.9908,
+      "learning_rate": 4.35001833516685e-06,
+      "loss": 0.9699,
       "step": 2630
     },
     {
       "epoch": 13.07,
-      "learning_rate": 1.9251925192519253e-06,
-      "loss": 0.9901,
+      "learning_rate": 4.345434543454346e-06,
+      "loss": 1.0243,
       "step": 2640
     },
     {
       "epoch": 13.12,
-      "learning_rate": 1.9114411441144115e-06,
-      "loss": 0.9948,
+      "learning_rate": 4.340850751741842e-06,
+      "loss": 1.0212,
       "step": 2650
     },
     {
       "epoch": 13.17,
-      "learning_rate": 1.8976897689768977e-06,
-      "loss": 1.0064,
+      "learning_rate": 4.336266960029337e-06,
+      "loss": 1.0573,
       "step": 2660
     },
     {
       "epoch": 13.22,
-      "learning_rate": 1.883938393839384e-06,
-      "loss": 1.0087,
+      "learning_rate": 4.331683168316832e-06,
+      "loss": 0.991,
       "step": 2670
     },
     {
       "epoch": 13.27,
-      "learning_rate": 1.8701870187018704e-06,
-      "loss": 0.9959,
+      "learning_rate": 4.327099376604327e-06,
+      "loss": 0.9973,
       "step": 2680
     },
     {
       "epoch": 13.32,
-      "learning_rate": 1.8564356435643566e-06,
-      "loss": 0.9834,
+      "learning_rate": 4.322515584891823e-06,
+      "loss": 0.9595,
       "step": 2690
     },
     {
       "epoch": 13.37,
-      "learning_rate": 1.8426842684268428e-06,
-      "loss": 1.0346,
+      "learning_rate": 4.317931793179319e-06,
+      "loss": 1.0311,
       "step": 2700
     },
     {
       "epoch": 13.42,
-      "learning_rate": 1.828932893289329e-06,
-      "loss": 0.9487,
+      "learning_rate": 4.313348001466814e-06,
+      "loss": 1.0002,
       "step": 2710
     },
     {
       "epoch": 13.47,
-      "learning_rate": 1.8151815181518153e-06,
-      "loss": 0.9952,
+      "learning_rate": 4.308764209754309e-06,
+      "loss": 1.0047,
       "step": 2720
     },
     {
       "epoch": 13.51,
-      "learning_rate": 1.8014301430143017e-06,
-      "loss": 1.0243,
+      "learning_rate": 4.304180418041805e-06,
+      "loss": 1.0282,
       "step": 2730
     },
     {
       "epoch": 13.56,
-      "learning_rate": 1.787678767876788e-06,
-      "loss": 0.9498,
+      "learning_rate": 4.2995966263293e-06,
+      "loss": 1.0308,
       "step": 2740
     },
     {
       "epoch": 13.61,
-      "learning_rate": 1.7739273927392742e-06,
-      "loss": 0.9573,
+      "learning_rate": 4.295012834616795e-06,
+      "loss": 0.9929,
       "step": 2750
     },
     {
       "epoch": 13.66,
-      "learning_rate": 1.7601760176017604e-06,
-      "loss": 0.9838,
+      "learning_rate": 4.29042904290429e-06,
+      "loss": 1.0382,
       "step": 2760
     },
     {
       "epoch": 13.71,
-      "learning_rate": 1.7464246424642466e-06,
-      "loss": 1.0254,
+      "learning_rate": 4.2858452511917865e-06,
+      "loss": 0.9726,
       "step": 2770
     },
     {
       "epoch": 13.76,
-      "learning_rate": 1.7326732673267326e-06,
-      "loss": 1.0186,
+      "learning_rate": 4.281261459479282e-06,
+      "loss": 1.0059,
       "step": 2780
     },
     {
       "epoch": 13.81,
-      "learning_rate": 1.7189218921892189e-06,
-      "loss": 1.032,
+      "learning_rate": 4.276677667766777e-06,
+      "loss": 1.0,
       "step": 2790
     },
     {
       "epoch": 13.86,
-      "learning_rate": 1.7051705170517053e-06,
-      "loss": 1.0525,
+      "learning_rate": 4.272093876054272e-06,
+      "loss": 1.0118,
       "step": 2800
     },
     {
       "epoch": 13.91,
-      "learning_rate": 1.6914191419141915e-06,
-      "loss": 1.0422,
+      "learning_rate": 4.267510084341768e-06,
+      "loss": 0.9912,
       "step": 2810
     },
     {
       "epoch": 13.96,
-      "learning_rate": 1.6776677667766778e-06,
-      "loss": 1.0064,
+      "learning_rate": 4.262926292629263e-06,
+      "loss": 0.9784,
       "step": 2820
     },
     {
       "epoch": 14.0,
-      "eval_accuracy": 0.6649251132009752,
-      "eval_loss": 0.8916007876396179,
-      "eval_runtime": 38.5634,
-      "eval_samples_per_second": 74.449,
-      "eval_steps_per_second": 2.334,
+      "eval_accuracy": 0.6757227446882619,
+      "eval_loss": 0.867244303226471,
+      "eval_runtime": 37.7798,
+      "eval_samples_per_second": 75.993,
+      "eval_steps_per_second": 2.382,
       "step": 2828
     },
     {
       "epoch": 14.01,
-      "learning_rate": 1.663916391639164e-06,
-      "loss": 1.0109,
+      "learning_rate": 4.258342500916759e-06,
+      "loss": 0.9895,
       "step": 2830
     },
     {
       "epoch": 14.06,
-      "learning_rate": 1.6501650165016502e-06,
-      "loss": 1.016,
+      "learning_rate": 4.253758709204254e-06,
+      "loss": 1.0208,
       "step": 2840
     },
     {
       "epoch": 14.11,
-      "learning_rate": 1.6364136413641364e-06,
-      "loss": 1.0287,
+      "learning_rate": 4.2491749174917495e-06,
+      "loss": 0.9673,
       "step": 2850
     },
     {
       "epoch": 14.16,
-      "learning_rate": 1.6226622662266229e-06,
-      "loss": 1.0048,
+      "learning_rate": 4.244591125779245e-06,
+      "loss": 1.0232,
       "step": 2860
     },
     {
       "epoch": 14.21,
-      "learning_rate": 1.6089108910891091e-06,
-      "loss": 1.0639,
+      "learning_rate": 4.24000733406674e-06,
+      "loss": 0.9917,
       "step": 2870
     },
     {
       "epoch": 14.26,
-      "learning_rate": 1.5951595159515953e-06,
-      "loss": 0.9976,
+      "learning_rate": 4.235423542354235e-06,
+      "loss": 0.9954,
       "step": 2880
     },
     {
       "epoch": 14.31,
-      "learning_rate": 1.5814081408140816e-06,
-      "loss": 0.9993,
+      "learning_rate": 4.2308397506417306e-06,
+      "loss": 1.0239,
       "step": 2890
     },
     {
       "epoch": 14.36,
-      "learning_rate": 1.5676567656765678e-06,
-      "loss": 1.045,
+      "learning_rate": 4.226255958929227e-06,
+      "loss": 1.0194,
       "step": 2900
     },
     {
       "epoch": 14.41,
-      "learning_rate": 1.5539053905390542e-06,
-      "loss": 1.0057,
+      "learning_rate": 4.221672167216722e-06,
+      "loss": 1.0078,
       "step": 2910
     },
     {
       "epoch": 14.46,
-      "learning_rate": 1.5401540154015405e-06,
-      "loss": 1.0304,
+      "learning_rate": 4.217088375504217e-06,
+      "loss": 0.9911,
       "step": 2920
     },
     {
       "epoch": 14.5,
-      "learning_rate": 1.5264026402640265e-06,
-      "loss": 1.0412,
+      "learning_rate": 4.2125045837917125e-06,
+      "loss": 1.0034,
       "step": 2930
     },
     {
       "epoch": 14.55,
-      "learning_rate": 1.5126512651265127e-06,
-      "loss": 1.0259,
+      "learning_rate": 4.207920792079208e-06,
+      "loss": 0.9614,
       "step": 2940
     },
     {
       "epoch": 14.6,
-      "learning_rate": 1.498899889988999e-06,
-      "loss": 1.0207,
+      "learning_rate": 4.203337000366703e-06,
+      "loss": 0.9772,
       "step": 2950
     },
     {
       "epoch": 14.65,
-      "learning_rate": 1.4851485148514852e-06,
-      "loss": 0.9956,
+      "learning_rate": 4.198753208654199e-06,
+      "loss": 0.9929,
       "step": 2960
     },
     {
       "epoch": 14.7,
-      "learning_rate": 1.4713971397139714e-06,
-      "loss": 0.9838,
+      "learning_rate": 4.194169416941694e-06,
+      "loss": 0.9751,
       "step": 2970
     },
     {
       "epoch": 14.75,
-      "learning_rate": 1.4576457645764578e-06,
-      "loss": 1.0084,
+      "learning_rate": 4.18958562522919e-06,
+      "loss": 0.994,
       "step": 2980
     },
     {
       "epoch": 14.8,
-      "learning_rate": 1.443894389438944e-06,
-      "loss": 0.9956,
+      "learning_rate": 4.185001833516685e-06,
+      "loss": 0.9394,
       "step": 2990
     },
     {
       "epoch": 14.85,
-      "learning_rate": 1.4301430143014303e-06,
-      "loss": 0.9595,
+      "learning_rate": 4.180418041804181e-06,
+      "loss": 1.0047,
       "step": 3000
     },
     {
       "epoch": 14.9,
-      "learning_rate": 1.4163916391639165e-06,
-      "loss": 0.9391,
+      "learning_rate": 4.175834250091676e-06,
+      "loss": 0.9899,
       "step": 3010
     },
     {
       "epoch": 14.95,
-      "learning_rate": 1.4026402640264027e-06,
-      "loss": 1.0213,
+      "learning_rate": 4.171250458379172e-06,
+      "loss": 1.0034,
       "step": 3020
     },
     {
       "epoch": 15.0,
-      "learning_rate": 1.3888888888888892e-06,
-      "loss": 0.9858,
+      "learning_rate": 4.166666666666667e-06,
+      "loss": 1.0439,
       "step": 3030
     },
     {
       "epoch": 15.0,
-      "eval_accuracy": 0.6732845698362939,
-      "eval_loss": 0.8882250785827637,
-      "eval_runtime": 38.5965,
-      "eval_samples_per_second": 74.385,
-      "eval_steps_per_second": 2.332,
+      "eval_accuracy": 0.6750261233019854,
+      "eval_loss": 0.8766063451766968,
+      "eval_runtime": 37.8313,
+      "eval_samples_per_second": 75.89,
+      "eval_steps_per_second": 2.379,
       "step": 3030
     },
     {
       "epoch": 15.05,
-      "learning_rate": 1.3751375137513754e-06,
-      "loss": 0.9837,
+      "learning_rate": 4.162082874954162e-06,
+      "loss": 1.0191,
       "step": 3040
     },
     {
       "epoch": 15.1,
-      "learning_rate": 1.3613861386138616e-06,
-      "loss": 0.9723,
+      "learning_rate": 4.157499083241658e-06,
+      "loss": 0.9551,
       "step": 3050
     },
     {
       "epoch": 15.15,
-      "learning_rate": 1.3476347634763478e-06,
-      "loss": 1.018,
+      "learning_rate": 4.1529152915291535e-06,
+      "loss": 0.9798,
       "step": 3060
     },
     {
       "epoch": 15.2,
-      "learning_rate": 1.333883388338834e-06,
-      "loss": 1.0612,
+      "learning_rate": 4.148331499816649e-06,
+      "loss": 0.9764,
       "step": 3070
     },
     {
       "epoch": 15.25,
-      "learning_rate": 1.32013201320132e-06,
-      "loss": 1.0081,
+      "learning_rate": 4.143747708104144e-06,
+      "loss": 1.0136,
       "step": 3080
     },
     {
       "epoch": 15.3,
-      "learning_rate": 1.3063806380638063e-06,
-      "loss": 1.0096,
+      "learning_rate": 4.139163916391639e-06,
+      "loss": 0.9354,
       "step": 3090
     },
     {
       "epoch": 15.35,
-      "learning_rate": 1.2926292629262925e-06,
-      "loss": 0.9442,
+      "learning_rate": 4.1345801246791354e-06,
+      "loss": 0.9896,
       "step": 3100
     },
     {
       "epoch": 15.4,
-      "learning_rate": 1.278877887788779e-06,
-      "loss": 1.0434,
+      "learning_rate": 4.129996332966631e-06,
+      "loss": 0.9964,
       "step": 3110
     },
     {
       "epoch": 15.45,
-      "learning_rate": 1.2651265126512652e-06,
-      "loss": 1.025,
+      "learning_rate": 4.125412541254126e-06,
+      "loss": 1.0298,
       "step": 3120
     },
     {
       "epoch": 15.5,
-      "learning_rate": 1.2513751375137514e-06,
-      "loss": 1.0102,
+      "learning_rate": 4.120828749541621e-06,
+      "loss": 0.9645,
       "step": 3130
     },
     {
       "epoch": 15.54,
-      "learning_rate": 1.2376237623762377e-06,
-      "loss": 0.9965,
+      "learning_rate": 4.1162449578291165e-06,
+      "loss": 1.0145,
       "step": 3140
     },
     {
       "epoch": 15.59,
-      "learning_rate": 1.2238723872387239e-06,
-      "loss": 1.0295,
+      "learning_rate": 4.111661166116613e-06,
+      "loss": 0.9588,
       "step": 3150
     },
     {
       "epoch": 15.64,
-      "learning_rate": 1.2101210121012103e-06,
-      "loss": 0.988,
+      "learning_rate": 4.107077374404108e-06,
+      "loss": 0.9552,
       "step": 3160
     },
     {
       "epoch": 15.69,
-      "learning_rate": 1.1963696369636966e-06,
-      "loss": 1.001,
+      "learning_rate": 4.102493582691603e-06,
+      "loss": 1.0058,
       "step": 3170
     },
     {
       "epoch": 15.74,
-      "learning_rate": 1.1826182618261828e-06,
-      "loss": 1.0232,
+      "learning_rate": 4.0979097909790984e-06,
+      "loss": 0.9558,
       "step": 3180
     },
     {
       "epoch": 15.79,
-      "learning_rate": 1.168866886688669e-06,
-      "loss": 1.0148,
+      "learning_rate": 4.093325999266594e-06,
+      "loss": 0.995,
       "step": 3190
     },
     {
       "epoch": 15.84,
-      "learning_rate": 1.1551155115511552e-06,
-      "loss": 0.9523,
+      "learning_rate": 4.088742207554089e-06,
+      "loss": 0.9753,
       "step": 3200
     },
     {
       "epoch": 15.89,
-      "learning_rate": 1.1413641364136415e-06,
-      "loss": 1.0571,
+      "learning_rate": 4.084158415841584e-06,
+      "loss": 1.0178,
       "step": 3210
     },
     {
       "epoch": 15.94,
-      "learning_rate": 1.1276127612761277e-06,
-      "loss": 0.973,
+      "learning_rate": 4.07957462412908e-06,
+      "loss": 0.9889,
       "step": 3220
     },
     {
       "epoch": 15.99,
-      "learning_rate": 1.113861386138614e-06,
-      "loss": 0.9863,
+      "learning_rate": 4.074990832416576e-06,
+      "loss": 0.9782,
       "step": 3230
     },
     {
       "epoch": 16.0,
-      "eval_accuracy": 0.6739811912225705,
-      "eval_loss": 0.8849846720695496,
-      "eval_runtime": 38.9067,
-      "eval_samples_per_second": 73.792,
-      "eval_steps_per_second": 2.313,
+      "eval_accuracy": 0.6746778126088471,
+      "eval_loss": 0.8658467531204224,
+      "eval_runtime": 37.7341,
+      "eval_samples_per_second": 76.085,
+      "eval_steps_per_second": 2.385,
       "step": 3232
     },
     {
       "epoch": 16.04,
-      "learning_rate": 1.1001100110011001e-06,
-      "loss": 0.9616,
+      "learning_rate": 4.070407040704071e-06,
+      "loss": 0.9658,
       "step": 3240
     },
     {
       "epoch": 16.09,
-      "learning_rate": 1.0863586358635866e-06,
-      "loss": 1.0094,
+      "learning_rate": 4.065823248991566e-06,
+      "loss": 0.9724,
       "step": 3250
     },
     {
       "epoch": 16.14,
-      "learning_rate": 1.0726072607260726e-06,
-      "loss": 0.9672,
+      "learning_rate": 4.061239457279061e-06,
+      "loss": 1.0293,
       "step": 3260
     },
     {
       "epoch": 16.19,
-      "learning_rate": 1.0588558855885588e-06,
-      "loss": 1.0321,
+      "learning_rate": 4.056655665566557e-06,
+      "loss": 1.0026,
       "step": 3270
     },
     {
       "epoch": 16.24,
-      "learning_rate": 1.0451045104510453e-06,
-      "loss": 1.0313,
+      "learning_rate": 4.052071873854053e-06,
+      "loss": 0.9738,
       "step": 3280
     },
     {
       "epoch": 16.29,
-      "learning_rate": 1.0313531353135315e-06,
-      "loss": 1.0116,
+      "learning_rate": 4.047488082141548e-06,
+      "loss": 0.9857,
       "step": 3290
     },
     {
       "epoch": 16.34,
-      "learning_rate": 1.0176017601760177e-06,
-      "loss": 0.9632,
+      "learning_rate": 4.042904290429043e-06,
+      "loss": 1.0062,
       "step": 3300
     },
     {
       "epoch": 16.39,
-      "learning_rate": 1.003850385038504e-06,
-      "loss": 1.0021,
+      "learning_rate": 4.038320498716539e-06,
+      "loss": 0.9816,
       "step": 3310
     },
     {
       "epoch": 16.44,
-      "learning_rate": 9.900990099009902e-07,
-      "loss": 0.9827,
+      "learning_rate": 4.033736707004034e-06,
+      "loss": 0.9615,
       "step": 3320
     },
     {
       "epoch": 16.49,
-      "learning_rate": 9.763476347634764e-07,
-      "loss": 0.9916,
+      "learning_rate": 4.029152915291529e-06,
+      "loss": 0.9772,
       "step": 3330
     },
     {
       "epoch": 16.53,
-      "learning_rate": 9.625962596259626e-07,
-      "loss": 0.9863,
+      "learning_rate": 4.024569123579024e-06,
+      "loss": 0.9882,
       "step": 3340
     },
     {
       "epoch": 16.58,
-      "learning_rate": 9.488448844884489e-07,
-      "loss": 0.9428,
+      "learning_rate": 4.0199853318665205e-06,
+      "loss": 0.9629,
       "step": 3350
     },
     {
       "epoch": 16.63,
-      "learning_rate": 9.350935093509352e-07,
-      "loss": 0.9882,
+      "learning_rate": 4.015401540154016e-06,
+      "loss": 0.9704,
       "step": 3360
     },
     {
       "epoch": 16.68,
-      "learning_rate": 9.213421342134214e-07,
-      "loss": 1.0206,
+      "learning_rate": 4.010817748441511e-06,
+      "loss": 0.9873,
       "step": 3370
     },
     {
       "epoch": 16.73,
-      "learning_rate": 9.075907590759076e-07,
-      "loss": 0.9906,
+      "learning_rate": 4.006233956729006e-06,
+      "loss": 0.9565,
       "step": 3380
     },
     {
       "epoch": 16.78,
-      "learning_rate": 8.93839383938394e-07,
-      "loss": 0.9814,
+      "learning_rate": 4.001650165016502e-06,
+      "loss": 0.9671,
       "step": 3390
     },
     {
       "epoch": 16.83,
-      "learning_rate": 8.800880088008802e-07,
-      "loss": 0.9964,
+      "learning_rate": 3.997066373303997e-06,
+      "loss": 0.9612,
       "step": 3400
     },
     {
       "epoch": 16.88,
-      "learning_rate": 8.663366336633663e-07,
-      "loss": 0.9854,
+      "learning_rate": 3.992482581591493e-06,
+      "loss": 0.974,
       "step": 3410
     },
     {
       "epoch": 16.93,
-      "learning_rate": 8.525852585258527e-07,
-      "loss": 1.0075,
+      "learning_rate": 3.987898789878988e-06,
+      "loss": 0.9577,
       "step": 3420
     },
     {
       "epoch": 16.98,
-      "learning_rate": 8.388338833883389e-07,
-      "loss": 1.0058,
+      "learning_rate": 3.9833149981664835e-06,
+      "loss": 0.9664,
       "step": 3430
     },
     {
       "epoch": 17.0,
-      "eval_accuracy": 0.6746778126088471,
-      "eval_loss": 0.8856182098388672,
-      "eval_runtime": 38.4783,
-      "eval_samples_per_second": 74.614,
-      "eval_steps_per_second": 2.339,
+      "eval_accuracy": 0.6764193660745385,
+      "eval_loss": 0.8595666885375977,
+      "eval_runtime": 37.7928,
+      "eval_samples_per_second": 75.967,
+      "eval_steps_per_second": 2.381,
       "step": 3434
     },
     {
       "epoch": 17.03,
-      "learning_rate": 8.250825082508251e-07,
-      "loss": 1.0082,
+      "learning_rate": 3.978731206453979e-06,
+      "loss": 0.9693,
       "step": 3440
     },
     {
       "epoch": 17.08,
-      "learning_rate": 8.113311331133114e-07,
-      "loss": 0.9692,
+      "learning_rate": 3.974147414741474e-06,
+      "loss": 0.9806,
       "step": 3450
     },
     {
       "epoch": 17.13,
-      "learning_rate": 7.975797579757977e-07,
-      "loss": 0.9329,
+      "learning_rate": 3.969563623028969e-06,
+      "loss": 0.9563,
       "step": 3460
     },
     {
       "epoch": 17.18,
-      "learning_rate": 7.838283828382839e-07,
-      "loss": 0.9938,
+      "learning_rate": 3.9649798313164654e-06,
+      "loss": 0.9308,
       "step": 3470
     },
     {
       "epoch": 17.23,
-      "learning_rate": 7.700770077007702e-07,
-      "loss": 0.9911,
+      "learning_rate": 3.960396039603961e-06,
+      "loss": 0.9858,
       "step": 3480
     },
     {
       "epoch": 17.28,
-      "learning_rate": 7.563256325632564e-07,
-      "loss": 1.0087,
+      "learning_rate": 3.955812247891456e-06,
+      "loss": 1.0189,
       "step": 3490
     },
     {
       "epoch": 17.33,
-      "learning_rate": 7.425742574257426e-07,
-      "loss": 0.9697,
+      "learning_rate": 3.951228456178951e-06,
+      "loss": 0.9754,
       "step": 3500
     },
     {
       "epoch": 17.38,
-      "learning_rate": 7.288228822882289e-07,
-      "loss": 0.973,
+      "learning_rate": 3.9466446644664465e-06,
+      "loss": 0.9648,
       "step": 3510
     },
     {
       "epoch": 17.43,
-      "learning_rate": 7.150715071507151e-07,
-      "loss": 1.0056,
+      "learning_rate": 3.942060872753943e-06,
+      "loss": 0.9555,
       "step": 3520
     },
     {
       "epoch": 17.48,
-      "learning_rate": 7.013201320132014e-07,
-      "loss": 0.9723,
+      "learning_rate": 3.937477081041438e-06,
+      "loss": 0.9821,
       "step": 3530
     },
     {
       "epoch": 17.52,
-      "learning_rate": 6.875687568756877e-07,
-      "loss": 0.9905,
+      "learning_rate": 3.932893289328933e-06,
+      "loss": 0.9465,
       "step": 3540
     },
     {
       "epoch": 17.57,
-      "learning_rate": 6.738173817381739e-07,
-      "loss": 1.0209,
+      "learning_rate": 3.928309497616428e-06,
+      "loss": 1.0026,
       "step": 3550
     },
     {
       "epoch": 17.62,
-      "learning_rate": 6.6006600660066e-07,
-      "loss": 0.9991,
+      "learning_rate": 3.923725705903924e-06,
+      "loss": 0.9656,
       "step": 3560
     },
     {
       "epoch": 17.67,
-      "learning_rate": 6.463146314631463e-07,
-      "loss": 1.0039,
+      "learning_rate": 3.91914191419142e-06,
+      "loss": 0.9419,
       "step": 3570
     },
     {
       "epoch": 17.72,
-      "learning_rate": 6.325632563256326e-07,
-      "loss": 0.946,
+      "learning_rate": 3.914558122478915e-06,
+      "loss": 0.9322,
       "step": 3580
     },
     {
       "epoch": 17.77,
-      "learning_rate": 6.188118811881188e-07,
-      "loss": 0.9709,
+      "learning_rate": 3.90997433076641e-06,
+      "loss": 0.9747,
       "step": 3590
     },
     {
       "epoch": 17.82,
-      "learning_rate": 6.050605060506052e-07,
-      "loss": 1.0213,
+      "learning_rate": 3.905390539053906e-06,
+      "loss": 0.9895,
       "step": 3600
     },
     {
       "epoch": 17.87,
-      "learning_rate": 5.913091309130914e-07,
-      "loss": 1.0016,
+      "learning_rate": 3.900806747341402e-06,
+      "loss": 0.9908,
       "step": 3610
     },
     {
       "epoch": 17.92,
-      "learning_rate": 5.775577557755776e-07,
-      "loss": 0.9889,
+      "learning_rate": 3.896222955628897e-06,
+      "loss": 0.9692,
       "step": 3620
     },
     {
       "epoch": 17.97,
-      "learning_rate": 5.638063806380638e-07,
-      "loss": 0.9637,
+      "learning_rate": 3.891639163916392e-06,
+      "loss": 1.0132,
       "step": 3630
     },
     {
       "epoch": 18.0,
-      "eval_accuracy": 0.6722396377568791,
-      "eval_loss": 0.8851516842842102,
-      "eval_runtime": 38.5349,
-      "eval_samples_per_second": 74.504,
-      "eval_steps_per_second": 2.336,
+      "eval_accuracy": 0.6805990943921978,
+      "eval_loss": 0.8490851521492004,
+      "eval_runtime": 37.9147,
+      "eval_samples_per_second": 75.723,
+      "eval_steps_per_second": 2.374,
       "step": 3636
     },
     {
       "epoch": 18.02,
-      "learning_rate": 5.500550055005501e-07,
-      "loss": 0.9832,
+      "learning_rate": 3.8870553722038875e-06,
+      "loss": 0.9588,
       "step": 3640
     },
     {
       "epoch": 18.07,
-      "learning_rate": 5.363036303630363e-07,
-      "loss": 0.9492,
+      "learning_rate": 3.882471580491383e-06,
+      "loss": 0.9421,
       "step": 3650
     },
     {
       "epoch": 18.12,
-      "learning_rate": 5.225522552255226e-07,
-      "loss": 0.9676,
+      "learning_rate": 3.877887788778878e-06,
+      "loss": 0.9018,
       "step": 3660
     },
     {
       "epoch": 18.17,
-      "learning_rate": 5.088008800880089e-07,
-      "loss": 0.991,
+      "learning_rate": 3.873303997066374e-06,
+      "loss": 0.9781,
       "step": 3670
     },
     {
       "epoch": 18.22,
-      "learning_rate": 4.950495049504951e-07,
-      "loss": 1.0016,
+      "learning_rate": 3.8687202053538694e-06,
+      "loss": 0.9531,
       "step": 3680
     },
     {
       "epoch": 18.27,
-      "learning_rate": 4.812981298129813e-07,
-      "loss": 1.0346,
+      "learning_rate": 3.864136413641365e-06,
+      "loss": 0.9852,
       "step": 3690
     },
     {
       "epoch": 18.32,
-      "learning_rate": 4.675467546754676e-07,
-      "loss": 1.0101,
+      "learning_rate": 3.85955262192886e-06,
+      "loss": 0.9337,
       "step": 3700
     },
     {
       "epoch": 18.37,
-      "learning_rate": 4.537953795379538e-07,
-      "loss": 1.0226,
+      "learning_rate": 3.854968830216355e-06,
+      "loss": 0.9562,
       "step": 3710
     },
     {
       "epoch": 18.42,
-      "learning_rate": 4.400440044004401e-07,
-      "loss": 1.0054,
+      "learning_rate": 3.8503850385038505e-06,
+      "loss": 0.9701,
       "step": 3720
     },
     {
       "epoch": 18.47,
-      "learning_rate": 4.2629262926292633e-07,
-      "loss": 0.994,
+      "learning_rate": 3.845801246791347e-06,
+      "loss": 0.9603,
       "step": 3730
     },
     {
       "epoch": 18.51,
-      "learning_rate": 4.1254125412541255e-07,
-      "loss": 1.0177,
+      "learning_rate": 3.841217455078842e-06,
+      "loss": 0.9581,
       "step": 3740
     },
     {
       "epoch": 18.56,
-      "learning_rate": 3.9878987898789883e-07,
-      "loss": 0.97,
+      "learning_rate": 3.836633663366337e-06,
+      "loss": 0.9454,
       "step": 3750
     },
     {
       "epoch": 18.61,
-      "learning_rate": 3.850385038503851e-07,
-      "loss": 0.9796,
+      "learning_rate": 3.8320498716538324e-06,
+      "loss": 0.9866,
       "step": 3760
     },
     {
       "epoch": 18.66,
-      "learning_rate": 3.712871287128713e-07,
-      "loss": 1.0034,
+      "learning_rate": 3.827466079941328e-06,
+      "loss": 0.9492,
       "step": 3770
     },
     {
       "epoch": 18.71,
-      "learning_rate": 3.5753575357535757e-07,
-      "loss": 0.9773,
+      "learning_rate": 3.822882288228823e-06,
+      "loss": 0.9461,
       "step": 3780
     },
     {
       "epoch": 18.76,
-      "learning_rate": 3.4378437843784385e-07,
-      "loss": 0.984,
+      "learning_rate": 3.818298496516318e-06,
+      "loss": 0.9636,
       "step": 3790
     },
     {
       "epoch": 18.81,
-      "learning_rate": 3.3003300330033e-07,
-      "loss": 0.9513,
+      "learning_rate": 3.8137147048038143e-06,
+      "loss": 0.9631,
       "step": 3800
     },
     {
       "epoch": 18.86,
-      "learning_rate": 3.162816281628163e-07,
-      "loss": 0.9628,
+      "learning_rate": 3.8091309130913096e-06,
+      "loss": 1.0091,
       "step": 3810
     },
     {
       "epoch": 18.91,
-      "learning_rate": 3.025302530253026e-07,
-      "loss": 0.9573,
+      "learning_rate": 3.804547121378805e-06,
+      "loss": 0.9455,
       "step": 3820
     },
     {
       "epoch": 18.96,
-      "learning_rate": 2.887788778877888e-07,
-      "loss": 0.9803,
+      "learning_rate": 3.7999633296663e-06,
+      "loss": 0.9703,
       "step": 3830
     },
     {
       "epoch": 19.0,
-      "eval_accuracy": 0.6753744339951236,
-      "eval_loss": 0.8829284906387329,
-      "eval_runtime": 38.4857,
-      "eval_samples_per_second": 74.599,
-      "eval_steps_per_second": 2.339,
+      "eval_accuracy": 0.6826889585510275,
+      "eval_loss": 0.8537988662719727,
+      "eval_runtime": 37.6927,
+      "eval_samples_per_second": 76.169,
+      "eval_steps_per_second": 2.388,
       "step": 3838
     },
     {
       "epoch": 19.01,
-      "learning_rate": 2.7502750275027504e-07,
-      "loss": 0.9945,
+      "learning_rate": 3.7953795379537954e-06,
+      "loss": 0.9781,
       "step": 3840
     },
     {
       "epoch": 19.06,
-      "learning_rate": 2.612761276127613e-07,
-      "loss": 1.0255,
+      "learning_rate": 3.7907957462412907e-06,
+      "loss": 0.9335,
       "step": 3850
     },
     {
       "epoch": 19.11,
-      "learning_rate": 2.4752475247524754e-07,
-      "loss": 0.966,
+      "learning_rate": 3.786211954528787e-06,
+      "loss": 0.9222,
       "step": 3860
     },
     {
       "epoch": 19.16,
-      "learning_rate": 2.337733773377338e-07,
-      "loss": 0.9895,
+      "learning_rate": 3.781628162816282e-06,
+      "loss": 0.9688,
       "step": 3870
     },
     {
       "epoch": 19.21,
-      "learning_rate": 2.2002200220022005e-07,
-      "loss": 0.9694,
+      "learning_rate": 3.7770443711037773e-06,
+      "loss": 0.9208,
       "step": 3880
     },
     {
       "epoch": 19.26,
-      "learning_rate": 2.0627062706270628e-07,
-      "loss": 0.9622,
+      "learning_rate": 3.7724605793912726e-06,
+      "loss": 0.9563,
       "step": 3890
     },
     {
       "epoch": 19.31,
-      "learning_rate": 1.9251925192519256e-07,
-      "loss": 0.9951,
+      "learning_rate": 3.767876787678768e-06,
+      "loss": 0.9485,
       "step": 3900
     },
     {
       "epoch": 19.36,
-      "learning_rate": 1.7876787678767878e-07,
-      "loss": 0.9759,
+      "learning_rate": 3.7632929959662636e-06,
+      "loss": 0.9915,
       "step": 3910
     },
     {
       "epoch": 19.41,
-      "learning_rate": 1.65016501650165e-07,
-      "loss": 1.019,
+      "learning_rate": 3.7587092042537593e-06,
+      "loss": 0.95,
       "step": 3920
     },
     {
       "epoch": 19.46,
-      "learning_rate": 1.512651265126513e-07,
-      "loss": 1.0533,
+      "learning_rate": 3.7541254125412545e-06,
+      "loss": 0.9322,
       "step": 3930
     },
     {
       "epoch": 19.5,
-      "learning_rate": 1.3751375137513752e-07,
-      "loss": 0.9828,
+      "learning_rate": 3.74954162082875e-06,
+      "loss": 0.9313,
       "step": 3940
     },
     {
       "epoch": 19.55,
-      "learning_rate": 1.2376237623762377e-07,
-      "loss": 1.0073,
+      "learning_rate": 3.7449578291162455e-06,
+      "loss": 0.9666,
       "step": 3950
     },
     {
       "epoch": 19.6,
-      "learning_rate": 1.1001100110011003e-07,
-      "loss": 0.9639,
+      "learning_rate": 3.7403740374037408e-06,
+      "loss": 0.9548,
       "step": 3960
     },
     {
       "epoch": 19.65,
-      "learning_rate": 9.625962596259628e-08,
-      "loss": 0.9632,
+      "learning_rate": 3.735790245691236e-06,
+      "loss": 1.0175,
       "step": 3970
     },
     {
       "epoch": 19.7,
-      "learning_rate": 8.25082508250825e-08,
-      "loss": 0.9732,
+      "learning_rate": 3.7312064539787313e-06,
+      "loss": 0.923,
       "step": 3980
     },
     {
       "epoch": 19.75,
-      "learning_rate": 6.875687568756876e-08,
-      "loss": 0.9832,
+      "learning_rate": 3.726622662266227e-06,
+      "loss": 0.967,
       "step": 3990
     },
     {
       "epoch": 19.8,
-      "learning_rate": 5.500550055005501e-08,
-      "loss": 1.0138,
+      "learning_rate": 3.7220388705537227e-06,
+      "loss": 0.9704,
       "step": 4000
     },
     {
       "epoch": 19.85,
-      "learning_rate": 4.125412541254125e-08,
-      "loss": 0.9895,
+      "learning_rate": 3.717455078841218e-06,
+      "loss": 0.9529,
       "step": 4010
     },
     {
       "epoch": 19.9,
-      "learning_rate": 2.7502750275027506e-08,
-      "loss": 0.9244,
+      "learning_rate": 3.712871287128713e-06,
+      "loss": 0.9551,
       "step": 4020
     },
     {
       "epoch": 19.95,
-      "learning_rate": 1.3751375137513753e-08,
-      "loss": 0.9557,
+      "learning_rate": 3.7082874954162085e-06,
+      "loss": 0.9782,
       "step": 4030
     },
     {
       "epoch": 20.0,
-      "learning_rate": 0.0,
-      "loss": 0.9356,
+      "learning_rate": 3.7037037037037037e-06,
+      "loss": 0.9399,
       "step": 4040
     },
     {
       "epoch": 20.0,
-      "eval_accuracy": 0.6788575409265064,
-      "eval_loss": 0.8812496662139893,
-      "eval_runtime": 38.3816,
-      "eval_samples_per_second": 74.801,
-      "eval_steps_per_second": 2.345,
+      "eval_accuracy": 0.6875653082549634,
+      "eval_loss": 0.8451740145683289,
+      "eval_runtime": 37.8133,
+      "eval_samples_per_second": 75.926,
+      "eval_steps_per_second": 2.38,
       "step": 4040
     },
     {
-      "epoch": 20.0,
-      "step": 4040,
-      "total_flos": 4.0029439499828675e+19,
-      "train_loss": 1.115002474690428,
-      "train_runtime": 19507.0594,
-      "train_samples_per_second": 26.491,
-      "train_steps_per_second": 0.207
+      "epoch": 20.05,
+      "learning_rate": 3.6991199119912e-06,
+      "loss": 0.9642,
+      "step": 4050
+    },
+    {
+      "epoch": 20.1,
+      "learning_rate": 3.694536120278695e-06,
+      "loss": 0.9373,
+      "step": 4060
+    },
+    {
+      "epoch": 20.15,
+      "learning_rate": 3.6899523285661904e-06,
+      "loss": 0.9331,
+      "step": 4070
+    },
+    {
+      "epoch": 20.2,
+      "learning_rate": 3.6853685368536857e-06,
+      "loss": 1.0111,
+      "step": 4080
+    },
+    {
+      "epoch": 20.25,
+      "learning_rate": 3.680784745141181e-06,
+      "loss": 0.9586,
+      "step": 4090
+    },
+    {
+      "epoch": 20.3,
+      "learning_rate": 3.676200953428676e-06,
+      "loss": 0.935,
+      "step": 4100
+    },
+    {
+      "epoch": 20.35,
+      "learning_rate": 3.6716171617161715e-06,
+      "loss": 0.919,
+      "step": 4110
+    },
+    {
+      "epoch": 20.4,
+      "learning_rate": 3.6670333700036676e-06,
+      "loss": 0.9147,
+      "step": 4120
+    },
+    {
+      "epoch": 20.45,
+      "learning_rate": 3.662449578291163e-06,
+      "loss": 0.8863,
+      "step": 4130
+    },
+    {
+      "epoch": 20.5,
+      "learning_rate": 3.657865786578658e-06,
+      "loss": 0.9427,
+      "step": 4140
+    },
+    {
+      "epoch": 20.54,
+      "learning_rate": 3.6532819948661534e-06,
+      "loss": 0.9168,
+      "step": 4150
+    },
+    {
+      "epoch": 20.59,
+      "learning_rate": 3.648698203153649e-06,
+      "loss": 0.9444,
+      "step": 4160
+    },
+    {
+      "epoch": 20.64,
+      "learning_rate": 3.6441144114411443e-06,
+      "loss": 0.9332,
+      "step": 4170
+    },
+    {
+      "epoch": 20.69,
+      "learning_rate": 3.63953061972864e-06,
+      "loss": 0.9046,
+      "step": 4180
+    },
+    {
+      "epoch": 20.74,
+      "learning_rate": 3.6349468280161353e-06,
+      "loss": 0.9499,
+      "step": 4190
+    },
+    {
+      "epoch": 20.79,
+      "learning_rate": 3.6303630363036306e-06,
+      "loss": 0.97,
+      "step": 4200
+    },
+    {
+      "epoch": 20.84,
+      "learning_rate": 3.6257792445911263e-06,
+      "loss": 0.9953,
+      "step": 4210
+    },
+    {
+      "epoch": 20.89,
+      "learning_rate": 3.6211954528786215e-06,
+      "loss": 0.9649,
+      "step": 4220
+    },
+    {
+      "epoch": 20.94,
+      "learning_rate": 3.616611661166117e-06,
+      "loss": 0.9314,
+      "step": 4230
+    },
+    {
+      "epoch": 20.99,
+      "learning_rate": 3.612027869453612e-06,
+      "loss": 0.9299,
+      "step": 4240
+    },
+    {
+      "epoch": 21.0,
+      "eval_accuracy": 0.6903517938000696,
+      "eval_loss": 0.8419761061668396,
+      "eval_runtime": 37.462,
+      "eval_samples_per_second": 76.638,
+      "eval_steps_per_second": 2.402,
+      "step": 4242
+    },
+    {
+      "epoch": 21.04,
+      "learning_rate": 3.6074440777411078e-06,
+      "loss": 0.9489,
+      "step": 4250
+    },
+    {
+      "epoch": 21.09,
+      "learning_rate": 3.6028602860286034e-06,
+      "loss": 0.8842,
+      "step": 4260
+    },
+    {
+      "epoch": 21.14,
+      "learning_rate": 3.5982764943160987e-06,
+      "loss": 0.9134,
+      "step": 4270
+    },
+    {
+      "epoch": 21.19,
+      "learning_rate": 3.593692702603594e-06,
+      "loss": 0.9087,
+      "step": 4280
+    },
+    {
+      "epoch": 21.24,
+      "learning_rate": 3.5891089108910892e-06,
+      "loss": 0.9688,
+      "step": 4290
+    },
+    {
+      "epoch": 21.29,
+      "learning_rate": 3.5845251191785845e-06,
+      "loss": 0.9176,
+      "step": 4300
+    },
+    {
+      "epoch": 21.34,
+      "learning_rate": 3.5799413274660806e-06,
+      "loss": 0.8795,
+      "step": 4310
+    },
+    {
+      "epoch": 21.39,
+      "learning_rate": 3.575357535753576e-06,
+      "loss": 0.9479,
+      "step": 4320
+    },
+    {
+      "epoch": 21.44,
+      "learning_rate": 3.570773744041071e-06,
+      "loss": 0.9494,
+      "step": 4330
+    },
+    {
+      "epoch": 21.49,
+      "learning_rate": 3.5661899523285664e-06,
+      "loss": 0.9834,
+      "step": 4340
+    },
+    {
+      "epoch": 21.53,
+      "learning_rate": 3.5616061606160617e-06,
+      "loss": 0.9358,
+      "step": 4350
+    },
+    {
+      "epoch": 21.58,
+      "learning_rate": 3.557022368903557e-06,
+      "loss": 0.9179,
+      "step": 4360
+    },
+    {
+      "epoch": 21.63,
+      "learning_rate": 3.552438577191053e-06,
+      "loss": 0.9287,
+      "step": 4370
+    },
+    {
+      "epoch": 21.68,
+      "learning_rate": 3.5478547854785484e-06,
+      "loss": 0.941,
+      "step": 4380
+    },
+    {
+      "epoch": 21.73,
+      "learning_rate": 3.5432709937660436e-06,
+      "loss": 0.9228,
+      "step": 4390
+    },
+    {
+      "epoch": 21.78,
+      "learning_rate": 3.538687202053539e-06,
+      "loss": 0.99,
+      "step": 4400
+    },
+    {
+      "epoch": 21.83,
+      "learning_rate": 3.534103410341034e-06,
+      "loss": 0.9405,
+      "step": 4410
+    },
+    {
+      "epoch": 21.88,
+      "learning_rate": 3.52951961862853e-06,
+      "loss": 0.9118,
+      "step": 4420
+    },
+    {
+      "epoch": 21.93,
+      "learning_rate": 3.524935826916025e-06,
+      "loss": 0.9547,
+      "step": 4430
+    },
+    {
+      "epoch": 21.98,
+      "learning_rate": 3.520352035203521e-06,
+      "loss": 0.9815,
+      "step": 4440
+    },
+    {
+      "epoch": 22.0,
+      "eval_accuracy": 0.6872169975618252,
+      "eval_loss": 0.8417115211486816,
+      "eval_runtime": 37.6008,
+      "eval_samples_per_second": 76.355,
+      "eval_steps_per_second": 2.394,
+      "step": 4444
+    },
+    {
+      "epoch": 22.03,
+      "learning_rate": 3.515768243491016e-06,
+      "loss": 0.9513,
+      "step": 4450
+    },
+    {
+      "epoch": 22.08,
+      "learning_rate": 3.5111844517785113e-06,
+      "loss": 0.9174,
+      "step": 4460
+    },
+    {
+      "epoch": 22.13,
+      "learning_rate": 3.506600660066007e-06,
+      "loss": 0.9308,
+      "step": 4470
+    },
+    {
+      "epoch": 22.18,
+      "learning_rate": 3.5020168683535023e-06,
+      "loss": 0.8863,
+      "step": 4480
+    },
+    {
+      "epoch": 22.23,
+      "learning_rate": 3.4974330766409976e-06,
+      "loss": 0.9378,
+      "step": 4490
+    },
+    {
+      "epoch": 22.28,
+      "learning_rate": 3.4928492849284933e-06,
+      "loss": 0.9508,
+      "step": 4500
+    },
+    {
+      "epoch": 22.33,
+      "learning_rate": 3.4882654932159885e-06,
+      "loss": 0.935,
+      "step": 4510
+    },
+    {
+      "epoch": 22.38,
+      "learning_rate": 3.4836817015034842e-06,
+      "loss": 0.8982,
+      "step": 4520
+    },
+    {
+      "epoch": 22.43,
+      "learning_rate": 3.4790979097909795e-06,
+      "loss": 0.9364,
+      "step": 4530
+    },
+    {
+      "epoch": 22.48,
+      "learning_rate": 3.4745141180784748e-06,
+      "loss": 0.9365,
+      "step": 4540
+    },
+    {
+      "epoch": 22.52,
+      "learning_rate": 3.46993032636597e-06,
+      "loss": 0.9348,
+      "step": 4550
+    },
+    {
+      "epoch": 22.57,
+      "learning_rate": 3.4653465346534653e-06,
+      "loss": 0.9349,
+      "step": 4560
+    },
+    {
+      "epoch": 22.62,
+      "learning_rate": 3.4607627429409614e-06,
+      "loss": 0.9728,
+      "step": 4570
+    },
+    {
+      "epoch": 22.67,
+      "learning_rate": 3.4561789512284567e-06,
+      "loss": 0.926,
+      "step": 4580
+    },
+    {
+      "epoch": 22.72,
+      "learning_rate": 3.451595159515952e-06,
+      "loss": 0.9449,
+      "step": 4590
+    },
+    {
+      "epoch": 22.77,
+      "learning_rate": 3.447011367803447e-06,
+      "loss": 0.9512,
+      "step": 4600
+    },
+    {
+      "epoch": 22.82,
+      "learning_rate": 3.4424275760909425e-06,
+      "loss": 0.9102,
+      "step": 4610
+    },
+    {
+      "epoch": 22.87,
+      "learning_rate": 3.4378437843784377e-06,
+      "loss": 0.9333,
+      "step": 4620
+    },
+    {
+      "epoch": 22.92,
+      "learning_rate": 3.433259992665934e-06,
+      "loss": 0.9412,
+      "step": 4630
+    },
+    {
+      "epoch": 22.97,
+      "learning_rate": 3.428676200953429e-06,
+      "loss": 0.9029,
+      "step": 4640
+    },
+    {
+      "epoch": 23.0,
+      "eval_accuracy": 0.6900034831069314,
+      "eval_loss": 0.8379387855529785,
+      "eval_runtime": 37.5549,
+      "eval_samples_per_second": 76.448,
+      "eval_steps_per_second": 2.396,
+      "step": 4646
+    },
+    {
+      "epoch": 23.02,
+      "learning_rate": 3.4240924092409244e-06,
+      "loss": 0.9224,
+      "step": 4650
+    },
+    {
+      "epoch": 23.07,
+      "learning_rate": 3.4195086175284197e-06,
+      "loss": 0.9421,
+      "step": 4660
+    },
+    {
+      "epoch": 23.12,
+      "learning_rate": 3.414924825815915e-06,
+      "loss": 0.8946,
+      "step": 4670
+    },
+    {
+      "epoch": 23.17,
+      "learning_rate": 3.4103410341034106e-06,
+      "loss": 0.92,
+      "step": 4680
+    },
+    {
+      "epoch": 23.22,
+      "learning_rate": 3.405757242390906e-06,
+      "loss": 0.9069,
+      "step": 4690
+    },
+    {
+      "epoch": 23.27,
+      "learning_rate": 3.4011734506784016e-06,
+      "loss": 0.9002,
+      "step": 4700
+    },
+    {
+      "epoch": 23.32,
+      "learning_rate": 3.396589658965897e-06,
+      "loss": 0.9086,
+      "step": 4710
+    },
+    {
+      "epoch": 23.37,
+      "learning_rate": 3.392005867253392e-06,
+      "loss": 0.9369,
+      "step": 4720
+    },
+    {
+      "epoch": 23.42,
+      "learning_rate": 3.387422075540888e-06,
+      "loss": 0.9632,
+      "step": 4730
+    },
+    {
+      "epoch": 23.47,
+      "learning_rate": 3.382838283828383e-06,
+      "loss": 0.9147,
+      "step": 4740
+    },
+    {
+      "epoch": 23.51,
+      "learning_rate": 3.3782544921158783e-06,
+      "loss": 0.9129,
+      "step": 4750
+    },
+    {
+      "epoch": 23.56,
+      "learning_rate": 3.373670700403374e-06,
+      "loss": 0.9556,
+      "step": 4760
+    },
+    {
+      "epoch": 23.61,
+      "learning_rate": 3.3690869086908697e-06,
+      "loss": 0.912,
+      "step": 4770
+    },
+    {
+      "epoch": 23.66,
+      "learning_rate": 3.364503116978365e-06,
+      "loss": 0.9415,
+      "step": 4780
+    },
+    {
+      "epoch": 23.71,
+      "learning_rate": 3.3599193252658603e-06,
+      "loss": 0.9376,
+      "step": 4790
+    },
+    {
+      "epoch": 23.76,
+      "learning_rate": 3.3553355335533555e-06,
+      "loss": 0.9602,
+      "step": 4800
+    },
+    {
+      "epoch": 23.81,
+      "learning_rate": 3.350751741840851e-06,
+      "loss": 0.9566,
+      "step": 4810
+    },
+    {
+      "epoch": 23.86,
+      "learning_rate": 3.346167950128346e-06,
+      "loss": 0.8946,
+      "step": 4820
+    },
+    {
+      "epoch": 23.91,
+      "learning_rate": 3.341584158415842e-06,
+      "loss": 0.8979,
+      "step": 4830
+    },
+    {
+      "epoch": 23.96,
+      "learning_rate": 3.3370003667033374e-06,
+      "loss": 0.9142,
+      "step": 4840
+    },
+    {
+      "epoch": 24.0,
+      "eval_accuracy": 0.6896551724137931,
+      "eval_loss": 0.8336252570152283,
+      "eval_runtime": 37.6833,
+      "eval_samples_per_second": 76.188,
+      "eval_steps_per_second": 2.388,
+      "step": 4848
+    },
+    {
+      "epoch": 24.01,
+      "learning_rate": 3.3324165749908327e-06,
+      "loss": 0.8926,
+      "step": 4850
+    },
+    {
+      "epoch": 24.06,
+      "learning_rate": 3.327832783278328e-06,
+      "loss": 0.9298,
+      "step": 4860
+    },
+    {
+      "epoch": 24.11,
+      "learning_rate": 3.3232489915658233e-06,
+      "loss": 0.9186,
+      "step": 4870
+    },
+    {
+      "epoch": 24.16,
+      "learning_rate": 3.3186651998533185e-06,
+      "loss": 0.9284,
+      "step": 4880
+    },
+    {
+      "epoch": 24.21,
+      "learning_rate": 3.3140814081408146e-06,
+      "loss": 0.8984,
+      "step": 4890
+    },
+    {
+      "epoch": 24.26,
+      "learning_rate": 3.30949761642831e-06,
+      "loss": 0.9123,
+      "step": 4900
+    },
+    {
+      "epoch": 24.31,
+      "learning_rate": 3.304913824715805e-06,
+      "loss": 0.9153,
+      "step": 4910
+    },
+    {
+      "epoch": 24.36,
+      "learning_rate": 3.3003300330033004e-06,
+      "loss": 0.9091,
+      "step": 4920
+    },
+    {
+      "epoch": 24.41,
+      "learning_rate": 3.2957462412907957e-06,
+      "loss": 0.9014,
+      "step": 4930
+    },
+    {
+      "epoch": 24.46,
+      "learning_rate": 3.2911624495782914e-06,
+      "loss": 0.8522,
+      "step": 4940
+    },
+    {
+      "epoch": 24.5,
+      "learning_rate": 3.286578657865787e-06,
+      "loss": 0.8821,
+      "step": 4950
+    },
+    {
+      "epoch": 24.55,
+      "learning_rate": 3.2819948661532824e-06,
+      "loss": 0.9143,
+      "step": 4960
+    },
+    {
+      "epoch": 24.6,
+      "learning_rate": 3.2774110744407776e-06,
+      "loss": 0.9349,
+      "step": 4970
+    },
+    {
+      "epoch": 24.65,
+      "learning_rate": 3.272827282728273e-06,
+      "loss": 0.8848,
+      "step": 4980
+    },
+    {
+      "epoch": 24.7,
+      "learning_rate": 3.2682434910157686e-06,
+      "loss": 0.9585,
+      "step": 4990
+    },
+    {
+      "epoch": 24.75,
+      "learning_rate": 3.263659699303264e-06,
+      "loss": 0.9204,
+      "step": 5000
+    },
+    {
+      "epoch": 24.8,
+      "learning_rate": 3.259075907590759e-06,
+      "loss": 0.9144,
+      "step": 5010
+    },
+    {
+      "epoch": 24.85,
+      "learning_rate": 3.254492115878255e-06,
+      "loss": 0.9504,
+      "step": 5020
+    },
+    {
+      "epoch": 24.9,
+      "learning_rate": 3.2499083241657505e-06,
+      "loss": 0.9342,
+      "step": 5030
+    },
+    {
+      "epoch": 24.95,
+      "learning_rate": 3.2453245324532458e-06,
+      "loss": 0.8876,
+      "step": 5040
+    },
+    {
+      "epoch": 25.0,
+      "learning_rate": 3.240740740740741e-06,
+      "loss": 0.8695,
+      "step": 5050
+    },
+    {
+      "epoch": 25.0,
+      "eval_accuracy": 0.6938349007314525,
+      "eval_loss": 0.8311742544174194,
+      "eval_runtime": 37.7877,
+      "eval_samples_per_second": 75.977,
+      "eval_steps_per_second": 2.382,
+      "step": 5050
+    },
+    {
+      "epoch": 25.05,
+      "learning_rate": 3.2361569490282363e-06,
+      "loss": 0.9352,
+      "step": 5060
+    },
+    {
+      "epoch": 25.1,
+      "learning_rate": 3.2315731573157316e-06,
+      "loss": 0.9414,
+      "step": 5070
+    },
+    {
+      "epoch": 25.15,
+      "learning_rate": 3.2269893656032277e-06,
+      "loss": 0.8946,
+      "step": 5080
+    },
+    {
+      "epoch": 25.2,
+      "learning_rate": 3.222405573890723e-06,
+      "loss": 0.9137,
+      "step": 5090
+    },
+    {
+      "epoch": 25.25,
+      "learning_rate": 3.2178217821782182e-06,
+      "loss": 0.9131,
+      "step": 5100
+    },
+    {
+      "epoch": 25.3,
+      "learning_rate": 3.2132379904657135e-06,
+      "loss": 0.8882,
+      "step": 5110
+    },
+    {
+      "epoch": 25.35,
+      "learning_rate": 3.2086541987532088e-06,
+      "loss": 0.924,
+      "step": 5120
+    },
+    {
+      "epoch": 25.4,
+      "learning_rate": 3.204070407040704e-06,
+      "loss": 0.913,
+      "step": 5130
+    },
+    {
+      "epoch": 25.45,
+      "learning_rate": 3.1994866153281993e-06,
+      "loss": 0.9007,
+      "step": 5140
+    },
+    {
+      "epoch": 25.5,
+      "learning_rate": 3.1949028236156954e-06,
+      "loss": 0.9319,
+      "step": 5150
+    },
+    {
+      "epoch": 25.54,
+      "learning_rate": 3.1903190319031907e-06,
+      "loss": 0.9237,
+      "step": 5160
+    },
+    {
+      "epoch": 25.59,
+      "learning_rate": 3.185735240190686e-06,
+      "loss": 0.9605,
+      "step": 5170
+    },
+    {
+      "epoch": 25.64,
+      "learning_rate": 3.1811514484781812e-06,
+      "loss": 0.912,
+      "step": 5180
+    },
+    {
+      "epoch": 25.69,
+      "learning_rate": 3.1765676567656765e-06,
+      "loss": 0.9243,
+      "step": 5190
+    },
+    {
+      "epoch": 25.74,
+      "learning_rate": 3.171983865053172e-06,
+      "loss": 0.8899,
+      "step": 5200
+    },
+    {
+      "epoch": 25.79,
+      "learning_rate": 3.167400073340668e-06,
+      "loss": 0.9175,
+      "step": 5210
+    },
+    {
+      "epoch": 25.84,
+      "learning_rate": 3.162816281628163e-06,
+      "loss": 0.8995,
+      "step": 5220
+    },
+    {
+      "epoch": 25.89,
+      "learning_rate": 3.1582324899156584e-06,
+      "loss": 0.9025,
+      "step": 5230
+    },
+    {
+      "epoch": 25.94,
+      "learning_rate": 3.153648698203154e-06,
+      "loss": 0.8972,
+      "step": 5240
+    },
+    {
+      "epoch": 25.99,
+      "learning_rate": 3.1490649064906494e-06,
+      "loss": 0.8791,
+      "step": 5250
+    },
+    {
+      "epoch": 26.0,
+      "eval_accuracy": 0.6941832114245907,
+      "eval_loss": 0.832297146320343,
+      "eval_runtime": 37.7358,
+      "eval_samples_per_second": 76.082,
+      "eval_steps_per_second": 2.385,
+      "step": 5252
+    },
+    {
+      "epoch": 26.04,
+      "learning_rate": 3.1444811147781446e-06,
+      "loss": 0.8714,
+      "step": 5260
+    },
+    {
+      "epoch": 26.09,
+      "learning_rate": 3.13989732306564e-06,
+      "loss": 0.9245,
+      "step": 5270
+    },
+    {
+      "epoch": 26.14,
+      "learning_rate": 3.1353135313531356e-06,
+      "loss": 0.8888,
+      "step": 5280
+    },
+    {
+      "epoch": 26.19,
+      "learning_rate": 3.1307297396406313e-06,
+      "loss": 0.9172,
+      "step": 5290
+    },
+    {
+      "epoch": 26.24,
+      "learning_rate": 3.1261459479281265e-06,
+      "loss": 0.8627,
+      "step": 5300
+    },
+    {
+      "epoch": 26.29,
+      "learning_rate": 3.121562156215622e-06,
+      "loss": 0.8754,
+      "step": 5310
+    },
+    {
+      "epoch": 26.34,
+      "learning_rate": 3.116978364503117e-06,
+      "loss": 0.9562,
+      "step": 5320
+    },
+    {
+      "epoch": 26.39,
+      "learning_rate": 3.1123945727906123e-06,
+      "loss": 0.9458,
+      "step": 5330
+    },
+    {
+      "epoch": 26.44,
+      "learning_rate": 3.1078107810781085e-06,
+      "loss": 0.879,
+      "step": 5340
+    },
+    {
+      "epoch": 26.49,
+      "learning_rate": 3.1032269893656037e-06,
+      "loss": 0.8834,
+      "step": 5350
+    },
+    {
+      "epoch": 26.53,
+      "learning_rate": 3.098643197653099e-06,
+      "loss": 0.9298,
+      "step": 5360
+    },
+    {
+      "epoch": 26.58,
+      "learning_rate": 3.0940594059405943e-06,
+      "loss": 0.9169,
+      "step": 5370
+    },
+    {
+      "epoch": 26.63,
+      "learning_rate": 3.0894756142280895e-06,
+      "loss": 0.9178,
+      "step": 5380
+    },
+    {
+      "epoch": 26.68,
+      "learning_rate": 3.084891822515585e-06,
+      "loss": 0.8689,
+      "step": 5390
+    },
+    {
+      "epoch": 26.73,
+      "learning_rate": 3.080308030803081e-06,
+      "loss": 0.9095,
+      "step": 5400
+    },
+    {
+      "epoch": 26.78,
+      "learning_rate": 3.075724239090576e-06,
+      "loss": 0.8799,
+      "step": 5410
+    },
+    {
+      "epoch": 26.83,
+      "learning_rate": 3.0711404473780715e-06,
+      "loss": 0.9089,
+      "step": 5420
+    },
+    {
+      "epoch": 26.88,
+      "learning_rate": 3.0665566556655667e-06,
+      "loss": 0.9122,
+      "step": 5430
+    },
+    {
+      "epoch": 26.93,
+      "learning_rate": 3.061972863953062e-06,
+      "loss": 0.8795,
+      "step": 5440
+    },
+    {
+      "epoch": 26.98,
+      "learning_rate": 3.0573890722405577e-06,
+      "loss": 0.923,
+      "step": 5450
+    },
+    {
+      "epoch": 27.0,
+      "eval_accuracy": 0.6955764541971439,
+      "eval_loss": 0.8244138360023499,
+      "eval_runtime": 37.7662,
+      "eval_samples_per_second": 76.02,
+      "eval_steps_per_second": 2.383,
+      "step": 5454
+    },
+    {
+      "epoch": 27.03,
+      "learning_rate": 3.052805280528053e-06,
+      "loss": 0.9176,
+      "step": 5460
+    },
+    {
+      "epoch": 27.08,
+      "learning_rate": 3.0482214888155486e-06,
+      "loss": 0.8875,
+      "step": 5470
+    },
+    {
+      "epoch": 27.13,
+      "learning_rate": 3.043637697103044e-06,
+      "loss": 0.9124,
+      "step": 5480
+    },
+    {
+      "epoch": 27.18,
+      "learning_rate": 3.039053905390539e-06,
+      "loss": 0.8766,
+      "step": 5490
+    },
+    {
+      "epoch": 27.23,
+      "learning_rate": 3.034470113678035e-06,
+      "loss": 0.9123,
+      "step": 5500
+    },
+    {
+      "epoch": 27.28,
+      "learning_rate": 3.02988632196553e-06,
+      "loss": 0.9254,
+      "step": 5510
+    },
+    {
+      "epoch": 27.33,
+      "learning_rate": 3.0253025302530254e-06,
+      "loss": 0.9415,
+      "step": 5520
+    },
+    {
+      "epoch": 27.38,
+      "learning_rate": 3.020718738540521e-06,
+      "loss": 0.8788,
+      "step": 5530
+    },
+    {
+      "epoch": 27.43,
+      "learning_rate": 3.0161349468280164e-06,
+      "loss": 0.875,
+      "step": 5540
+    },
+    {
+      "epoch": 27.48,
+      "learning_rate": 3.011551155115512e-06,
+      "loss": 0.9281,
+      "step": 5550
+    },
+    {
+      "epoch": 27.52,
+      "learning_rate": 3.0069673634030073e-06,
+      "loss": 0.9061,
+      "step": 5560
+    },
+    {
+      "epoch": 27.57,
+      "learning_rate": 3.0023835716905026e-06,
+      "loss": 0.856,
+      "step": 5570
+    },
+    {
+      "epoch": 27.62,
+      "learning_rate": 2.997799779977998e-06,
+      "loss": 0.921,
+      "step": 5580
+    },
+    {
+      "epoch": 27.67,
+      "learning_rate": 2.993215988265493e-06,
+      "loss": 0.8996,
+      "step": 5590
+    },
+    {
+      "epoch": 27.72,
+      "learning_rate": 2.9886321965529892e-06,
+      "loss": 0.8877,
+      "step": 5600
+    },
+    {
+      "epoch": 27.77,
+      "learning_rate": 2.9840484048404845e-06,
+      "loss": 0.9071,
+      "step": 5610
+    },
+    {
+      "epoch": 27.82,
+      "learning_rate": 2.9794646131279798e-06,
+      "loss": 0.92,
+      "step": 5620
+    },
+    {
+      "epoch": 27.87,
+      "learning_rate": 2.974880821415475e-06,
+      "loss": 0.8611,
+      "step": 5630
+    },
+    {
+      "epoch": 27.92,
+      "learning_rate": 2.9702970297029703e-06,
+      "loss": 0.8682,
+      "step": 5640
+    },
+    {
+      "epoch": 27.97,
+      "learning_rate": 2.9657132379904656e-06,
+      "loss": 0.8866,
+      "step": 5650
+    },
+    {
+      "epoch": 28.0,
+      "eval_accuracy": 0.696969696969697,
+      "eval_loss": 0.8260791897773743,
+      "eval_runtime": 37.6877,
+      "eval_samples_per_second": 76.179,
+      "eval_steps_per_second": 2.388,
+      "step": 5656
+    },
+    {
+      "epoch": 28.02,
+      "learning_rate": 2.9611294462779617e-06,
+      "loss": 0.9302,
+      "step": 5660
+    },
+    {
+      "epoch": 28.07,
+      "learning_rate": 2.956545654565457e-06,
+      "loss": 0.9175,
+      "step": 5670
+    },
+    {
+      "epoch": 28.12,
+      "learning_rate": 2.9519618628529522e-06,
+      "loss": 0.8833,
+      "step": 5680
+    },
+    {
+      "epoch": 28.17,
+      "learning_rate": 2.9473780711404475e-06,
+      "loss": 0.9456,
+      "step": 5690
+    },
+    {
+      "epoch": 28.22,
+      "learning_rate": 2.9427942794279428e-06,
+      "loss": 0.8942,
+      "step": 5700
+    },
+    {
+      "epoch": 28.27,
+      "learning_rate": 2.9382104877154385e-06,
+      "loss": 0.8874,
+      "step": 5710
+    },
+    {
+      "epoch": 28.32,
+      "learning_rate": 2.9336266960029337e-06,
+      "loss": 0.9504,
+      "step": 5720
+    },
+    {
+      "epoch": 28.37,
+      "learning_rate": 2.9290429042904294e-06,
+      "loss": 0.8749,
+      "step": 5730
+    },
+    {
+      "epoch": 28.42,
+      "learning_rate": 2.9244591125779247e-06,
+      "loss": 0.9183,
+      "step": 5740
+    },
+    {
+      "epoch": 28.47,
+      "learning_rate": 2.91987532086542e-06,
+      "loss": 0.9061,
+      "step": 5750
+    },
+    {
+      "epoch": 28.51,
+      "learning_rate": 2.9152915291529156e-06,
+      "loss": 0.916,
+      "step": 5760
+    },
+    {
+      "epoch": 28.56,
+      "learning_rate": 2.910707737440411e-06,
+      "loss": 0.8612,
+      "step": 5770
+    },
+    {
+      "epoch": 28.61,
+      "learning_rate": 2.906123945727906e-06,
+      "loss": 0.9104,
+      "step": 5780
+    },
+    {
+      "epoch": 28.66,
+      "learning_rate": 2.901540154015402e-06,
+      "loss": 0.8735,
+      "step": 5790
+    },
+    {
+      "epoch": 28.71,
+      "learning_rate": 2.896956362302897e-06,
+      "loss": 0.8609,
+      "step": 5800
+    },
+    {
+      "epoch": 28.76,
+      "learning_rate": 2.892372570590393e-06,
+      "loss": 0.8393,
+      "step": 5810
+    },
+    {
+      "epoch": 28.81,
+      "learning_rate": 2.887788778877888e-06,
+      "loss": 0.8938,
+      "step": 5820
+    },
+    {
+      "epoch": 28.86,
+      "learning_rate": 2.8832049871653834e-06,
+      "loss": 0.8842,
+      "step": 5830
+    },
+    {
+      "epoch": 28.91,
+      "learning_rate": 2.8786211954528786e-06,
+      "loss": 0.9184,
+      "step": 5840
+    },
+    {
+      "epoch": 28.96,
+      "learning_rate": 2.8740374037403747e-06,
+      "loss": 0.9319,
+      "step": 5850
+    },
+    {
+      "epoch": 29.0,
+      "eval_accuracy": 0.6990595611285266,
+      "eval_loss": 0.8254513740539551,
+      "eval_runtime": 37.8605,
+      "eval_samples_per_second": 75.831,
+      "eval_steps_per_second": 2.377,
+      "step": 5858
+    },
+    {
+      "epoch": 29.01,
+      "learning_rate": 2.86945361202787e-06,
+      "loss": 0.8666,
+      "step": 5860
+    },
+    {
+      "epoch": 29.06,
+      "learning_rate": 2.8648698203153653e-06,
+      "loss": 0.8971,
+      "step": 5870
+    },
+    {
+      "epoch": 29.11,
+      "learning_rate": 2.8602860286028605e-06,
+      "loss": 0.8854,
+      "step": 5880
+    },
+    {
+      "epoch": 29.16,
+      "learning_rate": 2.855702236890356e-06,
+      "loss": 0.8986,
+      "step": 5890
+    },
+    {
+      "epoch": 29.21,
+      "learning_rate": 2.851118445177851e-06,
+      "loss": 0.8653,
+      "step": 5900
+    },
+    {
+      "epoch": 29.26,
+      "learning_rate": 2.8465346534653464e-06,
+      "loss": 0.8694,
+      "step": 5910
+    },
+    {
+      "epoch": 29.31,
+      "learning_rate": 2.8419508617528425e-06,
+      "loss": 0.8983,
+      "step": 5920
+    },
+    {
+      "epoch": 29.36,
+      "learning_rate": 2.8373670700403377e-06,
+      "loss": 0.8863,
+      "step": 5930
+    },
+    {
+      "epoch": 29.41,
+      "learning_rate": 2.832783278327833e-06,
+      "loss": 0.8887,
+      "step": 5940
+    },
+    {
+      "epoch": 29.46,
+      "learning_rate": 2.8281994866153283e-06,
+      "loss": 0.8863,
+      "step": 5950
+    },
+    {
+      "epoch": 29.5,
+      "learning_rate": 2.8236156949028235e-06,
+      "loss": 0.9107,
+      "step": 5960
+    },
+    {
+      "epoch": 29.55,
+      "learning_rate": 2.8190319031903192e-06,
+      "loss": 0.9022,
+      "step": 5970
+    },
+    {
+      "epoch": 29.6,
+      "learning_rate": 2.814448111477815e-06,
+      "loss": 0.8382,
+      "step": 5980
+    },
+    {
+      "epoch": 29.65,
+      "learning_rate": 2.80986431976531e-06,
+      "loss": 0.8621,
+      "step": 5990
+    },
+    {
+      "epoch": 29.7,
+      "learning_rate": 2.8052805280528055e-06,
+      "loss": 0.9332,
+      "step": 6000
+    },
+    {
+      "epoch": 29.75,
+      "learning_rate": 2.8006967363403007e-06,
+      "loss": 0.8981,
+      "step": 6010
+    },
+    {
+      "epoch": 29.8,
+      "learning_rate": 2.7961129446277964e-06,
+      "loss": 0.8737,
+      "step": 6020
+    },
+    {
+      "epoch": 29.85,
+      "learning_rate": 2.7915291529152917e-06,
+      "loss": 0.8785,
+      "step": 6030
+    },
+    {
+      "epoch": 29.9,
+      "learning_rate": 2.786945361202787e-06,
+      "loss": 0.8908,
+      "step": 6040
+    },
+    {
+      "epoch": 29.95,
+      "learning_rate": 2.7823615694902826e-06,
+      "loss": 0.89,
+      "step": 6050
+    },
+    {
+      "epoch": 30.0,
+      "learning_rate": 2.7777777777777783e-06,
+      "loss": 0.9019,
+      "step": 6060
+    },
+    {
+      "epoch": 30.0,
+      "eval_accuracy": 0.7049808429118773,
+      "eval_loss": 0.8159587979316711,
+      "eval_runtime": 37.7241,
+      "eval_samples_per_second": 76.105,
+      "eval_steps_per_second": 2.386,
+      "step": 6060
+    },
+    {
+      "epoch": 30.05,
+      "learning_rate": 2.7731939860652736e-06,
+      "loss": 0.883,
+      "step": 6070
+    },
+    {
+      "epoch": 30.1,
+      "learning_rate": 2.768610194352769e-06,
+      "loss": 0.8574,
+      "step": 6080
+    },
+    {
+      "epoch": 30.15,
+      "learning_rate": 2.764026402640264e-06,
+      "loss": 0.824,
+      "step": 6090
+    },
+    {
+      "epoch": 30.2,
+      "learning_rate": 2.7594426109277594e-06,
+      "loss": 0.9086,
+      "step": 6100
+    },
+    {
+      "epoch": 30.25,
+      "learning_rate": 2.7548588192152555e-06,
+      "loss": 0.8721,
+      "step": 6110
+    },
+    {
+      "epoch": 30.3,
+      "learning_rate": 2.750275027502751e-06,
+      "loss": 0.8944,
+      "step": 6120
+    },
+    {
+      "epoch": 30.35,
+      "learning_rate": 2.745691235790246e-06,
+      "loss": 0.9499,
+      "step": 6130
+    },
+    {
+      "epoch": 30.4,
+      "learning_rate": 2.7411074440777413e-06,
+      "loss": 0.9045,
+      "step": 6140
+    },
+    {
+      "epoch": 30.45,
+      "learning_rate": 2.7365236523652366e-06,
+      "loss": 0.9083,
+      "step": 6150
+    },
+    {
+      "epoch": 30.5,
+      "learning_rate": 2.731939860652732e-06,
+      "loss": 0.8615,
+      "step": 6160
+    },
+    {
+      "epoch": 30.54,
+      "learning_rate": 2.727356068940227e-06,
+      "loss": 0.8891,
+      "step": 6170
+    },
+    {
+      "epoch": 30.59,
+      "learning_rate": 2.7227722772277232e-06,
+      "loss": 0.9233,
+      "step": 6180
+    },
+    {
+      "epoch": 30.64,
+      "learning_rate": 2.7181884855152185e-06,
+      "loss": 0.8996,
+      "step": 6190
+    },
+    {
+      "epoch": 30.69,
+      "learning_rate": 2.7136046938027138e-06,
+      "loss": 0.8747,
+      "step": 6200
+    },
+    {
+      "epoch": 30.74,
+      "learning_rate": 2.709020902090209e-06,
+      "loss": 0.8759,
+      "step": 6210
+    },
+    {
+      "epoch": 30.79,
+      "learning_rate": 2.7044371103777043e-06,
+      "loss": 0.9016,
+      "step": 6220
+    },
+    {
+      "epoch": 30.84,
+      "learning_rate": 2.6998533186652e-06,
+      "loss": 0.8735,
+      "step": 6230
+    },
+    {
+      "epoch": 30.89,
+      "learning_rate": 2.6952695269526957e-06,
+      "loss": 0.8563,
+      "step": 6240
+    },
+    {
+      "epoch": 30.94,
+      "learning_rate": 2.690685735240191e-06,
+      "loss": 0.9059,
+      "step": 6250
+    },
+    {
+      "epoch": 30.99,
+      "learning_rate": 2.6861019435276862e-06,
+      "loss": 0.8785,
+      "step": 6260
+    },
+    {
+      "epoch": 31.0,
+      "eval_accuracy": 0.7070707070707071,
+      "eval_loss": 0.8169212341308594,
+      "eval_runtime": 37.7132,
+      "eval_samples_per_second": 76.127,
+      "eval_steps_per_second": 2.386,
+      "step": 6262
+    },
+    {
+      "epoch": 31.04,
+      "learning_rate": 2.681518151815182e-06,
+      "loss": 0.8882,
+      "step": 6270
+    },
+    {
+      "epoch": 31.09,
+      "learning_rate": 2.676934360102677e-06,
+      "loss": 0.8748,
+      "step": 6280
+    },
+    {
+      "epoch": 31.14,
+      "learning_rate": 2.6723505683901725e-06,
+      "loss": 0.8372,
+      "step": 6290
+    },
+    {
+      "epoch": 31.19,
+      "learning_rate": 2.667766776677668e-06,
+      "loss": 0.9076,
+      "step": 6300
+    },
+    {
+      "epoch": 31.24,
+      "learning_rate": 2.6631829849651634e-06,
+      "loss": 0.9,
+      "step": 6310
+    },
+    {
+      "epoch": 31.29,
+      "learning_rate": 2.658599193252659e-06,
+      "loss": 0.8593,
+      "step": 6320
+    },
+    {
+      "epoch": 31.34,
+      "learning_rate": 2.6540154015401544e-06,
+      "loss": 0.8675,
+      "step": 6330
+    },
+    {
+      "epoch": 31.39,
+      "learning_rate": 2.6494316098276496e-06,
+      "loss": 0.889,
+      "step": 6340
+    },
+    {
+      "epoch": 31.44,
+      "learning_rate": 2.644847818115145e-06,
+      "loss": 0.8607,
+      "step": 6350
+    },
+    {
+      "epoch": 31.49,
+      "learning_rate": 2.64026402640264e-06,
+      "loss": 0.9069,
+      "step": 6360
+    },
+    {
+      "epoch": 31.53,
+      "learning_rate": 2.6356802346901363e-06,
+      "loss": 0.8469,
+      "step": 6370
+    },
+    {
+      "epoch": 31.58,
+      "learning_rate": 2.6310964429776316e-06,
+      "loss": 0.9334,
+      "step": 6380
+    },
+    {
+      "epoch": 31.63,
+      "learning_rate": 2.626512651265127e-06,
+      "loss": 0.859,
+      "step": 6390
+    },
+    {
+      "epoch": 31.68,
+      "learning_rate": 2.621928859552622e-06,
+      "loss": 0.9614,
+      "step": 6400
+    },
+    {
+      "epoch": 31.73,
+      "learning_rate": 2.6173450678401174e-06,
+      "loss": 0.893,
+      "step": 6410
+    },
+    {
+      "epoch": 31.78,
+      "learning_rate": 2.6127612761276126e-06,
+      "loss": 0.8778,
+      "step": 6420
+    },
+    {
+      "epoch": 31.83,
+      "learning_rate": 2.6081774844151087e-06,
+      "loss": 0.8632,
+      "step": 6430
+    },
+    {
+      "epoch": 31.88,
+      "learning_rate": 2.603593692702604e-06,
+      "loss": 0.7968,
+      "step": 6440
+    },
+    {
+      "epoch": 31.93,
+      "learning_rate": 2.5990099009900993e-06,
+      "loss": 0.9204,
+      "step": 6450
+    },
+    {
+      "epoch": 31.98,
+      "learning_rate": 2.5944261092775946e-06,
+      "loss": 0.8859,
+      "step": 6460
+    },
+    {
+      "epoch": 32.0,
+      "eval_accuracy": 0.7039359108324625,
+      "eval_loss": 0.8178386092185974,
+      "eval_runtime": 37.6559,
+      "eval_samples_per_second": 76.243,
+      "eval_steps_per_second": 2.39,
+      "step": 6464
+    },
+    {
+      "epoch": 32.03,
+      "learning_rate": 2.58984231756509e-06,
+      "loss": 0.8777,
+      "step": 6470
+    },
+    {
+      "epoch": 32.08,
+      "learning_rate": 2.585258525852585e-06,
+      "loss": 0.8731,
+      "step": 6480
+    },
+    {
+      "epoch": 32.13,
+      "learning_rate": 2.5806747341400808e-06,
+      "loss": 0.872,
+      "step": 6490
+    },
+    {
+      "epoch": 32.18,
+      "learning_rate": 2.5760909424275765e-06,
+      "loss": 0.8673,
+      "step": 6500
+    },
+    {
+      "epoch": 32.23,
+      "learning_rate": 2.5715071507150717e-06,
+      "loss": 0.9014,
+      "step": 6510
+    },
+    {
+      "epoch": 32.28,
+      "learning_rate": 2.566923359002567e-06,
+      "loss": 0.8257,
+      "step": 6520
+    },
+    {
+      "epoch": 32.33,
+      "learning_rate": 2.5623395672900627e-06,
+      "loss": 0.9027,
+      "step": 6530
+    },
+    {
+      "epoch": 32.38,
+      "learning_rate": 2.557755775577558e-06,
+      "loss": 0.9013,
+      "step": 6540
+    },
+    {
+      "epoch": 32.43,
+      "learning_rate": 2.5531719838650532e-06,
+      "loss": 0.8832,
+      "step": 6550
+    },
+    {
+      "epoch": 32.48,
+      "learning_rate": 2.548588192152549e-06,
+      "loss": 0.8314,
+      "step": 6560
+    },
+    {
+      "epoch": 32.52,
+      "learning_rate": 2.544004400440044e-06,
+      "loss": 0.8633,
+      "step": 6570
+    },
+    {
+      "epoch": 32.57,
+      "learning_rate": 2.53942060872754e-06,
+      "loss": 0.9049,
+      "step": 6580
+    },
+    {
+      "epoch": 32.62,
+      "learning_rate": 2.534836817015035e-06,
+      "loss": 0.862,
+      "step": 6590
+    },
+    {
+      "epoch": 32.67,
+      "learning_rate": 2.5302530253025304e-06,
+      "loss": 0.8941,
+      "step": 6600
+    },
+    {
+      "epoch": 32.72,
+      "learning_rate": 2.5256692335900257e-06,
+      "loss": 0.9173,
+      "step": 6610
+    },
+    {
+      "epoch": 32.77,
+      "learning_rate": 2.521085441877521e-06,
+      "loss": 0.8386,
+      "step": 6620
+    },
+    {
+      "epoch": 32.82,
+      "learning_rate": 2.516501650165017e-06,
+      "loss": 0.8704,
+      "step": 6630
+    },
+    {
+      "epoch": 32.87,
+      "learning_rate": 2.5119178584525123e-06,
+      "loss": 0.8695,
+      "step": 6640
+    },
+    {
+      "epoch": 32.92,
+      "learning_rate": 2.5073340667400076e-06,
+      "loss": 0.8994,
+      "step": 6650
+    },
+    {
+      "epoch": 32.97,
+      "learning_rate": 2.502750275027503e-06,
+      "loss": 0.8464,
+      "step": 6660
+    },
+    {
+      "epoch": 33.0,
+      "eval_accuracy": 0.7091605712295368,
+      "eval_loss": 0.8146793246269226,
+      "eval_runtime": 37.6306,
+      "eval_samples_per_second": 76.294,
+      "eval_steps_per_second": 2.392,
+      "step": 6666
+    },
+    {
+      "epoch": 33.02,
+      "learning_rate": 2.4981664833149986e-06,
+      "loss": 0.8774,
+      "step": 6670
+    },
+    {
+      "epoch": 33.07,
+      "learning_rate": 2.493582691602494e-06,
+      "loss": 0.8684,
+      "step": 6680
+    },
+    {
+      "epoch": 33.12,
+      "learning_rate": 2.488998899889989e-06,
+      "loss": 0.877,
+      "step": 6690
+    },
+    {
+      "epoch": 33.17,
+      "learning_rate": 2.4844151081774844e-06,
+      "loss": 0.8037,
+      "step": 6700
+    },
+    {
+      "epoch": 33.22,
+      "learning_rate": 2.47983131646498e-06,
+      "loss": 0.8617,
+      "step": 6710
+    },
+    {
+      "epoch": 33.27,
+      "learning_rate": 2.4752475247524753e-06,
+      "loss": 0.8513,
+      "step": 6720
+    },
+    {
+      "epoch": 33.32,
+      "learning_rate": 2.4706637330399706e-06,
+      "loss": 0.872,
+      "step": 6730
+    },
+    {
+      "epoch": 33.37,
+      "learning_rate": 2.4660799413274663e-06,
+      "loss": 0.8877,
+      "step": 6740
+    },
+    {
+      "epoch": 33.42,
+      "learning_rate": 2.4614961496149616e-06,
+      "loss": 0.939,
+      "step": 6750
+    },
+    {
+      "epoch": 33.47,
+      "learning_rate": 2.456912357902457e-06,
+      "loss": 0.8427,
+      "step": 6760
+    },
+    {
+      "epoch": 33.51,
+      "learning_rate": 2.4523285661899525e-06,
+      "loss": 0.8354,
+      "step": 6770
+    },
+    {
+      "epoch": 33.56,
+      "learning_rate": 2.4477447744774478e-06,
+      "loss": 0.8828,
+      "step": 6780
+    },
+    {
+      "epoch": 33.61,
+      "learning_rate": 2.4431609827649435e-06,
+      "loss": 0.8594,
+      "step": 6790
+    },
+    {
+      "epoch": 33.66,
+      "learning_rate": 2.4385771910524387e-06,
+      "loss": 0.849,
+      "step": 6800
+    },
+    {
+      "epoch": 33.71,
+      "learning_rate": 2.4339933993399344e-06,
+      "loss": 0.8784,
+      "step": 6810
+    },
+    {
+      "epoch": 33.76,
+      "learning_rate": 2.4294096076274297e-06,
+      "loss": 0.867,
+      "step": 6820
+    },
+    {
+      "epoch": 33.81,
+      "learning_rate": 2.424825815914925e-06,
+      "loss": 0.8735,
+      "step": 6830
+    },
+    {
+      "epoch": 33.86,
+      "learning_rate": 2.4202420242024207e-06,
+      "loss": 0.8622,
+      "step": 6840
+    },
+    {
+      "epoch": 33.91,
+      "learning_rate": 2.415658232489916e-06,
+      "loss": 0.8477,
+      "step": 6850
+    },
+    {
+      "epoch": 33.96,
+      "learning_rate": 2.411074440777411e-06,
+      "loss": 0.9143,
+      "step": 6860
+    },
+    {
+      "epoch": 34.0,
+      "eval_accuracy": 0.7028909787530477,
+      "eval_loss": 0.823244035243988,
+      "eval_runtime": 37.8151,
+      "eval_samples_per_second": 75.922,
+      "eval_steps_per_second": 2.38,
+      "step": 6868
+    },
+    {
+      "epoch": 34.01,
+      "learning_rate": 2.406490649064907e-06,
+      "loss": 0.8361,
+      "step": 6870
+    },
+    {
+      "epoch": 34.06,
+      "learning_rate": 2.401906857352402e-06,
+      "loss": 0.8612,
+      "step": 6880
+    },
+    {
+      "epoch": 34.11,
+      "learning_rate": 2.3973230656398974e-06,
+      "loss": 0.8923,
+      "step": 6890
+    },
+    {
+      "epoch": 34.16,
+      "learning_rate": 2.392739273927393e-06,
+      "loss": 0.9236,
+      "step": 6900
+    },
+    {
+      "epoch": 34.21,
+      "learning_rate": 2.3881554822148884e-06,
+      "loss": 0.8326,
+      "step": 6910
+    },
+    {
+      "epoch": 34.26,
+      "learning_rate": 2.3835716905023836e-06,
+      "loss": 0.8763,
+      "step": 6920
+    },
+    {
+      "epoch": 34.31,
+      "learning_rate": 2.3789878987898793e-06,
+      "loss": 0.8365,
+      "step": 6930
+    },
+    {
+      "epoch": 34.36,
+      "learning_rate": 2.3744041070773746e-06,
+      "loss": 0.8618,
+      "step": 6940
+    },
+    {
+      "epoch": 34.41,
+      "learning_rate": 2.36982031536487e-06,
+      "loss": 0.8766,
+      "step": 6950
+    },
+    {
+      "epoch": 34.46,
+      "learning_rate": 2.3652365236523656e-06,
+      "loss": 0.8877,
+      "step": 6960
+    },
+    {
+      "epoch": 34.5,
+      "learning_rate": 2.360652731939861e-06,
+      "loss": 0.8772,
+      "step": 6970
+    },
+    {
+      "epoch": 34.55,
+      "learning_rate": 2.356068940227356e-06,
+      "loss": 0.8489,
+      "step": 6980
+    },
+    {
+      "epoch": 34.6,
+      "learning_rate": 2.3514851485148514e-06,
+      "loss": 0.8171,
+      "step": 6990
+    },
+    {
+      "epoch": 34.65,
+      "learning_rate": 2.346901356802347e-06,
+      "loss": 0.8617,
+      "step": 7000
+    },
+    {
+      "epoch": 34.7,
+      "learning_rate": 2.3423175650898423e-06,
+      "loss": 0.9032,
+      "step": 7010
+    },
+    {
+      "epoch": 34.75,
+      "learning_rate": 2.337733773377338e-06,
+      "loss": 0.8746,
+      "step": 7020
+    },
+    {
+      "epoch": 34.8,
+      "learning_rate": 2.3331499816648333e-06,
+      "loss": 0.8782,
+      "step": 7030
+    },
+    {
+      "epoch": 34.85,
+      "learning_rate": 2.3285661899523286e-06,
+      "loss": 0.8676,
+      "step": 7040
+    },
+    {
+      "epoch": 34.9,
+      "learning_rate": 2.3239823982398242e-06,
+      "loss": 0.847,
+      "step": 7050
+    },
+    {
+      "epoch": 34.95,
+      "learning_rate": 2.3193986065273195e-06,
+      "loss": 0.8918,
+      "step": 7060
+    },
+    {
+      "epoch": 35.0,
+      "learning_rate": 2.314814814814815e-06,
+      "loss": 0.8506,
+      "step": 7070
+    },
+    {
+      "epoch": 35.0,
+      "eval_accuracy": 0.703239289446186,
+      "eval_loss": 0.815805971622467,
+      "eval_runtime": 37.7997,
+      "eval_samples_per_second": 75.953,
+      "eval_steps_per_second": 2.381,
+      "step": 7070
+    },
+    {
+      "epoch": 35.05,
+      "learning_rate": 2.3102310231023105e-06,
+      "loss": 0.8363,
+      "step": 7080
+    },
+    {
+      "epoch": 35.1,
+      "learning_rate": 2.3056472313898057e-06,
+      "loss": 0.8568,
+      "step": 7090
+    },
+    {
+      "epoch": 35.15,
+      "learning_rate": 2.3010634396773014e-06,
+      "loss": 0.8517,
+      "step": 7100
+    },
+    {
+      "epoch": 35.2,
+      "learning_rate": 2.2964796479647967e-06,
+      "loss": 0.8648,
+      "step": 7110
+    },
+    {
+      "epoch": 35.25,
+      "learning_rate": 2.2918958562522924e-06,
+      "loss": 0.8348,
+      "step": 7120
+    },
+    {
+      "epoch": 35.3,
+      "learning_rate": 2.2873120645397877e-06,
+      "loss": 0.8946,
+      "step": 7130
+    },
+    {
+      "epoch": 35.35,
+      "learning_rate": 2.282728272827283e-06,
+      "loss": 0.8464,
+      "step": 7140
+    },
+    {
+      "epoch": 35.4,
+      "learning_rate": 2.278144481114778e-06,
+      "loss": 0.8244,
+      "step": 7150
+    },
+    {
+      "epoch": 35.45,
+      "learning_rate": 2.273560689402274e-06,
+      "loss": 0.8465,
+      "step": 7160
+    },
+    {
+      "epoch": 35.5,
+      "learning_rate": 2.268976897689769e-06,
+      "loss": 0.881,
+      "step": 7170
+    },
+    {
+      "epoch": 35.54,
+      "learning_rate": 2.2643931059772644e-06,
+      "loss": 0.8788,
+      "step": 7180
+    },
+    {
+      "epoch": 35.59,
+      "learning_rate": 2.25980931426476e-06,
+      "loss": 0.8747,
+      "step": 7190
+    },
+    {
+      "epoch": 35.64,
+      "learning_rate": 2.2552255225522554e-06,
+      "loss": 0.8471,
+      "step": 7200
+    },
+    {
+      "epoch": 35.69,
+      "learning_rate": 2.2506417308397507e-06,
+      "loss": 0.8444,
+      "step": 7210
+    },
+    {
+      "epoch": 35.74,
+      "learning_rate": 2.2460579391272463e-06,
+      "loss": 0.8586,
+      "step": 7220
+    },
+    {
+      "epoch": 35.79,
+      "learning_rate": 2.2414741474147416e-06,
+      "loss": 0.8402,
+      "step": 7230
+    },
+    {
+      "epoch": 35.84,
+      "learning_rate": 2.236890355702237e-06,
+      "loss": 0.8907,
+      "step": 7240
+    },
+    {
+      "epoch": 35.89,
+      "learning_rate": 2.2323065639897326e-06,
+      "loss": 0.819,
+      "step": 7250
+    },
+    {
+      "epoch": 35.94,
+      "learning_rate": 2.227722772277228e-06,
+      "loss": 0.8506,
+      "step": 7260
+    },
+    {
+      "epoch": 35.99,
+      "learning_rate": 2.223138980564723e-06,
+      "loss": 0.9084,
+      "step": 7270
+    },
+    {
+      "epoch": 36.0,
+      "eval_accuracy": 0.7056774642981539,
+      "eval_loss": 0.8166046142578125,
+      "eval_runtime": 37.8249,
+      "eval_samples_per_second": 75.902,
+      "eval_steps_per_second": 2.379,
+      "step": 7272
+    },
+    {
+      "epoch": 36.04,
+      "learning_rate": 2.218555188852219e-06,
+      "loss": 0.8835,
+      "step": 7280
+    },
+    {
+      "epoch": 36.09,
+      "learning_rate": 2.213971397139714e-06,
+      "loss": 0.8816,
+      "step": 7290
+    },
+    {
+      "epoch": 36.14,
+      "learning_rate": 2.2093876054272093e-06,
+      "loss": 0.909,
+      "step": 7300
+    },
+    {
+      "epoch": 36.19,
+      "learning_rate": 2.204803813714705e-06,
+      "loss": 0.8118,
+      "step": 7310
+    },
+    {
+      "epoch": 36.24,
+      "learning_rate": 2.2002200220022003e-06,
+      "loss": 0.8809,
+      "step": 7320
+    },
+    {
+      "epoch": 36.29,
+      "learning_rate": 2.195636230289696e-06,
+      "loss": 0.8522,
+      "step": 7330
+    },
+    {
+      "epoch": 36.34,
+      "learning_rate": 2.1910524385771913e-06,
+      "loss": 0.8754,
+      "step": 7340
+    },
+    {
+      "epoch": 36.39,
+      "learning_rate": 2.186468646864687e-06,
+      "loss": 0.8451,
+      "step": 7350
+    },
+    {
+      "epoch": 36.44,
+      "learning_rate": 2.181884855152182e-06,
+      "loss": 0.8857,
+      "step": 7360
+    },
+    {
+      "epoch": 36.49,
+      "learning_rate": 2.1773010634396775e-06,
+      "loss": 0.8542,
+      "step": 7370
+    },
+    {
+      "epoch": 36.53,
+      "learning_rate": 2.172717271727173e-06,
+      "loss": 0.8372,
+      "step": 7380
+    },
+    {
+      "epoch": 36.58,
+      "learning_rate": 2.1681334800146684e-06,
+      "loss": 0.8584,
+      "step": 7390
+    },
+    {
+      "epoch": 36.63,
+      "learning_rate": 2.1635496883021637e-06,
+      "loss": 0.9064,
+      "step": 7400
+    },
+    {
+      "epoch": 36.68,
+      "learning_rate": 2.1589658965896594e-06,
+      "loss": 0.882,
+      "step": 7410
+    },
+    {
+      "epoch": 36.73,
+      "learning_rate": 2.1543821048771547e-06,
+      "loss": 0.838,
+      "step": 7420
+    },
+    {
+      "epoch": 36.78,
+      "learning_rate": 2.14979831316465e-06,
+      "loss": 0.8857,
+      "step": 7430
+    },
+    {
+      "epoch": 36.83,
+      "learning_rate": 2.145214521452145e-06,
+      "loss": 0.8118,
+      "step": 7440
+    },
+    {
+      "epoch": 36.88,
+      "learning_rate": 2.140630729739641e-06,
+      "loss": 0.8648,
+      "step": 7450
+    },
+    {
+      "epoch": 36.93,
+      "learning_rate": 2.136046938027136e-06,
+      "loss": 0.8966,
+      "step": 7460
+    },
+    {
+      "epoch": 36.98,
+      "learning_rate": 2.1314631463146314e-06,
+      "loss": 0.8616,
+      "step": 7470
+    },
+    {
+      "epoch": 37.0,
+      "eval_accuracy": 0.7088122605363985,
+      "eval_loss": 0.8132399916648865,
+      "eval_runtime": 37.6541,
+      "eval_samples_per_second": 76.247,
+      "eval_steps_per_second": 2.39,
+      "step": 7474
+    },
+    {
+      "epoch": 37.03,
+      "learning_rate": 2.126879354602127e-06,
+      "loss": 0.8186,
+      "step": 7480
+    },
+    {
+      "epoch": 37.08,
+      "learning_rate": 2.1222955628896224e-06,
+      "loss": 0.8562,
+      "step": 7490
+    },
+    {
+      "epoch": 37.13,
+      "learning_rate": 2.1177117711771177e-06,
+      "loss": 0.8221,
+      "step": 7500
+    },
+    {
+      "epoch": 37.18,
+      "learning_rate": 2.1131279794646133e-06,
+      "loss": 0.7865,
+      "step": 7510
+    },
+    {
+      "epoch": 37.23,
+      "learning_rate": 2.1085441877521086e-06,
+      "loss": 0.895,
+      "step": 7520
+    },
+    {
+      "epoch": 37.28,
+      "learning_rate": 2.103960396039604e-06,
+      "loss": 0.8637,
+      "step": 7530
+    },
+    {
+      "epoch": 37.33,
+      "learning_rate": 2.0993766043270996e-06,
+      "loss": 0.8334,
+      "step": 7540
+    },
+    {
+      "epoch": 37.38,
+      "learning_rate": 2.094792812614595e-06,
+      "loss": 0.8621,
+      "step": 7550
+    },
+    {
+      "epoch": 37.43,
+      "learning_rate": 2.0902090209020905e-06,
+      "loss": 0.8691,
+      "step": 7560
+    },
+    {
+      "epoch": 37.48,
+      "learning_rate": 2.085625229189586e-06,
+      "loss": 0.8956,
+      "step": 7570
+    },
+    {
+      "epoch": 37.52,
+      "learning_rate": 2.081041437477081e-06,
+      "loss": 0.8487,
+      "step": 7580
+    },
+    {
+      "epoch": 37.57,
+      "learning_rate": 2.0764576457645768e-06,
+      "loss": 0.8405,
+      "step": 7590
+    },
+    {
+      "epoch": 37.62,
+      "learning_rate": 2.071873854052072e-06,
+      "loss": 0.8562,
+      "step": 7600
+    },
+    {
+      "epoch": 37.67,
+      "learning_rate": 2.0672900623395677e-06,
+      "loss": 0.856,
+      "step": 7610
+    },
+    {
+      "epoch": 37.72,
+      "learning_rate": 2.062706270627063e-06,
+      "loss": 0.8249,
+      "step": 7620
+    },
+    {
+      "epoch": 37.77,
+      "learning_rate": 2.0581224789145583e-06,
+      "loss": 0.8173,
+      "step": 7630
+    },
+    {
+      "epoch": 37.82,
+      "learning_rate": 2.053538687202054e-06,
+      "loss": 0.8675,
+      "step": 7640
+    },
+    {
+      "epoch": 37.87,
+      "learning_rate": 2.0489548954895492e-06,
+      "loss": 0.8931,
+      "step": 7650
+    },
+    {
+      "epoch": 37.92,
+      "learning_rate": 2.0443711037770445e-06,
+      "loss": 0.8245,
+      "step": 7660
+    },
+    {
+      "epoch": 37.97,
+      "learning_rate": 2.03978731206454e-06,
+      "loss": 0.8656,
+      "step": 7670
+    },
+    {
+      "epoch": 38.0,
+      "eval_accuracy": 0.7046325322187391,
+      "eval_loss": 0.8155147433280945,
+      "eval_runtime": 37.8324,
+      "eval_samples_per_second": 75.887,
+      "eval_steps_per_second": 2.379,
+      "step": 7676
+    },
+    {
+      "epoch": 38.02,
+      "learning_rate": 2.0352035203520354e-06,
+      "loss": 0.817,
+      "step": 7680
+    },
+    {
+      "epoch": 38.07,
+      "learning_rate": 2.0306197286395307e-06,
+      "loss": 0.7992,
+      "step": 7690
+    },
+    {
+      "epoch": 38.12,
+      "learning_rate": 2.0260359369270264e-06,
+      "loss": 0.8208,
+      "step": 7700
+    },
+    {
+      "epoch": 38.17,
+      "learning_rate": 2.0214521452145217e-06,
+      "loss": 0.8909,
+      "step": 7710
+    },
+    {
+      "epoch": 38.22,
+      "learning_rate": 2.016868353502017e-06,
+      "loss": 0.8477,
+      "step": 7720
+    },
+    {
+      "epoch": 38.27,
+      "learning_rate": 2.012284561789512e-06,
+      "loss": 0.8801,
+      "step": 7730
+    },
+    {
+      "epoch": 38.32,
+      "learning_rate": 2.007700770077008e-06,
+      "loss": 0.8435,
+      "step": 7740
+    },
+    {
+      "epoch": 38.37,
+      "learning_rate": 2.003116978364503e-06,
+      "loss": 0.8418,
+      "step": 7750
+    },
+    {
+      "epoch": 38.42,
+      "learning_rate": 1.9985331866519984e-06,
+      "loss": 0.8422,
+      "step": 7760
+    },
+    {
+      "epoch": 38.47,
+      "learning_rate": 1.993949394939494e-06,
+      "loss": 0.8618,
+      "step": 7770
+    },
+    {
+      "epoch": 38.51,
+      "learning_rate": 1.9893656032269894e-06,
+      "loss": 0.8501,
+      "step": 7780
+    },
+    {
+      "epoch": 38.56,
+      "learning_rate": 1.9847818115144847e-06,
+      "loss": 0.835,
+      "step": 7790
+    },
+    {
+      "epoch": 38.61,
+      "learning_rate": 1.9801980198019803e-06,
+      "loss": 0.8525,
+      "step": 7800
+    },
+    {
+      "epoch": 38.66,
+      "learning_rate": 1.9756142280894756e-06,
+      "loss": 0.8364,
+      "step": 7810
+    },
+    {
+      "epoch": 38.71,
+      "learning_rate": 1.9710304363769713e-06,
+      "loss": 0.8333,
+      "step": 7820
+    },
+    {
+      "epoch": 38.76,
+      "learning_rate": 1.9664466446644666e-06,
+      "loss": 0.8417,
+      "step": 7830
+    },
+    {
+      "epoch": 38.81,
+      "learning_rate": 1.961862852951962e-06,
+      "loss": 0.8346,
+      "step": 7840
+    },
+    {
+      "epoch": 38.86,
+      "learning_rate": 1.9572790612394575e-06,
+      "loss": 0.8618,
+      "step": 7850
+    },
+    {
+      "epoch": 38.91,
+      "learning_rate": 1.952695269526953e-06,
+      "loss": 0.911,
+      "step": 7860
+    },
+    {
+      "epoch": 38.96,
+      "learning_rate": 1.9481114778144485e-06,
+      "loss": 0.8238,
+      "step": 7870
+    },
+    {
+      "epoch": 39.0,
+      "eval_accuracy": 0.7063740856844305,
+      "eval_loss": 0.8169887661933899,
+      "eval_runtime": 37.6839,
+      "eval_samples_per_second": 76.186,
+      "eval_steps_per_second": 2.388,
+      "step": 7878
+    },
+    {
+      "epoch": 39.01,
+      "learning_rate": 1.9435276861019438e-06,
+      "loss": 0.8682,
+      "step": 7880
+    },
+    {
+      "epoch": 39.06,
+      "learning_rate": 1.938943894389439e-06,
+      "loss": 0.8212,
+      "step": 7890
+    },
+    {
+      "epoch": 39.11,
+      "learning_rate": 1.9343601026769347e-06,
+      "loss": 0.809,
+      "step": 7900
+    },
+    {
+      "epoch": 39.16,
+      "learning_rate": 1.92977631096443e-06,
+      "loss": 0.8731,
+      "step": 7910
+    },
+    {
+      "epoch": 39.21,
+      "learning_rate": 1.9251925192519253e-06,
+      "loss": 0.83,
+      "step": 7920
+    },
+    {
+      "epoch": 39.26,
+      "learning_rate": 1.920608727539421e-06,
+      "loss": 0.8216,
+      "step": 7930
+    },
+    {
+      "epoch": 39.31,
+      "learning_rate": 1.9160249358269162e-06,
+      "loss": 0.8366,
+      "step": 7940
+    },
+    {
+      "epoch": 39.36,
+      "learning_rate": 1.9114411441144115e-06,
+      "loss": 0.8469,
+      "step": 7950
+    },
+    {
+      "epoch": 39.41,
+      "learning_rate": 1.9068573524019072e-06,
+      "loss": 0.7865,
+      "step": 7960
+    },
+    {
+      "epoch": 39.46,
+      "learning_rate": 1.9022735606894024e-06,
+      "loss": 0.8855,
+      "step": 7970
+    },
+    {
+      "epoch": 39.5,
+      "learning_rate": 1.8976897689768977e-06,
+      "loss": 0.853,
+      "step": 7980
+    },
+    {
+      "epoch": 39.55,
+      "learning_rate": 1.8931059772643934e-06,
+      "loss": 0.85,
+      "step": 7990
+    },
+    {
+      "epoch": 39.6,
+      "learning_rate": 1.8885221855518887e-06,
+      "loss": 0.8387,
+      "step": 8000
+    },
+    {
+      "epoch": 39.65,
+      "learning_rate": 1.883938393839384e-06,
+      "loss": 0.844,
+      "step": 8010
+    },
+    {
+      "epoch": 39.7,
+      "learning_rate": 1.8793546021268796e-06,
+      "loss": 0.8838,
+      "step": 8020
+    },
+    {
+      "epoch": 39.75,
+      "learning_rate": 1.874770810414375e-06,
+      "loss": 0.8451,
+      "step": 8030
+    },
+    {
+      "epoch": 39.8,
+      "learning_rate": 1.8701870187018704e-06,
+      "loss": 0.8624,
+      "step": 8040
+    },
+    {
+      "epoch": 39.85,
+      "learning_rate": 1.8656032269893656e-06,
+      "loss": 0.8157,
+      "step": 8050
+    },
+    {
+      "epoch": 39.9,
+      "learning_rate": 1.8610194352768613e-06,
+      "loss": 0.8618,
+      "step": 8060
+    },
+    {
+      "epoch": 39.95,
+      "learning_rate": 1.8564356435643566e-06,
+      "loss": 0.8535,
+      "step": 8070
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 1.8518518518518519e-06,
+      "loss": 0.8673,
+      "step": 8080
+    },
+    {
+      "epoch": 40.0,
+      "eval_accuracy": 0.7091605712295368,
+      "eval_loss": 0.8190204501152039,
+      "eval_runtime": 37.7893,
+      "eval_samples_per_second": 75.974,
+      "eval_steps_per_second": 2.382,
+      "step": 8080
+    },
+    {
+      "epoch": 40.05,
+      "learning_rate": 1.8472680601393476e-06,
+      "loss": 0.8216,
+      "step": 8090
+    },
+    {
+      "epoch": 40.1,
+      "learning_rate": 1.8426842684268428e-06,
+      "loss": 0.8405,
+      "step": 8100
+    },
+    {
+      "epoch": 40.15,
+      "learning_rate": 1.838100476714338e-06,
+      "loss": 0.9004,
+      "step": 8110
+    },
+    {
+      "epoch": 40.2,
+      "learning_rate": 1.8335166850018338e-06,
+      "loss": 0.8615,
+      "step": 8120
+    },
+    {
+      "epoch": 40.25,
+      "learning_rate": 1.828932893289329e-06,
+      "loss": 0.8324,
+      "step": 8130
+    },
+    {
+      "epoch": 40.3,
+      "learning_rate": 1.8243491015768245e-06,
+      "loss": 0.8305,
+      "step": 8140
+    },
+    {
+      "epoch": 40.35,
+      "learning_rate": 1.81976530986432e-06,
+      "loss": 0.8807,
+      "step": 8150
+    },
+    {
+      "epoch": 40.4,
+      "learning_rate": 1.8151815181518153e-06,
+      "loss": 0.8511,
+      "step": 8160
+    },
+    {
+      "epoch": 40.45,
+      "learning_rate": 1.8105977264393108e-06,
+      "loss": 0.9093,
+      "step": 8170
+    },
+    {
+      "epoch": 40.5,
+      "learning_rate": 1.806013934726806e-06,
+      "loss": 0.8506,
+      "step": 8180
+    },
+    {
+      "epoch": 40.54,
+      "learning_rate": 1.8014301430143017e-06,
+      "loss": 0.8493,
+      "step": 8190
+    },
+    {
+      "epoch": 40.59,
+      "learning_rate": 1.796846351301797e-06,
+      "loss": 0.8685,
+      "step": 8200
+    },
+    {
+      "epoch": 40.64,
+      "learning_rate": 1.7922625595892923e-06,
+      "loss": 0.8352,
+      "step": 8210
+    },
+    {
+      "epoch": 40.69,
+      "learning_rate": 1.787678767876788e-06,
+      "loss": 0.8496,
+      "step": 8220
+    },
+    {
+      "epoch": 40.74,
+      "learning_rate": 1.7830949761642832e-06,
+      "loss": 0.8376,
+      "step": 8230
+    },
+    {
+      "epoch": 40.79,
+      "learning_rate": 1.7785111844517785e-06,
+      "loss": 0.8103,
+      "step": 8240
+    },
+    {
+      "epoch": 40.84,
+      "learning_rate": 1.7739273927392742e-06,
+      "loss": 0.8858,
+      "step": 8250
+    },
+    {
+      "epoch": 40.89,
+      "learning_rate": 1.7693436010267694e-06,
+      "loss": 0.8044,
+      "step": 8260
+    },
+    {
+      "epoch": 40.94,
+      "learning_rate": 1.764759809314265e-06,
+      "loss": 0.8165,
+      "step": 8270
+    },
+    {
+      "epoch": 40.99,
+      "learning_rate": 1.7601760176017604e-06,
+      "loss": 0.8624,
+      "step": 8280
+    },
+    {
+      "epoch": 41.0,
+      "eval_accuracy": 0.709508881922675,
+      "eval_loss": 0.8126907348632812,
+      "eval_runtime": 37.7914,
+      "eval_samples_per_second": 75.97,
+      "eval_steps_per_second": 2.381,
+      "step": 8282
+    },
+    {
+      "epoch": 41.04,
+      "learning_rate": 1.7555922258892557e-06,
+      "loss": 0.856,
+      "step": 8290
+    },
+    {
+      "epoch": 41.09,
+      "learning_rate": 1.7510084341767512e-06,
+      "loss": 0.8634,
+      "step": 8300
+    },
+    {
+      "epoch": 41.14,
+      "learning_rate": 1.7464246424642466e-06,
+      "loss": 0.8399,
+      "step": 8310
+    },
+    {
+      "epoch": 41.19,
+      "learning_rate": 1.7418408507517421e-06,
+      "loss": 0.835,
+      "step": 8320
+    },
+    {
+      "epoch": 41.24,
+      "learning_rate": 1.7372570590392374e-06,
+      "loss": 0.885,
+      "step": 8330
+    },
+    {
+      "epoch": 41.29,
+      "learning_rate": 1.7326732673267326e-06,
+      "loss": 0.856,
+      "step": 8340
+    },
+    {
+      "epoch": 41.34,
+      "learning_rate": 1.7280894756142283e-06,
+      "loss": 0.7977,
+      "step": 8350
+    },
+    {
+      "epoch": 41.39,
+      "learning_rate": 1.7235056839017236e-06,
+      "loss": 0.8663,
+      "step": 8360
+    },
+    {
+      "epoch": 41.44,
+      "learning_rate": 1.7189218921892189e-06,
+      "loss": 0.849,
+      "step": 8370
+    },
+    {
+      "epoch": 41.49,
+      "learning_rate": 1.7143381004767146e-06,
+      "loss": 0.8644,
+      "step": 8380
+    },
+    {
+      "epoch": 41.53,
+      "learning_rate": 1.7097543087642098e-06,
+      "loss": 0.8576,
+      "step": 8390
+    },
+    {
+      "epoch": 41.58,
+      "learning_rate": 1.7051705170517053e-06,
+      "loss": 0.8698,
+      "step": 8400
+    },
+    {
+      "epoch": 41.63,
+      "learning_rate": 1.7005867253392008e-06,
+      "loss": 0.8004,
+      "step": 8410
+    },
+    {
+      "epoch": 41.68,
+      "learning_rate": 1.696002933626696e-06,
+      "loss": 0.8615,
+      "step": 8420
+    },
+    {
+      "epoch": 41.73,
+      "learning_rate": 1.6914191419141915e-06,
+      "loss": 0.8368,
+      "step": 8430
+    },
+    {
+      "epoch": 41.78,
+      "learning_rate": 1.686835350201687e-06,
+      "loss": 0.8636,
+      "step": 8440
+    },
+    {
+      "epoch": 41.83,
+      "learning_rate": 1.6822515584891825e-06,
+      "loss": 0.8291,
+      "step": 8450
+    },
+    {
+      "epoch": 41.88,
+      "learning_rate": 1.6776677667766778e-06,
+      "loss": 0.7972,
+      "step": 8460
+    },
+    {
+      "epoch": 41.93,
+      "learning_rate": 1.673083975064173e-06,
+      "loss": 0.8293,
+      "step": 8470
+    },
+    {
+      "epoch": 41.98,
+      "learning_rate": 1.6685001833516687e-06,
+      "loss": 0.8261,
+      "step": 8480
+    },
+    {
+      "epoch": 42.0,
+      "eval_accuracy": 0.7112504353883664,
+      "eval_loss": 0.8112697005271912,
+      "eval_runtime": 38.1653,
+      "eval_samples_per_second": 75.225,
+      "eval_steps_per_second": 2.358,
+      "step": 8484
+    },
+    {
+      "epoch": 42.03,
+      "learning_rate": 1.663916391639164e-06,
+      "loss": 0.8719,
+      "step": 8490
+    },
+    {
+      "epoch": 42.08,
+      "learning_rate": 1.6593325999266593e-06,
+      "loss": 0.8415,
+      "step": 8500
+    },
+    {
+      "epoch": 42.13,
+      "learning_rate": 1.654748808214155e-06,
+      "loss": 0.9061,
+      "step": 8510
+    },
+    {
+      "epoch": 42.18,
+      "learning_rate": 1.6501650165016502e-06,
+      "loss": 0.8486,
+      "step": 8520
+    },
+    {
+      "epoch": 42.23,
+      "learning_rate": 1.6455812247891457e-06,
+      "loss": 0.8179,
+      "step": 8530
+    },
+    {
+      "epoch": 42.28,
+      "learning_rate": 1.6409974330766412e-06,
+      "loss": 0.8173,
+      "step": 8540
+    },
+    {
+      "epoch": 42.33,
+      "learning_rate": 1.6364136413641364e-06,
+      "loss": 0.8388,
+      "step": 8550
+    },
+    {
+      "epoch": 42.38,
+      "learning_rate": 1.631829849651632e-06,
+      "loss": 0.813,
+      "step": 8560
+    },
+    {
+      "epoch": 42.43,
+      "learning_rate": 1.6272460579391274e-06,
+      "loss": 0.8688,
+      "step": 8570
+    },
+    {
+      "epoch": 42.48,
+      "learning_rate": 1.6226622662266229e-06,
+      "loss": 0.8339,
+      "step": 8580
+    },
+    {
+      "epoch": 42.52,
+      "learning_rate": 1.6180784745141182e-06,
+      "loss": 0.8137,
+      "step": 8590
+    },
+    {
+      "epoch": 42.57,
+      "learning_rate": 1.6134946828016138e-06,
+      "loss": 0.8667,
+      "step": 8600
+    },
+    {
+      "epoch": 42.62,
+      "learning_rate": 1.6089108910891091e-06,
+      "loss": 0.8718,
+      "step": 8610
+    },
+    {
+      "epoch": 42.67,
+      "learning_rate": 1.6043270993766044e-06,
+      "loss": 0.7705,
+      "step": 8620
+    },
+    {
+      "epoch": 42.72,
+      "learning_rate": 1.5997433076640996e-06,
+      "loss": 0.8177,
+      "step": 8630
+    },
+    {
+      "epoch": 42.77,
+      "learning_rate": 1.5951595159515953e-06,
+      "loss": 0.78,
+      "step": 8640
+    },
+    {
+      "epoch": 42.82,
+      "learning_rate": 1.5905757242390906e-06,
+      "loss": 0.8428,
+      "step": 8650
+    },
+    {
+      "epoch": 42.87,
+      "learning_rate": 1.585991932526586e-06,
+      "loss": 0.8571,
+      "step": 8660
+    },
+    {
+      "epoch": 42.92,
+      "learning_rate": 1.5814081408140816e-06,
+      "loss": 0.8797,
+      "step": 8670
+    },
+    {
+      "epoch": 42.97,
+      "learning_rate": 1.576824349101577e-06,
+      "loss": 0.8218,
+      "step": 8680
+    },
+    {
+      "epoch": 43.0,
+      "eval_accuracy": 0.709508881922675,
+      "eval_loss": 0.8149790167808533,
+      "eval_runtime": 37.9706,
+      "eval_samples_per_second": 75.611,
+      "eval_steps_per_second": 2.37,
+      "step": 8686
+    },
+    {
+      "epoch": 43.02,
+      "learning_rate": 1.5722405573890723e-06,
+      "loss": 0.8103,
+      "step": 8690
+    },
+    {
+      "epoch": 43.07,
+      "learning_rate": 1.5676567656765678e-06,
+      "loss": 0.8328,
+      "step": 8700
+    },
+    {
+      "epoch": 43.12,
+      "learning_rate": 1.5630729739640633e-06,
+      "loss": 0.8026,
+      "step": 8710
+    },
+    {
+      "epoch": 43.17,
+      "learning_rate": 1.5584891822515585e-06,
+      "loss": 0.8303,
+      "step": 8720
+    },
+    {
+      "epoch": 43.22,
+      "learning_rate": 1.5539053905390542e-06,
+      "loss": 0.8525,
+      "step": 8730
+    },
+    {
+      "epoch": 43.27,
+      "learning_rate": 1.5493215988265495e-06,
+      "loss": 0.8424,
+      "step": 8740
+    },
+    {
+      "epoch": 43.32,
+      "learning_rate": 1.5447378071140448e-06,
+      "loss": 0.8636,
+      "step": 8750
+    },
+    {
+      "epoch": 43.37,
+      "learning_rate": 1.5401540154015405e-06,
+      "loss": 0.8601,
+      "step": 8760
+    },
+    {
+      "epoch": 43.42,
+      "learning_rate": 1.5355702236890357e-06,
+      "loss": 0.8043,
+      "step": 8770
+    },
+    {
+      "epoch": 43.47,
+      "learning_rate": 1.530986431976531e-06,
+      "loss": 0.8512,
+      "step": 8780
+    },
+    {
+      "epoch": 43.51,
+      "learning_rate": 1.5264026402640265e-06,
+      "loss": 0.8059,
+      "step": 8790
+    },
+    {
+      "epoch": 43.56,
+      "learning_rate": 1.521818848551522e-06,
+      "loss": 0.8328,
+      "step": 8800
+    },
+    {
+      "epoch": 43.61,
+      "learning_rate": 1.5172350568390174e-06,
+      "loss": 0.8488,
+      "step": 8810
+    },
+    {
+      "epoch": 43.66,
+      "learning_rate": 1.5126512651265127e-06,
+      "loss": 0.8616,
+      "step": 8820
+    },
+    {
+      "epoch": 43.71,
+      "learning_rate": 1.5080674734140082e-06,
+      "loss": 0.8165,
+      "step": 8830
+    },
+    {
+      "epoch": 43.76,
+      "learning_rate": 1.5034836817015037e-06,
+      "loss": 0.8696,
+      "step": 8840
+    },
+    {
+      "epoch": 43.81,
+      "learning_rate": 1.498899889988999e-06,
+      "loss": 0.827,
+      "step": 8850
+    },
+    {
+      "epoch": 43.86,
+      "learning_rate": 1.4943160982764946e-06,
+      "loss": 0.8208,
+      "step": 8860
+    },
+    {
+      "epoch": 43.91,
+      "learning_rate": 1.4897323065639899e-06,
+      "loss": 0.8694,
+      "step": 8870
+    },
+    {
+      "epoch": 43.96,
+      "learning_rate": 1.4851485148514852e-06,
+      "loss": 0.8584,
+      "step": 8880
+    },
+    {
+      "epoch": 44.0,
+      "eval_accuracy": 0.7070707070707071,
+      "eval_loss": 0.8169552683830261,
+      "eval_runtime": 37.8546,
+      "eval_samples_per_second": 75.843,
+      "eval_steps_per_second": 2.378,
+      "step": 8888
+    },
+    {
+      "epoch": 44.01,
+      "learning_rate": 1.4805647231389808e-06,
+      "loss": 0.8055,
+      "step": 8890
+    },
+    {
+      "epoch": 44.06,
+      "learning_rate": 1.4759809314264761e-06,
+      "loss": 0.8627,
+      "step": 8900
+    },
+    {
+      "epoch": 44.11,
+      "learning_rate": 1.4713971397139714e-06,
+      "loss": 0.8734,
+      "step": 8910
+    },
+    {
+      "epoch": 44.16,
+      "learning_rate": 1.4668133480014669e-06,
+      "loss": 0.816,
+      "step": 8920
+    },
+    {
+      "epoch": 44.21,
+      "learning_rate": 1.4622295562889623e-06,
+      "loss": 0.8324,
+      "step": 8930
+    },
+    {
+      "epoch": 44.26,
+      "learning_rate": 1.4576457645764578e-06,
+      "loss": 0.7644,
+      "step": 8940
+    },
+    {
+      "epoch": 44.31,
+      "learning_rate": 1.453061972863953e-06,
+      "loss": 0.8084,
+      "step": 8950
+    },
+    {
+      "epoch": 44.36,
+      "learning_rate": 1.4484781811514486e-06,
+      "loss": 0.8343,
+      "step": 8960
+    },
+    {
+      "epoch": 44.41,
+      "learning_rate": 1.443894389438944e-06,
+      "loss": 0.8128,
+      "step": 8970
+    },
+    {
+      "epoch": 44.46,
+      "learning_rate": 1.4393105977264393e-06,
+      "loss": 0.8548,
+      "step": 8980
+    },
+    {
+      "epoch": 44.5,
+      "learning_rate": 1.434726806013935e-06,
+      "loss": 0.82,
+      "step": 8990
+    },
+    {
+      "epoch": 44.55,
+      "learning_rate": 1.4301430143014303e-06,
+      "loss": 0.8398,
+      "step": 9000
+    },
+    {
+      "epoch": 44.6,
+      "learning_rate": 1.4255592225889255e-06,
+      "loss": 0.8233,
+      "step": 9010
+    },
+    {
+      "epoch": 44.65,
+      "learning_rate": 1.4209754308764212e-06,
+      "loss": 0.8474,
+      "step": 9020
+    },
+    {
+      "epoch": 44.7,
+      "learning_rate": 1.4163916391639165e-06,
+      "loss": 0.8367,
+      "step": 9030
+    },
+    {
+      "epoch": 44.75,
+      "learning_rate": 1.4118078474514118e-06,
+      "loss": 0.8626,
+      "step": 9040
+    },
+    {
+      "epoch": 44.8,
+      "learning_rate": 1.4072240557389075e-06,
+      "loss": 0.8925,
+      "step": 9050
+    },
+    {
+      "epoch": 44.85,
+      "learning_rate": 1.4026402640264027e-06,
+      "loss": 0.8658,
+      "step": 9060
+    },
+    {
+      "epoch": 44.9,
+      "learning_rate": 1.3980564723138982e-06,
+      "loss": 0.8116,
+      "step": 9070
+    },
+    {
+      "epoch": 44.95,
+      "learning_rate": 1.3934726806013935e-06,
+      "loss": 0.7605,
+      "step": 9080
+    },
+    {
+      "epoch": 45.0,
+      "learning_rate": 1.3888888888888892e-06,
+      "loss": 0.8156,
+      "step": 9090
+    },
+    {
+      "epoch": 45.0,
+      "eval_accuracy": 0.711947056774643,
+      "eval_loss": 0.8117088079452515,
+      "eval_runtime": 37.7564,
+      "eval_samples_per_second": 76.04,
+      "eval_steps_per_second": 2.384,
+      "step": 9090
+    },
+    {
+      "epoch": 45.05,
+      "learning_rate": 1.3843050971763844e-06,
+      "loss": 0.835,
+      "step": 9100
+    },
+    {
+      "epoch": 45.1,
+      "learning_rate": 1.3797213054638797e-06,
+      "loss": 0.8481,
+      "step": 9110
+    },
+    {
+      "epoch": 45.15,
+      "learning_rate": 1.3751375137513754e-06,
+      "loss": 0.8351,
+      "step": 9120
+    },
+    {
+      "epoch": 45.2,
+      "learning_rate": 1.3705537220388707e-06,
+      "loss": 0.8075,
+      "step": 9130
+    },
+    {
+      "epoch": 45.25,
+      "learning_rate": 1.365969930326366e-06,
+      "loss": 0.8719,
+      "step": 9140
+    },
+    {
+      "epoch": 45.3,
+      "learning_rate": 1.3613861386138616e-06,
+      "loss": 0.8616,
+      "step": 9150
+    },
+    {
+      "epoch": 45.35,
+      "learning_rate": 1.3568023469013569e-06,
+      "loss": 0.8253,
+      "step": 9160
+    },
+    {
+      "epoch": 45.4,
+      "learning_rate": 1.3522185551888522e-06,
+      "loss": 0.8071,
+      "step": 9170
+    },
+    {
+      "epoch": 45.45,
+      "learning_rate": 1.3476347634763478e-06,
+      "loss": 0.877,
+      "step": 9180
+    },
+    {
+      "epoch": 45.5,
+      "learning_rate": 1.3430509717638431e-06,
+      "loss": 0.8273,
+      "step": 9190
+    },
+    {
+      "epoch": 45.54,
+      "learning_rate": 1.3384671800513386e-06,
+      "loss": 0.8264,
+      "step": 9200
+    },
+    {
+      "epoch": 45.59,
+      "learning_rate": 1.333883388338834e-06,
+      "loss": 0.872,
+      "step": 9210
+    },
+    {
+      "epoch": 45.64,
+      "learning_rate": 1.3292995966263296e-06,
+      "loss": 0.8435,
+      "step": 9220
+    },
+    {
+      "epoch": 45.69,
+      "learning_rate": 1.3247158049138248e-06,
+      "loss": 0.8388,
+      "step": 9230
+    },
+    {
+      "epoch": 45.74,
+      "learning_rate": 1.32013201320132e-06,
+      "loss": 0.8162,
+      "step": 9240
+    },
+    {
+      "epoch": 45.79,
+      "learning_rate": 1.3155482214888158e-06,
+      "loss": 0.8261,
+      "step": 9250
+    },
+    {
+      "epoch": 45.84,
+      "learning_rate": 1.310964429776311e-06,
+      "loss": 0.8444,
+      "step": 9260
+    },
+    {
+      "epoch": 45.89,
+      "learning_rate": 1.3063806380638063e-06,
+      "loss": 0.8326,
+      "step": 9270
+    },
+    {
+      "epoch": 45.94,
+      "learning_rate": 1.301796846351302e-06,
+      "loss": 0.827,
+      "step": 9280
+    },
+    {
+      "epoch": 45.99,
+      "learning_rate": 1.2972130546387973e-06,
+      "loss": 0.8075,
+      "step": 9290
+    },
+    {
+      "epoch": 46.0,
+      "eval_accuracy": 0.7115987460815048,
+      "eval_loss": 0.8132619261741638,
+      "eval_runtime": 37.8814,
+      "eval_samples_per_second": 75.789,
+      "eval_steps_per_second": 2.376,
+      "step": 9292
+    },
+    {
+      "epoch": 46.04,
+      "learning_rate": 1.2926292629262925e-06,
+      "loss": 0.7653,
+      "step": 9300
+    },
+    {
+      "epoch": 46.09,
+      "learning_rate": 1.2880454712137882e-06,
+      "loss": 0.8229,
+      "step": 9310
+    },
+    {
+      "epoch": 46.14,
+      "learning_rate": 1.2834616795012835e-06,
+      "loss": 0.8213,
+      "step": 9320
+    },
+    {
+      "epoch": 46.19,
+      "learning_rate": 1.278877887788779e-06,
+      "loss": 0.8245,
+      "step": 9330
+    },
+    {
+      "epoch": 46.24,
+      "learning_rate": 1.2742940960762745e-06,
+      "loss": 0.8627,
+      "step": 9340
+    },
+    {
+      "epoch": 46.29,
+      "learning_rate": 1.26971030436377e-06,
+      "loss": 0.8067,
+      "step": 9350
+    },
+    {
+      "epoch": 46.34,
+      "learning_rate": 1.2651265126512652e-06,
+      "loss": 0.8071,
+      "step": 9360
+    },
+    {
+      "epoch": 46.39,
+      "learning_rate": 1.2605427209387605e-06,
+      "loss": 0.7775,
+      "step": 9370
+    },
+    {
+      "epoch": 46.44,
+      "learning_rate": 1.2559589292262562e-06,
+      "loss": 0.8895,
+      "step": 9380
+    },
+    {
+      "epoch": 46.49,
+      "learning_rate": 1.2513751375137514e-06,
+      "loss": 0.829,
+      "step": 9390
+    },
+    {
+      "epoch": 46.53,
+      "learning_rate": 1.246791345801247e-06,
+      "loss": 0.8201,
+      "step": 9400
+    },
+    {
+      "epoch": 46.58,
+      "learning_rate": 1.2422075540887422e-06,
+      "loss": 0.8351,
+      "step": 9410
+    },
+    {
+      "epoch": 46.63,
+      "learning_rate": 1.2376237623762377e-06,
+      "loss": 0.8236,
+      "step": 9420
+    },
+    {
+      "epoch": 46.68,
+      "learning_rate": 1.2330399706637331e-06,
+      "loss": 0.8536,
+      "step": 9430
+    },
+    {
+      "epoch": 46.73,
+      "learning_rate": 1.2284561789512284e-06,
+      "loss": 0.8178,
+      "step": 9440
+    },
+    {
+      "epoch": 46.78,
+      "learning_rate": 1.2238723872387239e-06,
+      "loss": 0.8819,
+      "step": 9450
+    },
+    {
+      "epoch": 46.83,
+      "learning_rate": 1.2192885955262194e-06,
+      "loss": 0.8492,
+      "step": 9460
+    },
+    {
+      "epoch": 46.88,
+      "learning_rate": 1.2147048038137149e-06,
+      "loss": 0.8554,
+      "step": 9470
+    },
+    {
+      "epoch": 46.93,
+      "learning_rate": 1.2101210121012103e-06,
+      "loss": 0.8257,
+      "step": 9480
+    },
+    {
+      "epoch": 46.98,
+      "learning_rate": 1.2055372203887056e-06,
+      "loss": 0.8382,
+      "step": 9490
+    },
+    {
+      "epoch": 47.0,
+      "eval_accuracy": 0.7088122605363985,
+      "eval_loss": 0.8145635724067688,
+      "eval_runtime": 37.9073,
+      "eval_samples_per_second": 75.737,
+      "eval_steps_per_second": 2.374,
+      "step": 9494
+    },
+    {
+      "epoch": 47.03,
+      "learning_rate": 1.200953428676201e-06,
+      "loss": 0.8044,
+      "step": 9500
+    },
+    {
+      "epoch": 47.08,
+      "learning_rate": 1.1963696369636966e-06,
+      "loss": 0.8457,
+      "step": 9510
+    },
+    {
+      "epoch": 47.13,
+      "learning_rate": 1.1917858452511918e-06,
+      "loss": 0.8025,
+      "step": 9520
+    },
+    {
+      "epoch": 47.18,
+      "learning_rate": 1.1872020535386873e-06,
+      "loss": 0.8737,
+      "step": 9530
+    },
+    {
+      "epoch": 47.23,
+      "learning_rate": 1.1826182618261828e-06,
+      "loss": 0.8178,
+      "step": 9540
+    },
+    {
+      "epoch": 47.28,
+      "learning_rate": 1.178034470113678e-06,
+      "loss": 0.8104,
+      "step": 9550
+    },
+    {
+      "epoch": 47.33,
+      "learning_rate": 1.1734506784011735e-06,
+      "loss": 0.8155,
+      "step": 9560
+    },
+    {
+      "epoch": 47.38,
+      "learning_rate": 1.168866886688669e-06,
+      "loss": 0.8428,
+      "step": 9570
+    },
+    {
+      "epoch": 47.43,
+      "learning_rate": 1.1642830949761643e-06,
+      "loss": 0.8712,
+      "step": 9580
+    },
+    {
+      "epoch": 47.48,
+      "learning_rate": 1.1596993032636598e-06,
+      "loss": 0.8089,
+      "step": 9590
+    },
+    {
+      "epoch": 47.52,
+      "learning_rate": 1.1551155115511552e-06,
+      "loss": 0.8192,
+      "step": 9600
+    },
+    {
+      "epoch": 47.57,
+      "learning_rate": 1.1505317198386507e-06,
+      "loss": 0.7877,
+      "step": 9610
+    },
+    {
+      "epoch": 47.62,
+      "learning_rate": 1.1459479281261462e-06,
+      "loss": 0.8402,
+      "step": 9620
+    },
+    {
+      "epoch": 47.67,
+      "learning_rate": 1.1413641364136415e-06,
+      "loss": 0.7893,
+      "step": 9630
+    },
+    {
+      "epoch": 47.72,
+      "learning_rate": 1.136780344701137e-06,
+      "loss": 0.7682,
+      "step": 9640
+    },
+    {
+      "epoch": 47.77,
+      "learning_rate": 1.1321965529886322e-06,
+      "loss": 0.8198,
+      "step": 9650
+    },
+    {
+      "epoch": 47.82,
+      "learning_rate": 1.1276127612761277e-06,
+      "loss": 0.8195,
+      "step": 9660
+    },
+    {
+      "epoch": 47.87,
+      "learning_rate": 1.1230289695636232e-06,
+      "loss": 0.8447,
+      "step": 9670
+    },
+    {
+      "epoch": 47.92,
+      "learning_rate": 1.1184451778511184e-06,
+      "loss": 0.8063,
+      "step": 9680
+    },
+    {
+      "epoch": 47.97,
+      "learning_rate": 1.113861386138614e-06,
+      "loss": 0.7501,
+      "step": 9690
+    },
+    {
+      "epoch": 48.0,
+      "eval_accuracy": 0.7112504353883664,
+      "eval_loss": 0.8096461892127991,
+      "eval_runtime": 37.7711,
+      "eval_samples_per_second": 76.011,
+      "eval_steps_per_second": 2.383,
+      "step": 9696
+    },
+    {
+      "epoch": 48.02,
+      "learning_rate": 1.1092775944261094e-06,
+      "loss": 0.8664,
+      "step": 9700
+    },
+    {
+      "epoch": 48.07,
+      "learning_rate": 1.1046938027136047e-06,
+      "loss": 0.8197,
+      "step": 9710
+    },
+    {
+      "epoch": 48.12,
+      "learning_rate": 1.1001100110011001e-06,
+      "loss": 0.8066,
+      "step": 9720
+    },
+    {
+      "epoch": 48.17,
+      "learning_rate": 1.0955262192885956e-06,
+      "loss": 0.7948,
+      "step": 9730
+    },
+    {
+      "epoch": 48.22,
+      "learning_rate": 1.090942427576091e-06,
+      "loss": 0.8346,
+      "step": 9740
+    },
+    {
+      "epoch": 48.27,
+      "learning_rate": 1.0863586358635866e-06,
+      "loss": 0.8366,
+      "step": 9750
+    },
+    {
+      "epoch": 48.32,
+      "learning_rate": 1.0817748441510819e-06,
+      "loss": 0.7899,
+      "step": 9760
+    },
+    {
+      "epoch": 48.37,
+      "learning_rate": 1.0771910524385773e-06,
+      "loss": 0.8406,
+      "step": 9770
+    },
+    {
+      "epoch": 48.42,
+      "learning_rate": 1.0726072607260726e-06,
+      "loss": 0.7741,
+      "step": 9780
+    },
+    {
+      "epoch": 48.47,
+      "learning_rate": 1.068023469013568e-06,
+      "loss": 0.8229,
+      "step": 9790
+    },
+    {
+      "epoch": 48.51,
+      "learning_rate": 1.0634396773010636e-06,
+      "loss": 0.8054,
+      "step": 9800
+    },
+    {
+      "epoch": 48.56,
+      "learning_rate": 1.0588558855885588e-06,
+      "loss": 0.8461,
+      "step": 9810
+    },
+    {
+      "epoch": 48.61,
+      "learning_rate": 1.0542720938760543e-06,
+      "loss": 0.8402,
+      "step": 9820
+    },
+    {
+      "epoch": 48.66,
+      "learning_rate": 1.0496883021635498e-06,
+      "loss": 0.8557,
+      "step": 9830
+    },
+    {
+      "epoch": 48.71,
+      "learning_rate": 1.0451045104510453e-06,
+      "loss": 0.7887,
+      "step": 9840
+    },
+    {
+      "epoch": 48.76,
+      "learning_rate": 1.0405207187385405e-06,
+      "loss": 0.8348,
+      "step": 9850
+    },
+    {
+      "epoch": 48.81,
+      "learning_rate": 1.035936927026036e-06,
+      "loss": 0.8399,
+      "step": 9860
+    },
+    {
+      "epoch": 48.86,
+      "learning_rate": 1.0313531353135315e-06,
+      "loss": 0.8448,
+      "step": 9870
+    },
+    {
+      "epoch": 48.91,
+      "learning_rate": 1.026769343601027e-06,
+      "loss": 0.8333,
+      "step": 9880
+    },
+    {
+      "epoch": 48.96,
+      "learning_rate": 1.0221855518885222e-06,
+      "loss": 0.7859,
+      "step": 9890
+    },
+    {
+      "epoch": 49.0,
+      "eval_accuracy": 0.7081156391501219,
+      "eval_loss": 0.8102298378944397,
+      "eval_runtime": 37.9646,
+      "eval_samples_per_second": 75.623,
+      "eval_steps_per_second": 2.371,
+      "step": 9898
+    },
+    {
+      "epoch": 49.01,
+      "learning_rate": 1.0176017601760177e-06,
+      "loss": 0.8149,
+      "step": 9900
+    },
+    {
+      "epoch": 49.06,
+      "learning_rate": 1.0130179684635132e-06,
+      "loss": 0.826,
+      "step": 9910
+    },
+    {
+      "epoch": 49.11,
+      "learning_rate": 1.0084341767510085e-06,
+      "loss": 0.8114,
+      "step": 9920
+    },
+    {
+      "epoch": 49.16,
+      "learning_rate": 1.003850385038504e-06,
+      "loss": 0.7683,
+      "step": 9930
+    },
+    {
+      "epoch": 49.21,
+      "learning_rate": 9.992665933259992e-07,
+      "loss": 0.9005,
+      "step": 9940
+    },
+    {
+      "epoch": 49.26,
+      "learning_rate": 9.946828016134947e-07,
+      "loss": 0.8769,
+      "step": 9950
+    },
+    {
+      "epoch": 49.31,
+      "learning_rate": 9.900990099009902e-07,
+      "loss": 0.8237,
+      "step": 9960
+    },
+    {
+      "epoch": 49.36,
+      "learning_rate": 9.855152181884857e-07,
+      "loss": 0.8088,
+      "step": 9970
+    },
+    {
+      "epoch": 49.41,
+      "learning_rate": 9.80931426475981e-07,
+      "loss": 0.7938,
+      "step": 9980
+    },
+    {
+      "epoch": 49.46,
+      "learning_rate": 9.763476347634764e-07,
+      "loss": 0.8102,
+      "step": 9990
+    },
+    {
+      "epoch": 49.5,
+      "learning_rate": 9.717638430509719e-07,
+      "loss": 0.8354,
+      "step": 10000
+    },
+    {
+      "epoch": 49.55,
+      "learning_rate": 9.671800513384674e-07,
+      "loss": 0.8149,
+      "step": 10010
+    },
+    {
+      "epoch": 49.6,
+      "learning_rate": 9.625962596259626e-07,
+      "loss": 0.8101,
+      "step": 10020
+    },
+    {
+      "epoch": 49.65,
+      "learning_rate": 9.580124679134581e-07,
+      "loss": 0.8505,
+      "step": 10030
+    },
+    {
+      "epoch": 49.7,
+      "learning_rate": 9.534286762009536e-07,
+      "loss": 0.8208,
+      "step": 10040
+    },
+    {
+      "epoch": 49.75,
+      "learning_rate": 9.488448844884489e-07,
+      "loss": 0.809,
+      "step": 10050
+    },
+    {
+      "epoch": 49.8,
+      "learning_rate": 9.442610927759443e-07,
+      "loss": 0.8357,
+      "step": 10060
+    },
+    {
+      "epoch": 49.85,
+      "learning_rate": 9.396773010634398e-07,
+      "loss": 0.826,
+      "step": 10070
+    },
+    {
+      "epoch": 49.9,
+      "learning_rate": 9.350935093509352e-07,
+      "loss": 0.828,
+      "step": 10080
+    },
+    {
+      "epoch": 49.95,
+      "learning_rate": 9.305097176384307e-07,
+      "loss": 0.7783,
+      "step": 10090
+    },
+    {
+      "epoch": 50.0,
+      "learning_rate": 9.259259259259259e-07,
+      "loss": 0.8195,
+      "step": 10100
+    },
+    {
+      "epoch": 50.0,
+      "eval_accuracy": 0.7084639498432602,
+      "eval_loss": 0.8121119141578674,
+      "eval_runtime": 37.9784,
+      "eval_samples_per_second": 75.596,
+      "eval_steps_per_second": 2.37,
+      "step": 10100
+    },
+    {
+      "epoch": 50.05,
+      "learning_rate": 9.213421342134214e-07,
+      "loss": 0.7962,
+      "step": 10110
+    },
+    {
+      "epoch": 50.1,
+      "learning_rate": 9.167583425009169e-07,
+      "loss": 0.8515,
+      "step": 10120
+    },
+    {
+      "epoch": 50.15,
+      "learning_rate": 9.121745507884123e-07,
+      "loss": 0.8399,
+      "step": 10130
+    },
+    {
+      "epoch": 50.2,
+      "learning_rate": 9.075907590759076e-07,
+      "loss": 0.7957,
+      "step": 10140
+    },
+    {
+      "epoch": 50.25,
+      "learning_rate": 9.03006967363403e-07,
+      "loss": 0.8231,
+      "step": 10150
+    },
+    {
+      "epoch": 50.3,
+      "learning_rate": 8.984231756508985e-07,
+      "loss": 0.8647,
+      "step": 10160
+    },
+    {
+      "epoch": 50.35,
+      "learning_rate": 8.93839383938394e-07,
+      "loss": 0.8273,
+      "step": 10170
+    },
+    {
+      "epoch": 50.4,
+      "learning_rate": 8.892555922258892e-07,
+      "loss": 0.844,
+      "step": 10180
+    },
+    {
+      "epoch": 50.45,
+      "learning_rate": 8.846718005133847e-07,
+      "loss": 0.8458,
+      "step": 10190
+    },
+    {
+      "epoch": 50.5,
+      "learning_rate": 8.800880088008802e-07,
+      "loss": 0.771,
+      "step": 10200
+    },
+    {
+      "epoch": 50.54,
+      "learning_rate": 8.755042170883756e-07,
+      "loss": 0.7902,
+      "step": 10210
+    },
+    {
+      "epoch": 50.59,
+      "learning_rate": 8.709204253758711e-07,
+      "loss": 0.8249,
+      "step": 10220
+    },
+    {
+      "epoch": 50.64,
+      "learning_rate": 8.663366336633663e-07,
+      "loss": 0.7911,
+      "step": 10230
+    },
+    {
+      "epoch": 50.69,
+      "learning_rate": 8.617528419508618e-07,
+      "loss": 0.8097,
+      "step": 10240
+    },
+    {
+      "epoch": 50.74,
+      "learning_rate": 8.571690502383573e-07,
+      "loss": 0.8227,
+      "step": 10250
+    },
+    {
+      "epoch": 50.79,
+      "learning_rate": 8.525852585258527e-07,
+      "loss": 0.8189,
+      "step": 10260
+    },
+    {
+      "epoch": 50.84,
+      "learning_rate": 8.48001466813348e-07,
+      "loss": 0.8313,
+      "step": 10270
+    },
+    {
+      "epoch": 50.89,
+      "learning_rate": 8.434176751008435e-07,
+      "loss": 0.8324,
+      "step": 10280
+    },
+    {
+      "epoch": 50.94,
+      "learning_rate": 8.388338833883389e-07,
+      "loss": 0.7716,
+      "step": 10290
+    },
+    {
+      "epoch": 50.99,
+      "learning_rate": 8.342500916758344e-07,
+      "loss": 0.8397,
+      "step": 10300
+    },
+    {
+      "epoch": 51.0,
+      "eval_accuracy": 0.7098571926158133,
+      "eval_loss": 0.8119975924491882,
+      "eval_runtime": 37.7051,
+      "eval_samples_per_second": 76.144,
+      "eval_steps_per_second": 2.387,
+      "step": 10302
+    },
+    {
+      "epoch": 51.04,
+      "learning_rate": 8.296662999633296e-07,
+      "loss": 0.8302,
+      "step": 10310
+    },
+    {
+      "epoch": 51.09,
+      "learning_rate": 8.250825082508251e-07,
+      "loss": 0.7669,
+      "step": 10320
+    },
+    {
+      "epoch": 51.14,
+      "learning_rate": 8.204987165383206e-07,
+      "loss": 0.7764,
+      "step": 10330
+    },
+    {
+      "epoch": 51.19,
+      "learning_rate": 8.15914924825816e-07,
+      "loss": 0.7947,
+      "step": 10340
+    },
+    {
+      "epoch": 51.24,
+      "learning_rate": 8.113311331133114e-07,
+      "loss": 0.8261,
+      "step": 10350
+    },
+    {
+      "epoch": 51.29,
+      "learning_rate": 8.067473414008069e-07,
+      "loss": 0.8254,
+      "step": 10360
+    },
+    {
+      "epoch": 51.34,
+      "learning_rate": 8.021635496883022e-07,
+      "loss": 0.7947,
+      "step": 10370
+    },
+    {
+      "epoch": 51.39,
+      "learning_rate": 7.975797579757977e-07,
+      "loss": 0.8059,
+      "step": 10380
+    },
+    {
+      "epoch": 51.44,
+      "learning_rate": 7.92995966263293e-07,
+      "loss": 0.8458,
+      "step": 10390
+    },
+    {
+      "epoch": 51.49,
+      "learning_rate": 7.884121745507885e-07,
+      "loss": 0.835,
+      "step": 10400
+    },
+    {
+      "epoch": 51.53,
+      "learning_rate": 7.838283828382839e-07,
+      "loss": 0.8525,
+      "step": 10410
+    },
+    {
+      "epoch": 51.58,
+      "learning_rate": 7.792445911257793e-07,
+      "loss": 0.8166,
+      "step": 10420
+    },
+    {
+      "epoch": 51.63,
+      "learning_rate": 7.746607994132747e-07,
+      "loss": 0.8349,
+      "step": 10430
+    },
+    {
+      "epoch": 51.68,
+      "learning_rate": 7.700770077007702e-07,
+      "loss": 0.8185,
+      "step": 10440
+    },
+    {
+      "epoch": 51.73,
+      "learning_rate": 7.654932159882655e-07,
+      "loss": 0.7536,
+      "step": 10450
+    },
+    {
+      "epoch": 51.78,
+      "learning_rate": 7.60909424275761e-07,
+      "loss": 0.8196,
+      "step": 10460
+    },
+    {
+      "epoch": 51.83,
+      "learning_rate": 7.563256325632564e-07,
+      "loss": 0.8156,
+      "step": 10470
+    },
+    {
+      "epoch": 51.88,
+      "learning_rate": 7.517418408507518e-07,
+      "loss": 0.8065,
+      "step": 10480
+    },
+    {
+      "epoch": 51.93,
+      "learning_rate": 7.471580491382473e-07,
+      "loss": 0.7963,
+      "step": 10490
+    },
+    {
+      "epoch": 51.98,
+      "learning_rate": 7.425742574257426e-07,
+      "loss": 0.8561,
+      "step": 10500
+    },
+    {
+      "epoch": 52.0,
+      "eval_accuracy": 0.7126436781609196,
+      "eval_loss": 0.8088911175727844,
+      "eval_runtime": 37.7933,
+      "eval_samples_per_second": 75.966,
+      "eval_steps_per_second": 2.381,
+      "step": 10504
+    },
+    {
+      "epoch": 52.03,
+      "learning_rate": 7.379904657132381e-07,
+      "loss": 0.8136,
+      "step": 10510
+    },
+    {
+      "epoch": 52.08,
+      "learning_rate": 7.334066740007334e-07,
+      "loss": 0.7859,
+      "step": 10520
+    },
+    {
+      "epoch": 52.13,
+      "learning_rate": 7.288228822882289e-07,
+      "loss": 0.8152,
+      "step": 10530
+    },
+    {
+      "epoch": 52.18,
+      "learning_rate": 7.242390905757243e-07,
+      "loss": 0.8257,
+      "step": 10540
+    },
+    {
+      "epoch": 52.23,
+      "learning_rate": 7.196552988632197e-07,
+      "loss": 0.7502,
+      "step": 10550
+    },
+    {
+      "epoch": 52.28,
+      "learning_rate": 7.150715071507151e-07,
+      "loss": 0.823,
+      "step": 10560
+    },
+    {
+      "epoch": 52.33,
+      "learning_rate": 7.104877154382106e-07,
+      "loss": 0.8275,
+      "step": 10570
+    },
+    {
+      "epoch": 52.38,
+      "learning_rate": 7.059039237257059e-07,
+      "loss": 0.838,
+      "step": 10580
+    },
+    {
+      "epoch": 52.43,
+      "learning_rate": 7.013201320132014e-07,
+      "loss": 0.7744,
+      "step": 10590
+    },
+    {
+      "epoch": 52.48,
+      "learning_rate": 6.967363403006967e-07,
+      "loss": 0.8118,
+      "step": 10600
+    },
+    {
+      "epoch": 52.52,
+      "learning_rate": 6.921525485881922e-07,
+      "loss": 0.8174,
+      "step": 10610
+    },
+    {
+      "epoch": 52.57,
+      "learning_rate": 6.875687568756877e-07,
+      "loss": 0.835,
+      "step": 10620
+    },
+    {
+      "epoch": 52.62,
+      "learning_rate": 6.82984965163183e-07,
+      "loss": 0.8685,
+      "step": 10630
+    },
+    {
+      "epoch": 52.67,
+      "learning_rate": 6.784011734506784e-07,
+      "loss": 0.7815,
+      "step": 10640
+    },
+    {
+      "epoch": 52.72,
+      "learning_rate": 6.738173817381739e-07,
+      "loss": 0.8223,
+      "step": 10650
+    },
+    {
+      "epoch": 52.77,
+      "learning_rate": 6.692335900256693e-07,
+      "loss": 0.8294,
+      "step": 10660
+    },
+    {
+      "epoch": 52.82,
+      "learning_rate": 6.646497983131648e-07,
+      "loss": 0.8011,
+      "step": 10670
+    },
+    {
+      "epoch": 52.87,
+      "learning_rate": 6.6006600660066e-07,
+      "loss": 0.8576,
+      "step": 10680
+    },
+    {
+      "epoch": 52.92,
+      "learning_rate": 6.554822148881555e-07,
+      "loss": 0.8278,
+      "step": 10690
+    },
+    {
+      "epoch": 52.97,
+      "learning_rate": 6.50898423175651e-07,
+      "loss": 0.8082,
+      "step": 10700
+    },
+    {
+      "epoch": 53.0,
+      "eval_accuracy": 0.7133402995471961,
+      "eval_loss": 0.8090488910675049,
+      "eval_runtime": 37.9515,
+      "eval_samples_per_second": 75.649,
+      "eval_steps_per_second": 2.371,
+      "step": 10706
+    },
+    {
+      "epoch": 53.02,
+      "learning_rate": 6.463146314631463e-07,
+      "loss": 0.827,
+      "step": 10710
+    },
+    {
+      "epoch": 53.07,
+      "learning_rate": 6.417308397506418e-07,
+      "loss": 0.7886,
+      "step": 10720
+    },
+    {
+      "epoch": 53.12,
+      "learning_rate": 6.371470480381372e-07,
+      "loss": 0.815,
+      "step": 10730
+    },
+    {
+      "epoch": 53.17,
+      "learning_rate": 6.325632563256326e-07,
+      "loss": 0.7834,
+      "step": 10740
+    },
+    {
+      "epoch": 53.22,
+      "learning_rate": 6.279794646131281e-07,
+      "loss": 0.8121,
+      "step": 10750
+    },
+    {
+      "epoch": 53.27,
+      "learning_rate": 6.233956729006235e-07,
+      "loss": 0.8029,
+      "step": 10760
+    },
+    {
+      "epoch": 53.32,
+      "learning_rate": 6.188118811881188e-07,
+      "loss": 0.7956,
+      "step": 10770
+    },
+    {
+      "epoch": 53.37,
+      "learning_rate": 6.142280894756142e-07,
+      "loss": 0.826,
+      "step": 10780
+    },
+    {
+      "epoch": 53.42,
+      "learning_rate": 6.096442977631097e-07,
+      "loss": 0.7969,
+      "step": 10790
+    },
+    {
+      "epoch": 53.47,
+      "learning_rate": 6.050605060506052e-07,
+      "loss": 0.7705,
+      "step": 10800
+    },
+    {
+      "epoch": 53.51,
+      "learning_rate": 6.004767143381005e-07,
+      "loss": 0.8815,
+      "step": 10810
+    },
+    {
+      "epoch": 53.56,
+      "learning_rate": 5.958929226255959e-07,
+      "loss": 0.7885,
+      "step": 10820
+    },
+    {
+      "epoch": 53.61,
+      "learning_rate": 5.913091309130914e-07,
+      "loss": 0.7814,
+      "step": 10830
+    },
+    {
+      "epoch": 53.66,
+      "learning_rate": 5.867253392005868e-07,
+      "loss": 0.8209,
+      "step": 10840
+    },
+    {
+      "epoch": 53.71,
+      "learning_rate": 5.821415474880821e-07,
+      "loss": 0.8433,
+      "step": 10850
+    },
+    {
+      "epoch": 53.76,
+      "learning_rate": 5.775577557755776e-07,
+      "loss": 0.7888,
+      "step": 10860
+    },
+    {
+      "epoch": 53.81,
+      "learning_rate": 5.729739640630731e-07,
+      "loss": 0.784,
+      "step": 10870
+    },
+    {
+      "epoch": 53.86,
+      "learning_rate": 5.683901723505685e-07,
+      "loss": 0.7682,
+      "step": 10880
+    },
+    {
+      "epoch": 53.91,
+      "learning_rate": 5.638063806380638e-07,
+      "loss": 0.8472,
+      "step": 10890
+    },
+    {
+      "epoch": 53.96,
+      "learning_rate": 5.592225889255592e-07,
+      "loss": 0.8574,
+      "step": 10900
+    },
+    {
+      "epoch": 54.0,
+      "eval_accuracy": 0.7105538140020898,
+      "eval_loss": 0.8086779117584229,
+      "eval_runtime": 37.8423,
+      "eval_samples_per_second": 75.867,
+      "eval_steps_per_second": 2.378,
+      "step": 10908
+    },
+    {
+      "epoch": 54.01,
+      "learning_rate": 5.546387972130547e-07,
+      "loss": 0.813,
+      "step": 10910
+    },
+    {
+      "epoch": 54.06,
+      "learning_rate": 5.500550055005501e-07,
+      "loss": 0.7833,
+      "step": 10920
+    },
+    {
+      "epoch": 54.11,
+      "learning_rate": 5.454712137880456e-07,
+      "loss": 0.8359,
+      "step": 10930
+    },
+    {
+      "epoch": 54.16,
+      "learning_rate": 5.408874220755409e-07,
+      "loss": 0.8052,
+      "step": 10940
+    },
+    {
+      "epoch": 54.21,
+      "learning_rate": 5.363036303630363e-07,
+      "loss": 0.7958,
+      "step": 10950
+    },
+    {
+      "epoch": 54.26,
+      "learning_rate": 5.317198386505318e-07,
+      "loss": 0.8152,
+      "step": 10960
+    },
+    {
+      "epoch": 54.31,
+      "learning_rate": 5.271360469380272e-07,
+      "loss": 0.7896,
+      "step": 10970
+    },
+    {
+      "epoch": 54.36,
+      "learning_rate": 5.225522552255226e-07,
+      "loss": 0.8023,
+      "step": 10980
+    },
+    {
+      "epoch": 54.41,
+      "learning_rate": 5.17968463513018e-07,
+      "loss": 0.7844,
+      "step": 10990
+    },
+    {
+      "epoch": 54.46,
+      "learning_rate": 5.133846718005135e-07,
+      "loss": 0.796,
+      "step": 11000
+    },
+    {
+      "epoch": 54.5,
+      "learning_rate": 5.088008800880089e-07,
+      "loss": 0.8383,
+      "step": 11010
+    },
+    {
+      "epoch": 54.55,
+      "learning_rate": 5.042170883755042e-07,
+      "loss": 0.8616,
+      "step": 11020
+    },
+    {
+      "epoch": 54.6,
+      "learning_rate": 4.996332966629996e-07,
+      "loss": 0.8329,
+      "step": 11030
+    },
+    {
+      "epoch": 54.65,
+      "learning_rate": 4.950495049504951e-07,
+      "loss": 0.7927,
+      "step": 11040
+    },
+    {
+      "epoch": 54.7,
+      "learning_rate": 4.904657132379905e-07,
+      "loss": 0.8107,
+      "step": 11050
+    },
+    {
+      "epoch": 54.75,
+      "learning_rate": 4.858819215254859e-07,
+      "loss": 0.8191,
+      "step": 11060
+    },
+    {
+      "epoch": 54.8,
+      "learning_rate": 4.812981298129813e-07,
+      "loss": 0.7835,
+      "step": 11070
+    },
+    {
+      "epoch": 54.85,
+      "learning_rate": 4.767143381004768e-07,
+      "loss": 0.7889,
+      "step": 11080
+    },
+    {
+      "epoch": 54.9,
+      "learning_rate": 4.7213054638797217e-07,
+      "loss": 0.8122,
+      "step": 11090
+    },
+    {
+      "epoch": 54.95,
+      "learning_rate": 4.675467546754676e-07,
+      "loss": 0.7858,
+      "step": 11100
+    },
+    {
+      "epoch": 55.0,
+      "learning_rate": 4.6296296296296297e-07,
+      "loss": 0.8611,
+      "step": 11110
+    },
+    {
+      "epoch": 55.0,
+      "eval_accuracy": 0.7091605712295368,
+      "eval_loss": 0.8092777132987976,
+      "eval_runtime": 37.9093,
+      "eval_samples_per_second": 75.733,
+      "eval_steps_per_second": 2.374,
+      "step": 11110
+    },
+    {
+      "epoch": 55.05,
+      "learning_rate": 4.5837917125045845e-07,
+      "loss": 0.7974,
+      "step": 11120
+    },
+    {
+      "epoch": 55.1,
+      "learning_rate": 4.537953795379538e-07,
+      "loss": 0.8413,
+      "step": 11130
+    },
+    {
+      "epoch": 55.15,
+      "learning_rate": 4.4921158782544925e-07,
+      "loss": 0.8022,
+      "step": 11140
+    },
+    {
+      "epoch": 55.2,
+      "learning_rate": 4.446277961129446e-07,
+      "loss": 0.8098,
+      "step": 11150
+    },
+    {
+      "epoch": 55.25,
+      "learning_rate": 4.400440044004401e-07,
+      "loss": 0.8127,
+      "step": 11160
+    },
+    {
+      "epoch": 55.3,
+      "learning_rate": 4.3546021268793553e-07,
+      "loss": 0.7874,
+      "step": 11170
+    },
+    {
+      "epoch": 55.35,
+      "learning_rate": 4.308764209754309e-07,
+      "loss": 0.7934,
+      "step": 11180
+    },
+    {
+      "epoch": 55.4,
+      "learning_rate": 4.2629262926292633e-07,
+      "loss": 0.8165,
+      "step": 11190
+    },
+    {
+      "epoch": 55.45,
+      "learning_rate": 4.2170883755042175e-07,
+      "loss": 0.8315,
+      "step": 11200
+    },
+    {
+      "epoch": 55.5,
+      "learning_rate": 4.171250458379172e-07,
+      "loss": 0.8293,
+      "step": 11210
+    },
+    {
+      "epoch": 55.54,
+      "learning_rate": 4.1254125412541255e-07,
+      "loss": 0.7952,
+      "step": 11220
+    },
+    {
+      "epoch": 55.59,
+      "learning_rate": 4.07957462412908e-07,
+      "loss": 0.8268,
+      "step": 11230
+    },
+    {
+      "epoch": 55.64,
+      "learning_rate": 4.0337367070040346e-07,
+      "loss": 0.7734,
+      "step": 11240
+    },
+    {
+      "epoch": 55.69,
+      "learning_rate": 3.9878987898789883e-07,
+      "loss": 0.7453,
+      "step": 11250
+    },
+    {
+      "epoch": 55.74,
+      "learning_rate": 3.9420608727539426e-07,
+      "loss": 0.8319,
+      "step": 11260
+    },
+    {
+      "epoch": 55.79,
+      "learning_rate": 3.8962229556288964e-07,
+      "loss": 0.7978,
+      "step": 11270
+    },
+    {
+      "epoch": 55.84,
+      "learning_rate": 3.850385038503851e-07,
+      "loss": 0.8549,
+      "step": 11280
+    },
+    {
+      "epoch": 55.89,
+      "learning_rate": 3.804547121378805e-07,
+      "loss": 0.8052,
+      "step": 11290
+    },
+    {
+      "epoch": 55.94,
+      "learning_rate": 3.758709204253759e-07,
+      "loss": 0.8086,
+      "step": 11300
+    },
+    {
+      "epoch": 55.99,
+      "learning_rate": 3.712871287128713e-07,
+      "loss": 0.8886,
+      "step": 11310
+    },
+    {
+      "epoch": 56.0,
+      "eval_accuracy": 0.7091605712295368,
+      "eval_loss": 0.8099872469902039,
+      "eval_runtime": 37.9807,
+      "eval_samples_per_second": 75.591,
+      "eval_steps_per_second": 2.37,
+      "step": 11312
+    },
+    {
+      "epoch": 56.04,
+      "learning_rate": 3.667033370003667e-07,
+      "loss": 0.7899,
+      "step": 11320
+    },
+    {
+      "epoch": 56.09,
+      "learning_rate": 3.6211954528786214e-07,
+      "loss": 0.8107,
+      "step": 11330
+    },
+    {
+      "epoch": 56.14,
+      "learning_rate": 3.5753575357535757e-07,
+      "loss": 0.8012,
+      "step": 11340
+    },
+    {
+      "epoch": 56.19,
+      "learning_rate": 3.5295196186285294e-07,
+      "loss": 0.7763,
+      "step": 11350
+    },
+    {
+      "epoch": 56.24,
+      "learning_rate": 3.4836817015034837e-07,
+      "loss": 0.7733,
+      "step": 11360
+    },
+    {
+      "epoch": 56.29,
+      "learning_rate": 3.4378437843784385e-07,
+      "loss": 0.7821,
+      "step": 11370
+    },
+    {
+      "epoch": 56.34,
+      "learning_rate": 3.392005867253392e-07,
+      "loss": 0.8309,
+      "step": 11380
+    },
+    {
+      "epoch": 56.39,
+      "learning_rate": 3.3461679501283465e-07,
+      "loss": 0.842,
+      "step": 11390
+    },
+    {
+      "epoch": 56.44,
+      "learning_rate": 3.3003300330033e-07,
+      "loss": 0.7914,
+      "step": 11400
+    },
+    {
+      "epoch": 56.49,
+      "learning_rate": 3.254492115878255e-07,
+      "loss": 0.7616,
+      "step": 11410
+    },
+    {
+      "epoch": 56.53,
+      "learning_rate": 3.208654198753209e-07,
+      "loss": 0.7778,
+      "step": 11420
+    },
+    {
+      "epoch": 56.58,
+      "learning_rate": 3.162816281628163e-07,
+      "loss": 0.8547,
+      "step": 11430
+    },
+    {
+      "epoch": 56.63,
+      "learning_rate": 3.1169783645031173e-07,
+      "loss": 0.8007,
+      "step": 11440
+    },
+    {
+      "epoch": 56.68,
+      "learning_rate": 3.071140447378071e-07,
+      "loss": 0.8262,
+      "step": 11450
+    },
+    {
+      "epoch": 56.73,
+      "learning_rate": 3.025302530253026e-07,
+      "loss": 0.8248,
+      "step": 11460
+    },
+    {
+      "epoch": 56.78,
+      "learning_rate": 2.9794646131279796e-07,
+      "loss": 0.8365,
+      "step": 11470
+    },
+    {
+      "epoch": 56.83,
+      "learning_rate": 2.933626696002934e-07,
+      "loss": 0.7722,
+      "step": 11480
+    },
+    {
+      "epoch": 56.88,
+      "learning_rate": 2.887788778877888e-07,
+      "loss": 0.8469,
+      "step": 11490
+    },
+    {
+      "epoch": 56.93,
+      "learning_rate": 2.8419508617528424e-07,
+      "loss": 0.7565,
+      "step": 11500
+    },
+    {
+      "epoch": 56.98,
+      "learning_rate": 2.796112944627796e-07,
+      "loss": 0.7857,
+      "step": 11510
+    },
+    {
+      "epoch": 57.0,
+      "eval_accuracy": 0.7133402995471961,
+      "eval_loss": 0.8085740208625793,
+      "eval_runtime": 37.9493,
+      "eval_samples_per_second": 75.654,
+      "eval_steps_per_second": 2.372,
+      "step": 11514
+    },
+    {
+      "epoch": 57.03,
+      "learning_rate": 2.7502750275027504e-07,
+      "loss": 0.7953,
+      "step": 11520
+    },
+    {
+      "epoch": 57.08,
+      "learning_rate": 2.7044371103777046e-07,
+      "loss": 0.8295,
+      "step": 11530
+    },
+    {
+      "epoch": 57.13,
+      "learning_rate": 2.658599193252659e-07,
+      "loss": 0.7976,
+      "step": 11540
+    },
+    {
+      "epoch": 57.18,
+      "learning_rate": 2.612761276127613e-07,
+      "loss": 0.8533,
+      "step": 11550
+    },
+    {
+      "epoch": 57.23,
+      "learning_rate": 2.5669233590025674e-07,
+      "loss": 0.8084,
+      "step": 11560
+    },
+    {
+      "epoch": 57.28,
+      "learning_rate": 2.521085441877521e-07,
+      "loss": 0.8624,
+      "step": 11570
+    },
+    {
+      "epoch": 57.33,
+      "learning_rate": 2.4752475247524754e-07,
+      "loss": 0.7803,
+      "step": 11580
+    },
+    {
+      "epoch": 57.38,
+      "learning_rate": 2.4294096076274297e-07,
+      "loss": 0.8188,
+      "step": 11590
+    },
+    {
+      "epoch": 57.43,
+      "learning_rate": 2.383571690502384e-07,
+      "loss": 0.7965,
+      "step": 11600
+    },
+    {
+      "epoch": 57.48,
+      "learning_rate": 2.337733773377338e-07,
+      "loss": 0.8207,
+      "step": 11610
+    },
+    {
+      "epoch": 57.52,
+      "learning_rate": 2.2918958562522922e-07,
+      "loss": 0.7905,
+      "step": 11620
+    },
+    {
+      "epoch": 57.57,
+      "learning_rate": 2.2460579391272462e-07,
+      "loss": 0.8254,
+      "step": 11630
+    },
+    {
+      "epoch": 57.62,
+      "learning_rate": 2.2002200220022005e-07,
+      "loss": 0.8399,
+      "step": 11640
+    },
+    {
+      "epoch": 57.67,
+      "learning_rate": 2.1543821048771545e-07,
+      "loss": 0.7487,
+      "step": 11650
+    },
+    {
+      "epoch": 57.72,
+      "learning_rate": 2.1085441877521088e-07,
+      "loss": 0.8066,
+      "step": 11660
+    },
+    {
+      "epoch": 57.77,
+      "learning_rate": 2.0627062706270628e-07,
+      "loss": 0.8546,
+      "step": 11670
+    },
+    {
+      "epoch": 57.82,
+      "learning_rate": 2.0168683535020173e-07,
+      "loss": 0.8128,
+      "step": 11680
+    },
+    {
+      "epoch": 57.87,
+      "learning_rate": 1.9710304363769713e-07,
+      "loss": 0.8101,
+      "step": 11690
+    },
+    {
+      "epoch": 57.92,
+      "learning_rate": 1.9251925192519256e-07,
+      "loss": 0.8032,
+      "step": 11700
+    },
+    {
+      "epoch": 57.97,
+      "learning_rate": 1.8793546021268796e-07,
+      "loss": 0.8467,
+      "step": 11710
+    },
+    {
+      "epoch": 58.0,
+      "eval_accuracy": 0.711947056774643,
+      "eval_loss": 0.8083097338676453,
+      "eval_runtime": 37.9449,
+      "eval_samples_per_second": 75.662,
+      "eval_steps_per_second": 2.372,
+      "step": 11716
+    },
+    {
+      "epoch": 58.02,
+      "learning_rate": 1.8335166850018336e-07,
+      "loss": 0.8067,
+      "step": 11720
+    },
+    {
+      "epoch": 58.07,
+      "learning_rate": 1.7876787678767878e-07,
+      "loss": 0.7722,
+      "step": 11730
+    },
+    {
+      "epoch": 58.12,
+      "learning_rate": 1.7418408507517418e-07,
+      "loss": 0.809,
+      "step": 11740
+    },
+    {
+      "epoch": 58.17,
+      "learning_rate": 1.696002933626696e-07,
+      "loss": 0.8205,
+      "step": 11750
+    },
+    {
+      "epoch": 58.22,
+      "learning_rate": 1.65016501650165e-07,
+      "loss": 0.7859,
+      "step": 11760
+    },
+    {
+      "epoch": 58.27,
+      "learning_rate": 1.6043270993766044e-07,
+      "loss": 0.7905,
+      "step": 11770
+    },
+    {
+      "epoch": 58.32,
+      "learning_rate": 1.5584891822515586e-07,
+      "loss": 0.7877,
+      "step": 11780
+    },
+    {
+      "epoch": 58.37,
+      "learning_rate": 1.512651265126513e-07,
+      "loss": 0.8101,
+      "step": 11790
+    },
+    {
+      "epoch": 58.42,
+      "learning_rate": 1.466813348001467e-07,
+      "loss": 0.8343,
+      "step": 11800
+    },
+    {
+      "epoch": 58.47,
+      "learning_rate": 1.4209754308764212e-07,
+      "loss": 0.813,
+      "step": 11810
+    },
+    {
+      "epoch": 58.51,
+      "learning_rate": 1.3751375137513752e-07,
+      "loss": 0.8187,
+      "step": 11820
+    },
+    {
+      "epoch": 58.56,
+      "learning_rate": 1.3292995966263294e-07,
+      "loss": 0.8228,
+      "step": 11830
+    },
+    {
+      "epoch": 58.61,
+      "learning_rate": 1.2834616795012837e-07,
+      "loss": 0.7594,
+      "step": 11840
+    },
+    {
+      "epoch": 58.66,
+      "learning_rate": 1.2376237623762377e-07,
+      "loss": 0.8163,
+      "step": 11850
+    },
+    {
+      "epoch": 58.71,
+      "learning_rate": 1.191785845251192e-07,
+      "loss": 0.7826,
+      "step": 11860
+    },
+    {
+      "epoch": 58.76,
+      "learning_rate": 1.1459479281261461e-07,
+      "loss": 0.764,
+      "step": 11870
+    },
+    {
+      "epoch": 58.81,
+      "learning_rate": 1.1001100110011003e-07,
+      "loss": 0.8094,
+      "step": 11880
+    },
+    {
+      "epoch": 58.86,
+      "learning_rate": 1.0542720938760544e-07,
+      "loss": 0.8254,
+      "step": 11890
+    },
+    {
+      "epoch": 58.91,
+      "learning_rate": 1.0084341767510087e-07,
+      "loss": 0.8663,
+      "step": 11900
+    },
+    {
+      "epoch": 58.96,
+      "learning_rate": 9.625962596259628e-08,
+      "loss": 0.795,
+      "step": 11910
+    },
+    {
+      "epoch": 59.0,
+      "eval_accuracy": 0.711947056774643,
+      "eval_loss": 0.8083323240280151,
+      "eval_runtime": 37.6073,
+      "eval_samples_per_second": 76.342,
+      "eval_steps_per_second": 2.393,
+      "step": 11918
+    },
+    {
+      "epoch": 59.01,
+      "learning_rate": 9.167583425009168e-08,
+      "loss": 0.8674,
+      "step": 11920
+    },
+    {
+      "epoch": 59.06,
+      "learning_rate": 8.709204253758709e-08,
+      "loss": 0.7984,
+      "step": 11930
+    },
+    {
+      "epoch": 59.11,
+      "learning_rate": 8.25082508250825e-08,
+      "loss": 0.8149,
+      "step": 11940
+    },
+    {
+      "epoch": 59.16,
+      "learning_rate": 7.792445911257793e-08,
+      "loss": 0.8072,
+      "step": 11950
+    },
+    {
+      "epoch": 59.21,
+      "learning_rate": 7.334066740007335e-08,
+      "loss": 0.8066,
+      "step": 11960
+    },
+    {
+      "epoch": 59.26,
+      "learning_rate": 6.875687568756876e-08,
+      "loss": 0.7986,
+      "step": 11970
+    },
+    {
+      "epoch": 59.31,
+      "learning_rate": 6.417308397506419e-08,
+      "loss": 0.8466,
+      "step": 11980
+    },
+    {
+      "epoch": 59.36,
+      "learning_rate": 5.95892922625596e-08,
+      "loss": 0.8068,
+      "step": 11990
+    },
+    {
+      "epoch": 59.41,
+      "learning_rate": 5.500550055005501e-08,
+      "loss": 0.7913,
+      "step": 12000
+    },
+    {
+      "epoch": 59.46,
+      "learning_rate": 5.042170883755043e-08,
+      "loss": 0.8158,
+      "step": 12010
+    },
+    {
+      "epoch": 59.5,
+      "learning_rate": 4.583791712504584e-08,
+      "loss": 0.8054,
+      "step": 12020
+    },
+    {
+      "epoch": 59.55,
+      "learning_rate": 4.125412541254125e-08,
+      "loss": 0.7698,
+      "step": 12030
+    },
+    {
+      "epoch": 59.6,
+      "learning_rate": 3.667033370003667e-08,
+      "loss": 0.8143,
+      "step": 12040
+    },
+    {
+      "epoch": 59.65,
+      "learning_rate": 3.208654198753209e-08,
+      "loss": 0.8176,
+      "step": 12050
+    },
+    {
+      "epoch": 59.7,
+      "learning_rate": 2.7502750275027506e-08,
+      "loss": 0.8113,
+      "step": 12060
+    },
+    {
+      "epoch": 59.75,
+      "learning_rate": 2.291895856252292e-08,
+      "loss": 0.7801,
+      "step": 12070
+    },
+    {
+      "epoch": 59.8,
+      "learning_rate": 1.8335166850018336e-08,
+      "loss": 0.8079,
+      "step": 12080
+    },
+    {
+      "epoch": 59.85,
+      "learning_rate": 1.3751375137513753e-08,
+      "loss": 0.8231,
+      "step": 12090
+    },
+    {
+      "epoch": 59.9,
+      "learning_rate": 9.167583425009168e-09,
+      "loss": 0.77,
+      "step": 12100
+    },
+    {
+      "epoch": 59.95,
+      "learning_rate": 4.583791712504584e-09,
+      "loss": 0.7914,
+      "step": 12110
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 0.0,
+      "loss": 0.7975,
+      "step": 12120
+    },
+    {
+      "epoch": 60.0,
+      "eval_accuracy": 0.7133402995471961,
+      "eval_loss": 0.8079430460929871,
+      "eval_runtime": 37.8041,
+      "eval_samples_per_second": 75.944,
+      "eval_steps_per_second": 2.381,
+      "step": 12120
+    },
+    {
+      "epoch": 60.0,
+      "step": 12120,
+      "total_flos": 1.2008831849948586e+20,
+      "train_loss": 0.9543137445308194,
+      "train_runtime": 57078.0116,
+      "train_samples_per_second": 27.161,
+      "train_steps_per_second": 0.212
     }
   ],
   "logging_steps": 10,
-  "max_steps": 4040,
-  "num_train_epochs": 20,
+  "max_steps": 12120,
+  "num_train_epochs": 60,
   "save_steps": 500,
-  "total_flos": 4.0029439499828675e+19,
+  "total_flos": 1.2008831849948586e+20,
   "trial_name": null,
   "trial_params": null
 }