diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,6 +1,6 @@
 {
   "best_metric": 0.8695652173913043,
-  "best_model_checkpoint": "CTMAE-P2-V4-S1/checkpoint-2489",
+  "best_model_checkpoint": "CTMAE-P2-V4-S1/checkpoint-5109",
   "epoch": 49.01246153846154,
   "eval_steps": 500,
   "global_step": 6500,
@@ -10,5029 +10,5029 @@
   "log_history": [
     {
       "epoch": 0.0015384615384615385,
-      "grad_norm": 6.0519328117370605,
+      "grad_norm": 5.231192111968994,
       "learning_rate": 1.5384615384615387e-07,
-      "loss": 0.7005,
+      "loss": 0.6572,
       "step": 10
     },
     {
       "epoch": 0.003076923076923077,
-      "grad_norm": 1.9966832399368286,
+      "grad_norm": 2.369296073913574,
       "learning_rate": 3.0769230769230774e-07,
-      "loss": 0.6944,
+      "loss": 0.6691,
       "step": 20
     },
     {
       "epoch": 0.004615384615384616,
-      "grad_norm": 2.0280706882476807,
+      "grad_norm": 1.869748830795288,
       "learning_rate": 4.615384615384616e-07,
-      "loss": 0.6865,
+      "loss": 0.6819,
       "step": 30
     },
     {
       "epoch": 0.006153846153846154,
-      "grad_norm": 3.468716621398926,
+      "grad_norm": 3.1415841579437256,
       "learning_rate": 6.153846153846155e-07,
-      "loss": 0.677,
+      "loss": 0.6515,
       "step": 40
     },
     {
       "epoch": 0.007692307692307693,
-      "grad_norm": 6.667259693145752,
+      "grad_norm": 7.882626533508301,
       "learning_rate": 7.692307692307694e-07,
-      "loss": 0.6879,
+      "loss": 0.6806,
       "step": 50
     },
     {
       "epoch": 0.009230769230769232,
-      "grad_norm": 2.7107460498809814,
+      "grad_norm": 4.328676223754883,
       "learning_rate": 9.230769230769232e-07,
-      "loss": 0.6369,
+      "loss": 0.6101,
       "step": 60
     },
     {
       "epoch": 0.010769230769230769,
-      "grad_norm": 5.00808048248291,
+      "grad_norm": 5.356165409088135,
       "learning_rate": 1.076923076923077e-06,
-      "loss": 0.6337,
+      "loss": 0.6203,
       "step": 70
     },
     {
       "epoch": 0.012307692307692308,
-      "grad_norm": 7.976367950439453,
+      "grad_norm": 9.514527320861816,
       "learning_rate": 1.230769230769231e-06,
-      "loss": 0.5953,
+      "loss": 0.5716,
       "step": 80
     },
     {
       "epoch": 0.013846153846153847,
-      "grad_norm": 5.392799377441406,
+      "grad_norm": 5.624985694885254,
       "learning_rate": 1.3846153846153848e-06,
-      "loss": 0.5653,
+      "loss": 0.5457,
       "step": 90
     },
     {
       "epoch": 0.015384615384615385,
-      "grad_norm": 5.476315498352051,
+      "grad_norm": 9.588476181030273,
       "learning_rate": 1.5384615384615387e-06,
-      "loss": 0.6058,
+      "loss": 0.6216,
       "step": 100
     },
     {
       "epoch": 0.016923076923076923,
-      "grad_norm": 6.422147750854492,
+      "grad_norm": 8.239978790283203,
       "learning_rate": 1.6923076923076926e-06,
-      "loss": 0.6694,
+      "loss": 0.6664,
       "step": 110
     },
     {
       "epoch": 0.018461538461538463,
-      "grad_norm": 4.2546281814575195,
+      "grad_norm": 4.471070766448975,
       "learning_rate": 1.8461538461538465e-06,
-      "loss": 0.3622,
+      "loss": 0.3663,
       "step": 120
     },
     {
       "epoch": 0.02,
-      "grad_norm": 6.692765235900879,
+      "grad_norm": 9.988832473754883,
       "learning_rate": 2.0000000000000003e-06,
-      "loss": 0.6432,
+      "loss": 0.6464,
       "step": 130
     },
     {
       "epoch": 0.020153846153846154,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 0.8218211531639099,
-      "eval_runtime": 15.023,
-      "eval_samples_per_second": 3.062,
-      "eval_steps_per_second": 1.531,
+      "eval_loss": 0.8455027937889099,
+      "eval_runtime": 16.4489,
+      "eval_samples_per_second": 2.797,
+      "eval_steps_per_second": 1.398,
       "step": 131
     },
     {
       "epoch": 1.0013846153846153,
-      "grad_norm": 6.260814189910889,
+      "grad_norm": 6.516556262969971,
       "learning_rate": 2.153846153846154e-06,
-      "loss": 0.4914,
+      "loss": 0.5165,
       "step": 140
     },
     {
       "epoch": 1.002923076923077,
-      "grad_norm": 6.606039524078369,
+      "grad_norm": 5.840267658233643,
       "learning_rate": 2.307692307692308e-06,
-      "loss": 0.5981,
+      "loss": 0.5904,
       "step": 150
     },
     {
       "epoch": 1.0044615384615385,
-      "grad_norm": 5.284712314605713,
+      "grad_norm": 5.51566743850708,
       "learning_rate": 2.461538461538462e-06,
-      "loss": 0.9034,
+      "loss": 0.852,
       "step": 160
     },
     {
       "epoch": 1.006,
-      "grad_norm": 3.669586658477783,
+      "grad_norm": 5.656895160675049,
       "learning_rate": 2.615384615384616e-06,
-      "loss": 0.6945,
+      "loss": 0.6753,
       "step": 170
     },
     {
       "epoch": 1.0075384615384615,
-      "grad_norm": 7.1405510902404785,
+      "grad_norm": 4.901610851287842,
       "learning_rate": 2.7692307692307697e-06,
-      "loss": 0.6841,
+      "loss": 0.6832,
       "step": 180
     },
     {
       "epoch": 1.009076923076923,
-      "grad_norm": 16.444698333740234,
+      "grad_norm": 13.044291496276855,
       "learning_rate": 2.9230769230769236e-06,
-      "loss": 0.6695,
+      "loss": 0.6882,
       "step": 190
     },
     {
       "epoch": 1.0106153846153847,
-      "grad_norm": 3.4212985038757324,
+      "grad_norm": 6.622377395629883,
       "learning_rate": 3.0769230769230774e-06,
-      "loss": 0.7139,
+      "loss": 0.6862,
       "step": 200
     },
     {
       "epoch": 1.0121538461538462,
-      "grad_norm": 2.3612794876098633,
+      "grad_norm": 2.9806430339813232,
       "learning_rate": 3.2307692307692313e-06,
-      "loss": 0.6926,
+      "loss": 0.6587,
       "step": 210
     },
     {
       "epoch": 1.0136923076923077,
-      "grad_norm": 25.119522094726562,
+      "grad_norm": 12.589070320129395,
       "learning_rate": 3.384615384615385e-06,
-      "loss": 0.868,
+      "loss": 0.7686,
       "step": 220
     },
     {
       "epoch": 1.0152307692307692,
-      "grad_norm": 19.717683792114258,
+      "grad_norm": 25.83409881591797,
       "learning_rate": 3.538461538461539e-06,
-      "loss": 0.3797,
+      "loss": 0.326,
       "step": 230
     },
     {
       "epoch": 1.0167692307692309,
-      "grad_norm": 20.694677352905273,
+      "grad_norm": 23.816606521606445,
       "learning_rate": 3.692307692307693e-06,
-      "loss": 1.3772,
+      "loss": 1.2987,
       "step": 240
     },
     {
       "epoch": 1.0183076923076924,
-      "grad_norm": 1.1371208429336548,
+      "grad_norm": 1.1379772424697876,
       "learning_rate": 3.846153846153847e-06,
-      "loss": 0.9887,
+      "loss": 0.9808,
       "step": 250
     },
     {
       "epoch": 1.0198461538461538,
-      "grad_norm": 0.3055202066898346,
+      "grad_norm": 0.29118087887763977,
       "learning_rate": 4.000000000000001e-06,
-      "loss": 0.4243,
+      "loss": 0.4219,
       "step": 260
     },
     {
       "epoch": 1.0201538461538462,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 2.131279706954956,
-      "eval_runtime": 13.9504,
-      "eval_samples_per_second": 3.297,
-      "eval_steps_per_second": 1.649,
+      "eval_loss": 2.1244828701019287,
+      "eval_runtime": 15.389,
+      "eval_samples_per_second": 2.989,
+      "eval_steps_per_second": 1.495,
       "step": 262
     },
     {
       "epoch": 2.0012307692307694,
-      "grad_norm": 19.514455795288086,
+      "grad_norm": 60.733497619628906,
       "learning_rate": 4.1538461538461545e-06,
-      "loss": 1.8913,
+      "loss": 1.9812,
       "step": 270
     },
     {
       "epoch": 2.0027692307692306,
-      "grad_norm": 1.755078673362732,
+      "grad_norm": 0.9404956698417664,
       "learning_rate": 4.307692307692308e-06,
-      "loss": 0.9929,
+      "loss": 1.1022,
       "step": 280
     },
     {
       "epoch": 2.0043076923076923,
-      "grad_norm": 16.61043930053711,
+      "grad_norm": 13.872081756591797,
       "learning_rate": 4.461538461538462e-06,
-      "loss": 1.1395,
+      "loss": 1.1193,
       "step": 290
     },
     {
       "epoch": 2.005846153846154,
-      "grad_norm": 7.139761447906494,
+      "grad_norm": 3.828695058822632,
       "learning_rate": 4.615384615384616e-06,
-      "loss": 0.6417,
+      "loss": 0.6174,
       "step": 300
     },
     {
       "epoch": 2.0073846153846153,
-      "grad_norm": 7.5372395515441895,
+      "grad_norm": 5.568641185760498,
       "learning_rate": 4.76923076923077e-06,
-      "loss": 0.5172,
+      "loss": 0.5336,
       "step": 310
     },
     {
       "epoch": 2.008923076923077,
-      "grad_norm": 13.364325523376465,
+      "grad_norm": 22.636089324951172,
       "learning_rate": 4.923076923076924e-06,
-      "loss": 0.571,
+      "loss": 0.6174,
       "step": 320
     },
     {
       "epoch": 2.0104615384615383,
-      "grad_norm": 16.03362464904785,
+      "grad_norm": 20.548023223876953,
       "learning_rate": 5.076923076923077e-06,
-      "loss": 0.7023,
+      "loss": 0.7026,
       "step": 330
     },
     {
       "epoch": 2.012,
-      "grad_norm": 1.2387570142745972,
+      "grad_norm": 0.8045111298561096,
       "learning_rate": 5.230769230769232e-06,
-      "loss": 1.0144,
+      "loss": 1.0882,
       "step": 340
     },
     {
       "epoch": 2.0135384615384617,
-      "grad_norm": 0.9126087427139282,
+      "grad_norm": 3.381517171859741,
       "learning_rate": 5.384615384615385e-06,
-      "loss": 0.7244,
+      "loss": 0.7211,
       "step": 350
     },
     {
       "epoch": 2.015076923076923,
-      "grad_norm": 15.709758758544922,
+      "grad_norm": 37.99066162109375,
       "learning_rate": 5.538461538461539e-06,
-      "loss": 1.1351,
+      "loss": 1.1209,
       "step": 360
     },
     {
       "epoch": 2.0166153846153847,
-      "grad_norm": 1.8802133798599243,
+      "grad_norm": 9.296101570129395,
       "learning_rate": 5.692307692307692e-06,
-      "loss": 1.0503,
+      "loss": 1.0455,
       "step": 370
     },
     {
       "epoch": 2.018153846153846,
-      "grad_norm": 0.583645224571228,
+      "grad_norm": 1.0250704288482666,
       "learning_rate": 5.846153846153847e-06,
-      "loss": 0.4027,
+      "loss": 0.3871,
       "step": 380
     },
     {
       "epoch": 2.0196923076923077,
-      "grad_norm": 30.528959274291992,
+      "grad_norm": 44.13499450683594,
       "learning_rate": 6e-06,
-      "loss": 1.0867,
+      "loss": 1.1348,
       "step": 390
     },
     {
       "epoch": 2.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.9686189889907837,
-      "eval_runtime": 13.9707,
-      "eval_samples_per_second": 3.293,
-      "eval_steps_per_second": 1.646,
+      "eval_loss": 2.1032562255859375,
+      "eval_runtime": 15.3715,
+      "eval_samples_per_second": 2.993,
+      "eval_steps_per_second": 1.496,
       "step": 393
     },
     {
       "epoch": 3.001076923076923,
-      "grad_norm": 0.5618914365768433,
+      "grad_norm": 0.8588966131210327,
       "learning_rate": 6.153846153846155e-06,
-      "loss": 1.1478,
+      "loss": 1.2127,
       "step": 400
     },
     {
       "epoch": 3.0026153846153845,
-      "grad_norm": 14.351670265197754,
+      "grad_norm": 15.664327621459961,
       "learning_rate": 6.307692307692308e-06,
-      "loss": 0.6091,
+      "loss": 0.6752,
       "step": 410
     },
     {
       "epoch": 3.004153846153846,
-      "grad_norm": 1.4651223421096802,
+      "grad_norm": 0.8547345399856567,
       "learning_rate": 6.461538461538463e-06,
-      "loss": 0.9063,
+      "loss": 1.0194,
       "step": 420
     },
     {
       "epoch": 3.005692307692308,
-      "grad_norm": 1.2253414392471313,
+      "grad_norm": 1.0899734497070312,
       "learning_rate": 6.615384615384616e-06,
-      "loss": 1.2102,
+      "loss": 1.1903,
       "step": 430
     },
     {
       "epoch": 3.007230769230769,
-      "grad_norm": 1.232196569442749,
+      "grad_norm": 2.424067497253418,
       "learning_rate": 6.76923076923077e-06,
-      "loss": 0.5107,
+      "loss": 0.5215,
       "step": 440
     },
     {
       "epoch": 3.008769230769231,
-      "grad_norm": 15.324777603149414,
+      "grad_norm": 16.104740142822266,
       "learning_rate": 6.923076923076923e-06,
-      "loss": 1.1337,
+      "loss": 1.1537,
       "step": 450
     },
     {
       "epoch": 3.010307692307692,
-      "grad_norm": 42.23984909057617,
+      "grad_norm": 28.782960891723633,
       "learning_rate": 7.076923076923078e-06,
-      "loss": 1.1615,
+      "loss": 1.2909,
       "step": 460
     },
     {
       "epoch": 3.011846153846154,
-      "grad_norm": 18.432376861572266,
+      "grad_norm": 16.444820404052734,
       "learning_rate": 7.230769230769231e-06,
-      "loss": 0.691,
+      "loss": 0.7383,
       "step": 470
     },
     {
       "epoch": 3.0133846153846156,
-      "grad_norm": 0.8665348887443542,
+      "grad_norm": 0.6579256057739258,
       "learning_rate": 7.384615384615386e-06,
-      "loss": 0.8319,
+      "loss": 0.905,
       "step": 480
     },
     {
       "epoch": 3.014923076923077,
-      "grad_norm": 12.59626293182373,
+      "grad_norm": 13.267662048339844,
       "learning_rate": 7.538461538461539e-06,
-      "loss": 1.3428,
+      "loss": 1.3404,
       "step": 490
     },
     {
       "epoch": 3.0164615384615385,
-      "grad_norm": 1.335548996925354,
+      "grad_norm": 1.343061089515686,
       "learning_rate": 7.692307692307694e-06,
-      "loss": 0.6631,
+      "loss": 0.6575,
       "step": 500
     },
     {
       "epoch": 3.018,
-      "grad_norm": 12.23017692565918,
+      "grad_norm": 13.570049285888672,
       "learning_rate": 7.846153846153847e-06,
-      "loss": 1.4626,
+      "loss": 1.5361,
       "step": 510
     },
     {
       "epoch": 3.0195384615384615,
-      "grad_norm": 4.052307605743408,
+      "grad_norm": 5.5172834396362305,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.7155,
+      "loss": 0.7407,
       "step": 520
     },
     {
       "epoch": 3.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 0.8454405665397644,
-      "eval_runtime": 13.1526,
-      "eval_samples_per_second": 3.497,
-      "eval_steps_per_second": 1.749,
+      "eval_loss": 0.8222675919532776,
+      "eval_runtime": 15.2244,
+      "eval_samples_per_second": 3.021,
+      "eval_steps_per_second": 1.511,
       "step": 524
     },
     {
       "epoch": 4.000923076923077,
-      "grad_norm": 2.5005366802215576,
+      "grad_norm": 2.6641976833343506,
       "learning_rate": 8.153846153846154e-06,
-      "loss": 0.3124,
+      "loss": 0.3329,
       "step": 530
     },
     {
       "epoch": 4.002461538461539,
-      "grad_norm": 17.99958610534668,
+      "grad_norm": 12.581295013427734,
       "learning_rate": 8.307692307692309e-06,
-      "loss": 0.88,
+      "loss": 0.8316,
       "step": 540
     },
     {
       "epoch": 4.004,
-      "grad_norm": 21.894472122192383,
+      "grad_norm": 12.648645401000977,
       "learning_rate": 8.461538461538462e-06,
-      "loss": 0.7552,
+      "loss": 0.8462,
       "step": 550
     },
     {
       "epoch": 4.005538461538461,
-      "grad_norm": 4.645376205444336,
+      "grad_norm": 8.547640800476074,
       "learning_rate": 8.615384615384617e-06,
-      "loss": 1.2199,
+      "loss": 1.4294,
       "step": 560
     },
     {
       "epoch": 4.007076923076923,
-      "grad_norm": 4.8306145668029785,
+      "grad_norm": 3.8689968585968018,
       "learning_rate": 8.76923076923077e-06,
-      "loss": 0.6159,
+      "loss": 0.5714,
       "step": 570
     },
     {
       "epoch": 4.008615384615385,
-      "grad_norm": 7.024281978607178,
+      "grad_norm": 6.020469665527344,
       "learning_rate": 8.923076923076925e-06,
-      "loss": 0.5809,
+      "loss": 0.5943,
       "step": 580
     },
     {
       "epoch": 4.010153846153846,
-      "grad_norm": 1.274326205253601,
+      "grad_norm": 1.7420231103897095,
       "learning_rate": 9.076923076923078e-06,
-      "loss": 0.9819,
+      "loss": 0.8042,
       "step": 590
     },
     {
       "epoch": 4.011692307692308,
-      "grad_norm": 11.132002830505371,
+      "grad_norm": 9.07840633392334,
       "learning_rate": 9.230769230769232e-06,
-      "loss": 0.7859,
+      "loss": 0.648,
       "step": 600
     },
     {
       "epoch": 4.013230769230769,
-      "grad_norm": 12.536384582519531,
+      "grad_norm": 16.2183837890625,
       "learning_rate": 9.384615384615385e-06,
-      "loss": 0.6264,
+      "loss": 0.5598,
       "step": 610
     },
     {
       "epoch": 4.014769230769231,
-      "grad_norm": 12.76213550567627,
+      "grad_norm": 14.91943073272705,
       "learning_rate": 9.53846153846154e-06,
-      "loss": 1.412,
+      "loss": 1.6148,
       "step": 620
     },
     {
       "epoch": 4.016307692307692,
-      "grad_norm": 3.126918315887451,
+      "grad_norm": 3.1478230953216553,
       "learning_rate": 9.692307692307693e-06,
-      "loss": 0.6571,
+      "loss": 0.7639,
       "step": 630
     },
     {
       "epoch": 4.017846153846154,
-      "grad_norm": 11.6869478225708,
+      "grad_norm": 13.324525833129883,
       "learning_rate": 9.846153846153848e-06,
-      "loss": 0.3807,
+      "loss": 0.3795,
       "step": 640
     },
     {
       "epoch": 4.019384615384616,
-      "grad_norm": 0.48865261673927307,
+      "grad_norm": 0.5761486887931824,
       "learning_rate": 1e-05,
-      "loss": 1.2743,
+      "loss": 1.2791,
       "step": 650
     },
     {
       "epoch": 4.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.4546117782592773,
-      "eval_runtime": 13.1792,
-      "eval_samples_per_second": 3.49,
-      "eval_steps_per_second": 1.745,
+      "eval_loss": 1.5891335010528564,
+      "eval_runtime": 15.2455,
+      "eval_samples_per_second": 3.017,
+      "eval_steps_per_second": 1.509,
       "step": 655
     },
     {
       "epoch": 5.00076923076923,
-      "grad_norm": 0.90949547290802,
+      "grad_norm": 0.8028017282485962,
       "learning_rate": 9.982905982905984e-06,
-      "loss": 0.726,
+      "loss": 0.6859,
       "step": 660
     },
     {
       "epoch": 5.002307692307692,
-      "grad_norm": 0.4137256145477295,
+      "grad_norm": 0.44530415534973145,
       "learning_rate": 9.965811965811966e-06,
-      "loss": 0.9188,
+      "loss": 0.9753,
       "step": 670
     },
     {
       "epoch": 5.003846153846154,
-      "grad_norm": 9.0480318069458,
+      "grad_norm": 11.103082656860352,
       "learning_rate": 9.94871794871795e-06,
-      "loss": 1.2085,
+      "loss": 1.199,
       "step": 680
     },
     {
       "epoch": 5.0053846153846155,
-      "grad_norm": 16.014402389526367,
+      "grad_norm": 30.11835289001465,
       "learning_rate": 9.931623931623933e-06,
-      "loss": 0.7161,
+      "loss": 0.7103,
       "step": 690
     },
     {
       "epoch": 5.006923076923077,
-      "grad_norm": 1.0252034664154053,
+      "grad_norm": 0.8249630928039551,
       "learning_rate": 9.914529914529915e-06,
-      "loss": 0.9142,
+      "loss": 1.0671,
       "step": 700
     },
     {
       "epoch": 5.008461538461538,
-      "grad_norm": 1.1601945161819458,
+      "grad_norm": 1.280652642250061,
       "learning_rate": 9.897435897435899e-06,
-      "loss": 0.7552,
+      "loss": 0.8142,
       "step": 710
     },
     {
       "epoch": 5.01,
-      "grad_norm": 8.401622772216797,
+      "grad_norm": 11.131413459777832,
       "learning_rate": 9.880341880341882e-06,
-      "loss": 0.8651,
+      "loss": 0.8559,
       "step": 720
     },
     {
       "epoch": 5.0115384615384615,
-      "grad_norm": 1.329608678817749,
+      "grad_norm": 1.8706928491592407,
       "learning_rate": 9.863247863247864e-06,
-      "loss": 0.7054,
+      "loss": 0.6559,
       "step": 730
     },
     {
       "epoch": 5.013076923076923,
-      "grad_norm": 12.047654151916504,
+      "grad_norm": 17.81602668762207,
       "learning_rate": 9.846153846153848e-06,
-      "loss": 0.7337,
+      "loss": 0.791,
       "step": 740
     },
     {
       "epoch": 5.014615384615385,
-      "grad_norm": 8.120343208312988,
+      "grad_norm": 15.149870872497559,
       "learning_rate": 9.82905982905983e-06,
-      "loss": 0.818,
+      "loss": 0.961,
       "step": 750
     },
     {
       "epoch": 5.016153846153846,
-      "grad_norm": 1.3821797370910645,
+      "grad_norm": 1.6111587285995483,
       "learning_rate": 9.811965811965812e-06,
-      "loss": 0.8734,
+      "loss": 1.0334,
       "step": 760
     },
     {
       "epoch": 5.0176923076923075,
-      "grad_norm": 0.37048640847206116,
+      "grad_norm": 0.36878788471221924,
       "learning_rate": 9.794871794871795e-06,
-      "loss": 0.7733,
+      "loss": 0.856,
       "step": 770
     },
     {
       "epoch": 5.019230769230769,
-      "grad_norm": 0.5471999645233154,
+      "grad_norm": 0.4154496490955353,
       "learning_rate": 9.777777777777779e-06,
-      "loss": 0.7952,
+      "loss": 0.8313,
       "step": 780
     },
     {
       "epoch": 5.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.4257819652557373,
-      "eval_runtime": 13.1427,
-      "eval_samples_per_second": 3.5,
-      "eval_steps_per_second": 1.75,
+      "eval_loss": 1.4842634201049805,
+      "eval_runtime": 15.3443,
+      "eval_samples_per_second": 2.998,
+      "eval_steps_per_second": 1.499,
       "step": 786
     },
     {
       "epoch": 6.000615384615385,
-      "grad_norm": 8.841777801513672,
+      "grad_norm": 11.131230354309082,
       "learning_rate": 9.76068376068376e-06,
-      "loss": 0.8073,
+      "loss": 0.8276,
       "step": 790
     },
     {
       "epoch": 6.002153846153846,
-      "grad_norm": 0.8512263298034668,
+      "grad_norm": 0.8737971186637878,
       "learning_rate": 9.743589743589744e-06,
-      "loss": 0.954,
+      "loss": 0.9384,
       "step": 800
     },
     {
       "epoch": 6.003692307692308,
-      "grad_norm": 0.698432207107544,
+      "grad_norm": 0.7840532660484314,
       "learning_rate": 9.726495726495728e-06,
-      "loss": 0.918,
+      "loss": 0.8842,
       "step": 810
     },
     {
       "epoch": 6.005230769230769,
-      "grad_norm": 1.1482590436935425,
+      "grad_norm": 1.5278918743133545,
       "learning_rate": 9.70940170940171e-06,
-      "loss": 1.1294,
+      "loss": 1.1051,
       "step": 820
     },
     {
       "epoch": 6.006769230769231,
-      "grad_norm": 0.6176318526268005,
+      "grad_norm": 0.3794391453266144,
       "learning_rate": 9.692307692307693e-06,
-      "loss": 0.538,
+      "loss": 0.5665,
       "step": 830
     },
     {
       "epoch": 6.008307692307692,
-      "grad_norm": 16.801511764526367,
+      "grad_norm": 17.662307739257812,
       "learning_rate": 9.675213675213677e-06,
-      "loss": 1.0966,
+      "loss": 1.0918,
       "step": 840
     },
     {
       "epoch": 6.009846153846154,
-      "grad_norm": 0.6106650829315186,
+      "grad_norm": 0.5820503234863281,
       "learning_rate": 9.658119658119659e-06,
-      "loss": 0.6375,
+      "loss": 0.6169,
       "step": 850
     },
     {
       "epoch": 6.011384615384616,
-      "grad_norm": 1.5337104797363281,
+      "grad_norm": 1.5412383079528809,
       "learning_rate": 9.641025641025642e-06,
-      "loss": 1.0615,
+      "loss": 1.1419,
       "step": 860
     },
     {
       "epoch": 6.012923076923077,
-      "grad_norm": 1.1498308181762695,
+      "grad_norm": 1.7280353307724,
       "learning_rate": 9.623931623931626e-06,
-      "loss": 0.5056,
+      "loss": 0.4997,
       "step": 870
     },
     {
       "epoch": 6.014461538461538,
-      "grad_norm": 1.3140219449996948,
+      "grad_norm": 1.3223026990890503,
       "learning_rate": 9.606837606837607e-06,
-      "loss": 0.8228,
+      "loss": 0.7593,
       "step": 880
     },
     {
       "epoch": 6.016,
-      "grad_norm": 0.12818345427513123,
+      "grad_norm": 0.32935842871665955,
       "learning_rate": 9.589743589743591e-06,
-      "loss": 0.2663,
+      "loss": 0.2595,
       "step": 890
     },
     {
       "epoch": 6.017538461538462,
-      "grad_norm": 1.8986399173736572,
+      "grad_norm": 0.42623624205589294,
       "learning_rate": 9.572649572649575e-06,
-      "loss": 1.5698,
+      "loss": 1.5931,
       "step": 900
     },
     {
       "epoch": 6.019076923076923,
-      "grad_norm": 3.083142042160034,
+      "grad_norm": 1.9000190496444702,
       "learning_rate": 9.555555555555556e-06,
-      "loss": 0.7364,
+      "loss": 1.0405,
       "step": 910
     },
     {
       "epoch": 6.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.2200851440429688,
-      "eval_runtime": 11.8352,
-      "eval_samples_per_second": 3.887,
-      "eval_steps_per_second": 1.943,
+      "eval_loss": 1.59943687915802,
+      "eval_runtime": 14.3376,
+      "eval_samples_per_second": 3.208,
+      "eval_steps_per_second": 1.604,
       "step": 917
     },
     {
       "epoch": 7.000461538461538,
-      "grad_norm": 0.997331976890564,
+      "grad_norm": 0.6575685143470764,
       "learning_rate": 9.53846153846154e-06,
-      "loss": 0.5659,
+      "loss": 0.6591,
       "step": 920
     },
     {
       "epoch": 7.002,
-      "grad_norm": 1.0925484895706177,
+      "grad_norm": 0.7103784680366516,
       "learning_rate": 9.521367521367522e-06,
-      "loss": 0.8216,
+      "loss": 0.9918,
       "step": 930
     },
     {
       "epoch": 7.0035384615384615,
-      "grad_norm": 0.649455726146698,
+      "grad_norm": 0.5410154461860657,
       "learning_rate": 9.504273504273504e-06,
-      "loss": 0.9357,
+      "loss": 0.9483,
       "step": 940
     },
     {
       "epoch": 7.005076923076923,
-      "grad_norm": 10.865209579467773,
+      "grad_norm": 18.393903732299805,
       "learning_rate": 9.487179487179487e-06,
-      "loss": 0.786,
+      "loss": 0.8155,
       "step": 950
     },
     {
       "epoch": 7.006615384615385,
-      "grad_norm": 0.565341055393219,
+      "grad_norm": 0.3362399935722351,
       "learning_rate": 9.470085470085471e-06,
-      "loss": 1.0838,
+      "loss": 1.1243,
       "step": 960
     },
     {
       "epoch": 7.008153846153846,
-      "grad_norm": 11.902983665466309,
+      "grad_norm": 11.037163734436035,
       "learning_rate": 9.452991452991453e-06,
-      "loss": 1.0737,
+      "loss": 1.0532,
       "step": 970
     },
     {
       "epoch": 7.0096923076923074,
-      "grad_norm": 0.8235186338424683,
+      "grad_norm": 1.363205909729004,
       "learning_rate": 9.435897435897436e-06,
-      "loss": 0.705,
+      "loss": 0.5738,
       "step": 980
     },
     {
       "epoch": 7.011230769230769,
-      "grad_norm": 0.6383252739906311,
+      "grad_norm": 0.8769473433494568,
       "learning_rate": 9.41880341880342e-06,
-      "loss": 1.0167,
+      "loss": 0.9794,
       "step": 990
     },
     {
       "epoch": 7.012769230769231,
-      "grad_norm": 0.4198904037475586,
+      "grad_norm": 0.2998683750629425,
       "learning_rate": 9.401709401709402e-06,
-      "loss": 0.4029,
+      "loss": 0.4462,
       "step": 1000
     },
     {
       "epoch": 7.014307692307693,
-      "grad_norm": 0.6520665287971497,
+      "grad_norm": 0.7185797691345215,
       "learning_rate": 9.384615384615385e-06,
-      "loss": 0.8384,
+      "loss": 0.9716,
       "step": 1010
     },
     {
       "epoch": 7.015846153846154,
-      "grad_norm": 0.3294767737388611,
+      "grad_norm": 1.1516577005386353,
       "learning_rate": 9.367521367521369e-06,
-      "loss": 1.1558,
+      "loss": 1.2309,
       "step": 1020
     },
     {
       "epoch": 7.017384615384615,
-      "grad_norm": 12.820337295532227,
+      "grad_norm": 15.251852035522461,
       "learning_rate": 9.35042735042735e-06,
-      "loss": 0.6545,
+      "loss": 0.6877,
       "step": 1030
     },
     {
       "epoch": 7.018923076923077,
-      "grad_norm": 1.148991584777832,
+      "grad_norm": 0.695763349533081,
       "learning_rate": 9.333333333333334e-06,
-      "loss": 1.4976,
+      "loss": 1.5845,
       "step": 1040
     },
     {
       "epoch": 7.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 0.9824587106704712,
-      "eval_runtime": 11.7676,
-      "eval_samples_per_second": 3.909,
-      "eval_steps_per_second": 1.955,
+      "eval_loss": 1.0656368732452393,
+      "eval_runtime": 14.3595,
+      "eval_samples_per_second": 3.203,
+      "eval_steps_per_second": 1.602,
       "step": 1048
     },
     {
       "epoch": 8.000307692307691,
-      "grad_norm": 2.197751045227051,
+      "grad_norm": 2.3898580074310303,
       "learning_rate": 9.316239316239318e-06,
-      "loss": 0.6061,
+      "loss": 0.6178,
       "step": 1050
     },
     {
       "epoch": 8.001846153846154,
-      "grad_norm": 0.49791476130485535,
+      "grad_norm": 0.4845191538333893,
       "learning_rate": 9.2991452991453e-06,
-      "loss": 0.6114,
+      "loss": 0.5488,
       "step": 1060
     },
     {
       "epoch": 8.003384615384615,
-      "grad_norm": 0.7892986536026001,
+      "grad_norm": 0.49041852355003357,
       "learning_rate": 9.282051282051283e-06,
-      "loss": 1.132,
+      "loss": 1.1417,
       "step": 1070
     },
     {
       "epoch": 8.004923076923077,
-      "grad_norm": 0.17233861982822418,
+      "grad_norm": 0.20023146271705627,
       "learning_rate": 9.264957264957267e-06,
-      "loss": 0.2331,
+      "loss": 0.2078,
       "step": 1080
     },
     {
       "epoch": 8.006461538461538,
-      "grad_norm": 27.1361083984375,
+      "grad_norm": 27.033252716064453,
       "learning_rate": 9.247863247863249e-06,
-      "loss": 1.9122,
+      "loss": 2.0153,
       "step": 1090
     },
     {
       "epoch": 8.008,
-      "grad_norm": 3.272629976272583,
+      "grad_norm": 1.9710267782211304,
       "learning_rate": 9.230769230769232e-06,
-      "loss": 0.9478,
+      "loss": 1.1385,
       "step": 1100
     },
     {
       "epoch": 8.009538461538462,
-      "grad_norm": 2.749512195587158,
+      "grad_norm": 1.4016526937484741,
       "learning_rate": 9.213675213675214e-06,
-      "loss": 0.8206,
+      "loss": 0.9465,
       "step": 1110
     },
     {
       "epoch": 8.011076923076923,
-      "grad_norm": 2.485612392425537,
+      "grad_norm": 2.237942934036255,
       "learning_rate": 9.196581196581196e-06,
-      "loss": 0.4048,
+      "loss": 0.3707,
       "step": 1120
     },
     {
       "epoch": 8.012615384615385,
-      "grad_norm": 0.7969468235969543,
+      "grad_norm": 0.6307343244552612,
       "learning_rate": 9.17948717948718e-06,
-      "loss": 0.5164,
+      "loss": 0.5705,
       "step": 1130
     },
     {
       "epoch": 8.014153846153846,
-      "grad_norm": 19.155437469482422,
+      "grad_norm": 22.620988845825195,
       "learning_rate": 9.162393162393163e-06,
-      "loss": 1.4568,
+      "loss": 1.5419,
       "step": 1140
     },
     {
       "epoch": 8.015692307692309,
-      "grad_norm": 9.772501945495605,
+      "grad_norm": 15.816442489624023,
       "learning_rate": 9.145299145299145e-06,
-      "loss": 0.5541,
+      "loss": 0.53,
       "step": 1150
     },
     {
       "epoch": 8.01723076923077,
-      "grad_norm": 0.7671765685081482,
+      "grad_norm": 0.6915560364723206,
       "learning_rate": 9.128205128205129e-06,
-      "loss": 0.7012,
+      "loss": 0.7185,
       "step": 1160
     },
     {
       "epoch": 8.01876923076923,
-      "grad_norm": 0.7729051113128662,
+      "grad_norm": 0.9734246730804443,
       "learning_rate": 9.111111111111112e-06,
-      "loss": 0.7766,
+      "loss": 0.7645,
       "step": 1170
     },
     {
       "epoch": 8.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.4507434368133545,
-      "eval_runtime": 12.5475,
-      "eval_samples_per_second": 3.666,
-      "eval_steps_per_second": 1.833,
+      "eval_loss": 1.7169452905654907,
+      "eval_runtime": 14.2676,
+      "eval_samples_per_second": 3.224,
+      "eval_steps_per_second": 1.612,
       "step": 1179
     },
     {
       "epoch": 9.000153846153847,
-      "grad_norm": 0.7271968126296997,
+      "grad_norm": 0.8301605582237244,
       "learning_rate": 9.094017094017094e-06,
-      "loss": 0.8353,
+      "loss": 0.7664,
       "step": 1180
     },
     {
       "epoch": 9.001692307692307,
-      "grad_norm": 10.993083000183105,
+      "grad_norm": 14.224320411682129,
       "learning_rate": 9.076923076923078e-06,
-      "loss": 0.8165,
+      "loss": 0.9788,
       "step": 1190
     },
     {
       "epoch": 9.00323076923077,
-      "grad_norm": 0.34806424379348755,
+      "grad_norm": 0.36506837606430054,
       "learning_rate": 9.059829059829061e-06,
-      "loss": 0.8572,
+      "loss": 0.9742,
       "step": 1200
     },
     {
       "epoch": 9.00476923076923,
-      "grad_norm": 7.650801181793213,
+      "grad_norm": 9.940993309020996,
       "learning_rate": 9.042735042735043e-06,
-      "loss": 1.6574,
+      "loss": 1.8968,
       "step": 1210
     },
     {
       "epoch": 9.006307692307692,
-      "grad_norm": 1.8591915369033813,
+      "grad_norm": 1.4142972230911255,
       "learning_rate": 9.025641025641027e-06,
-      "loss": 0.7499,
+      "loss": 0.798,
       "step": 1220
     },
     {
       "epoch": 9.007846153846154,
-      "grad_norm": 5.587526798248291,
+      "grad_norm": 7.723425388336182,
       "learning_rate": 9.00854700854701e-06,
-      "loss": 0.6896,
+      "loss": 0.7826,
       "step": 1230
     },
     {
       "epoch": 9.009384615384615,
-      "grad_norm": 1.4475104808807373,
+      "grad_norm": 1.1705164909362793,
       "learning_rate": 8.991452991452992e-06,
-      "loss": 0.4126,
+      "loss": 0.3885,
       "step": 1240
     },
     {
       "epoch": 9.010923076923078,
-      "grad_norm": 9.723190307617188,
+      "grad_norm": 22.563297271728516,
       "learning_rate": 8.974358974358976e-06,
-      "loss": 0.2476,
+      "loss": 0.2718,
       "step": 1250
     },
     {
       "epoch": 9.012461538461539,
-      "grad_norm": 0.20038339495658875,
+      "grad_norm": 0.1944069117307663,
       "learning_rate": 8.957264957264959e-06,
-      "loss": 0.6484,
+      "loss": 0.7178,
       "step": 1260
     },
     {
       "epoch": 9.014,
-      "grad_norm": 0.46997225284576416,
+      "grad_norm": 0.48125672340393066,
       "learning_rate": 8.940170940170941e-06,
-      "loss": 1.1973,
+      "loss": 1.4657,
       "step": 1270
     },
     {
       "epoch": 9.015538461538462,
-      "grad_norm": 7.039483547210693,
+      "grad_norm": 8.400080680847168,
       "learning_rate": 8.923076923076925e-06,
-      "loss": 1.7299,
+      "loss": 1.9048,
       "step": 1280
     },
     {
       "epoch": 9.017076923076923,
-      "grad_norm": 8.236711502075195,
+      "grad_norm": 9.659461975097656,
       "learning_rate": 8.905982905982906e-06,
-      "loss": 0.5039,
+      "loss": 0.547,
       "step": 1290
     },
     {
       "epoch": 9.018615384615385,
-      "grad_norm": 8.963602066040039,
+      "grad_norm": 8.35889720916748,
       "learning_rate": 8.888888888888888e-06,
-      "loss": 0.5795,
+      "loss": 0.5754,
       "step": 1300
     },
     {
       "epoch": 9.020153846153846,
-      "grad_norm": 2.075558662414551,
+      "grad_norm": 2.34293270111084,
       "learning_rate": 8.871794871794872e-06,
-      "loss": 0.8254,
+      "loss": 0.9242,
       "step": 1310
     },
     {
       "epoch": 9.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.0844687223434448,
-      "eval_runtime": 12.4975,
-      "eval_samples_per_second": 3.681,
-      "eval_steps_per_second": 1.84,
+      "eval_loss": 1.1991312503814697,
+      "eval_runtime": 13.2631,
+      "eval_samples_per_second": 3.468,
+      "eval_steps_per_second": 1.734,
       "step": 1310
     },
     {
       "epoch": 10.00153846153846,
-      "grad_norm": 1.5464600324630737,
+      "grad_norm": 1.7834680080413818,
       "learning_rate": 8.854700854700855e-06,
-      "loss": 0.6458,
+      "loss": 0.6969,
       "step": 1320
     },
     {
       "epoch": 10.003076923076923,
-      "grad_norm": 9.952284812927246,
+      "grad_norm": 9.991689682006836,
       "learning_rate": 8.837606837606837e-06,
-      "loss": 0.7211,
+      "loss": 0.8392,
       "step": 1330
     },
     {
       "epoch": 10.004615384615384,
-      "grad_norm": 0.5603816509246826,
+      "grad_norm": 0.44776904582977295,
       "learning_rate": 8.820512820512821e-06,
-      "loss": 0.5634,
+      "loss": 0.6171,
       "step": 1340
     },
     {
       "epoch": 10.006153846153847,
-      "grad_norm": 17.749595642089844,
+      "grad_norm": 17.601219177246094,
       "learning_rate": 8.803418803418804e-06,
-      "loss": 1.0856,
+      "loss": 1.3078,
       "step": 1350
     },
     {
       "epoch": 10.007692307692308,
-      "grad_norm": 16.027507781982422,
+      "grad_norm": 16.06428337097168,
       "learning_rate": 8.786324786324786e-06,
-      "loss": 0.7795,
+      "loss": 0.6834,
       "step": 1360
     },
     {
       "epoch": 10.009230769230768,
-      "grad_norm": 0.3579050600528717,
+      "grad_norm": 1.1437227725982666,
       "learning_rate": 8.76923076923077e-06,
-      "loss": 0.3561,
+      "loss": 0.4063,
       "step": 1370
     },
     {
       "epoch": 10.010769230769231,
-      "grad_norm": 25.084381103515625,
+      "grad_norm": 42.95561981201172,
       "learning_rate": 8.752136752136753e-06,
-      "loss": 1.1388,
+      "loss": 1.3579,
       "step": 1380
     },
     {
       "epoch": 10.012307692307692,
-      "grad_norm": 1.1872502565383911,
+      "grad_norm": 0.9299348592758179,
       "learning_rate": 8.735042735042735e-06,
-      "loss": 0.8342,
+      "loss": 0.9216,
       "step": 1390
     },
     {
       "epoch": 10.013846153846155,
-      "grad_norm": 11.217093467712402,
+      "grad_norm": 18.4874210357666,
       "learning_rate": 8.717948717948719e-06,
-      "loss": 0.6937,
+      "loss": 0.7695,
       "step": 1400
     },
     {
       "epoch": 10.015384615384615,
-      "grad_norm": 8.533183097839355,
+      "grad_norm": 16.704692840576172,
       "learning_rate": 8.700854700854702e-06,
-      "loss": 1.1324,
+      "loss": 1.5006,
       "step": 1410
     },
     {
       "epoch": 10.016923076923076,
-      "grad_norm": 8.81098747253418,
+      "grad_norm": 18.067237854003906,
       "learning_rate": 8.683760683760684e-06,
-      "loss": 1.4037,
+      "loss": 1.5833,
       "step": 1420
     },
     {
       "epoch": 10.018461538461539,
-      "grad_norm": 1.5235704183578491,
+      "grad_norm": 1.3544650077819824,
       "learning_rate": 8.666666666666668e-06,
-      "loss": 0.3098,
+      "loss": 0.3055,
       "step": 1430
     },
     {
       "epoch": 10.02,
-      "grad_norm": 0.6626542210578918,
+      "grad_norm": 0.18036764860153198,
       "learning_rate": 8.649572649572651e-06,
-      "loss": 0.7267,
+      "loss": 0.9458,
       "step": 1440
     },
     {
       "epoch": 10.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.270902395248413,
-      "eval_runtime": 12.4316,
-      "eval_samples_per_second": 3.7,
-      "eval_steps_per_second": 1.85,
+      "eval_loss": 1.6688299179077148,
+      "eval_runtime": 14.2959,
+      "eval_samples_per_second": 3.218,
+      "eval_steps_per_second": 1.609,
       "step": 1441
     },
     {
       "epoch": 11.001384615384616,
-      "grad_norm": 16.215713500976562,
+      "grad_norm": 21.070219039916992,
       "learning_rate": 8.632478632478633e-06,
-      "loss": 1.2159,
+      "loss": 1.2651,
       "step": 1450
     },
     {
       "epoch": 11.002923076923077,
-      "grad_norm": 0.3360421061515808,
+      "grad_norm": 0.4505758583545685,
       "learning_rate": 8.615384615384617e-06,
-      "loss": 0.7116,
+      "loss": 0.6572,
       "step": 1460
     },
     {
       "epoch": 11.004461538461538,
-      "grad_norm": 8.118979454040527,
+      "grad_norm": 14.809083938598633,
       "learning_rate": 8.598290598290599e-06,
-      "loss": 0.6784,
+      "loss": 0.7536,
       "step": 1470
     },
     {
       "epoch": 11.006,
-      "grad_norm": 1.1598076820373535,
+      "grad_norm": 1.0160809755325317,
       "learning_rate": 8.58119658119658e-06,
-      "loss": 1.2064,
+      "loss": 1.1752,
       "step": 1480
     },
     {
       "epoch": 11.007538461538461,
-      "grad_norm": 8.073029518127441,
+      "grad_norm": 22.986021041870117,
       "learning_rate": 8.564102564102564e-06,
-      "loss": 1.0324,
+      "loss": 1.0872,
       "step": 1490
     },
     {
       "epoch": 11.009076923076924,
-      "grad_norm": 2.737980604171753,
+      "grad_norm": 3.7158827781677246,
       "learning_rate": 8.547008547008548e-06,
-      "loss": 0.6106,
+      "loss": 0.6267,
       "step": 1500
     },
     {
       "epoch": 11.010615384615384,
-      "grad_norm": 1.8243392705917358,
+      "grad_norm": 1.3689260482788086,
       "learning_rate": 8.52991452991453e-06,
-      "loss": 0.4851,
+      "loss": 0.4821,
       "step": 1510
     },
     {
       "epoch": 11.012153846153845,
-      "grad_norm": 2.195685863494873,
+      "grad_norm": 0.8232589364051819,
       "learning_rate": 8.512820512820513e-06,
-      "loss": 0.2933,
+      "loss": 0.2661,
       "step": 1520
     },
     {
       "epoch": 11.013692307692308,
-      "grad_norm": 11.78964900970459,
+      "grad_norm": 14.713327407836914,
       "learning_rate": 8.495726495726497e-06,
-      "loss": 1.4347,
+      "loss": 1.5607,
       "step": 1530
     },
     {
       "epoch": 11.015230769230769,
-      "grad_norm": 7.746746063232422,
+      "grad_norm": 5.836471080780029,
       "learning_rate": 8.478632478632479e-06,
-      "loss": 0.6918,
+      "loss": 0.5947,
       "step": 1540
     },
     {
       "epoch": 11.016769230769231,
-      "grad_norm": 0.4824214577674866,
+      "grad_norm": 1.2984594106674194,
       "learning_rate": 8.461538461538462e-06,
-      "loss": 0.3756,
+      "loss": 0.2553,
       "step": 1550
     },
     {
       "epoch": 11.018307692307692,
-      "grad_norm": 7.721612453460693,
+      "grad_norm": 8.778816223144531,
       "learning_rate": 8.444444444444446e-06,
-      "loss": 0.9131,
+      "loss": 0.8028,
       "step": 1560
     },
     {
       "epoch": 11.019846153846153,
-      "grad_norm": 1.2157827615737915,
+      "grad_norm": 2.403917074203491,
       "learning_rate": 8.427350427350428e-06,
-      "loss": 0.6193,
+      "loss": 0.7963,
       "step": 1570
     },
     {
       "epoch": 11.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.2886906862258911,
-      "eval_runtime": 12.4564,
-      "eval_samples_per_second": 3.693,
-      "eval_steps_per_second": 1.846,
+      "eval_loss": 1.4273507595062256,
+      "eval_runtime": 15.4322,
+      "eval_samples_per_second": 2.981,
+      "eval_steps_per_second": 1.49,
       "step": 1572
     },
     {
       "epoch": 12.00123076923077,
-      "grad_norm": 15.795467376708984,
+      "grad_norm": 12.588306427001953,
       "learning_rate": 8.410256410256411e-06,
-      "loss": 0.5223,
+      "loss": 0.5775,
       "step": 1580
     },
     {
       "epoch": 12.00276923076923,
-      "grad_norm": 0.2842244803905487,
+      "grad_norm": 0.550173282623291,
       "learning_rate": 8.393162393162395e-06,
-      "loss": 0.5231,
+      "loss": 0.5131,
       "step": 1590
     },
     {
       "epoch": 12.004307692307693,
-      "grad_norm": 18.358034133911133,
+      "grad_norm": 18.136802673339844,
       "learning_rate": 8.376068376068377e-06,
-      "loss": 0.8485,
+      "loss": 0.8369,
       "step": 1600
     },
     {
       "epoch": 12.005846153846154,
-      "grad_norm": 18.451539993286133,
+      "grad_norm": 32.2064208984375,
       "learning_rate": 8.35897435897436e-06,
-      "loss": 1.1096,
+      "loss": 1.2122,
       "step": 1610
     },
     {
       "epoch": 12.007384615384616,
-      "grad_norm": 14.058935165405273,
+      "grad_norm": 18.612905502319336,
       "learning_rate": 8.341880341880344e-06,
-      "loss": 0.4075,
+      "loss": 0.4848,
       "step": 1620
     },
     {
       "epoch": 12.008923076923077,
-      "grad_norm": 3.4443199634552,
+      "grad_norm": 1.330094337463379,
       "learning_rate": 8.324786324786326e-06,
-      "loss": 0.831,
+      "loss": 0.8161,
       "step": 1630
     },
     {
       "epoch": 12.010461538461538,
-      "grad_norm": 0.17183153331279755,
+      "grad_norm": 0.7111147046089172,
       "learning_rate": 8.307692307692309e-06,
-      "loss": 1.2585,
+      "loss": 1.1669,
       "step": 1640
     },
     {
       "epoch": 12.012,
-      "grad_norm": 0.2953018546104431,
+      "grad_norm": 1.2424266338348389,
       "learning_rate": 8.290598290598293e-06,
-      "loss": 0.4976,
+      "loss": 0.5004,
       "step": 1650
     },
     {
       "epoch": 12.013538461538461,
-      "grad_norm": 14.40168571472168,
+      "grad_norm": 48.19282531738281,
       "learning_rate": 8.273504273504273e-06,
-      "loss": 1.2874,
+      "loss": 1.3147,
       "step": 1660
     },
     {
       "epoch": 12.015076923076924,
-      "grad_norm": 1.2708631753921509,
+      "grad_norm": 1.2261056900024414,
       "learning_rate": 8.256410256410256e-06,
-      "loss": 0.9277,
+      "loss": 0.8555,
       "step": 1670
     },
     {
       "epoch": 12.016615384615385,
-      "grad_norm": 7.030950546264648,
+      "grad_norm": 23.954980850219727,
       "learning_rate": 8.23931623931624e-06,
-      "loss": 0.8064,
+      "loss": 0.904,
       "step": 1680
     },
     {
       "epoch": 12.018153846153846,
-      "grad_norm": 5.090956687927246,
+      "grad_norm": 6.341338157653809,
       "learning_rate": 8.222222222222222e-06,
-      "loss": 0.8262,
+      "loss": 0.8733,
       "step": 1690
     },
     {
       "epoch": 12.019692307692308,
-      "grad_norm": 6.321172714233398,
+      "grad_norm": 13.682536125183105,
       "learning_rate": 8.205128205128205e-06,
-      "loss": 0.637,
+      "loss": 0.6725,
       "step": 1700
     },
     {
       "epoch": 12.020153846153846,
-      "eval_accuracy": 0.5869565217391305,
-      "eval_loss": 0.6063858270645142,
-      "eval_runtime": 11.7832,
-      "eval_samples_per_second": 3.904,
-      "eval_steps_per_second": 1.952,
+      "eval_accuracy": 0.5652173913043478,
+      "eval_loss": 0.6368175148963928,
+      "eval_runtime": 14.2821,
+      "eval_samples_per_second": 3.221,
+      "eval_steps_per_second": 1.61,
       "step": 1703
     },
     {
       "epoch": 13.001076923076923,
-      "grad_norm": 6.917983531951904,
+      "grad_norm": 7.995401382446289,
       "learning_rate": 8.188034188034189e-06,
-      "loss": 0.371,
+      "loss": 0.3643,
       "step": 1710
     },
     {
       "epoch": 13.002615384615385,
-      "grad_norm": 2.6089928150177,
+      "grad_norm": 0.815584659576416,
       "learning_rate": 8.17094017094017e-06,
-      "loss": 0.3077,
+      "loss": 0.2884,
       "step": 1720
     },
     {
       "epoch": 13.004153846153846,
-      "grad_norm": 0.46078240871429443,
+      "grad_norm": 0.2550635039806366,
       "learning_rate": 8.153846153846154e-06,
-      "loss": 0.7638,
+      "loss": 0.8071,
       "step": 1730
     },
     {
       "epoch": 13.005692307692307,
-      "grad_norm": 11.816447257995605,
+      "grad_norm": 12.024386405944824,
       "learning_rate": 8.136752136752138e-06,
-      "loss": 0.9692,
+      "loss": 1.0065,
       "step": 1740
     },
     {
       "epoch": 13.00723076923077,
-      "grad_norm": 3.5472028255462646,
+      "grad_norm": 1.1342324018478394,
       "learning_rate": 8.11965811965812e-06,
-      "loss": 0.4919,
+      "loss": 0.5521,
       "step": 1750
     },
     {
       "epoch": 13.00876923076923,
-      "grad_norm": 2.8847992420196533,
+      "grad_norm": 5.240642547607422,
       "learning_rate": 8.102564102564103e-06,
-      "loss": 1.0981,
+      "loss": 1.0931,
       "step": 1760
     },
     {
       "epoch": 13.010307692307693,
-      "grad_norm": 7.087960720062256,
+      "grad_norm": 13.917285919189453,
       "learning_rate": 8.085470085470087e-06,
-      "loss": 0.7239,
+      "loss": 0.9356,
       "step": 1770
     },
     {
       "epoch": 13.011846153846154,
-      "grad_norm": 2.2250142097473145,
+      "grad_norm": 3.561969041824341,
       "learning_rate": 8.068376068376069e-06,
-      "loss": 0.5984,
+      "loss": 0.624,
       "step": 1780
     },
     {
       "epoch": 13.013384615384615,
-      "grad_norm": 12.928844451904297,
+      "grad_norm": 10.098306655883789,
       "learning_rate": 8.051282051282052e-06,
-      "loss": 0.6177,
+      "loss": 0.8013,
       "step": 1790
     },
     {
       "epoch": 13.014923076923077,
-      "grad_norm": 39.16981887817383,
+      "grad_norm": 26.3494815826416,
       "learning_rate": 8.034188034188036e-06,
-      "loss": 0.9743,
+      "loss": 0.899,
       "step": 1800
     },
     {
       "epoch": 13.016461538461538,
-      "grad_norm": 1.7411730289459229,
+      "grad_norm": 6.523705959320068,
       "learning_rate": 8.017094017094018e-06,
-      "loss": 0.9101,
+      "loss": 0.824,
       "step": 1810
     },
     {
       "epoch": 13.018,
-      "grad_norm": 0.1714581549167633,
+      "grad_norm": 0.18379585444927216,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.5956,
+      "loss": 0.4197,
       "step": 1820
     },
     {
       "epoch": 13.019538461538462,
-      "grad_norm": 18.013525009155273,
+      "grad_norm": 27.26563835144043,
       "learning_rate": 7.982905982905985e-06,
-      "loss": 0.7321,
+      "loss": 0.6568,
       "step": 1830
     },
     {
       "epoch": 13.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 0.9141055941581726,
-      "eval_runtime": 11.8108,
-      "eval_samples_per_second": 3.895,
-      "eval_steps_per_second": 1.947,
+      "eval_loss": 1.1090037822723389,
+      "eval_runtime": 14.3123,
+      "eval_samples_per_second": 3.214,
+      "eval_steps_per_second": 1.607,
       "step": 1834
     },
     {
       "epoch": 14.000923076923076,
-      "grad_norm": 43.33615493774414,
+      "grad_norm": 35.11122131347656,
       "learning_rate": 7.965811965811967e-06,
-      "loss": 1.1067,
+      "loss": 1.036,
       "step": 1840
     },
     {
       "epoch": 14.002461538461539,
-      "grad_norm": 13.135071754455566,
+      "grad_norm": 11.35055160522461,
       "learning_rate": 7.948717948717949e-06,
-      "loss": 0.5216,
+      "loss": 0.5101,
       "step": 1850
     },
     {
       "epoch": 14.004,
-      "grad_norm": 2.0259616374969482,
+      "grad_norm": 0.5444856882095337,
       "learning_rate": 7.931623931623932e-06,
-      "loss": 0.4612,
+      "loss": 0.577,
       "step": 1860
     },
     {
       "epoch": 14.005538461538462,
-      "grad_norm": 8.404983520507812,
+      "grad_norm": 16.805994033813477,
       "learning_rate": 7.914529914529914e-06,
-      "loss": 0.6618,
+      "loss": 0.6073,
       "step": 1870
     },
     {
       "epoch": 14.007076923076923,
-      "grad_norm": 10.403939247131348,
+      "grad_norm": 27.786434173583984,
       "learning_rate": 7.897435897435898e-06,
-      "loss": 0.4837,
+      "loss": 0.4355,
       "step": 1880
     },
     {
       "epoch": 14.008615384615384,
-      "grad_norm": 3.110192060470581,
+      "grad_norm": 6.101508617401123,
       "learning_rate": 7.880341880341881e-06,
-      "loss": 0.2619,
+      "loss": 0.3788,
       "step": 1890
     },
     {
       "epoch": 14.010153846153846,
-      "grad_norm": 61.4323844909668,
+      "grad_norm": 15.304099082946777,
       "learning_rate": 7.863247863247863e-06,
-      "loss": 0.5425,
+      "loss": 0.4415,
       "step": 1900
     },
     {
       "epoch": 14.011692307692307,
-      "grad_norm": 37.54103469848633,
+      "grad_norm": 38.16310119628906,
       "learning_rate": 7.846153846153847e-06,
-      "loss": 0.7557,
+      "loss": 0.6878,
       "step": 1910
     },
     {
       "epoch": 14.01323076923077,
-      "grad_norm": 17.662887573242188,
+      "grad_norm": 26.13823699951172,
       "learning_rate": 7.82905982905983e-06,
-      "loss": 1.6872,
+      "loss": 1.3675,
       "step": 1920
     },
     {
       "epoch": 14.01476923076923,
-      "grad_norm": 6.54232120513916,
+      "grad_norm": 10.720767974853516,
       "learning_rate": 7.811965811965812e-06,
-      "loss": 0.9641,
+      "loss": 0.9347,
       "step": 1930
     },
     {
       "epoch": 14.016307692307691,
-      "grad_norm": 2.722644090652466,
+      "grad_norm": 5.6711249351501465,
       "learning_rate": 7.794871794871796e-06,
-      "loss": 0.0621,
+      "loss": 0.1223,
       "step": 1940
     },
     {
       "epoch": 14.017846153846154,
-      "grad_norm": 0.08367615938186646,
+      "grad_norm": 0.05928150564432144,
       "learning_rate": 7.77777777777778e-06,
-      "loss": 1.3665,
+      "loss": 1.1779,
       "step": 1950
     },
     {
       "epoch": 14.019384615384615,
-      "grad_norm": 0.41343918442726135,
+      "grad_norm": 0.44221627712249756,
       "learning_rate": 7.760683760683761e-06,
-      "loss": 1.0134,
+      "loss": 1.0769,
       "step": 1960
     },
     {
       "epoch": 14.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.6379895210266113,
-      "eval_runtime": 12.2392,
-      "eval_samples_per_second": 3.758,
-      "eval_steps_per_second": 1.879,
+      "eval_loss": 1.7989178895950317,
+      "eval_runtime": 14.2522,
+      "eval_samples_per_second": 3.228,
+      "eval_steps_per_second": 1.614,
       "step": 1965
     },
     {
       "epoch": 15.000769230769231,
-      "grad_norm": 0.2228805422782898,
+      "grad_norm": 0.055725689977407455,
       "learning_rate": 7.743589743589745e-06,
-      "loss": 0.8081,
+      "loss": 0.8038,
       "step": 1970
     },
     {
       "epoch": 15.002307692307692,
-      "grad_norm": 0.21026729047298431,
+      "grad_norm": 0.0996011346578598,
       "learning_rate": 7.726495726495728e-06,
-      "loss": 0.4037,
+      "loss": 0.5037,
       "step": 1980
     },
     {
       "epoch": 15.003846153846155,
-      "grad_norm": 26.223190307617188,
+      "grad_norm": 46.82464599609375,
       "learning_rate": 7.70940170940171e-06,
-      "loss": 1.2921,
+      "loss": 1.2062,
       "step": 1990
     },
     {
       "epoch": 15.005384615384616,
-      "grad_norm": 15.445429801940918,
+      "grad_norm": 28.789623260498047,
       "learning_rate": 7.692307692307694e-06,
-      "loss": 1.4975,
+      "loss": 1.5169,
       "step": 2000
     },
     {
       "epoch": 15.006923076923076,
-      "grad_norm": 4.351694583892822,
+      "grad_norm": 4.082338333129883,
       "learning_rate": 7.675213675213677e-06,
-      "loss": 0.7678,
+      "loss": 1.2396,
       "step": 2010
     },
     {
       "epoch": 15.008461538461539,
-      "grad_norm": 16.3181095123291,
+      "grad_norm": 3.1324188709259033,
       "learning_rate": 7.658119658119659e-06,
-      "loss": 0.3639,
+      "loss": 0.3347,
       "step": 2020
     },
     {
       "epoch": 15.01,
-      "grad_norm": 79.23714447021484,
+      "grad_norm": 21.909648895263672,
       "learning_rate": 7.641025641025641e-06,
-      "loss": 0.7732,
+      "loss": 0.8502,
       "step": 2030
     },
     {
       "epoch": 15.011538461538462,
-      "grad_norm": 0.3189145028591156,
+      "grad_norm": 0.24542367458343506,
       "learning_rate": 7.6239316239316244e-06,
-      "loss": 0.285,
+      "loss": 0.3358,
       "step": 2040
     },
     {
       "epoch": 15.013076923076923,
-      "grad_norm": 0.07052574306726456,
+      "grad_norm": 0.24301926791667938,
       "learning_rate": 7.606837606837607e-06,
-      "loss": 0.5103,
+      "loss": 0.7185,
       "step": 2050
     },
     {
       "epoch": 15.014615384615384,
-      "grad_norm": 8.32340145111084,
+      "grad_norm": 1.348420262336731,
       "learning_rate": 7.58974358974359e-06,
-      "loss": 0.232,
+      "loss": 0.2519,
       "step": 2060
     },
     {
       "epoch": 15.016153846153847,
-      "grad_norm": 30.055326461791992,
+      "grad_norm": 44.476741790771484,
       "learning_rate": 7.572649572649573e-06,
-      "loss": 0.603,
+      "loss": 0.6645,
       "step": 2070
     },
     {
       "epoch": 15.017692307692307,
-      "grad_norm": 21.265649795532227,
+      "grad_norm": 17.186023712158203,
       "learning_rate": 7.555555555555556e-06,
-      "loss": 1.3278,
+      "loss": 1.2717,
       "step": 2080
     },
     {
       "epoch": 15.01923076923077,
-      "grad_norm": 13.572966575622559,
+      "grad_norm": 9.872564315795898,
       "learning_rate": 7.538461538461539e-06,
-      "loss": 0.2139,
+      "loss": 0.2657,
       "step": 2090
     },
     {
       "epoch": 15.020153846153846,
-      "eval_accuracy": 0.6956521739130435,
-      "eval_loss": 0.6075901985168457,
-      "eval_runtime": 12.5118,
-      "eval_samples_per_second": 3.677,
-      "eval_steps_per_second": 1.838,
+      "eval_accuracy": 0.6304347826086957,
+      "eval_loss": 0.7522389888763428,
+      "eval_runtime": 15.3747,
+      "eval_samples_per_second": 2.992,
+      "eval_steps_per_second": 1.496,
       "step": 2096
     },
     {
       "epoch": 16.000615384615383,
-      "grad_norm": 6.703625679016113,
+      "grad_norm": 5.565269470214844,
       "learning_rate": 7.521367521367522e-06,
-      "loss": 0.5604,
+      "loss": 0.6323,
       "step": 2100
     },
     {
       "epoch": 16.002153846153846,
-      "grad_norm": 0.617182731628418,
+      "grad_norm": 5.698361396789551,
       "learning_rate": 7.504273504273505e-06,
-      "loss": 0.561,
+      "loss": 0.5126,
       "step": 2110
     },
     {
       "epoch": 16.003692307692308,
-      "grad_norm": 41.750701904296875,
+      "grad_norm": 58.876426696777344,
       "learning_rate": 7.487179487179488e-06,
-      "loss": 0.8217,
+      "loss": 0.8453,
       "step": 2120
     },
     {
       "epoch": 16.00523076923077,
-      "grad_norm": 0.29348522424697876,
+      "grad_norm": 1.2901254892349243,
       "learning_rate": 7.4700854700854706e-06,
-      "loss": 0.5181,
+      "loss": 0.3286,
       "step": 2130
     },
     {
       "epoch": 16.00676923076923,
-      "grad_norm": 33.394954681396484,
+      "grad_norm": 42.79770278930664,
       "learning_rate": 7.452991452991454e-06,
-      "loss": 0.8665,
+      "loss": 0.6111,
       "step": 2140
     },
     {
       "epoch": 16.008307692307692,
-      "grad_norm": 18.40253448486328,
+      "grad_norm": 18.82196044921875,
       "learning_rate": 7.435897435897437e-06,
-      "loss": 0.4002,
+      "loss": 0.8318,
       "step": 2150
     },
     {
       "epoch": 16.009846153846155,
-      "grad_norm": 61.75335693359375,
+      "grad_norm": 39.72857666015625,
       "learning_rate": 7.4188034188034196e-06,
-      "loss": 0.8184,
+      "loss": 0.7653,
       "step": 2160
     },
     {
       "epoch": 16.011384615384614,
-      "grad_norm": 30.64858627319336,
+      "grad_norm": 13.437899589538574,
       "learning_rate": 7.401709401709402e-06,
-      "loss": 0.9719,
+      "loss": 0.8242,
       "step": 2170
     },
     {
       "epoch": 16.012923076923077,
-      "grad_norm": 20.503198623657227,
+      "grad_norm": 9.645297050476074,
       "learning_rate": 7.384615384615386e-06,
-      "loss": 0.6894,
+      "loss": 0.5432,
       "step": 2180
     },
     {
       "epoch": 16.01446153846154,
-      "grad_norm": 0.41924965381622314,
+      "grad_norm": 1.7769389152526855,
       "learning_rate": 7.3675213675213686e-06,
-      "loss": 0.5238,
+      "loss": 0.3875,
       "step": 2190
     },
     {
       "epoch": 16.016,
-      "grad_norm": 0.09503482282161713,
+      "grad_norm": 1.114364504814148,
       "learning_rate": 7.350427350427351e-06,
-      "loss": 0.3226,
+      "loss": 0.3803,
       "step": 2200
     },
     {
       "epoch": 16.01753846153846,
-      "grad_norm": 15.260120391845703,
+      "grad_norm": 32.540401458740234,
       "learning_rate": 7.333333333333333e-06,
-      "loss": 1.7077,
+      "loss": 1.3934,
       "step": 2210
     },
     {
       "epoch": 16.019076923076923,
-      "grad_norm": 40.093467712402344,
+      "grad_norm": 12.514657974243164,
       "learning_rate": 7.316239316239317e-06,
-      "loss": 0.687,
+      "loss": 0.6413,
       "step": 2220
     },
     {
       "epoch": 16.020153846153846,
       "eval_accuracy": 0.5434782608695652,
-      "eval_loss": 1.1890361309051514,
-      "eval_runtime": 12.5109,
-      "eval_samples_per_second": 3.677,
-      "eval_steps_per_second": 1.838,
+      "eval_loss": 1.5062237977981567,
+      "eval_runtime": 12.2561,
+      "eval_samples_per_second": 3.753,
+      "eval_steps_per_second": 1.877,
       "step": 2227
     },
     {
       "epoch": 17.00046153846154,
-      "grad_norm": 22.440990447998047,
+      "grad_norm": 29.90598487854004,
       "learning_rate": 7.299145299145299e-06,
-      "loss": 0.7403,
+      "loss": 0.9334,
       "step": 2230
     },
     {
       "epoch": 17.002,
-      "grad_norm": 30.42022705078125,
+      "grad_norm": 19.59234046936035,
       "learning_rate": 7.282051282051282e-06,
-      "loss": 0.6101,
+      "loss": 0.5555,
       "step": 2240
     },
     {
       "epoch": 17.00353846153846,
-      "grad_norm": 28.41487693786621,
+      "grad_norm": 29.715717315673828,
       "learning_rate": 7.264957264957266e-06,
-      "loss": 0.6806,
+      "loss": 0.8931,
       "step": 2250
     },
     {
       "epoch": 17.005076923076924,
-      "grad_norm": 0.5429576635360718,
+      "grad_norm": 2.181522846221924,
       "learning_rate": 7.247863247863248e-06,
-      "loss": 0.7311,
+      "loss": 0.6539,
       "step": 2260
     },
     {
       "epoch": 17.006615384615383,
-      "grad_norm": 0.37554216384887695,
+      "grad_norm": 0.40200403332710266,
       "learning_rate": 7.230769230769231e-06,
-      "loss": 0.6523,
+      "loss": 0.6576,
       "step": 2270
     },
     {
       "epoch": 17.008153846153846,
-      "grad_norm": 12.936208724975586,
+      "grad_norm": 2.2605209350585938,
       "learning_rate": 7.213675213675214e-06,
-      "loss": 0.7911,
+      "loss": 0.6304,
       "step": 2280
     },
     {
       "epoch": 17.00969230769231,
-      "grad_norm": 20.977012634277344,
+      "grad_norm": 16.979948043823242,
       "learning_rate": 7.196581196581197e-06,
-      "loss": 0.2984,
+      "loss": 0.2713,
       "step": 2290
     },
     {
       "epoch": 17.01123076923077,
-      "grad_norm": 0.6222103834152222,
+      "grad_norm": 4.920323848724365,
       "learning_rate": 7.17948717948718e-06,
-      "loss": 0.3332,
+      "loss": 0.3698,
       "step": 2300
     },
     {
       "epoch": 17.01276923076923,
-      "grad_norm": 16.443161010742188,
+      "grad_norm": 28.805387496948242,
       "learning_rate": 7.162393162393163e-06,
-      "loss": 0.8002,
+      "loss": 0.8369,
       "step": 2310
     },
     {
       "epoch": 17.014307692307693,
-      "grad_norm": 0.1513749361038208,
+      "grad_norm": 0.8019755482673645,
       "learning_rate": 7.145299145299146e-06,
-      "loss": 0.7041,
+      "loss": 0.656,
       "step": 2320
     },
     {
       "epoch": 17.015846153846155,
-      "grad_norm": 0.09506427496671677,
+      "grad_norm": 0.03935155272483826,
       "learning_rate": 7.128205128205129e-06,
-      "loss": 0.6438,
+      "loss": 0.485,
       "step": 2330
     },
     {
       "epoch": 17.017384615384614,
-      "grad_norm": 14.386686325073242,
+      "grad_norm": 7.895570278167725,
       "learning_rate": 7.111111111111112e-06,
-      "loss": 0.3637,
+      "loss": 0.5234,
       "step": 2340
     },
     {
       "epoch": 17.018923076923077,
-      "grad_norm": 18.201074600219727,
+      "grad_norm": 16.784347534179688,
       "learning_rate": 7.0940170940170945e-06,
-      "loss": 0.8722,
+      "loss": 0.7469,
       "step": 2350
     },
     {
       "epoch": 17.020153846153846,
-      "eval_accuracy": 0.782608695652174,
-      "eval_loss": 0.5483578443527222,
-      "eval_runtime": 12.6125,
-      "eval_samples_per_second": 3.647,
-      "eval_steps_per_second": 1.824,
+      "eval_accuracy": 0.5869565217391305,
+      "eval_loss": 0.9306740164756775,
+      "eval_runtime": 12.275,
+      "eval_samples_per_second": 3.747,
+      "eval_steps_per_second": 1.874,
       "step": 2358
     },
     {
       "epoch": 18.000307692307693,
-      "grad_norm": 28.085227966308594,
+      "grad_norm": 55.89947509765625,
       "learning_rate": 7.076923076923078e-06,
-      "loss": 0.4885,
+      "loss": 0.5654,
       "step": 2360
     },
     {
       "epoch": 18.001846153846152,
-      "grad_norm": 46.327796936035156,
+      "grad_norm": 79.32176971435547,
       "learning_rate": 7.059829059829061e-06,
-      "loss": 0.896,
+      "loss": 1.1086,
       "step": 2370
     },
     {
       "epoch": 18.003384615384615,
-      "grad_norm": 8.360269546508789,
+      "grad_norm": 24.28437042236328,
       "learning_rate": 7.0427350427350435e-06,
-      "loss": 0.3218,
+      "loss": 0.3862,
       "step": 2380
     },
     {
       "epoch": 18.004923076923077,
-      "grad_norm": 16.091289520263672,
+      "grad_norm": 19.3510684967041,
       "learning_rate": 7.025641025641025e-06,
-      "loss": 0.4428,
+      "loss": 0.4597,
       "step": 2390
     },
     {
       "epoch": 18.00646153846154,
-      "grad_norm": 0.039979055523872375,
+      "grad_norm": 0.014105038717389107,
       "learning_rate": 7.008547008547009e-06,
-      "loss": 0.1822,
+      "loss": 0.2573,
       "step": 2400
     },
     {
       "epoch": 18.008,
-      "grad_norm": 0.1585582047700882,
+      "grad_norm": 0.15911158919334412,
       "learning_rate": 6.991452991452992e-06,
-      "loss": 0.4568,
+      "loss": 0.5554,
       "step": 2410
     },
     {
       "epoch": 18.00953846153846,
-      "grad_norm": 25.942386627197266,
+      "grad_norm": 19.263647079467773,
       "learning_rate": 6.974358974358974e-06,
-      "loss": 0.5834,
+      "loss": 0.7673,
       "step": 2420
     },
     {
       "epoch": 18.011076923076924,
-      "grad_norm": 22.262611389160156,
+      "grad_norm": 7.370254993438721,
       "learning_rate": 6.957264957264958e-06,
-      "loss": 0.6944,
+      "loss": 0.736,
       "step": 2430
     },
     {
       "epoch": 18.012615384615383,
-      "grad_norm": 39.91915512084961,
+      "grad_norm": 39.19303512573242,
       "learning_rate": 6.940170940170941e-06,
-      "loss": 0.3754,
+      "loss": 0.3204,
       "step": 2440
     },
     {
       "epoch": 18.014153846153846,
-      "grad_norm": 12.724471092224121,
+      "grad_norm": 11.077559471130371,
       "learning_rate": 6.923076923076923e-06,
-      "loss": 0.4949,
+      "loss": 0.6953,
       "step": 2450
     },
     {
       "epoch": 18.01569230769231,
-      "grad_norm": 0.6947053670883179,
+      "grad_norm": 4.552363395690918,
       "learning_rate": 6.905982905982906e-06,
-      "loss": 0.5062,
+      "loss": 0.4272,
       "step": 2460
     },
     {
       "epoch": 18.017230769230768,
-      "grad_norm": 12.842805862426758,
+      "grad_norm": 2.9337615966796875,
       "learning_rate": 6.88888888888889e-06,
-      "loss": 1.3376,
+      "loss": 0.8707,
       "step": 2470
     },
     {
       "epoch": 18.01876923076923,
-      "grad_norm": 65.57331085205078,
+      "grad_norm": 28.357864379882812,
       "learning_rate": 6.871794871794872e-06,
-      "loss": 1.176,
+      "loss": 0.6873,
       "step": 2480
     },
     {
       "epoch": 18.020153846153846,
-      "eval_accuracy": 0.8695652173913043,
-      "eval_loss": 0.40178215503692627,
-      "eval_runtime": 11.8576,
-      "eval_samples_per_second": 3.879,
-      "eval_steps_per_second": 1.94,
+      "eval_accuracy": 0.717391304347826,
+      "eval_loss": 0.46706026792526245,
+      "eval_runtime": 12.3285,
+      "eval_samples_per_second": 3.731,
+      "eval_steps_per_second": 1.866,
       "step": 2489
     },
     {
       "epoch": 19.000153846153847,
-      "grad_norm": 23.745304107666016,
+      "grad_norm": 35.06557846069336,
       "learning_rate": 6.854700854700855e-06,
-      "loss": 0.5627,
+      "loss": 0.5413,
       "step": 2490
     },
     {
       "epoch": 19.00169230769231,
-      "grad_norm": 7.5470075607299805,
+      "grad_norm": 5.010317325592041,
       "learning_rate": 6.837606837606839e-06,
-      "loss": 0.5329,
+      "loss": 0.4907,
       "step": 2500
     },
     {
       "epoch": 19.00323076923077,
-      "grad_norm": 0.15952041745185852,
+      "grad_norm": 0.34195172786712646,
       "learning_rate": 6.820512820512821e-06,
-      "loss": 0.5293,
+      "loss": 0.4879,
       "step": 2510
     },
     {
       "epoch": 19.00476923076923,
-      "grad_norm": 22.13117790222168,
+      "grad_norm": 47.938201904296875,
       "learning_rate": 6.803418803418804e-06,
-      "loss": 1.0728,
+      "loss": 1.0135,
       "step": 2520
     },
     {
       "epoch": 19.006307692307693,
-      "grad_norm": 3.7636966705322266,
+      "grad_norm": 10.164339065551758,
       "learning_rate": 6.786324786324787e-06,
-      "loss": 0.6426,
+      "loss": 0.5712,
       "step": 2530
     },
     {
       "epoch": 19.007846153846153,
-      "grad_norm": 41.60896682739258,
+      "grad_norm": 17.56395721435547,
       "learning_rate": 6.76923076923077e-06,
-      "loss": 0.4748,
+      "loss": 0.5498,
       "step": 2540
     },
     {
       "epoch": 19.009384615384615,
-      "grad_norm": 2.7363779544830322,
+      "grad_norm": 4.655912399291992,
       "learning_rate": 6.752136752136753e-06,
-      "loss": 0.682,
+      "loss": 0.8304,
       "step": 2550
     },
     {
       "epoch": 19.010923076923078,
-      "grad_norm": 49.60470199584961,
+      "grad_norm": 11.02091121673584,
       "learning_rate": 6.735042735042736e-06,
-      "loss": 0.4322,
+      "loss": 0.3397,
       "step": 2560
     },
     {
       "epoch": 19.012461538461537,
-      "grad_norm": 0.17257747054100037,
+      "grad_norm": 0.09357502311468124,
       "learning_rate": 6.717948717948718e-06,
-      "loss": 0.298,
+      "loss": 0.3724,
       "step": 2570
     },
     {
       "epoch": 19.014,
-      "grad_norm": 45.82680892944336,
+      "grad_norm": 103.91709899902344,
       "learning_rate": 6.700854700854701e-06,
-      "loss": 0.77,
+      "loss": 0.9505,
       "step": 2580
     },
     {
       "epoch": 19.015538461538462,
-      "grad_norm": 68.6201171875,
+      "grad_norm": 38.2786750793457,
       "learning_rate": 6.683760683760684e-06,
-      "loss": 0.6183,
+      "loss": 0.5508,
       "step": 2590
     },
     {
       "epoch": 19.017076923076925,
-      "grad_norm": 7.198949337005615,
+      "grad_norm": 0.7876746654510498,
       "learning_rate": 6.666666666666667e-06,
-      "loss": 0.3093,
+      "loss": 0.3628,
       "step": 2600
     },
     {
       "epoch": 19.018615384615384,
-      "grad_norm": 0.10329917073249817,
+      "grad_norm": 0.5578277707099915,
       "learning_rate": 6.64957264957265e-06,
-      "loss": 0.5356,
+      "loss": 0.3031,
       "step": 2610
     },
     {
       "epoch": 19.020153846153846,
-      "grad_norm": 90.10623168945312,
+      "grad_norm": 87.09941101074219,
       "learning_rate": 6.632478632478633e-06,
-      "loss": 0.8042,
+      "loss": 0.8322,
       "step": 2620
     },
     {
       "epoch": 19.020153846153846,
-      "eval_accuracy": 0.717391304347826,
-      "eval_loss": 0.9282347559928894,
-      "eval_runtime": 11.7547,
-      "eval_samples_per_second": 3.913,
-      "eval_steps_per_second": 1.957,
+      "eval_accuracy": 0.6521739130434783,
+      "eval_loss": 1.0945212841033936,
+      "eval_runtime": 12.3077,
+      "eval_samples_per_second": 3.737,
+      "eval_steps_per_second": 1.869,
       "step": 2620
     },
     {
       "epoch": 20.001538461538463,
-      "grad_norm": 58.79783248901367,
+      "grad_norm": 31.272397994995117,
       "learning_rate": 6.615384615384616e-06,
-      "loss": 0.4324,
+      "loss": 0.4821,
       "step": 2630
     },
     {
       "epoch": 20.00307692307692,
-      "grad_norm": 0.12335860729217529,
+      "grad_norm": 10.369728088378906,
       "learning_rate": 6.598290598290598e-06,
-      "loss": 0.6378,
+      "loss": 0.2458,
       "step": 2640
     },
     {
       "epoch": 20.004615384615384,
-      "grad_norm": 1.2578246593475342,
+      "grad_norm": 0.37299492955207825,
       "learning_rate": 6.581196581196582e-06,
-      "loss": 0.5711,
+      "loss": 0.5177,
       "step": 2650
     },
     {
       "epoch": 20.006153846153847,
-      "grad_norm": 9.80817699432373,
+      "grad_norm": 3.4289143085479736,
       "learning_rate": 6.564102564102565e-06,
-      "loss": 0.2199,
+      "loss": 0.3925,
       "step": 2660
     },
     {
       "epoch": 20.00769230769231,
-      "grad_norm": 0.04995802044868469,
+      "grad_norm": 0.035517238080501556,
       "learning_rate": 6.547008547008547e-06,
-      "loss": 0.5418,
+      "loss": 0.7851,
       "step": 2670
     },
     {
       "epoch": 20.00923076923077,
-      "grad_norm": 76.49468231201172,
+      "grad_norm": 112.83161163330078,
       "learning_rate": 6.529914529914531e-06,
-      "loss": 0.4855,
+      "loss": 0.4781,
       "step": 2680
     },
     {
       "epoch": 20.01076923076923,
-      "grad_norm": 1.1559330224990845,
+      "grad_norm": 0.1585376113653183,
       "learning_rate": 6.512820512820514e-06,
-      "loss": 0.9054,
+      "loss": 0.8505,
       "step": 2690
     },
     {
       "epoch": 20.012307692307694,
-      "grad_norm": 91.41716003417969,
+      "grad_norm": 65.49649047851562,
       "learning_rate": 6.495726495726496e-06,
-      "loss": 0.3235,
+      "loss": 0.3276,
       "step": 2700
     },
     {
       "epoch": 20.013846153846153,
-      "grad_norm": 70.10804748535156,
+      "grad_norm": 21.698440551757812,
       "learning_rate": 6.478632478632479e-06,
-      "loss": 1.0521,
+      "loss": 0.8595,
       "step": 2710
     },
     {
       "epoch": 20.015384615384615,
-      "grad_norm": 0.5130220055580139,
+      "grad_norm": 0.9229622483253479,
       "learning_rate": 6.461538461538463e-06,
-      "loss": 0.4354,
+      "loss": 0.4589,
       "step": 2720
     },
     {
       "epoch": 20.016923076923078,
-      "grad_norm": 26.976499557495117,
+      "grad_norm": 55.14228820800781,
       "learning_rate": 6.444444444444445e-06,
-      "loss": 1.1985,
+      "loss": 0.9099,
       "step": 2730
     },
     {
       "epoch": 20.018461538461537,
-      "grad_norm": 49.21026611328125,
+      "grad_norm": 5.382133483886719,
       "learning_rate": 6.427350427350428e-06,
-      "loss": 1.6716,
+      "loss": 1.1102,
       "step": 2740
     },
     {
       "epoch": 20.02,
-      "grad_norm": 0.3829849660396576,
+      "grad_norm": 5.689920902252197,
       "learning_rate": 6.410256410256412e-06,
-      "loss": 0.3013,
+      "loss": 0.3899,
       "step": 2750
     },
     {
       "epoch": 20.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 0.5082508325576782,
-      "eval_runtime": 13.0921,
-      "eval_samples_per_second": 3.514,
-      "eval_steps_per_second": 1.757,
+      "eval_accuracy": 0.6739130434782609,
+      "eval_loss": 0.8506317734718323,
+      "eval_runtime": 12.2869,
+      "eval_samples_per_second": 3.744,
+      "eval_steps_per_second": 1.872,
       "step": 2751
     },
     {
       "epoch": 21.001384615384616,
-      "grad_norm": 10.179632186889648,
+      "grad_norm": 2.9915590286254883,
       "learning_rate": 6.3931623931623935e-06,
-      "loss": 0.6051,
+      "loss": 0.3736,
       "step": 2760
     },
     {
       "epoch": 21.00292307692308,
-      "grad_norm": 20.752038955688477,
+      "grad_norm": 88.86798095703125,
       "learning_rate": 6.376068376068376e-06,
-      "loss": 0.8644,
+      "loss": 1.1611,
       "step": 2770
     },
     {
       "epoch": 21.004461538461538,
-      "grad_norm": 0.5752691030502319,
+      "grad_norm": 6.824946880340576,
       "learning_rate": 6.358974358974359e-06,
-      "loss": 0.4362,
+      "loss": 0.6866,
       "step": 2780
     },
     {
       "epoch": 21.006,
-      "grad_norm": 1.0075253248214722,
+      "grad_norm": 0.5927864909172058,
       "learning_rate": 6.3418803418803425e-06,
-      "loss": 0.9587,
+      "loss": 1.2722,
       "step": 2790
     },
     {
       "epoch": 21.007538461538463,
-      "grad_norm": 1.7478026151657104,
+      "grad_norm": 5.484760284423828,
       "learning_rate": 6.324786324786325e-06,
-      "loss": 0.9172,
+      "loss": 0.7169,
       "step": 2800
     },
     {
       "epoch": 21.009076923076922,
-      "grad_norm": 59.94158935546875,
+      "grad_norm": 41.423397064208984,
       "learning_rate": 6.307692307692308e-06,
-      "loss": 0.3976,
+      "loss": 0.4645,
       "step": 2810
     },
     {
       "epoch": 21.010615384615384,
-      "grad_norm": 0.6232649683952332,
+      "grad_norm": 5.4901814460754395,
       "learning_rate": 6.290598290598291e-06,
-      "loss": 0.2702,
+      "loss": 0.2936,
       "step": 2820
     },
     {
       "epoch": 21.012153846153847,
-      "grad_norm": 0.06733821332454681,
+      "grad_norm": 0.04715157300233841,
       "learning_rate": 6.273504273504274e-06,
-      "loss": 0.7284,
+      "loss": 0.5224,
       "step": 2830
     },
     {
       "epoch": 21.013692307692306,
-      "grad_norm": 0.04839285835623741,
+      "grad_norm": 0.01829337514936924,
       "learning_rate": 6.256410256410257e-06,
-      "loss": 0.7976,
+      "loss": 0.6392,
       "step": 2840
     },
     {
       "epoch": 21.01523076923077,
-      "grad_norm": 0.7099334597587585,
+      "grad_norm": 0.25962960720062256,
       "learning_rate": 6.23931623931624e-06,
-      "loss": 0.0915,
+      "loss": 0.0898,
       "step": 2850
     },
     {
       "epoch": 21.01676923076923,
-      "grad_norm": 37.17452621459961,
+      "grad_norm": 68.61187744140625,
       "learning_rate": 6.222222222222223e-06,
-      "loss": 0.8673,
+      "loss": 0.7612,
       "step": 2860
     },
     {
       "epoch": 21.018307692307694,
-      "grad_norm": 4.46577262878418,
+      "grad_norm": 65.6745834350586,
       "learning_rate": 6.205128205128206e-06,
-      "loss": 0.9571,
+      "loss": 0.7687,
       "step": 2870
     },
     {
       "epoch": 21.019846153846153,
-      "grad_norm": 3.844195604324341,
+      "grad_norm": 79.904296875,
       "learning_rate": 6.188034188034189e-06,
-      "loss": 0.646,
+      "loss": 0.8377,
       "step": 2880
     },
     {
       "epoch": 21.020153846153846,
-      "eval_accuracy": 0.8260869565217391,
-      "eval_loss": 0.5288403630256653,
-      "eval_runtime": 11.7673,
-      "eval_samples_per_second": 3.909,
-      "eval_steps_per_second": 1.955,
+      "eval_accuracy": 0.6956521739130435,
+      "eval_loss": 0.8037568926811218,
+      "eval_runtime": 12.2724,
+      "eval_samples_per_second": 3.748,
+      "eval_steps_per_second": 1.874,
       "step": 2882
     },
     {
       "epoch": 22.00123076923077,
-      "grad_norm": 0.1899593472480774,
+      "grad_norm": 0.550711452960968,
       "learning_rate": 6.170940170940171e-06,
-      "loss": 0.9999,
+      "loss": 0.5705,
       "step": 2890
     },
     {
       "epoch": 22.002769230769232,
-      "grad_norm": 12.806218147277832,
+      "grad_norm": 80.71566772460938,
       "learning_rate": 6.153846153846155e-06,
-      "loss": 0.7985,
+      "loss": 0.9061,
       "step": 2900
     },
     {
       "epoch": 22.00430769230769,
-      "grad_norm": 20.41181182861328,
+      "grad_norm": 29.348573684692383,
       "learning_rate": 6.136752136752138e-06,
-      "loss": 0.5117,
+      "loss": 0.4016,
       "step": 2910
     },
     {
       "epoch": 22.005846153846154,
-      "grad_norm": 0.4226987957954407,
+      "grad_norm": 5.291245460510254,
       "learning_rate": 6.11965811965812e-06,
-      "loss": 0.2177,
+      "loss": 0.3618,
       "step": 2920
     },
     {
       "epoch": 22.007384615384616,
-      "grad_norm": 39.437103271484375,
+      "grad_norm": 51.70157241821289,
       "learning_rate": 6.102564102564104e-06,
-      "loss": 0.2983,
+      "loss": 0.2102,
       "step": 2930
     },
     {
       "epoch": 22.008923076923075,
-      "grad_norm": 0.6932996511459351,
+      "grad_norm": 0.5470988750457764,
       "learning_rate": 6.085470085470086e-06,
-      "loss": 0.3892,
+      "loss": 0.4396,
       "step": 2940
     },
     {
       "epoch": 22.010461538461538,
-      "grad_norm": 0.663569986820221,
+      "grad_norm": 2.904543876647949,
       "learning_rate": 6.0683760683760684e-06,
-      "loss": 0.5625,
+      "loss": 0.6828,
       "step": 2950
     },
     {
       "epoch": 22.012,
-      "grad_norm": 0.21277102828025818,
+      "grad_norm": 7.380136966705322,
       "learning_rate": 6.051282051282051e-06,
-      "loss": 0.4287,
+      "loss": 0.6294,
       "step": 2960
     },
     {
       "epoch": 22.013538461538463,
-      "grad_norm": 85.67404174804688,
+      "grad_norm": 45.839874267578125,
       "learning_rate": 6.034188034188035e-06,
-      "loss": 1.0995,
+      "loss": 1.1419,
       "step": 2970
     },
     {
       "epoch": 22.015076923076922,
-      "grad_norm": 103.44427490234375,
+      "grad_norm": 1.467175006866455,
       "learning_rate": 6.0170940170940174e-06,
-      "loss": 0.4695,
+      "loss": 0.5255,
       "step": 2980
     },
     {
       "epoch": 22.016615384615385,
-      "grad_norm": 0.0456969179213047,
+      "grad_norm": 0.06315338611602783,
       "learning_rate": 6e-06,
-      "loss": 0.4399,
+      "loss": 0.6435,
       "step": 2990
     },
     {
       "epoch": 22.018153846153847,
-      "grad_norm": 39.62498474121094,
+      "grad_norm": 99.21288299560547,
       "learning_rate": 5.982905982905983e-06,
-      "loss": 0.6504,
+      "loss": 0.7692,
       "step": 3000
     },
     {
       "epoch": 22.019692307692306,
-      "grad_norm": 18.204975128173828,
+      "grad_norm": 68.81067657470703,
       "learning_rate": 5.9658119658119664e-06,
-      "loss": 1.3053,
+      "loss": 0.7089,
       "step": 3010
     },
     {
       "epoch": 22.020153846153846,
       "eval_accuracy": 0.6521739130434783,
-      "eval_loss": 1.1223881244659424,
-      "eval_runtime": 11.8128,
-      "eval_samples_per_second": 3.894,
-      "eval_steps_per_second": 1.947,
+      "eval_loss": 0.8172503113746643,
+      "eval_runtime": 12.3335,
+      "eval_samples_per_second": 3.73,
+      "eval_steps_per_second": 1.865,
       "step": 3013
     },
     {
       "epoch": 23.001076923076923,
-      "grad_norm": 0.4275420904159546,
+      "grad_norm": 1.280930519104004,
       "learning_rate": 5.948717948717949e-06,
-      "loss": 0.2397,
+      "loss": 0.2027,
       "step": 3020
     },
     {
       "epoch": 23.002615384615385,
-      "grad_norm": 4.155840873718262,
+      "grad_norm": 17.986162185668945,
       "learning_rate": 5.931623931623932e-06,
-      "loss": 0.545,
+      "loss": 0.6504,
       "step": 3030
     },
     {
       "epoch": 23.004153846153844,
-      "grad_norm": 0.14375780522823334,
+      "grad_norm": 0.0962698683142662,
       "learning_rate": 5.914529914529915e-06,
-      "loss": 0.6312,
+      "loss": 0.5776,
       "step": 3040
     },
     {
       "epoch": 23.005692307692307,
-      "grad_norm": 0.049200814217329025,
+      "grad_norm": 0.04187112674117088,
       "learning_rate": 5.897435897435898e-06,
-      "loss": 0.6866,
+      "loss": 0.3429,
       "step": 3050
     },
     {
       "epoch": 23.00723076923077,
-      "grad_norm": 15.059101104736328,
+      "grad_norm": 57.80844497680664,
       "learning_rate": 5.880341880341881e-06,
-      "loss": 0.7464,
+      "loss": 0.6209,
       "step": 3060
     },
     {
       "epoch": 23.008769230769232,
-      "grad_norm": 37.120506286621094,
+      "grad_norm": 40.497493743896484,
       "learning_rate": 5.863247863247864e-06,
-      "loss": 0.2967,
+      "loss": 0.3255,
       "step": 3070
     },
     {
       "epoch": 23.01030769230769,
-      "grad_norm": 0.6163169741630554,
+      "grad_norm": 4.082076549530029,
       "learning_rate": 5.846153846153847e-06,
-      "loss": 0.4502,
+      "loss": 0.399,
       "step": 3080
     },
     {
       "epoch": 23.011846153846154,
-      "grad_norm": 79.16121673583984,
+      "grad_norm": 61.76959228515625,
       "learning_rate": 5.82905982905983e-06,
-      "loss": 0.9185,
+      "loss": 0.8943,
       "step": 3090
     },
     {
       "epoch": 23.013384615384616,
-      "grad_norm": 3.2994115352630615,
+      "grad_norm": 0.1485362946987152,
       "learning_rate": 5.8119658119658126e-06,
-      "loss": 0.4721,
+      "loss": 0.0424,
       "step": 3100
     },
     {
       "epoch": 23.014923076923075,
-      "grad_norm": 189.37454223632812,
+      "grad_norm": 19.416404724121094,
       "learning_rate": 5.794871794871796e-06,
-      "loss": 0.7812,
+      "loss": 0.4231,
       "step": 3110
     },
     {
       "epoch": 23.016461538461538,
-      "grad_norm": 0.10399169474840164,
+      "grad_norm": 0.07781932502985,
       "learning_rate": 5.777777777777778e-06,
-      "loss": 0.7956,
+      "loss": 0.7434,
       "step": 3120
     },
     {
       "epoch": 23.018,
-      "grad_norm": 0.03775927051901817,
+      "grad_norm": 0.03244343400001526,
       "learning_rate": 5.760683760683761e-06,
-      "loss": 0.292,
+      "loss": 0.2638,
       "step": 3130
     },
     {
       "epoch": 23.019538461538463,
-      "grad_norm": 55.82505798339844,
+      "grad_norm": 93.13057708740234,
       "learning_rate": 5.743589743589743e-06,
-      "loss": 0.9903,
+      "loss": 0.6477,
       "step": 3140
     },
     {
       "epoch": 23.020153846153846,
-      "eval_accuracy": 0.8043478260869565,
-      "eval_loss": 0.47060108184814453,
-      "eval_runtime": 11.9647,
-      "eval_samples_per_second": 3.845,
-      "eval_steps_per_second": 1.922,
+      "eval_accuracy": 0.782608695652174,
+      "eval_loss": 0.5238762497901917,
+      "eval_runtime": 12.3519,
+      "eval_samples_per_second": 3.724,
+      "eval_steps_per_second": 1.862,
       "step": 3144
     },
     {
       "epoch": 24.000923076923076,
-      "grad_norm": 1.2134654521942139,
+      "grad_norm": 6.272166728973389,
       "learning_rate": 5.726495726495727e-06,
-      "loss": 0.5882,
+      "loss": 0.2452,
       "step": 3150
     },
     {
       "epoch": 24.00246153846154,
-      "grad_norm": 9.802472114562988,
+      "grad_norm": 13.931325912475586,
       "learning_rate": 5.70940170940171e-06,
-      "loss": 0.1785,
+      "loss": 0.0921,
       "step": 3160
     },
     {
       "epoch": 24.004,
-      "grad_norm": 0.06365148723125458,
+      "grad_norm": 0.02355341799557209,
       "learning_rate": 5.692307692307692e-06,
-      "loss": 0.3565,
+      "loss": 0.2619,
       "step": 3170
     },
     {
       "epoch": 24.00553846153846,
-      "grad_norm": 19.367048263549805,
+      "grad_norm": 39.35104751586914,
       "learning_rate": 5.675213675213675e-06,
-      "loss": 1.3728,
+      "loss": 1.3229,
       "step": 3180
     },
     {
       "epoch": 24.007076923076923,
-      "grad_norm": 0.3147422671318054,
+      "grad_norm": 0.203730508685112,
       "learning_rate": 5.658119658119659e-06,
-      "loss": 0.7831,
+      "loss": 0.4974,
       "step": 3190
     },
     {
       "epoch": 24.008615384615386,
-      "grad_norm": 5.211735248565674,
+      "grad_norm": 1.321668267250061,
       "learning_rate": 5.641025641025641e-06,
-      "loss": 0.4236,
+      "loss": 0.5942,
       "step": 3200
     },
     {
       "epoch": 24.010153846153845,
-      "grad_norm": 0.2542964816093445,
+      "grad_norm": 0.12687939405441284,
       "learning_rate": 5.623931623931624e-06,
-      "loss": 0.1983,
+      "loss": 0.2655,
       "step": 3210
     },
     {
       "epoch": 24.011692307692307,
-      "grad_norm": 34.508331298828125,
+      "grad_norm": 0.13267870247364044,
       "learning_rate": 5.606837606837608e-06,
-      "loss": 0.3243,
+      "loss": 0.1771,
       "step": 3220
     },
     {
       "epoch": 24.01323076923077,
-      "grad_norm": 0.13946391642093658,
+      "grad_norm": 0.06731533259153366,
       "learning_rate": 5.58974358974359e-06,
-      "loss": 0.3546,
+      "loss": 0.5532,
       "step": 3230
     },
     {
       "epoch": 24.014769230769232,
-      "grad_norm": 78.96417999267578,
+      "grad_norm": 246.54559326171875,
       "learning_rate": 5.572649572649573e-06,
-      "loss": 0.8219,
+      "loss": 0.4319,
       "step": 3240
     },
     {
       "epoch": 24.01630769230769,
-      "grad_norm": 0.11380944401025772,
+      "grad_norm": 0.31551143527030945,
       "learning_rate": 5.555555555555557e-06,
-      "loss": 0.669,
+      "loss": 0.4138,
       "step": 3250
     },
     {
       "epoch": 24.017846153846154,
-      "grad_norm": 47.72481918334961,
+      "grad_norm": 38.96064376831055,
       "learning_rate": 5.538461538461539e-06,
-      "loss": 1.383,
+      "loss": 1.126,
       "step": 3260
     },
     {
       "epoch": 24.019384615384617,
-      "grad_norm": 30.822315216064453,
+      "grad_norm": 12.547762870788574,
       "learning_rate": 5.521367521367522e-06,
-      "loss": 0.7157,
+      "loss": 0.7071,
       "step": 3270
     },
     {
       "epoch": 24.020153846153846,
       "eval_accuracy": 0.782608695652174,
-      "eval_loss": 0.8162667751312256,
-      "eval_runtime": 11.7412,
-      "eval_samples_per_second": 3.918,
-      "eval_steps_per_second": 1.959,
+      "eval_loss": 0.757663369178772,
+      "eval_runtime": 13.3061,
+      "eval_samples_per_second": 3.457,
+      "eval_steps_per_second": 1.729,
       "step": 3275
     },
     {
       "epoch": 25.00076923076923,
-      "grad_norm": 0.24460583925247192,
+      "grad_norm": 0.1560353934764862,
       "learning_rate": 5.504273504273505e-06,
-      "loss": 0.9906,
+      "loss": 0.693,
       "step": 3280
     },
     {
       "epoch": 25.002307692307692,
-      "grad_norm": 81.8243179321289,
+      "grad_norm": 96.63082885742188,
       "learning_rate": 5.487179487179488e-06,
-      "loss": 0.4271,
+      "loss": 0.3708,
       "step": 3290
     },
     {
       "epoch": 25.003846153846155,
-      "grad_norm": 49.65717315673828,
+      "grad_norm": 30.65414810180664,
       "learning_rate": 5.470085470085471e-06,
-      "loss": 0.2202,
+      "loss": 0.0755,
       "step": 3300
     },
     {
       "epoch": 25.005384615384614,
-      "grad_norm": 0.29510897397994995,
+      "grad_norm": 1.7438099384307861,
       "learning_rate": 5.452991452991453e-06,
-      "loss": 0.5606,
+      "loss": 0.7645,
       "step": 3310
     },
     {
       "epoch": 25.006923076923076,
-      "grad_norm": 0.05217023566365242,
+      "grad_norm": 0.024945253506302834,
       "learning_rate": 5.435897435897436e-06,
-      "loss": 0.6757,
+      "loss": 0.593,
       "step": 3320
     },
     {
       "epoch": 25.00846153846154,
-      "grad_norm": 0.026966163888573647,
+      "grad_norm": 0.007312324829399586,
       "learning_rate": 5.418803418803419e-06,
-      "loss": 1.164,
+      "loss": 1.1078,
       "step": 3330
     },
     {
       "epoch": 25.01,
-      "grad_norm": 0.2099008560180664,
+      "grad_norm": 0.061362847685813904,
       "learning_rate": 5.401709401709402e-06,
-      "loss": 0.7998,
+      "loss": 0.7872,
       "step": 3340
     },
     {
       "epoch": 25.01153846153846,
-      "grad_norm": 3.2398674488067627,
+      "grad_norm": 0.2667884826660156,
       "learning_rate": 5.384615384615385e-06,
-      "loss": 0.7801,
+      "loss": 0.7606,
       "step": 3350
     },
     {
       "epoch": 25.013076923076923,
-      "grad_norm": 8.356966018676758,
+      "grad_norm": 0.3242189288139343,
       "learning_rate": 5.367521367521367e-06,
-      "loss": 0.4305,
+      "loss": 0.273,
       "step": 3360
     },
     {
       "epoch": 25.014615384615386,
-      "grad_norm": 84.3459701538086,
+      "grad_norm": 74.60770416259766,
       "learning_rate": 5.350427350427351e-06,
-      "loss": 0.391,
+      "loss": 0.7905,
       "step": 3370
     },
     {
       "epoch": 25.016153846153845,
-      "grad_norm": 2.4751970767974854,
+      "grad_norm": 0.7066124081611633,
       "learning_rate": 5.333333333333334e-06,
-      "loss": 0.3874,
+      "loss": 0.5153,
       "step": 3380
     },
     {
       "epoch": 25.017692307692307,
-      "grad_norm": 51.4213981628418,
+      "grad_norm": 139.34030151367188,
       "learning_rate": 5.316239316239316e-06,
-      "loss": 0.4588,
+      "loss": 0.2195,
       "step": 3390
     },
     {
       "epoch": 25.01923076923077,
-      "grad_norm": 0.3578292429447174,
+      "grad_norm": 0.05378177389502525,
       "learning_rate": 5.2991452991453e-06,
-      "loss": 0.3073,
+      "loss": 0.468,
       "step": 3400
     },
     {
       "epoch": 25.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.1408771276474,
-      "eval_runtime": 13.2155,
-      "eval_samples_per_second": 3.481,
-      "eval_steps_per_second": 1.74,
+      "eval_accuracy": 0.717391304347826,
+      "eval_loss": 1.135241150856018,
+      "eval_runtime": 14.2385,
+      "eval_samples_per_second": 3.231,
+      "eval_steps_per_second": 1.615,
       "step": 3406
     },
     {
       "epoch": 26.000615384615383,
-      "grad_norm": 0.9141789674758911,
+      "grad_norm": 0.021668357774615288,
       "learning_rate": 5.282051282051283e-06,
-      "loss": 0.3622,
+      "loss": 0.2559,
       "step": 3410
     },
     {
       "epoch": 26.002153846153846,
-      "grad_norm": 318.7442932128906,
+      "grad_norm": 0.5756977796554565,
       "learning_rate": 5.264957264957265e-06,
-      "loss": 0.4683,
+      "loss": 0.537,
       "step": 3420
     },
     {
       "epoch": 26.003692307692308,
-      "grad_norm": 65.43302917480469,
+      "grad_norm": 91.57279205322266,
       "learning_rate": 5.247863247863249e-06,
-      "loss": 0.9523,
+      "loss": 0.8496,
       "step": 3430
     },
     {
       "epoch": 26.00523076923077,
-      "grad_norm": 0.06784581393003464,
+      "grad_norm": 0.7005146145820618,
       "learning_rate": 5.230769230769232e-06,
-      "loss": 0.3191,
+      "loss": 0.2402,
       "step": 3440
     },
     {
       "epoch": 26.00676923076923,
-      "grad_norm": 35.10286331176758,
+      "grad_norm": 101.13854217529297,
       "learning_rate": 5.213675213675214e-06,
-      "loss": 1.0472,
+      "loss": 0.401,
       "step": 3450
     },
     {
       "epoch": 26.008307692307692,
-      "grad_norm": 78.8796157836914,
+      "grad_norm": 56.47507858276367,
       "learning_rate": 5.196581196581197e-06,
-      "loss": 0.5854,
+      "loss": 0.8318,
       "step": 3460
     },
     {
       "epoch": 26.009846153846155,
-      "grad_norm": 104.45633697509766,
+      "grad_norm": 119.30040740966797,
       "learning_rate": 5.179487179487181e-06,
-      "loss": 0.4423,
+      "loss": 0.9256,
       "step": 3470
     },
     {
       "epoch": 26.011384615384614,
-      "grad_norm": 0.09899577498435974,
+      "grad_norm": 0.10046964138746262,
       "learning_rate": 5.162393162393163e-06,
-      "loss": 0.6182,
+      "loss": 0.3626,
       "step": 3480
     },
     {
       "epoch": 26.012923076923077,
-      "grad_norm": 85.23323059082031,
+      "grad_norm": 91.69265747070312,
       "learning_rate": 5.145299145299145e-06,
-      "loss": 0.4479,
+      "loss": 0.3519,
       "step": 3490
     },
     {
       "epoch": 26.01446153846154,
-      "grad_norm": 84.03840637207031,
+      "grad_norm": 0.20490188896656036,
       "learning_rate": 5.128205128205128e-06,
-      "loss": 0.4672,
+      "loss": 0.4033,
       "step": 3500
     },
     {
       "epoch": 26.016,
-      "grad_norm": 60.73435974121094,
+      "grad_norm": 121.6290283203125,
       "learning_rate": 5.1111111111111115e-06,
-      "loss": 0.5186,
+      "loss": 0.3824,
       "step": 3510
     },
     {
       "epoch": 26.01753846153846,
-      "grad_norm": 0.2027343213558197,
+      "grad_norm": 0.06126533821225166,
       "learning_rate": 5.094017094017094e-06,
-      "loss": 0.2607,
+      "loss": 0.4969,
       "step": 3520
     },
     {
       "epoch": 26.019076923076923,
-      "grad_norm": 0.8834146857261658,
+      "grad_norm": 0.14738288521766663,
       "learning_rate": 5.076923076923077e-06,
-      "loss": 0.5069,
+      "loss": 0.6235,
       "step": 3530
     },
     {
       "epoch": 26.020153846153846,
-      "eval_accuracy": 0.8043478260869565,
-      "eval_loss": 0.5148446559906006,
-      "eval_runtime": 13.9275,
-      "eval_samples_per_second": 3.303,
-      "eval_steps_per_second": 1.651,
+      "eval_accuracy": 0.7608695652173914,
+      "eval_loss": 0.7960148453712463,
+      "eval_runtime": 14.2758,
+      "eval_samples_per_second": 3.222,
+      "eval_steps_per_second": 1.611,
       "step": 3537
     },
     {
       "epoch": 27.00046153846154,
-      "grad_norm": 0.4461596608161926,
+      "grad_norm": 1.279218077659607,
       "learning_rate": 5.05982905982906e-06,
-      "loss": 0.1949,
+      "loss": 0.0932,
       "step": 3540
     },
     {
       "epoch": 27.002,
-      "grad_norm": 12.315749168395996,
+      "grad_norm": 2.229957342147827,
       "learning_rate": 5.042735042735043e-06,
-      "loss": 0.139,
+      "loss": 0.1229,
       "step": 3550
     },
     {
       "epoch": 27.00353846153846,
-      "grad_norm": 0.21825408935546875,
+      "grad_norm": 0.5116841197013855,
       "learning_rate": 5.025641025641026e-06,
-      "loss": 0.5893,
+      "loss": 0.5097,
       "step": 3560
     },
     {
       "epoch": 27.005076923076924,
-      "grad_norm": 33.19759750366211,
+      "grad_norm": 114.2696304321289,
       "learning_rate": 5.008547008547009e-06,
-      "loss": 0.8881,
+      "loss": 0.8607,
       "step": 3570
     },
     {
       "epoch": 27.006615384615383,
-      "grad_norm": 30.013351440429688,
+      "grad_norm": 11.77940559387207,
       "learning_rate": 4.991452991452992e-06,
-      "loss": 0.4414,
+      "loss": 0.0523,
       "step": 3580
     },
     {
       "epoch": 27.008153846153846,
-      "grad_norm": 286.8537292480469,
+      "grad_norm": 7.5745463371276855,
       "learning_rate": 4.974358974358975e-06,
-      "loss": 0.6297,
+      "loss": 0.5678,
       "step": 3590
     },
     {
       "epoch": 27.00969230769231,
-      "grad_norm": 65.54366302490234,
+      "grad_norm": 1.557455062866211,
       "learning_rate": 4.957264957264958e-06,
-      "loss": 0.8037,
+      "loss": 0.426,
       "step": 3600
     },
     {
       "epoch": 27.01123076923077,
-      "grad_norm": 57.52277755737305,
+      "grad_norm": 37.66556930541992,
       "learning_rate": 4.940170940170941e-06,
-      "loss": 0.4425,
+      "loss": 1.2422,
       "step": 3610
     },
     {
       "epoch": 27.01276923076923,
-      "grad_norm": 0.022357992827892303,
+      "grad_norm": 0.04430201277136803,
       "learning_rate": 4.923076923076924e-06,
-      "loss": 0.3324,
+      "loss": 0.1195,
       "step": 3620
     },
     {
       "epoch": 27.014307692307693,
-      "grad_norm": 2.10927677154541,
+      "grad_norm": 0.05586925521492958,
       "learning_rate": 4.905982905982906e-06,
-      "loss": 0.1722,
+      "loss": 0.4615,
       "step": 3630
     },
     {
       "epoch": 27.015846153846155,
-      "grad_norm": 214.25904846191406,
+      "grad_norm": 15.102178573608398,
       "learning_rate": 4.888888888888889e-06,
-      "loss": 0.5133,
+      "loss": 0.744,
       "step": 3640
     },
     {
       "epoch": 27.017384615384614,
-      "grad_norm": 26.58594512939453,
+      "grad_norm": 191.96832275390625,
       "learning_rate": 4.871794871794872e-06,
-      "loss": 0.8377,
+      "loss": 0.3811,
       "step": 3650
     },
     {
       "epoch": 27.018923076923077,
-      "grad_norm": 0.19922487437725067,
+      "grad_norm": 0.024080609902739525,
       "learning_rate": 4.854700854700855e-06,
-      "loss": 0.2683,
+      "loss": 0.2239,
       "step": 3660
     },
     {
       "epoch": 27.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.2583907842636108,
-      "eval_runtime": 13.2151,
-      "eval_samples_per_second": 3.481,
-      "eval_steps_per_second": 1.74,
+      "eval_accuracy": 0.6956521739130435,
+      "eval_loss": 1.2848166227340698,
+      "eval_runtime": 14.57,
+      "eval_samples_per_second": 3.157,
+      "eval_steps_per_second": 1.579,
       "step": 3668
     },
     {
       "epoch": 28.000307692307693,
-      "grad_norm": 150.6471710205078,
+      "grad_norm": 48.32284164428711,
       "learning_rate": 4.837606837606838e-06,
-      "loss": 0.6211,
+      "loss": 0.4242,
       "step": 3670
     },
     {
       "epoch": 28.001846153846152,
-      "grad_norm": 56.26972579956055,
+      "grad_norm": 170.11517333984375,
       "learning_rate": 4.820512820512821e-06,
-      "loss": 0.7831,
+      "loss": 0.7591,
       "step": 3680
     },
     {
       "epoch": 28.003384615384615,
-      "grad_norm": 101.69254302978516,
+      "grad_norm": 229.9419708251953,
       "learning_rate": 4.803418803418804e-06,
-      "loss": 0.8847,
+      "loss": 0.5551,
       "step": 3690
     },
     {
       "epoch": 28.004923076923077,
-      "grad_norm": 55.50261306762695,
+      "grad_norm": 113.58826446533203,
       "learning_rate": 4.786324786324787e-06,
-      "loss": 0.5746,
+      "loss": 0.6995,
       "step": 3700
     },
     {
       "epoch": 28.00646153846154,
-      "grad_norm": 0.04360581934452057,
+      "grad_norm": 0.026729678735136986,
       "learning_rate": 4.76923076923077e-06,
-      "loss": 0.8075,
+      "loss": 0.377,
       "step": 3710
     },
     {
       "epoch": 28.008,
-      "grad_norm": 46.10295104980469,
+      "grad_norm": 205.2780303955078,
       "learning_rate": 4.752136752136752e-06,
-      "loss": 0.2659,
+      "loss": 0.1621,
       "step": 3720
     },
     {
       "epoch": 28.00953846153846,
-      "grad_norm": 0.35408440232276917,
+      "grad_norm": 0.05401019752025604,
       "learning_rate": 4.7350427350427355e-06,
-      "loss": 0.4686,
+      "loss": 0.3585,
       "step": 3730
     },
     {
       "epoch": 28.011076923076924,
-      "grad_norm": 0.19690901041030884,
+      "grad_norm": 1.323789119720459,
       "learning_rate": 4.717948717948718e-06,
-      "loss": 0.3786,
+      "loss": 0.3368,
       "step": 3740
     },
     {
       "epoch": 28.012615384615383,
-      "grad_norm": 0.33714887499809265,
+      "grad_norm": 0.10587374866008759,
       "learning_rate": 4.700854700854701e-06,
-      "loss": 0.0679,
+      "loss": 0.0565,
       "step": 3750
     },
     {
       "epoch": 28.014153846153846,
-      "grad_norm": 646.0299072265625,
+      "grad_norm": 131.05316162109375,
       "learning_rate": 4.6837606837606844e-06,
-      "loss": 0.2962,
+      "loss": 0.3678,
       "step": 3760
     },
     {
       "epoch": 28.01569230769231,
-      "grad_norm": 227.01092529296875,
+      "grad_norm": 21.758975982666016,
       "learning_rate": 4.666666666666667e-06,
-      "loss": 0.996,
+      "loss": 0.5154,
       "step": 3770
     },
     {
       "epoch": 28.017230769230768,
-      "grad_norm": 132.82162475585938,
+      "grad_norm": 7.62009334564209,
       "learning_rate": 4.64957264957265e-06,
-      "loss": 0.4102,
+      "loss": 0.1317,
       "step": 3780
     },
     {
       "epoch": 28.01876923076923,
-      "grad_norm": 0.0445413812994957,
+      "grad_norm": 0.23396851122379303,
       "learning_rate": 4.6324786324786334e-06,
-      "loss": 0.4506,
+      "loss": 0.3065,
       "step": 3790
     },
     {
       "epoch": 28.020153846153846,
-      "eval_accuracy": 0.782608695652174,
-      "eval_loss": 0.8241485953330994,
-      "eval_runtime": 13.2068,
-      "eval_samples_per_second": 3.483,
-      "eval_steps_per_second": 1.742,
+      "eval_accuracy": 0.7391304347826086,
+      "eval_loss": 1.0994690656661987,
+      "eval_runtime": 14.3055,
+      "eval_samples_per_second": 3.216,
+      "eval_steps_per_second": 1.608,
       "step": 3799
     },
     {
       "epoch": 29.000153846153847,
-      "grad_norm": 0.13771362602710724,
+      "grad_norm": 0.6101329922676086,
       "learning_rate": 4.615384615384616e-06,
-      "loss": 0.4563,
+      "loss": 0.3451,
       "step": 3800
     },
     {
       "epoch": 29.00169230769231,
-      "grad_norm": 0.031042367219924927,
+      "grad_norm": 0.007814668118953705,
       "learning_rate": 4.598290598290598e-06,
-      "loss": 0.8861,
+      "loss": 0.9179,
       "step": 3810
     },
     {
       "epoch": 29.00323076923077,
-      "grad_norm": 20.32213592529297,
+      "grad_norm": 44.847835540771484,
       "learning_rate": 4.581196581196582e-06,
-      "loss": 0.5048,
+      "loss": 0.3479,
       "step": 3820
     },
     {
       "epoch": 29.00476923076923,
-      "grad_norm": 135.0621337890625,
+      "grad_norm": 250.74884033203125,
       "learning_rate": 4.564102564102564e-06,
-      "loss": 0.5914,
+      "loss": 0.7657,
       "step": 3830
     },
     {
       "epoch": 29.006307692307693,
-      "grad_norm": 0.17822052538394928,
+      "grad_norm": 49.82850646972656,
       "learning_rate": 4.547008547008547e-06,
-      "loss": 1.1719,
+      "loss": 0.8607,
       "step": 3840
     },
     {
       "epoch": 29.007846153846153,
-      "grad_norm": 49.83673858642578,
+      "grad_norm": 0.8170290589332581,
       "learning_rate": 4.5299145299145306e-06,
-      "loss": 0.5184,
+      "loss": 0.3227,
       "step": 3850
     },
     {
       "epoch": 29.009384615384615,
-      "grad_norm": 0.018014727160334587,
+      "grad_norm": 0.02247241511940956,
       "learning_rate": 4.512820512820513e-06,
-      "loss": 0.4954,
+      "loss": 0.5761,
       "step": 3860
     },
     {
       "epoch": 29.010923076923078,
-      "grad_norm": 0.09828851372003555,
+      "grad_norm": 0.12600962817668915,
       "learning_rate": 4.495726495726496e-06,
-      "loss": 0.9552,
+      "loss": 0.8165,
       "step": 3870
     },
     {
       "epoch": 29.012461538461537,
-      "grad_norm": 5.368714809417725,
+      "grad_norm": 5.0875468254089355,
       "learning_rate": 4.4786324786324796e-06,
-      "loss": 0.5076,
+      "loss": 0.1522,
       "step": 3880
     },
     {
       "epoch": 29.014,
-      "grad_norm": 34.52876663208008,
+      "grad_norm": 189.6921844482422,
       "learning_rate": 4.461538461538462e-06,
-      "loss": 0.3437,
+      "loss": 0.2228,
       "step": 3890
     },
     {
       "epoch": 29.015538461538462,
-      "grad_norm": 0.25485557317733765,
+      "grad_norm": 0.3662845194339752,
       "learning_rate": 4.444444444444444e-06,
-      "loss": 0.6708,
+      "loss": 0.3342,
       "step": 3900
     },
     {
       "epoch": 29.017076923076925,
-      "grad_norm": 0.6703222990036011,
+      "grad_norm": 0.055656664073467255,
       "learning_rate": 4.427350427350428e-06,
-      "loss": 0.0826,
+      "loss": 0.1002,
       "step": 3910
     },
     {
       "epoch": 29.018615384615384,
-      "grad_norm": 25.28285026550293,
+      "grad_norm": 29.58397674560547,
       "learning_rate": 4.4102564102564104e-06,
-      "loss": 0.1152,
+      "loss": 0.1444,
       "step": 3920
     },
     {
       "epoch": 29.020153846153846,
-      "grad_norm": 0.022476214915513992,
+      "grad_norm": 0.031024612486362457,
       "learning_rate": 4.393162393162393e-06,
-      "loss": 0.6501,
+      "loss": 0.5603,
       "step": 3930
     },
     {
       "epoch": 29.020153846153846,
-      "eval_accuracy": 0.8260869565217391,
-      "eval_loss": 0.9973244071006775,
-      "eval_runtime": 11.7165,
-      "eval_samples_per_second": 3.926,
-      "eval_steps_per_second": 1.963,
+      "eval_accuracy": 0.7391304347826086,
+      "eval_loss": 0.8299373388290405,
+      "eval_runtime": 14.3114,
+      "eval_samples_per_second": 3.214,
+      "eval_steps_per_second": 1.607,
       "step": 3930
     },
     {
       "epoch": 30.001538461538463,
-      "grad_norm": 1.4469164609909058,
+      "grad_norm": 0.730601966381073,
       "learning_rate": 4.376068376068377e-06,
-      "loss": 0.7975,
+      "loss": 0.3666,
       "step": 3940
     },
     {
       "epoch": 30.00307692307692,
-      "grad_norm": 0.03349291905760765,
+      "grad_norm": 0.030513187870383263,
       "learning_rate": 4.358974358974359e-06,
-      "loss": 0.9181,
+      "loss": 0.2333,
       "step": 3950
     },
     {
       "epoch": 30.004615384615384,
-      "grad_norm": 1.3015611171722412,
+      "grad_norm": 238.04132080078125,
       "learning_rate": 4.341880341880342e-06,
-      "loss": 0.3424,
+      "loss": 0.4796,
       "step": 3960
     },
     {
       "epoch": 30.006153846153847,
-      "grad_norm": 0.5967833399772644,
+      "grad_norm": 194.7256317138672,
       "learning_rate": 4.324786324786326e-06,
-      "loss": 0.2202,
+      "loss": 0.2509,
       "step": 3970
     },
     {
       "epoch": 30.00769230769231,
-      "grad_norm": 0.056970831006765366,
+      "grad_norm": 0.010514265857636929,
       "learning_rate": 4.307692307692308e-06,
-      "loss": 0.215,
+      "loss": 0.5021,
       "step": 3980
     },
     {
       "epoch": 30.00923076923077,
-      "grad_norm": 0.009819090366363525,
+      "grad_norm": 0.006039201747626066,
       "learning_rate": 4.29059829059829e-06,
-      "loss": 0.287,
+      "loss": 0.0323,
       "step": 3990
     },
     {
       "epoch": 30.01076923076923,
-      "grad_norm": 3.7894606590270996,
+      "grad_norm": 1.242340087890625,
       "learning_rate": 4.273504273504274e-06,
-      "loss": 0.2059,
+      "loss": 0.3553,
       "step": 4000
     },
     {
       "epoch": 30.012307692307694,
-      "grad_norm": 0.008765391074120998,
+      "grad_norm": 0.02939273789525032,
       "learning_rate": 4.2564102564102566e-06,
-      "loss": 0.1785,
+      "loss": 0.115,
       "step": 4010
     },
     {
       "epoch": 30.013846153846153,
-      "grad_norm": 0.07702428847551346,
+      "grad_norm": 0.023452777415513992,
       "learning_rate": 4.239316239316239e-06,
-      "loss": 0.0549,
+      "loss": 0.3139,
       "step": 4020
     },
     {
       "epoch": 30.015384615384615,
-      "grad_norm": 0.10615424811840057,
+      "grad_norm": 0.011114229448139668,
       "learning_rate": 4.222222222222223e-06,
-      "loss": 0.3937,
+      "loss": 0.4254,
       "step": 4030
     },
     {
       "epoch": 30.016923076923078,
-      "grad_norm": 7.000075340270996,
+      "grad_norm": 210.99807739257812,
       "learning_rate": 4.2051282051282055e-06,
-      "loss": 0.8288,
+      "loss": 1.1873,
       "step": 4040
     },
     {
       "epoch": 30.018461538461537,
-      "grad_norm": 0.042215779423713684,
+      "grad_norm": 0.32026055455207825,
       "learning_rate": 4.188034188034188e-06,
-      "loss": 0.7491,
+      "loss": 1.1807,
       "step": 4050
     },
     {
       "epoch": 30.02,
-      "grad_norm": 0.014717997051775455,
+      "grad_norm": 0.039069026708602905,
       "learning_rate": 4.170940170940172e-06,
-      "loss": 0.6552,
+      "loss": 0.4514,
       "step": 4060
     },
     {
       "epoch": 30.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.0375455617904663,
-      "eval_runtime": 11.779,
-      "eval_samples_per_second": 3.905,
-      "eval_steps_per_second": 1.953,
+      "eval_accuracy": 0.8260869565217391,
+      "eval_loss": 0.5078631043434143,
+      "eval_runtime": 14.3127,
+      "eval_samples_per_second": 3.214,
+      "eval_steps_per_second": 1.607,
       "step": 4061
     },
     {
       "epoch": 31.001384615384616,
-      "grad_norm": 2.11873722076416,
+      "grad_norm": 105.17155456542969,
       "learning_rate": 4.1538461538461545e-06,
-      "loss": 0.0935,
+      "loss": 0.0627,
       "step": 4070
     },
     {
       "epoch": 31.00292307692308,
-      "grad_norm": 3.1624631881713867,
+      "grad_norm": 26.555856704711914,
       "learning_rate": 4.136752136752136e-06,
-      "loss": 0.2214,
+      "loss": 0.4188,
       "step": 4080
     },
     {
       "epoch": 31.004461538461538,
-      "grad_norm": 4.239480018615723,
+      "grad_norm": 0.021121378988027573,
       "learning_rate": 4.11965811965812e-06,
-      "loss": 0.4094,
+      "loss": 0.6885,
       "step": 4090
     },
     {
       "epoch": 31.006,
-      "grad_norm": 0.017483239993453026,
+      "grad_norm": 0.12947960197925568,
       "learning_rate": 4.102564102564103e-06,
-      "loss": 0.2046,
+      "loss": 0.1758,
       "step": 4100
     },
     {
       "epoch": 31.007538461538463,
-      "grad_norm": 104.08116149902344,
+      "grad_norm": 67.61315155029297,
       "learning_rate": 4.085470085470085e-06,
-      "loss": 0.4756,
+      "loss": 0.0586,
       "step": 4110
     },
     {
       "epoch": 31.009076923076922,
-      "grad_norm": 175.4596710205078,
+      "grad_norm": 4.274855613708496,
       "learning_rate": 4.068376068376069e-06,
-      "loss": 1.1802,
+      "loss": 0.7762,
       "step": 4120
     },
     {
       "epoch": 31.010615384615384,
-      "grad_norm": 47.34334182739258,
+      "grad_norm": 0.21890561282634735,
       "learning_rate": 4.051282051282052e-06,
-      "loss": 0.0158,
+      "loss": 0.0027,
       "step": 4130
     },
     {
       "epoch": 31.012153846153847,
-      "grad_norm": 0.1495905965566635,
+      "grad_norm": 199.3285369873047,
       "learning_rate": 4.034188034188034e-06,
-      "loss": 0.5666,
+      "loss": 0.8657,
       "step": 4140
     },
     {
       "epoch": 31.013692307692306,
-      "grad_norm": 102.30220794677734,
+      "grad_norm": 4.651693820953369,
       "learning_rate": 4.017094017094018e-06,
-      "loss": 0.2828,
+      "loss": 0.6412,
       "step": 4150
     },
     {
       "epoch": 31.01523076923077,
-      "grad_norm": 0.18496392667293549,
+      "grad_norm": 0.5808627009391785,
       "learning_rate": 4.000000000000001e-06,
-      "loss": 0.1033,
+      "loss": 0.2872,
       "step": 4160
     },
     {
       "epoch": 31.01676923076923,
-      "grad_norm": 24.492019653320312,
+      "grad_norm": 141.95333862304688,
       "learning_rate": 3.982905982905983e-06,
-      "loss": 0.4495,
+      "loss": 0.4645,
       "step": 4170
     },
     {
       "epoch": 31.018307692307694,
-      "grad_norm": 0.6317444443702698,
+      "grad_norm": 0.023080535233020782,
       "learning_rate": 3.965811965811966e-06,
-      "loss": 0.2928,
+      "loss": 0.1451,
       "step": 4180
     },
     {
       "epoch": 31.019846153846153,
-      "grad_norm": 26.888805389404297,
+      "grad_norm": 0.30592817068099976,
       "learning_rate": 3.948717948717949e-06,
-      "loss": 0.0676,
+      "loss": 0.3525,
       "step": 4190
     },
     {
       "epoch": 31.020153846153846,
-      "eval_accuracy": 0.8043478260869565,
-      "eval_loss": 0.915073573589325,
-      "eval_runtime": 12.5845,
-      "eval_samples_per_second": 3.655,
-      "eval_steps_per_second": 1.828,
+      "eval_accuracy": 0.717391304347826,
+      "eval_loss": 0.6860739588737488,
+      "eval_runtime": 14.2337,
+      "eval_samples_per_second": 3.232,
+      "eval_steps_per_second": 1.616,
       "step": 4192
     },
     {
       "epoch": 32.001230769230766,
-      "grad_norm": 52.63165283203125,
+      "grad_norm": 5.447160720825195,
       "learning_rate": 3.9316239316239315e-06,
-      "loss": 0.9533,
+      "loss": 0.3353,
       "step": 4200
     },
     {
       "epoch": 32.00276923076923,
-      "grad_norm": 0.004477484151721001,
+      "grad_norm": 0.004281382076442242,
       "learning_rate": 3.914529914529915e-06,
-      "loss": 0.0349,
+      "loss": 0.1991,
       "step": 4210
     },
     {
       "epoch": 32.00430769230769,
-      "grad_norm": 2.399238348007202,
+      "grad_norm": 0.5702845454216003,
       "learning_rate": 3.897435897435898e-06,
-      "loss": 0.5123,
+      "loss": 0.6549,
       "step": 4220
     },
     {
       "epoch": 32.00584615384616,
-      "grad_norm": 0.004892773926258087,
+      "grad_norm": 0.004613607656210661,
       "learning_rate": 3.8803418803418805e-06,
-      "loss": 0.1441,
+      "loss": 0.3367,
       "step": 4230
     },
     {
       "epoch": 32.007384615384616,
-      "grad_norm": 328.5547790527344,
+      "grad_norm": 27.875974655151367,
       "learning_rate": 3.863247863247864e-06,
-      "loss": 0.1297,
+      "loss": 0.1289,
       "step": 4240
     },
     {
       "epoch": 32.008923076923075,
-      "grad_norm": 0.249193474650383,
+      "grad_norm": 0.518906831741333,
       "learning_rate": 3.846153846153847e-06,
-      "loss": 0.0022,
+      "loss": 0.0353,
       "step": 4250
     },
     {
       "epoch": 32.01046153846154,
-      "grad_norm": 0.024616535753011703,
+      "grad_norm": 0.028710070997476578,
       "learning_rate": 3.8290598290598295e-06,
-      "loss": 0.4523,
+      "loss": 0.4762,
       "step": 4260
     },
     {
       "epoch": 32.012,
-      "grad_norm": 0.03703987970948219,
+      "grad_norm": 0.04342300817370415,
       "learning_rate": 3.8119658119658122e-06,
-      "loss": 0.4191,
+      "loss": 0.7021,
       "step": 4270
     },
     {
       "epoch": 32.01353846153846,
-      "grad_norm": 0.008937480859458447,
+      "grad_norm": 0.09270918369293213,
       "learning_rate": 3.794871794871795e-06,
-      "loss": 0.0589,
+      "loss": 0.0038,
       "step": 4280
     },
     {
       "epoch": 32.015076923076926,
-      "grad_norm": 0.0255438182502985,
+      "grad_norm": 0.020397480577230453,
       "learning_rate": 3.777777777777778e-06,
-      "loss": 0.2685,
+      "loss": 0.0029,
       "step": 4290
     },
     {
       "epoch": 32.016615384615385,
-      "grad_norm": 0.5818451046943665,
+      "grad_norm": 8.115623474121094,
       "learning_rate": 3.760683760683761e-06,
-      "loss": 0.2815,
+      "loss": 0.5877,
       "step": 4300
     },
     {
       "epoch": 32.018153846153844,
-      "grad_norm": 3.7883238792419434,
+      "grad_norm": 1.5437602996826172,
       "learning_rate": 3.743589743589744e-06,
-      "loss": 0.5135,
+      "loss": 0.3975,
       "step": 4310
     },
     {
       "epoch": 32.01969230769231,
-      "grad_norm": 0.006536950822919607,
+      "grad_norm": 0.018172316253185272,
       "learning_rate": 3.726495726495727e-06,
-      "loss": 0.0514,
+      "loss": 0.0011,
       "step": 4320
     },
     {
       "epoch": 32.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.3664484024047852,
-      "eval_runtime": 14.0178,
-      "eval_samples_per_second": 3.282,
-      "eval_steps_per_second": 1.641,
+      "eval_accuracy": 0.7391304347826086,
+      "eval_loss": 1.1932977437973022,
+      "eval_runtime": 12.2332,
+      "eval_samples_per_second": 3.76,
+      "eval_steps_per_second": 1.88,
       "step": 4323
     },
     {
       "epoch": 33.00107692307692,
-      "grad_norm": 0.07465940713882446,
+      "grad_norm": 0.05162428319454193,
       "learning_rate": 3.7094017094017098e-06,
-      "loss": 0.0032,
+      "loss": 0.0026,
       "step": 4330
     },
     {
       "epoch": 33.00261538461538,
-      "grad_norm": 0.08228836953639984,
+      "grad_norm": 107.2607192993164,
       "learning_rate": 3.692307692307693e-06,
-      "loss": 0.3491,
+      "loss": 0.3884,
       "step": 4340
     },
     {
       "epoch": 33.00415384615385,
-      "grad_norm": 33.137882232666016,
+      "grad_norm": 0.1267254799604416,
       "learning_rate": 3.6752136752136756e-06,
-      "loss": 0.8265,
+      "loss": 0.4135,
       "step": 4350
     },
     {
       "epoch": 33.00569230769231,
-      "grad_norm": 0.9730855822563171,
+      "grad_norm": 217.87318420410156,
       "learning_rate": 3.6581196581196584e-06,
-      "loss": 0.3718,
+      "loss": 0.4459,
       "step": 4360
     },
     {
       "epoch": 33.007230769230766,
-      "grad_norm": 94.9317626953125,
+      "grad_norm": 56.772464752197266,
       "learning_rate": 3.641025641025641e-06,
-      "loss": 1.0373,
+      "loss": 0.9378,
       "step": 4370
     },
     {
       "epoch": 33.00876923076923,
-      "grad_norm": 0.011030171066522598,
+      "grad_norm": 0.0087711401283741,
       "learning_rate": 3.623931623931624e-06,
-      "loss": 0.1513,
+      "loss": 0.1754,
       "step": 4380
     },
     {
       "epoch": 33.01030769230769,
-      "grad_norm": 2.059166431427002,
+      "grad_norm": 0.3356245756149292,
       "learning_rate": 3.606837606837607e-06,
-      "loss": 0.3012,
+      "loss": 0.0885,
       "step": 4390
     },
     {
       "epoch": 33.01184615384616,
-      "grad_norm": 0.2883024215698242,
+      "grad_norm": 0.015498592518270016,
       "learning_rate": 3.58974358974359e-06,
-      "loss": 0.3411,
+      "loss": 0.7416,
       "step": 4400
     },
     {
       "epoch": 33.013384615384616,
-      "grad_norm": 47.48052215576172,
+      "grad_norm": 128.6541748046875,
       "learning_rate": 3.572649572649573e-06,
-      "loss": 0.9255,
+      "loss": 1.085,
       "step": 4410
     },
     {
       "epoch": 33.014923076923075,
-      "grad_norm": 0.042310360819101334,
+      "grad_norm": 0.009648068808019161,
       "learning_rate": 3.555555555555556e-06,
-      "loss": 0.0015,
+      "loss": 0.2592,
       "step": 4420
     },
     {
       "epoch": 33.01646153846154,
-      "grad_norm": 0.2150733917951584,
+      "grad_norm": 0.02363857999444008,
       "learning_rate": 3.538461538461539e-06,
-      "loss": 0.2367,
+      "loss": 0.4846,
       "step": 4430
     },
     {
       "epoch": 33.018,
-      "grad_norm": 0.3859119117259979,
+      "grad_norm": 5.12437105178833,
       "learning_rate": 3.5213675213675218e-06,
-      "loss": 0.1022,
+      "loss": 0.1673,
       "step": 4440
     },
     {
       "epoch": 33.01953846153846,
-      "grad_norm": 0.15183915197849274,
+      "grad_norm": 0.10337384045124054,
       "learning_rate": 3.5042735042735045e-06,
-      "loss": 0.5447,
+      "loss": 0.4758,
       "step": 4450
     },
     {
       "epoch": 33.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.4551424980163574,
-      "eval_runtime": 13.2027,
-      "eval_samples_per_second": 3.484,
-      "eval_steps_per_second": 1.742,
+      "eval_accuracy": 0.782608695652174,
+      "eval_loss": 0.8102787137031555,
+      "eval_runtime": 12.3056,
+      "eval_samples_per_second": 3.738,
+      "eval_steps_per_second": 1.869,
       "step": 4454
     },
     {
       "epoch": 34.00092307692308,
-      "grad_norm": 0.4627760350704193,
+      "grad_norm": 0.9924832582473755,
       "learning_rate": 3.487179487179487e-06,
-      "loss": 0.9165,
+      "loss": 0.8534,
       "step": 4460
     },
     {
       "epoch": 34.00246153846154,
-      "grad_norm": 110.15799713134766,
+      "grad_norm": 266.9491271972656,
       "learning_rate": 3.4700854700854703e-06,
-      "loss": 0.3152,
+      "loss": 0.2155,
       "step": 4470
     },
     {
       "epoch": 34.004,
-      "grad_norm": 199.7411346435547,
+      "grad_norm": 81.45610046386719,
       "learning_rate": 3.452991452991453e-06,
-      "loss": 0.4348,
+      "loss": 0.0367,
       "step": 4480
     },
     {
       "epoch": 34.005538461538464,
-      "grad_norm": 0.009803686290979385,
+      "grad_norm": 0.007189776282757521,
       "learning_rate": 3.435897435897436e-06,
-      "loss": 0.0009,
+      "loss": 0.0019,
       "step": 4490
     },
     {
       "epoch": 34.00707692307692,
-      "grad_norm": 269.10552978515625,
+      "grad_norm": 4.55959939956665,
       "learning_rate": 3.4188034188034193e-06,
-      "loss": 0.4589,
+      "loss": 0.354,
       "step": 4500
     },
     {
       "epoch": 34.00861538461538,
-      "grad_norm": 0.06494735926389694,
+      "grad_norm": 173.19137573242188,
       "learning_rate": 3.401709401709402e-06,
-      "loss": 0.3366,
+      "loss": 0.4827,
       "step": 4510
     },
     {
       "epoch": 34.01015384615385,
-      "grad_norm": 214.33226013183594,
+      "grad_norm": 0.9809094667434692,
       "learning_rate": 3.384615384615385e-06,
-      "loss": 0.2165,
+      "loss": 0.016,
       "step": 4520
     },
     {
       "epoch": 34.01169230769231,
-      "grad_norm": 43.572837829589844,
+      "grad_norm": 0.1640501171350479,
       "learning_rate": 3.367521367521368e-06,
-      "loss": 0.7485,
+      "loss": 0.2968,
       "step": 4530
     },
     {
       "epoch": 34.013230769230766,
-      "grad_norm": 0.27556538581848145,
+      "grad_norm": 0.09363839030265808,
       "learning_rate": 3.3504273504273506e-06,
-      "loss": 0.5172,
+      "loss": 0.2213,
       "step": 4540
     },
     {
       "epoch": 34.01476923076923,
-      "grad_norm": 0.007643167860805988,
+      "grad_norm": 0.07453832775354385,
       "learning_rate": 3.3333333333333333e-06,
-      "loss": 0.4093,
+      "loss": 0.5347,
       "step": 4550
     },
     {
       "epoch": 34.01630769230769,
-      "grad_norm": 63.7769660949707,
+      "grad_norm": 1.27366042137146,
       "learning_rate": 3.3162393162393165e-06,
-      "loss": 0.3615,
+      "loss": 0.0888,
       "step": 4560
     },
     {
       "epoch": 34.01784615384615,
-      "grad_norm": 0.06752733886241913,
+      "grad_norm": 0.33367863297462463,
       "learning_rate": 3.299145299145299e-06,
-      "loss": 0.357,
+      "loss": 0.4626,
       "step": 4570
     },
     {
       "epoch": 34.01938461538462,
-      "grad_norm": 0.05651543661952019,
+      "grad_norm": 19.6287899017334,
       "learning_rate": 3.2820512820512823e-06,
-      "loss": 0.3701,
+      "loss": 0.5178,
       "step": 4580
     },
     {
       "epoch": 34.020153846153846,
       "eval_accuracy": 0.782608695652174,
-      "eval_loss": 1.3387607336044312,
-      "eval_runtime": 13.1827,
-      "eval_samples_per_second": 3.489,
-      "eval_steps_per_second": 1.745,
+      "eval_loss": 0.8117380738258362,
+      "eval_runtime": 12.9527,
+      "eval_samples_per_second": 3.551,
+      "eval_steps_per_second": 1.776,
       "step": 4585
     },
     {
       "epoch": 35.00076923076923,
-      "grad_norm": 0.06349892169237137,
+      "grad_norm": 0.019067266955971718,
       "learning_rate": 3.2649572649572655e-06,
-      "loss": 0.0061,
+      "loss": 0.0913,
       "step": 4590
     },
     {
       "epoch": 35.002307692307696,
-      "grad_norm": 0.1903676837682724,
+      "grad_norm": 0.030062997713685036,
       "learning_rate": 3.247863247863248e-06,
-      "loss": 0.0449,
+      "loss": 0.0059,
       "step": 4600
     },
     {
       "epoch": 35.003846153846155,
-      "grad_norm": 13.700048446655273,
+      "grad_norm": 192.97792053222656,
       "learning_rate": 3.2307692307692313e-06,
-      "loss": 0.3592,
+      "loss": 0.4131,
       "step": 4610
     },
     {
       "epoch": 35.005384615384614,
-      "grad_norm": 0.04581848159432411,
+      "grad_norm": 0.018678348511457443,
       "learning_rate": 3.213675213675214e-06,
-      "loss": 0.4179,
+      "loss": 0.4101,
       "step": 4620
     },
     {
       "epoch": 35.00692307692308,
-      "grad_norm": 165.4931640625,
+      "grad_norm": 198.5098114013672,
       "learning_rate": 3.1965811965811967e-06,
-      "loss": 0.3106,
+      "loss": 0.2959,
       "step": 4630
     },
     {
       "epoch": 35.00846153846154,
-      "grad_norm": 0.022043738514184952,
+      "grad_norm": 409.6368103027344,
       "learning_rate": 3.1794871794871795e-06,
-      "loss": 0.4508,
+      "loss": 0.3681,
       "step": 4640
     },
     {
       "epoch": 35.01,
-      "grad_norm": 0.03153197467327118,
+      "grad_norm": 0.03744253143668175,
       "learning_rate": 3.1623931623931626e-06,
-      "loss": 0.0166,
+      "loss": 0.0807,
       "step": 4650
     },
     {
       "epoch": 35.011538461538464,
-      "grad_norm": 3.3061561584472656,
+      "grad_norm": 0.089897520840168,
       "learning_rate": 3.1452991452991453e-06,
-      "loss": 0.7242,
+      "loss": 0.4647,
       "step": 4660
     },
     {
       "epoch": 35.01307692307692,
-      "grad_norm": 0.6018900871276855,
+      "grad_norm": 0.10863794386386871,
       "learning_rate": 3.1282051282051284e-06,
-      "loss": 0.1013,
+      "loss": 0.4155,
       "step": 4670
     },
     {
       "epoch": 35.01461538461538,
-      "grad_norm": 18.714059829711914,
+      "grad_norm": 17.10301971435547,
       "learning_rate": 3.1111111111111116e-06,
-      "loss": 0.0358,
+      "loss": 0.0211,
       "step": 4680
     },
     {
       "epoch": 35.01615384615385,
-      "grad_norm": 290.2351989746094,
+      "grad_norm": 0.2196597307920456,
       "learning_rate": 3.0940170940170943e-06,
-      "loss": 0.1013,
+      "loss": 0.149,
       "step": 4690
     },
     {
       "epoch": 35.01769230769231,
-      "grad_norm": 0.00717718992382288,
+      "grad_norm": 0.006109450478106737,
       "learning_rate": 3.0769230769230774e-06,
-      "loss": 0.3149,
+      "loss": 0.3733,
       "step": 4700
     },
     {
       "epoch": 35.01923076923077,
-      "grad_norm": 0.06576789915561676,
+      "grad_norm": 25.878141403198242,
       "learning_rate": 3.05982905982906e-06,
-      "loss": 0.6869,
+      "loss": 0.3839,
       "step": 4710
     },
     {
       "epoch": 35.020153846153846,
-      "eval_accuracy": 0.782608695652174,
-      "eval_loss": 1.3302055597305298,
-      "eval_runtime": 13.1813,
-      "eval_samples_per_second": 3.49,
-      "eval_steps_per_second": 1.745,
+      "eval_accuracy": 0.8260869565217391,
+      "eval_loss": 0.7044435739517212,
+      "eval_runtime": 12.89,
+      "eval_samples_per_second": 3.569,
+      "eval_steps_per_second": 1.784,
       "step": 4716
     },
     {
       "epoch": 36.00061538461539,
-      "grad_norm": 0.015848729759454727,
+      "grad_norm": 0.002284604124724865,
       "learning_rate": 3.042735042735043e-06,
-      "loss": 0.0065,
+      "loss": 0.2765,
       "step": 4720
     },
     {
       "epoch": 36.002153846153846,
-      "grad_norm": 0.10322362929582596,
+      "grad_norm": 0.07333678007125854,
       "learning_rate": 3.0256410256410256e-06,
-      "loss": 0.2323,
+      "loss": 0.1605,
       "step": 4730
     },
     {
       "epoch": 36.003692307692305,
-      "grad_norm": 0.03065655007958412,
+      "grad_norm": 0.005839182063937187,
       "learning_rate": 3.0085470085470087e-06,
-      "loss": 0.1782,
+      "loss": 0.2207,
       "step": 4740
     },
     {
       "epoch": 36.00523076923077,
-      "grad_norm": 0.0023964950814843178,
+      "grad_norm": 0.00254116952419281,
       "learning_rate": 2.9914529914529914e-06,
-      "loss": 0.3799,
+      "loss": 0.1382,
       "step": 4750
     },
     {
       "epoch": 36.00676923076923,
-      "grad_norm": 203.9849090576172,
+      "grad_norm": 1.1401203870773315,
       "learning_rate": 2.9743589743589746e-06,
-      "loss": 0.2273,
+      "loss": 0.0034,
       "step": 4760
     },
     {
       "epoch": 36.00830769230769,
-      "grad_norm": 0.5617519617080688,
+      "grad_norm": 0.14569438993930817,
       "learning_rate": 2.9572649572649577e-06,
-      "loss": 0.2777,
+      "loss": 0.3145,
       "step": 4770
     },
     {
       "epoch": 36.009846153846155,
-      "grad_norm": 0.004091217648237944,
+      "grad_norm": 0.0022192460019141436,
       "learning_rate": 2.9401709401709404e-06,
-      "loss": 0.1771,
+      "loss": 0.0096,
       "step": 4780
     },
     {
       "epoch": 36.011384615384614,
-      "grad_norm": 0.03037184663116932,
+      "grad_norm": 17.7734375,
       "learning_rate": 2.9230769230769236e-06,
-      "loss": 0.1418,
+      "loss": 0.3355,
       "step": 4790
     },
     {
       "epoch": 36.01292307692308,
-      "grad_norm": 0.013294171541929245,
+      "grad_norm": 0.020285405218601227,
       "learning_rate": 2.9059829059829063e-06,
-      "loss": 0.4339,
+      "loss": 0.2285,
       "step": 4800
     },
     {
       "epoch": 36.01446153846154,
-      "grad_norm": 0.0023209303617477417,
+      "grad_norm": 0.0021161625627428293,
       "learning_rate": 2.888888888888889e-06,
-      "loss": 0.2652,
+      "loss": 0.2925,
       "step": 4810
     },
     {
       "epoch": 36.016,
-      "grad_norm": 63.16322708129883,
+      "grad_norm": 4.639321804046631,
       "learning_rate": 2.8717948717948717e-06,
-      "loss": 0.5442,
+      "loss": 0.1859,
       "step": 4820
     },
     {
       "epoch": 36.017538461538464,
-      "grad_norm": 436.7990417480469,
+      "grad_norm": 182.3374786376953,
       "learning_rate": 2.854700854700855e-06,
-      "loss": 0.5163,
+      "loss": 0.6906,
       "step": 4830
     },
     {
       "epoch": 36.01907692307692,
-      "grad_norm": 0.8330660462379456,
+      "grad_norm": 0.006469838321208954,
       "learning_rate": 2.8376068376068376e-06,
-      "loss": 0.1371,
+      "loss": 0.0296,
       "step": 4840
     },
     {
       "epoch": 36.020153846153846,
-      "eval_accuracy": 0.782608695652174,
-      "eval_loss": 1.4859415292739868,
-      "eval_runtime": 12.7482,
-      "eval_samples_per_second": 3.608,
-      "eval_steps_per_second": 1.804,
+      "eval_accuracy": 0.8043478260869565,
+      "eval_loss": 0.7979722023010254,
+      "eval_runtime": 12.3235,
+      "eval_samples_per_second": 3.733,
+      "eval_steps_per_second": 1.866,
       "step": 4847
     },
     {
       "epoch": 37.000461538461536,
-      "grad_norm": 64.5765151977539,
+      "grad_norm": 138.13241577148438,
       "learning_rate": 2.8205128205128207e-06,
-      "loss": 0.3055,
+      "loss": 0.2583,
       "step": 4850
     },
     {
       "epoch": 37.002,
-      "grad_norm": 0.06066625565290451,
+      "grad_norm": 0.012276099063456059,
       "learning_rate": 2.803418803418804e-06,
-      "loss": 0.0016,
+      "loss": 0.2321,
       "step": 4860
     },
     {
       "epoch": 37.00353846153846,
-      "grad_norm": 0.007019346114248037,
+      "grad_norm": 0.007102313917130232,
       "learning_rate": 2.7863247863247866e-06,
-      "loss": 0.5366,
+      "loss": 0.1123,
       "step": 4870
     },
     {
       "epoch": 37.00507692307692,
-      "grad_norm": 0.008301621302962303,
+      "grad_norm": 0.004116872325539589,
       "learning_rate": 2.7692307692307697e-06,
-      "loss": 0.4092,
+      "loss": 0.0633,
       "step": 4880
     },
     {
       "epoch": 37.00661538461539,
-      "grad_norm": 0.020703766494989395,
+      "grad_norm": 0.011307389475405216,
       "learning_rate": 2.7521367521367524e-06,
-      "loss": 0.2475,
+      "loss": 0.0045,
       "step": 4890
     },
     {
       "epoch": 37.008153846153846,
-      "grad_norm": 2.0971076488494873,
+      "grad_norm": 0.013810048811137676,
       "learning_rate": 2.7350427350427355e-06,
-      "loss": 0.0706,
+      "loss": 0.0039,
       "step": 4900
     },
     {
       "epoch": 37.009692307692305,
-      "grad_norm": 9.441088676452637,
+      "grad_norm": 0.3569272756576538,
       "learning_rate": 2.717948717948718e-06,
-      "loss": 0.0394,
+      "loss": 0.0356,
       "step": 4910
     },
     {
       "epoch": 37.01123076923077,
-      "grad_norm": 385.3316345214844,
+      "grad_norm": 7.234623432159424,
       "learning_rate": 2.700854700854701e-06,
-      "loss": 0.2511,
+      "loss": 0.1737,
       "step": 4920
     },
     {
       "epoch": 37.01276923076923,
-      "grad_norm": 0.006842486094683409,
+      "grad_norm": 0.010227791033685207,
       "learning_rate": 2.6837606837606837e-06,
-      "loss": 0.7641,
+      "loss": 0.2023,
       "step": 4930
     },
     {
       "epoch": 37.01430769230769,
-      "grad_norm": 0.004666823893785477,
+      "grad_norm": 0.004664374515414238,
       "learning_rate": 2.666666666666667e-06,
-      "loss": 0.5452,
+      "loss": 0.5141,
       "step": 4940
     },
     {
       "epoch": 37.015846153846155,
-      "grad_norm": 0.49463582038879395,
+      "grad_norm": 9.983306884765625,
       "learning_rate": 2.64957264957265e-06,
-      "loss": 0.2126,
+      "loss": 0.6317,
       "step": 4950
     },
     {
       "epoch": 37.017384615384614,
-      "grad_norm": 5.189666271209717,
+      "grad_norm": 0.011219915002584457,
       "learning_rate": 2.6324786324786327e-06,
-      "loss": 0.2353,
+      "loss": 0.0153,
       "step": 4960
     },
     {
       "epoch": 37.01892307692308,
-      "grad_norm": 0.033904653042554855,
+      "grad_norm": 184.78318786621094,
       "learning_rate": 2.615384615384616e-06,
-      "loss": 0.5565,
+      "loss": 0.6077,
       "step": 4970
     },
     {
       "epoch": 37.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.410302996635437,
-      "eval_runtime": 12.6149,
-      "eval_samples_per_second": 3.646,
-      "eval_steps_per_second": 1.823,
+      "eval_accuracy": 0.8478260869565217,
+      "eval_loss": 0.8047866225242615,
+      "eval_runtime": 13.2741,
+      "eval_samples_per_second": 3.465,
+      "eval_steps_per_second": 1.733,
       "step": 4978
     },
     {
       "epoch": 38.00030769230769,
-      "grad_norm": 0.08889954537153244,
+      "grad_norm": 161.82296752929688,
       "learning_rate": 2.5982905982905985e-06,
-      "loss": 0.5995,
+      "loss": 0.1163,
       "step": 4980
     },
     {
       "epoch": 38.00184615384615,
-      "grad_norm": 0.17473474144935608,
+      "grad_norm": 0.009625504724681377,
       "learning_rate": 2.5811965811965817e-06,
-      "loss": 0.0306,
+      "loss": 0.0482,
       "step": 4990
     },
     {
       "epoch": 38.00338461538462,
-      "grad_norm": 209.5628204345703,
+      "grad_norm": 0.13098998367786407,
       "learning_rate": 2.564102564102564e-06,
-      "loss": 0.3034,
+      "loss": 0.03,
       "step": 5000
     },
     {
       "epoch": 38.00492307692308,
-      "grad_norm": 0.004574312828481197,
+      "grad_norm": 0.0021843614522367716,
       "learning_rate": 2.547008547008547e-06,
-      "loss": 0.3718,
+      "loss": 0.1062,
       "step": 5010
     },
     {
       "epoch": 38.00646153846154,
-      "grad_norm": 0.0036003710702061653,
+      "grad_norm": 0.002490828512236476,
       "learning_rate": 2.52991452991453e-06,
-      "loss": 0.3068,
+      "loss": 0.4272,
       "step": 5020
     },
     {
       "epoch": 38.008,
-      "grad_norm": 0.2461969256401062,
+      "grad_norm": 0.09035219252109528,
       "learning_rate": 2.512820512820513e-06,
-      "loss": 0.6456,
+      "loss": 0.6797,
       "step": 5030
     },
     {
       "epoch": 38.00953846153846,
-      "grad_norm": 0.10105012357234955,
+      "grad_norm": 0.0043014646507799625,
       "learning_rate": 2.495726495726496e-06,
-      "loss": 0.2028,
+      "loss": 0.2433,
       "step": 5040
     },
     {
       "epoch": 38.01107692307692,
-      "grad_norm": 35.23940658569336,
+      "grad_norm": 292.16546630859375,
       "learning_rate": 2.478632478632479e-06,
-      "loss": 0.3564,
+      "loss": 0.1033,
       "step": 5050
     },
     {
       "epoch": 38.01261538461539,
-      "grad_norm": 2.9968295097351074,
+      "grad_norm": 1.600675344467163,
       "learning_rate": 2.461538461538462e-06,
-      "loss": 0.3022,
+      "loss": 0.4145,
       "step": 5060
     },
     {
       "epoch": 38.014153846153846,
-      "grad_norm": 0.0421447791159153,
+      "grad_norm": 0.007035980932414532,
       "learning_rate": 2.4444444444444447e-06,
-      "loss": 0.001,
+      "loss": 0.0899,
       "step": 5070
     },
     {
       "epoch": 38.015692307692305,
-      "grad_norm": 0.06525139510631561,
+      "grad_norm": 0.03002479299902916,
       "learning_rate": 2.4273504273504274e-06,
-      "loss": 0.4745,
+      "loss": 0.5029,
       "step": 5080
     },
     {
       "epoch": 38.01723076923077,
-      "grad_norm": 0.09138324856758118,
+      "grad_norm": 0.04752730950713158,
       "learning_rate": 2.4102564102564105e-06,
-      "loss": 0.0038,
+      "loss": 0.1976,
       "step": 5090
     },
     {
       "epoch": 38.01876923076923,
-      "grad_norm": 0.06561964750289917,
+      "grad_norm": 0.055556993931531906,
       "learning_rate": 2.3931623931623937e-06,
-      "loss": 0.0022,
+      "loss": 0.0009,
       "step": 5100
     },
     {
       "epoch": 38.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.3428059816360474,
-      "eval_runtime": 13.1706,
-      "eval_samples_per_second": 3.493,
-      "eval_steps_per_second": 1.746,
+      "eval_accuracy": 0.8695652173913043,
+      "eval_loss": 0.7783644199371338,
+      "eval_runtime": 13.2785,
+      "eval_samples_per_second": 3.464,
+      "eval_steps_per_second": 1.732,
       "step": 5109
     },
     {
       "epoch": 39.00015384615384,
-      "grad_norm": 0.08478695154190063,
+      "grad_norm": 0.024660605937242508,
       "learning_rate": 2.376068376068376e-06,
-      "loss": 0.2608,
+      "loss": 0.3316,
       "step": 5110
     },
     {
       "epoch": 39.00169230769231,
-      "grad_norm": 0.8815059065818787,
+      "grad_norm": 258.7638244628906,
       "learning_rate": 2.358974358974359e-06,
-      "loss": 0.0008,
+      "loss": 0.1754,
       "step": 5120
     },
     {
       "epoch": 39.00323076923077,
-      "grad_norm": 0.08619926869869232,
+      "grad_norm": 5.474260330200195,
       "learning_rate": 2.3418803418803422e-06,
-      "loss": 0.0012,
+      "loss": 0.0066,
       "step": 5130
     },
     {
       "epoch": 39.00476923076923,
-      "grad_norm": 0.015669770538806915,
+      "grad_norm": 0.005695298779755831,
       "learning_rate": 2.324786324786325e-06,
-      "loss": 0.0009,
+      "loss": 0.0498,
       "step": 5140
     },
     {
       "epoch": 39.00630769230769,
-      "grad_norm": 0.07468448579311371,
+      "grad_norm": 0.005307150073349476,
       "learning_rate": 2.307692307692308e-06,
-      "loss": 0.003,
+      "loss": 0.0004,
       "step": 5150
     },
     {
       "epoch": 39.00784615384615,
-      "grad_norm": 0.09687216579914093,
+      "grad_norm": 0.05786298215389252,
       "learning_rate": 2.290598290598291e-06,
-      "loss": 0.0039,
+      "loss": 0.3589,
       "step": 5160
     },
     {
       "epoch": 39.00938461538462,
-      "grad_norm": 0.6945201754570007,
+      "grad_norm": 2.5580010414123535,
       "learning_rate": 2.2735042735042735e-06,
-      "loss": 0.355,
+      "loss": 0.3609,
       "step": 5170
     },
     {
       "epoch": 39.01092307692308,
-      "grad_norm": 0.0034933302085846663,
+      "grad_norm": 0.008595977909862995,
       "learning_rate": 2.2564102564102566e-06,
-      "loss": 0.9715,
+      "loss": 0.7624,
       "step": 5180
     },
     {
       "epoch": 39.01246153846154,
-      "grad_norm": 0.003874710062518716,
+      "grad_norm": 0.002324423985555768,
       "learning_rate": 2.2393162393162398e-06,
-      "loss": 0.0534,
+      "loss": 0.2605,
       "step": 5190
     },
     {
       "epoch": 39.014,
-      "grad_norm": 60.87498474121094,
+      "grad_norm": 74.57159423828125,
       "learning_rate": 2.222222222222222e-06,
-      "loss": 0.3278,
+      "loss": 0.5296,
       "step": 5200
     },
     {
       "epoch": 39.01553846153846,
-      "grad_norm": 0.1837165504693985,
+      "grad_norm": 0.003017917973920703,
       "learning_rate": 2.2051282051282052e-06,
-      "loss": 0.5495,
+      "loss": 0.4424,
       "step": 5210
     },
     {
       "epoch": 39.01707692307692,
-      "grad_norm": 0.1290096938610077,
+      "grad_norm": 0.36411431431770325,
       "learning_rate": 2.1880341880341884e-06,
-      "loss": 0.0861,
+      "loss": 0.0152,
       "step": 5220
     },
     {
       "epoch": 39.01861538461539,
-      "grad_norm": 0.02415439300239086,
+      "grad_norm": 0.005482113920152187,
       "learning_rate": 2.170940170940171e-06,
-      "loss": 0.1622,
+      "loss": 0.2472,
       "step": 5230
     },
     {
       "epoch": 39.020153846153846,
-      "grad_norm": 65.6265640258789,
+      "grad_norm": 390.8055114746094,
       "learning_rate": 2.153846153846154e-06,
-      "loss": 1.5897,
+      "loss": 0.6204,
       "step": 5240
     },
     {
       "epoch": 39.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.4124759435653687,
-      "eval_runtime": 11.8096,
-      "eval_samples_per_second": 3.895,
-      "eval_steps_per_second": 1.948,
+      "eval_accuracy": 0.8260869565217391,
+      "eval_loss": 0.8538339734077454,
+      "eval_runtime": 13.2825,
+      "eval_samples_per_second": 3.463,
+      "eval_steps_per_second": 1.732,
       "step": 5240
     },
     {
       "epoch": 40.00153846153846,
-      "grad_norm": 0.01301716361194849,
+      "grad_norm": 0.02487771585583687,
       "learning_rate": 2.136752136752137e-06,
-      "loss": 0.429,
+      "loss": 0.2536,
       "step": 5250
     },
     {
       "epoch": 40.003076923076925,
-      "grad_norm": 0.023441065102815628,
+      "grad_norm": 0.0057983254082500935,
       "learning_rate": 2.1196581196581196e-06,
-      "loss": 0.4187,
+      "loss": 0.0244,
       "step": 5260
     },
     {
       "epoch": 40.004615384615384,
-      "grad_norm": 0.004829849116504192,
+      "grad_norm": 0.002391011221334338,
       "learning_rate": 2.1025641025641028e-06,
-      "loss": 0.0711,
+      "loss": 0.3115,
       "step": 5270
     },
     {
       "epoch": 40.00615384615384,
-      "grad_norm": 0.045608654618263245,
+      "grad_norm": 0.04637107998132706,
       "learning_rate": 2.085470085470086e-06,
-      "loss": 0.0636,
+      "loss": 0.0012,
       "step": 5280
     },
     {
       "epoch": 40.00769230769231,
-      "grad_norm": 322.0433044433594,
+      "grad_norm": 258.67364501953125,
       "learning_rate": 2.068376068376068e-06,
-      "loss": 0.2149,
+      "loss": 0.1759,
       "step": 5290
     },
     {
       "epoch": 40.00923076923077,
-      "grad_norm": 42.855224609375,
+      "grad_norm": 55.54315185546875,
       "learning_rate": 2.0512820512820513e-06,
-      "loss": 0.4352,
+      "loss": 0.4044,
       "step": 5300
     },
     {
       "epoch": 40.01076923076923,
-      "grad_norm": 0.002952966606244445,
+      "grad_norm": 0.00581948971375823,
       "learning_rate": 2.0341880341880345e-06,
-      "loss": 0.6437,
+      "loss": 0.0459,
       "step": 5310
     },
     {
       "epoch": 40.012307692307694,
-      "grad_norm": 0.003094248939305544,
+      "grad_norm": 0.0035888785496354103,
       "learning_rate": 2.017094017094017e-06,
-      "loss": 0.0081,
+      "loss": 0.074,
       "step": 5320
     },
     {
       "epoch": 40.01384615384615,
-      "grad_norm": 0.003211932722479105,
+      "grad_norm": 0.004512140527367592,
       "learning_rate": 2.0000000000000003e-06,
-      "loss": 0.586,
+      "loss": 0.5589,
       "step": 5330
     },
     {
       "epoch": 40.01538461538462,
-      "grad_norm": 0.005071121733635664,
+      "grad_norm": 0.0035309726372361183,
       "learning_rate": 1.982905982905983e-06,
-      "loss": 0.0073,
+      "loss": 0.3519,
       "step": 5340
     },
     {
       "epoch": 40.01692307692308,
-      "grad_norm": 0.08057103306055069,
+      "grad_norm": 0.21208490431308746,
       "learning_rate": 1.9658119658119658e-06,
-      "loss": 0.0025,
+      "loss": 0.0178,
       "step": 5350
     },
     {
       "epoch": 40.01846153846154,
-      "grad_norm": 0.005270438268780708,
+      "grad_norm": 0.019819842651486397,
       "learning_rate": 1.948717948717949e-06,
-      "loss": 0.0004,
+      "loss": 0.0041,
       "step": 5360
     },
     {
       "epoch": 40.02,
-      "grad_norm": 0.008296512998640537,
+      "grad_norm": 0.008097551763057709,
       "learning_rate": 1.931623931623932e-06,
-      "loss": 0.1302,
+      "loss": 0.0006,
       "step": 5370
     },
     {
       "epoch": 40.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.3892539739608765,
-      "eval_runtime": 12.4205,
-      "eval_samples_per_second": 3.704,
-      "eval_steps_per_second": 1.852,
+      "eval_accuracy": 0.7608695652173914,
+      "eval_loss": 1.05526864528656,
+      "eval_runtime": 13.2614,
+      "eval_samples_per_second": 3.469,
+      "eval_steps_per_second": 1.734,
       "step": 5371
     },
     {
       "epoch": 41.001384615384616,
-      "grad_norm": 0.0030074138194322586,
+      "grad_norm": 0.0024010681081563234,
       "learning_rate": 1.9145299145299148e-06,
-      "loss": 0.1594,
+      "loss": 0.2204,
       "step": 5380
     },
     {
       "epoch": 41.002923076923075,
-      "grad_norm": 0.04759305715560913,
+      "grad_norm": 0.016204699873924255,
       "learning_rate": 1.8974358974358975e-06,
-      "loss": 0.0777,
+      "loss": 0.0439,
       "step": 5390
     },
     {
       "epoch": 41.00446153846154,
-      "grad_norm": 0.020955096930265427,
+      "grad_norm": 0.20511499047279358,
       "learning_rate": 1.8803418803418804e-06,
-      "loss": 0.0888,
+      "loss": 0.0055,
       "step": 5400
     },
     {
       "epoch": 41.006,
-      "grad_norm": 0.020802758634090424,
+      "grad_norm": 0.021914811804890633,
       "learning_rate": 1.8632478632478635e-06,
-      "loss": 0.0009,
+      "loss": 0.3349,
       "step": 5410
     },
     {
       "epoch": 41.00753846153846,
-      "grad_norm": 196.2333221435547,
+      "grad_norm": 14.952799797058105,
       "learning_rate": 1.8461538461538465e-06,
-      "loss": 0.4222,
+      "loss": 0.1452,
       "step": 5420
     },
     {
       "epoch": 41.009076923076925,
-      "grad_norm": 0.2782272398471832,
+      "grad_norm": 2.2803289890289307,
       "learning_rate": 1.8290598290598292e-06,
-      "loss": 0.3757,
+      "loss": 0.3572,
       "step": 5430
     },
     {
       "epoch": 41.010615384615384,
-      "grad_norm": 0.020732909440994263,
+      "grad_norm": 0.007608635351061821,
       "learning_rate": 1.811965811965812e-06,
-      "loss": 0.0009,
+      "loss": 0.0002,
       "step": 5440
     },
     {
       "epoch": 41.01215384615384,
-      "grad_norm": 0.012637576088309288,
+      "grad_norm": 0.004036715254187584,
       "learning_rate": 1.794871794871795e-06,
-      "loss": 0.4355,
+      "loss": 0.4369,
       "step": 5450
     },
     {
       "epoch": 41.01369230769231,
-      "grad_norm": 0.011620227247476578,
+      "grad_norm": 0.01055358536541462,
       "learning_rate": 1.777777777777778e-06,
-      "loss": 0.399,
+      "loss": 0.0004,
       "step": 5460
     },
     {
       "epoch": 41.01523076923077,
-      "grad_norm": 0.025496402755379677,
+      "grad_norm": 0.039770375937223434,
       "learning_rate": 1.7606837606837609e-06,
-      "loss": 0.1318,
+      "loss": 0.3951,
       "step": 5470
     },
     {
       "epoch": 41.01676923076923,
-      "grad_norm": 94.95100402832031,
+      "grad_norm": 152.85882568359375,
       "learning_rate": 1.7435897435897436e-06,
-      "loss": 0.3159,
+      "loss": 0.0219,
       "step": 5480
     },
     {
       "epoch": 41.018307692307694,
-      "grad_norm": 0.048655539751052856,
+      "grad_norm": 0.012421571649610996,
       "learning_rate": 1.7264957264957265e-06,
-      "loss": 0.2638,
+      "loss": 0.2717,
       "step": 5490
     },
     {
       "epoch": 41.01984615384615,
-      "grad_norm": 0.004611465148627758,
+      "grad_norm": 0.0027024794835597277,
       "learning_rate": 1.7094017094017097e-06,
-      "loss": 0.2463,
+      "loss": 0.2712,
       "step": 5500
     },
     {
       "epoch": 41.020153846153846,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.559918761253357,
-      "eval_runtime": 12.4494,
-      "eval_samples_per_second": 3.695,
-      "eval_steps_per_second": 1.847,
+      "eval_accuracy": 0.8260869565217391,
+      "eval_loss": 0.9511474967002869,
+      "eval_runtime": 14.5263,
+      "eval_samples_per_second": 3.167,
+      "eval_steps_per_second": 1.583,
       "step": 5502
     },
     {
       "epoch": 42.001230769230766,
-      "grad_norm": 0.023069551214575768,
+      "grad_norm": 0.09235262870788574,
       "learning_rate": 1.6923076923076926e-06,
-      "loss": 0.8881,
+      "loss": 0.3691,
       "step": 5510
     },
     {
       "epoch": 42.00276923076923,
-      "grad_norm": 0.006624761037528515,
+      "grad_norm": 0.007580755278468132,
       "learning_rate": 1.6752136752136753e-06,
-      "loss": 0.2337,
+      "loss": 0.425,
       "step": 5520
     },
     {
       "epoch": 42.00430769230769,
-      "grad_norm": 96.46836853027344,
+      "grad_norm": 2.2295045852661133,
       "learning_rate": 1.6581196581196582e-06,
-      "loss": 0.3171,
+      "loss": 0.3389,
       "step": 5530
     },
     {
       "epoch": 42.00584615384616,
-      "grad_norm": 317.9783935546875,
+      "grad_norm": 10.590227127075195,
       "learning_rate": 1.6410256410256412e-06,
-      "loss": 0.4246,
+      "loss": 0.0063,
       "step": 5540
     },
     {
       "epoch": 42.007384615384616,
-      "grad_norm": 0.0062882499769330025,
+      "grad_norm": 0.004151643253862858,
       "learning_rate": 1.623931623931624e-06,
-      "loss": 0.2339,
+      "loss": 0.205,
       "step": 5550
     },
     {
       "epoch": 42.008923076923075,
-      "grad_norm": 0.005598544143140316,
+      "grad_norm": 0.0024669792037457228,
       "learning_rate": 1.606837606837607e-06,
-      "loss": 0.4042,
+      "loss": 0.1462,
       "step": 5560
     },
     {
       "epoch": 42.01046153846154,
-      "grad_norm": 0.030421020463109016,
+      "grad_norm": 0.05891695246100426,
       "learning_rate": 1.5897435897435897e-06,
-      "loss": 0.0007,
+      "loss": 0.0029,
       "step": 5570
     },
     {
       "epoch": 42.012,
-      "grad_norm": 0.006722313351929188,
+      "grad_norm": 0.0049569690600037575,
       "learning_rate": 1.5726495726495727e-06,
-      "loss": 0.6109,
+      "loss": 0.7399,
       "step": 5580
     },
     {
       "epoch": 42.01353846153846,
-      "grad_norm": 0.05044174566864967,
+      "grad_norm": 0.007224519271403551,
       "learning_rate": 1.5555555555555558e-06,
-      "loss": 0.7525,
+      "loss": 0.3351,
       "step": 5590
     },
     {
       "epoch": 42.015076923076926,
-      "grad_norm": 0.08396083116531372,
+      "grad_norm": 0.03601997718214989,
       "learning_rate": 1.5384615384615387e-06,
-      "loss": 0.0008,
+      "loss": 0.0797,
       "step": 5600
     },
     {
       "epoch": 42.016615384615385,
-      "grad_norm": 0.0033397823572158813,
+      "grad_norm": 0.008051466196775436,
       "learning_rate": 1.5213675213675214e-06,
-      "loss": 0.1934,
+      "loss": 0.1576,
       "step": 5610
     },
     {
       "epoch": 42.018153846153844,
-      "grad_norm": 0.5943049192428589,
+      "grad_norm": 0.038704995065927505,
       "learning_rate": 1.5042735042735044e-06,
-      "loss": 0.0032,
+      "loss": 0.0667,
       "step": 5620
     },
     {
       "epoch": 42.01969230769231,
-      "grad_norm": 0.5589593648910522,
+      "grad_norm": 0.034863125532865524,
       "learning_rate": 1.4871794871794873e-06,
-      "loss": 0.0007,
+      "loss": 0.0579,
       "step": 5630
     },
     {
       "epoch": 42.020153846153846,
-      "eval_accuracy": 0.782608695652174,
-      "eval_loss": 1.3541553020477295,
-      "eval_runtime": 11.8006,
-      "eval_samples_per_second": 3.898,
-      "eval_steps_per_second": 1.949,
+      "eval_accuracy": 0.8478260869565217,
+      "eval_loss": 0.9046187996864319,
+      "eval_runtime": 14.2902,
+      "eval_samples_per_second": 3.219,
+      "eval_steps_per_second": 1.609,
       "step": 5633
     },
     {
       "epoch": 43.00107692307692,
-      "grad_norm": 0.016454918310046196,
+      "grad_norm": 0.0052635460160672665,
       "learning_rate": 1.4700854700854702e-06,
-      "loss": 0.0767,
+      "loss": 0.0014,
       "step": 5640
     },
     {
       "epoch": 43.00261538461538,
-      "grad_norm": 0.4291391968727112,
+      "grad_norm": 0.06584282964468002,
       "learning_rate": 1.4529914529914531e-06,
-      "loss": 0.1525,
+      "loss": 0.0013,
       "step": 5650
     },
     {
       "epoch": 43.00415384615385,
-      "grad_norm": 2.03802490234375,
+      "grad_norm": 8.854089736938477,
       "learning_rate": 1.4358974358974359e-06,
-      "loss": 0.0424,
+      "loss": 0.4235,
       "step": 5660
     },
     {
       "epoch": 43.00569230769231,
-      "grad_norm": 0.0391116738319397,
+      "grad_norm": 0.009933763183653355,
       "learning_rate": 1.4188034188034188e-06,
-      "loss": 0.0007,
+      "loss": 0.0003,
       "step": 5670
     },
     {
       "epoch": 43.007230769230766,
-      "grad_norm": 0.04126261919736862,
+      "grad_norm": 0.030539080500602722,
       "learning_rate": 1.401709401709402e-06,
-      "loss": 0.0006,
+      "loss": 0.0907,
       "step": 5680
     },
     {
       "epoch": 43.00876923076923,
-      "grad_norm": 0.15058164298534393,
+      "grad_norm": 0.008198861964046955,
       "learning_rate": 1.3846153846153848e-06,
-      "loss": 1.0264,
+      "loss": 0.6164,
       "step": 5690
     },
     {
       "epoch": 43.01030769230769,
-      "grad_norm": 25.521728515625,
+      "grad_norm": 222.2707061767578,
       "learning_rate": 1.3675213675213678e-06,
-      "loss": 0.0045,
+      "loss": 0.5585,
       "step": 5700
     },
     {
       "epoch": 43.01184615384616,
-      "grad_norm": 0.009670182131230831,
+      "grad_norm": 0.3781268000602722,
       "learning_rate": 1.3504273504273505e-06,
       "loss": 0.0003,
       "step": 5710
     },
     {
       "epoch": 43.013384615384616,
-      "grad_norm": 0.0412202887237072,
+      "grad_norm": 0.6270176768302917,
       "learning_rate": 1.3333333333333334e-06,
-      "loss": 0.0057,
+      "loss": 0.0022,
       "step": 5720
     },
     {
       "epoch": 43.014923076923075,
-      "grad_norm": 108.24889373779297,
+      "grad_norm": 277.80279541015625,
       "learning_rate": 1.3162393162393163e-06,
-      "loss": 0.4326,
+      "loss": 0.3854,
       "step": 5730
     },
     {
       "epoch": 43.01646153846154,
-      "grad_norm": 0.012433959171175957,
+      "grad_norm": 0.05374988541007042,
       "learning_rate": 1.2991452991452993e-06,
-      "loss": 0.2995,
+      "loss": 0.3497,
       "step": 5740
     },
     {
       "epoch": 43.018,
-      "grad_norm": 0.015085268765687943,
+      "grad_norm": 0.003168632509186864,
       "learning_rate": 1.282051282051282e-06,
-      "loss": 0.0008,
+      "loss": 0.0003,
       "step": 5750
     },
     {
       "epoch": 43.01953846153846,
-      "grad_norm": 0.08252725005149841,
+      "grad_norm": 0.014277983456850052,
       "learning_rate": 1.264957264957265e-06,
-      "loss": 0.0004,
+      "loss": 0.0003,
       "step": 5760
     },
     {
       "epoch": 43.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.8062047958374023,
-      "eval_runtime": 11.7555,
-      "eval_samples_per_second": 3.913,
-      "eval_steps_per_second": 1.957,
+      "eval_accuracy": 0.6956521739130435,
+      "eval_loss": 1.5328577756881714,
+      "eval_runtime": 14.2867,
+      "eval_samples_per_second": 3.22,
+      "eval_steps_per_second": 1.61,
       "step": 5764
     },
     {
       "epoch": 44.00092307692308,
-      "grad_norm": 325.7167053222656,
+      "grad_norm": 36.56468963623047,
       "learning_rate": 1.247863247863248e-06,
-      "loss": 0.8863,
+      "loss": 0.8606,
       "step": 5770
     },
     {
       "epoch": 44.00246153846154,
-      "grad_norm": 0.023112991824746132,
+      "grad_norm": 0.11488334834575653,
       "learning_rate": 1.230769230769231e-06,
-      "loss": 0.0018,
+      "loss": 0.0009,
       "step": 5780
     },
     {
       "epoch": 44.004,
-      "grad_norm": 0.03161883354187012,
+      "grad_norm": 0.3461620807647705,
       "learning_rate": 1.2136752136752137e-06,
-      "loss": 0.5382,
+      "loss": 0.6396,
       "step": 5790
     },
     {
       "epoch": 44.005538461538464,
-      "grad_norm": 0.226994588971138,
+      "grad_norm": 0.05253095179796219,
       "learning_rate": 1.1965811965811968e-06,
-      "loss": 0.0069,
+      "loss": 0.0004,
       "step": 5800
     },
     {
       "epoch": 44.00707692307692,
-      "grad_norm": 0.38113123178482056,
+      "grad_norm": 0.02474433183670044,
       "learning_rate": 1.1794871794871795e-06,
-      "loss": 0.0085,
+      "loss": 0.3365,
       "step": 5810
     },
     {
       "epoch": 44.00861538461538,
-      "grad_norm": 0.032291267067193985,
+      "grad_norm": 0.004439428448677063,
       "learning_rate": 1.1623931623931625e-06,
-      "loss": 0.0003,
+      "loss": 0.0001,
       "step": 5820
     },
     {
       "epoch": 44.01015384615385,
-      "grad_norm": 0.1154903769493103,
+      "grad_norm": 0.14496807754039764,
       "learning_rate": 1.1452991452991454e-06,
-      "loss": 0.0081,
+      "loss": 0.0003,
       "step": 5830
     },
     {
       "epoch": 44.01169230769231,
-      "grad_norm": 0.05736929178237915,
+      "grad_norm": 1.252375602722168,
       "learning_rate": 1.1282051282051283e-06,
-      "loss": 1.1695,
+      "loss": 0.5105,
       "step": 5840
     },
     {
       "epoch": 44.013230769230766,
-      "grad_norm": 0.016687337309122086,
+      "grad_norm": 8.587737083435059,
       "learning_rate": 1.111111111111111e-06,
-      "loss": 0.0583,
+      "loss": 0.0079,
       "step": 5850
     },
     {
       "epoch": 44.01476923076923,
-      "grad_norm": 0.0028132584411650896,
+      "grad_norm": 0.002821454079821706,
       "learning_rate": 1.0940170940170942e-06,
-      "loss": 0.129,
+      "loss": 0.3752,
       "step": 5860
     },
     {
       "epoch": 44.01630769230769,
-      "grad_norm": 0.007681978866457939,
+      "grad_norm": 0.0037518569733947515,
       "learning_rate": 1.076923076923077e-06,
-      "loss": 0.0005,
+      "loss": 0.1974,
       "step": 5870
     },
     {
       "epoch": 44.01784615384615,
-      "grad_norm": 4.105467796325684,
+      "grad_norm": 0.08341187983751297,
       "learning_rate": 1.0598290598290598e-06,
-      "loss": 0.0324,
+      "loss": 0.3219,
       "step": 5880
     },
     {
       "epoch": 44.01938461538462,
-      "grad_norm": 0.0026431982405483723,
+      "grad_norm": 0.0015944752376526594,
       "learning_rate": 1.042735042735043e-06,
-      "loss": 0.0803,
+      "loss": 0.0002,
       "step": 5890
     },
     {
       "epoch": 44.020153846153846,
       "eval_accuracy": 0.782608695652174,
-      "eval_loss": 1.5013920068740845,
-      "eval_runtime": 11.8321,
-      "eval_samples_per_second": 3.888,
-      "eval_steps_per_second": 1.944,
+      "eval_loss": 1.0590087175369263,
+      "eval_runtime": 12.272,
+      "eval_samples_per_second": 3.748,
+      "eval_steps_per_second": 1.874,
       "step": 5895
     },
     {
       "epoch": 45.00076923076923,
-      "grad_norm": 0.15906000137329102,
+      "grad_norm": 0.02641480229794979,
       "learning_rate": 1.0256410256410257e-06,
-      "loss": 0.0844,
+      "loss": 0.3025,
       "step": 5900
     },
     {
       "epoch": 45.002307692307696,
-      "grad_norm": 0.07680743932723999,
+      "grad_norm": 0.008294407278299332,
       "learning_rate": 1.0085470085470086e-06,
-      "loss": 0.0004,
+      "loss": 0.0002,
       "step": 5910
     },
     {
       "epoch": 45.003846153846155,
-      "grad_norm": 0.0016978027997538447,
+      "grad_norm": 0.002305571222677827,
       "learning_rate": 9.914529914529915e-07,
-      "loss": 0.0031,
+      "loss": 0.0002,
       "step": 5920
     },
     {
       "epoch": 45.005384615384614,
-      "grad_norm": 0.04445165395736694,
+      "grad_norm": 0.08077816665172577,
       "learning_rate": 9.743589743589745e-07,
-      "loss": 0.319,
+      "loss": 0.0018,
       "step": 5930
     },
     {
       "epoch": 45.00692307692308,
-      "grad_norm": 0.005968225654214621,
+      "grad_norm": 0.008085973560810089,
       "learning_rate": 9.572649572649574e-07,
-      "loss": 0.3625,
+      "loss": 0.3963,
       "step": 5940
     },
     {
       "epoch": 45.00846153846154,
-      "grad_norm": 0.010701755993068218,
+      "grad_norm": 0.004294309299439192,
       "learning_rate": 9.401709401709402e-07,
-      "loss": 0.0003,
+      "loss": 0.0004,
       "step": 5950
     },
     {
       "epoch": 45.01,
-      "grad_norm": 0.0020827685948461294,
+      "grad_norm": 0.004988194443285465,
       "learning_rate": 9.230769230769232e-07,
-      "loss": 0.0168,
+      "loss": 0.3549,
       "step": 5960
     },
     {
       "epoch": 45.011538461538464,
-      "grad_norm": 0.03637014701962471,
+      "grad_norm": 0.010682044550776482,
       "learning_rate": 9.05982905982906e-07,
       "loss": 0.0005,
       "step": 5970
     },
     {
       "epoch": 45.01307692307692,
-      "grad_norm": 0.002841139445081353,
+      "grad_norm": 0.0026554034557193518,
       "learning_rate": 8.88888888888889e-07,
-      "loss": 0.0003,
+      "loss": 0.0259,
       "step": 5980
     },
     {
       "epoch": 45.01461538461538,
-      "grad_norm": 0.05124405771493912,
+      "grad_norm": 0.01651105470955372,
       "learning_rate": 8.717948717948718e-07,
-      "loss": 0.0692,
+      "loss": 0.4298,
       "step": 5990
     },
     {
       "epoch": 45.01615384615385,
-      "grad_norm": 0.0027790500316768885,
+      "grad_norm": 0.0037173342425376177,
       "learning_rate": 8.547008547008548e-07,
-      "loss": 0.0007,
+      "loss": 0.0024,
       "step": 6000
     },
     {
       "epoch": 45.01769230769231,
-      "grad_norm": 9.085582733154297,
+      "grad_norm": 0.08719909936189651,
       "learning_rate": 8.376068376068377e-07,
-      "loss": 0.1762,
+      "loss": 0.3234,
       "step": 6010
     },
     {
       "epoch": 45.01923076923077,
-      "grad_norm": 0.016285724937915802,
+      "grad_norm": 0.0035305186174809933,
       "learning_rate": 8.205128205128206e-07,
-      "loss": 0.3332,
+      "loss": 0.4781,
       "step": 6020
     },
     {
       "epoch": 45.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.5840011835098267,
-      "eval_runtime": 11.9274,
-      "eval_samples_per_second": 3.857,
-      "eval_steps_per_second": 1.928,
+      "eval_accuracy": 0.8478260869565217,
+      "eval_loss": 0.8769366145133972,
+      "eval_runtime": 12.2431,
+      "eval_samples_per_second": 3.757,
+      "eval_steps_per_second": 1.879,
       "step": 6026
     },
     {
       "epoch": 46.00061538461539,
-      "grad_norm": 0.02442973293364048,
+      "grad_norm": 0.012966303154826164,
       "learning_rate": 8.034188034188035e-07,
-      "loss": 0.5798,
+      "loss": 0.0704,
       "step": 6030
     },
     {
       "epoch": 46.002153846153846,
-      "grad_norm": 0.011202111840248108,
+      "grad_norm": 0.02987567149102688,
       "learning_rate": 7.863247863247863e-07,
-      "loss": 0.0004,
+      "loss": 0.0069,
       "step": 6040
     },
     {
       "epoch": 46.003692307692305,
-      "grad_norm": 0.04835433140397072,
+      "grad_norm": 0.8032574653625488,
       "learning_rate": 7.692307692307694e-07,
-      "loss": 0.0371,
+      "loss": 0.0006,
       "step": 6050
     },
     {
       "epoch": 46.00523076923077,
-      "grad_norm": 0.036281924694776535,
+      "grad_norm": 530.5457763671875,
       "learning_rate": 7.521367521367522e-07,
-      "loss": 0.0007,
+      "loss": 0.0802,
       "step": 6060
     },
     {
       "epoch": 46.00676923076923,
-      "grad_norm": 0.00819159671664238,
+      "grad_norm": 0.02995951659977436,
       "learning_rate": 7.350427350427351e-07,
       "loss": 0.0002,
       "step": 6070
     },
     {
       "epoch": 46.00830769230769,
-      "grad_norm": 0.058177947998046875,
+      "grad_norm": 0.026577329263091087,
       "learning_rate": 7.179487179487179e-07,
-      "loss": 0.2703,
+      "loss": 0.5913,
       "step": 6080
     },
     {
       "epoch": 46.009846153846155,
-      "grad_norm": 0.005668714642524719,
+      "grad_norm": 0.01174863986670971,
       "learning_rate": 7.00854700854701e-07,
-      "loss": 0.2731,
+      "loss": 0.453,
       "step": 6090
     },
     {
       "epoch": 46.011384615384614,
-      "grad_norm": 0.002219714689999819,
+      "grad_norm": 0.002361681777983904,
       "learning_rate": 6.837606837606839e-07,
-      "loss": 0.0002,
+      "loss": 0.0034,
       "step": 6100
     },
     {
       "epoch": 46.01292307692308,
-      "grad_norm": 0.028290949761867523,
+      "grad_norm": 0.003955264110118151,
       "learning_rate": 6.666666666666667e-07,
-      "loss": 0.0008,
+      "loss": 0.2987,
       "step": 6110
     },
     {
       "epoch": 46.01446153846154,
-      "grad_norm": 0.06725189834833145,
+      "grad_norm": 0.37594401836395264,
       "learning_rate": 6.495726495726496e-07,
-      "loss": 0.4295,
+      "loss": 0.0005,
       "step": 6120
     },
     {
       "epoch": 46.016,
-      "grad_norm": 0.0370306521654129,
+      "grad_norm": 3.303600311279297,
       "learning_rate": 6.324786324786325e-07,
-      "loss": 0.2355,
+      "loss": 0.1983,
       "step": 6130
     },
     {
       "epoch": 46.017538461538464,
-      "grad_norm": 0.005414789542555809,
+      "grad_norm": 0.003181650536134839,
       "learning_rate": 6.153846153846155e-07,
-      "loss": 0.2553,
+      "loss": 0.2586,
       "step": 6140
     },
     {
       "epoch": 46.01907692307692,
-      "grad_norm": 0.6177451610565186,
+      "grad_norm": 328.0619812011719,
       "learning_rate": 5.982905982905984e-07,
-      "loss": 0.0004,
+      "loss": 0.0706,
       "step": 6150
     },
     {
       "epoch": 46.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.5470941066741943,
-      "eval_runtime": 11.9254,
-      "eval_samples_per_second": 3.857,
-      "eval_steps_per_second": 1.929,
+      "eval_accuracy": 0.8260869565217391,
+      "eval_loss": 0.9024959802627563,
+      "eval_runtime": 12.57,
+      "eval_samples_per_second": 3.66,
+      "eval_steps_per_second": 1.83,
       "step": 6157
     },
     {
       "epoch": 47.000461538461536,
-      "grad_norm": 0.36279433965682983,
+      "grad_norm": 0.17874684929847717,
       "learning_rate": 5.811965811965812e-07,
-      "loss": 0.1052,
+      "loss": 0.0003,
       "step": 6160
     },
     {
       "epoch": 47.002,
-      "grad_norm": 0.010881070047616959,
+      "grad_norm": 0.012226130813360214,
       "learning_rate": 5.641025641025642e-07,
-      "loss": 0.0509,
+      "loss": 0.3706,
       "step": 6170
     },
     {
       "epoch": 47.00353846153846,
-      "grad_norm": 0.0020965421572327614,
+      "grad_norm": 0.0017176754772663116,
       "learning_rate": 5.470085470085471e-07,
-      "loss": 0.4284,
+      "loss": 0.2268,
       "step": 6180
     },
     {
       "epoch": 47.00507692307692,
-      "grad_norm": 0.03328271955251694,
+      "grad_norm": 0.03336606174707413,
       "learning_rate": 5.299145299145299e-07,
-      "loss": 0.0003,
+      "loss": 0.0199,
       "step": 6190
     },
     {
       "epoch": 47.00661538461539,
-      "grad_norm": 0.014189296402037144,
+      "grad_norm": 0.007278658449649811,
       "learning_rate": 5.128205128205128e-07,
-      "loss": 0.0007,
+      "loss": 0.0025,
       "step": 6200
     },
     {
       "epoch": 47.008153846153846,
-      "grad_norm": 0.009917080402374268,
+      "grad_norm": 0.004852349869906902,
       "learning_rate": 4.957264957264958e-07,
-      "loss": 0.343,
+      "loss": 0.0046,
       "step": 6210
     },
     {
       "epoch": 47.009692307692305,
-      "grad_norm": 0.016622619703412056,
+      "grad_norm": 0.03278949484229088,
       "learning_rate": 4.786324786324787e-07,
-      "loss": 0.0019,
+      "loss": 0.0523,
       "step": 6220
     },
     {
       "epoch": 47.01123076923077,
-      "grad_norm": 0.01013423502445221,
+      "grad_norm": 0.020033802837133408,
       "learning_rate": 4.615384615384616e-07,
-      "loss": 0.4385,
+      "loss": 0.198,
       "step": 6230
     },
     {
       "epoch": 47.01276923076923,
-      "grad_norm": 0.0019633290357887745,
+      "grad_norm": 0.001353133120574057,
       "learning_rate": 4.444444444444445e-07,
-      "loss": 0.2154,
+      "loss": 0.0002,
       "step": 6240
     },
     {
       "epoch": 47.01430769230769,
-      "grad_norm": 119.6874771118164,
+      "grad_norm": 6.156984329223633,
       "learning_rate": 4.273504273504274e-07,
-      "loss": 0.0332,
+      "loss": 0.0043,
       "step": 6250
     },
     {
       "epoch": 47.015846153846155,
-      "grad_norm": 0.03719891607761383,
+      "grad_norm": 0.05321473255753517,
       "learning_rate": 4.102564102564103e-07,
-      "loss": 0.0003,
+      "loss": 0.0004,
       "step": 6260
     },
     {
       "epoch": 47.017384615384614,
-      "grad_norm": 1.4264271259307861,
+      "grad_norm": 0.06571904569864273,
       "learning_rate": 3.9316239316239316e-07,
-      "loss": 0.2852,
+      "loss": 0.2826,
       "step": 6270
     },
     {
       "epoch": 47.01892307692308,
-      "grad_norm": 0.01973658986389637,
+      "grad_norm": 0.05385449528694153,
       "learning_rate": 3.760683760683761e-07,
-      "loss": 0.0004,
+      "loss": 0.0024,
       "step": 6280
     },
     {
       "epoch": 47.020153846153846,
-      "eval_accuracy": 0.782608695652174,
-      "eval_loss": 1.398191213607788,
-      "eval_runtime": 13.3455,
-      "eval_samples_per_second": 3.447,
-      "eval_steps_per_second": 1.723,
+      "eval_accuracy": 0.8260869565217391,
+      "eval_loss": 0.8749639391899109,
+      "eval_runtime": 12.2597,
+      "eval_samples_per_second": 3.752,
+      "eval_steps_per_second": 1.876,
       "step": 6288
     },
     {
       "epoch": 48.00030769230769,
-      "grad_norm": 0.0016242864076048136,
+      "grad_norm": 0.0012860174756497145,
       "learning_rate": 3.5897435897435896e-07,
-      "loss": 0.7016,
+      "loss": 0.3718,
       "step": 6290
     },
     {
       "epoch": 48.00184615384615,
-      "grad_norm": 0.015563036315143108,
+      "grad_norm": 0.05020932853221893,
       "learning_rate": 3.4188034188034194e-07,
-      "loss": 0.5297,
+      "loss": 0.0032,
       "step": 6300
     },
     {
       "epoch": 48.00338461538462,
-      "grad_norm": 0.20225787162780762,
+      "grad_norm": 0.025825316086411476,
       "learning_rate": 3.247863247863248e-07,
-      "loss": 0.0009,
+      "loss": 0.0002,
       "step": 6310
     },
     {
       "epoch": 48.00492307692308,
-      "grad_norm": 0.008095812052488327,
+      "grad_norm": 0.032054170966148376,
       "learning_rate": 3.0769230769230774e-07,
-      "loss": 0.0005,
+      "loss": 0.001,
       "step": 6320
     },
     {
       "epoch": 48.00646153846154,
-      "grad_norm": 0.0020545681472867727,
+      "grad_norm": 0.0013651000335812569,
       "learning_rate": 2.905982905982906e-07,
-      "loss": 0.1397,
+      "loss": 0.0003,
       "step": 6330
     },
     {
       "epoch": 48.008,
-      "grad_norm": 0.03401735797524452,
+      "grad_norm": 0.008745058439671993,
       "learning_rate": 2.7350427350427354e-07,
-      "loss": 0.0047,
+      "loss": 0.1515,
       "step": 6340
     },
     {
       "epoch": 48.00953846153846,
-      "grad_norm": 1.6771128177642822,
+      "grad_norm": 6.276891708374023,
       "learning_rate": 2.564102564102564e-07,
-      "loss": 0.2786,
+      "loss": 0.0349,
       "step": 6350
     },
     {
       "epoch": 48.01107692307692,
-      "grad_norm": 0.0300904531031847,
+      "grad_norm": 0.028106998652219772,
       "learning_rate": 2.3931623931623934e-07,
-      "loss": 0.1159,
+      "loss": 0.0062,
       "step": 6360
     },
     {
       "epoch": 48.01261538461539,
-      "grad_norm": 0.002180424053221941,
+      "grad_norm": 0.0018312828615307808,
       "learning_rate": 2.2222222222222224e-07,
-      "loss": 0.2689,
+      "loss": 0.3374,
       "step": 6370
     },
     {
       "epoch": 48.014153846153846,
-      "grad_norm": 0.004255858715623617,
+      "grad_norm": 0.002776710083708167,
       "learning_rate": 2.0512820512820514e-07,
-      "loss": 0.0019,
+      "loss": 0.1546,
       "step": 6380
     },
     {
       "epoch": 48.015692307692305,
-      "grad_norm": 158.4202117919922,
+      "grad_norm": 0.14190010726451874,
       "learning_rate": 1.8803418803418804e-07,
-      "loss": 0.0496,
+      "loss": 0.0691,
       "step": 6390
     },
     {
       "epoch": 48.01723076923077,
-      "grad_norm": 0.05164070799946785,
+      "grad_norm": 0.024348756298422813,
       "learning_rate": 1.7094017094017097e-07,
-      "loss": 0.0574,
+      "loss": 0.4598,
       "step": 6400
     },
     {
       "epoch": 48.01876923076923,
-      "grad_norm": 0.09816153347492218,
+      "grad_norm": 0.366004079580307,
       "learning_rate": 1.5384615384615387e-07,
-      "loss": 0.0019,
+      "loss": 0.0153,
       "step": 6410
     },
     {
       "epoch": 48.020153846153846,
-      "eval_accuracy": 0.7391304347826086,
-      "eval_loss": 1.488898515701294,
-      "eval_runtime": 13.5505,
-      "eval_samples_per_second": 3.395,
-      "eval_steps_per_second": 1.697,
+      "eval_accuracy": 0.8695652173913043,
+      "eval_loss": 0.8642791509628296,
+      "eval_runtime": 12.3029,
+      "eval_samples_per_second": 3.739,
+      "eval_steps_per_second": 1.869,
       "step": 6419
     },
     {
       "epoch": 49.00015384615384,
-      "grad_norm": 0.028583820909261703,
+      "grad_norm": 221.2472686767578,
       "learning_rate": 1.3675213675213677e-07,
-      "loss": 0.0017,
+      "loss": 0.0774,
       "step": 6420
     },
     {
       "epoch": 49.00169230769231,
-      "grad_norm": 0.4499860107898712,
+      "grad_norm": 0.018726691603660583,
       "learning_rate": 1.1965811965811967e-07,
-      "loss": 0.0003,
+      "loss": 0.0001,
       "step": 6430
     },
     {
       "epoch": 49.00323076923077,
-      "grad_norm": 6.271653652191162,
+      "grad_norm": 193.82705688476562,
       "learning_rate": 1.0256410256410257e-07,
-      "loss": 0.2985,
+      "loss": 0.6276,
       "step": 6440
     },
     {
       "epoch": 49.00476923076923,
-      "grad_norm": 0.0023325886577367783,
+      "grad_norm": 0.0033707364927977324,
       "learning_rate": 8.547008547008549e-08,
-      "loss": 0.0946,
+      "loss": 0.0251,
       "step": 6450
     },
     {
       "epoch": 49.00630769230769,
-      "grad_norm": 476.3946228027344,
+      "grad_norm": 163.57437133789062,
       "learning_rate": 6.837606837606839e-08,
-      "loss": 0.1531,
+      "loss": 0.3097,
       "step": 6460
     },
     {
       "epoch": 49.00784615384615,
-      "grad_norm": 0.0066766985692083836,
+      "grad_norm": 0.0071478127501904964,
       "learning_rate": 5.1282051282051286e-08,
-      "loss": 0.0096,
+      "loss": 0.0032,
       "step": 6470
     },
     {
       "epoch": 49.00938461538462,
-      "grad_norm": 14.471455574035645,
+      "grad_norm": 0.24642449617385864,
       "learning_rate": 3.418803418803419e-08,
-      "loss": 0.0033,
+      "loss": 0.0175,
       "step": 6480
     },
     {
       "epoch": 49.01092307692308,
-      "grad_norm": 0.01039290800690651,
+      "grad_norm": 0.008312939666211605,
       "learning_rate": 1.7094017094017096e-08,
-      "loss": 0.0098,
+      "loss": 0.1417,
       "step": 6490
     },
     {
       "epoch": 49.01246153846154,
-      "grad_norm": 0.005664715077728033,
+      "grad_norm": 0.0030096701811999083,
       "learning_rate": 0.0,
-      "loss": 0.0006,
+      "loss": 0.0004,
       "step": 6500
     },
     {
       "epoch": 49.01246153846154,
-      "eval_accuracy": 0.7608695652173914,
-      "eval_loss": 1.4876021146774292,
-      "eval_runtime": 14.8808,
-      "eval_samples_per_second": 3.091,
-      "eval_steps_per_second": 1.546,
+      "eval_accuracy": 0.8478260869565217,
+      "eval_loss": 0.8714874982833862,
+      "eval_runtime": 15.2476,
+      "eval_samples_per_second": 3.017,
+      "eval_steps_per_second": 1.508,
       "step": 6500
     },
     {
       "epoch": 49.01246153846154,
       "step": 6500,
       "total_flos": 5.686818180940844e+19,
-      "train_loss": 0.5316385031611157,
-      "train_runtime": 7434.7076,
-      "train_samples_per_second": 1.749,
-      "train_steps_per_second": 0.874
+      "train_loss": 0.5191439180429129,
+      "train_runtime": 8374.5594,
+      "train_samples_per_second": 1.552,
+      "train_steps_per_second": 0.776
     },
     {
       "epoch": 49.01246153846154,
       "eval_accuracy": 0.8695652173913043,
-      "eval_loss": 0.40178221464157104,
-      "eval_runtime": 13.1957,
-      "eval_samples_per_second": 3.486,
-      "eval_steps_per_second": 1.743,
+      "eval_loss": 0.7783644199371338,
+      "eval_runtime": 12.9745,
+      "eval_samples_per_second": 3.545,
+      "eval_steps_per_second": 1.773,
       "step": 6500
     },
     {
       "epoch": 49.01246153846154,
       "eval_accuracy": 0.8695652173913043,
-      "eval_loss": 0.40178215503692627,
-      "eval_runtime": 13.2103,
-      "eval_samples_per_second": 3.482,
-      "eval_steps_per_second": 1.741,
+      "eval_loss": 0.778364360332489,
+      "eval_runtime": 13.1898,
+      "eval_samples_per_second": 3.488,
+      "eval_steps_per_second": 1.744,
       "step": 6500
     }
   ],