diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.514367816091954,
-  "best_model_checkpoint": "videomae-base-finetuned-ucf101-subset/checkpoint-1824",
+  "best_metric": 0.8629629629629629,
+  "best_model_checkpoint": "videomae-base-finetuned-ucf101-subset/checkpoint-1664",
   "epoch": 119.00833333333334,
   "eval_steps": 500,
   "global_step": 1920,
@@ -10,2453 +10,2453 @@
   "log_history": [
     {
       "epoch": 0.005208333333333333,
-      "grad_norm": 5.254432678222656,
+      "grad_norm": 8.761439323425293,
       "learning_rate": 2.604166666666667e-06,
-      "loss": 4.2685,
+      "loss": 0.4529,
       "step": 10
     },
     {
       "epoch": 0.008333333333333333,
-      "eval_accuracy": 0.014367816091954023,
-      "eval_loss": 4.244288921356201,
-      "eval_runtime": 144.9444,
-      "eval_samples_per_second": 9.604,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.7074074074074074,
+      "eval_loss": 1.0265196561813354,
+      "eval_runtime": 64.9312,
+      "eval_samples_per_second": 4.158,
+      "eval_steps_per_second": 0.077,
       "step": 16
     },
     {
       "epoch": 1.0020833333333334,
-      "grad_norm": 4.573583602905273,
+      "grad_norm": 4.886209487915039,
       "learning_rate": 5.208333333333334e-06,
-      "loss": 4.2711,
+      "loss": 0.2986,
       "step": 20
     },
     {
       "epoch": 1.0072916666666667,
-      "grad_norm": 6.3157782554626465,
+      "grad_norm": 3.012814521789551,
       "learning_rate": 7.8125e-06,
-      "loss": 4.2462,
+      "loss": 0.2409,
       "step": 30
     },
     {
       "epoch": 1.0083333333333333,
-      "eval_accuracy": 0.016522988505747127,
-      "eval_loss": 4.222896575927734,
-      "eval_runtime": 126.0729,
-      "eval_samples_per_second": 11.041,
-      "eval_steps_per_second": 0.175,
+      "eval_accuracy": 0.762962962962963,
+      "eval_loss": 0.8731228113174438,
+      "eval_runtime": 55.22,
+      "eval_samples_per_second": 4.89,
+      "eval_steps_per_second": 0.091,
       "step": 32
     },
     {
       "epoch": 2.004166666666667,
-      "grad_norm": 5.368467330932617,
+      "grad_norm": 3.798407793045044,
       "learning_rate": 1.0416666666666668e-05,
-      "loss": 4.2193,
+      "loss": 0.21,
       "step": 40
     },
     {
       "epoch": 2.0083333333333333,
-      "eval_accuracy": 0.01293103448275862,
-      "eval_loss": 4.211052417755127,
-      "eval_runtime": 156.4771,
-      "eval_samples_per_second": 8.896,
-      "eval_steps_per_second": 0.141,
+      "eval_accuracy": 0.7481481481481481,
+      "eval_loss": 0.8199293613433838,
+      "eval_runtime": 55.7591,
+      "eval_samples_per_second": 4.842,
+      "eval_steps_per_second": 0.09,
       "step": 48
     },
     {
       "epoch": 3.0010416666666666,
-      "grad_norm": 4.4701738357543945,
+      "grad_norm": 2.993506669998169,
       "learning_rate": 1.3020833333333334e-05,
-      "loss": 4.2291,
+      "loss": 0.1638,
       "step": 50
     },
     {
       "epoch": 3.00625,
-      "grad_norm": 3.1355085372924805,
+      "grad_norm": 3.574033737182617,
       "learning_rate": 1.5625e-05,
-      "loss": 4.2251,
+      "loss": 0.149,
       "step": 60
     },
     {
       "epoch": 3.0083333333333333,
-      "eval_accuracy": 0.013649425287356323,
-      "eval_loss": 4.212449550628662,
-      "eval_runtime": 164.2057,
-      "eval_samples_per_second": 8.477,
-      "eval_steps_per_second": 0.134,
+      "eval_accuracy": 0.7592592592592593,
+      "eval_loss": 0.8314273357391357,
+      "eval_runtime": 52.8486,
+      "eval_samples_per_second": 5.109,
+      "eval_steps_per_second": 0.095,
       "step": 64
     },
     {
       "epoch": 4.003125,
-      "grad_norm": 3.2147018909454346,
+      "grad_norm": 1.6014480590820312,
       "learning_rate": 1.8229166666666668e-05,
-      "loss": 4.2197,
+      "loss": 0.1425,
       "step": 70
     },
     {
       "epoch": 4.008333333333334,
-      "grad_norm": 11.049864768981934,
+      "grad_norm": 19.053049087524414,
       "learning_rate": 2.0833333333333336e-05,
-      "loss": 4.231,
+      "loss": 0.1131,
       "step": 80
     },
     {
       "epoch": 4.008333333333334,
-      "eval_accuracy": 0.015804597701149427,
-      "eval_loss": 4.213141918182373,
-      "eval_runtime": 115.2514,
-      "eval_samples_per_second": 12.078,
-      "eval_steps_per_second": 0.191,
+      "eval_accuracy": 0.774074074074074,
+      "eval_loss": 0.7753406763076782,
+      "eval_runtime": 55.454,
+      "eval_samples_per_second": 4.869,
+      "eval_steps_per_second": 0.09,
       "step": 80
     },
     {
       "epoch": 5.005208333333333,
-      "grad_norm": 2.9415507316589355,
+      "grad_norm": 3.0630757808685303,
       "learning_rate": 2.34375e-05,
-      "loss": 4.2197,
+      "loss": 0.1177,
       "step": 90
     },
     {
       "epoch": 5.008333333333334,
-      "eval_accuracy": 0.015086206896551725,
-      "eval_loss": 4.20913553237915,
-      "eval_runtime": 132.3001,
-      "eval_samples_per_second": 10.522,
-      "eval_steps_per_second": 0.166,
+      "eval_accuracy": 0.7666666666666667,
+      "eval_loss": 0.7645485997200012,
+      "eval_runtime": 53.0065,
+      "eval_samples_per_second": 5.094,
+      "eval_steps_per_second": 0.094,
       "step": 96
     },
     {
       "epoch": 6.002083333333333,
-      "grad_norm": 2.928065776824951,
+      "grad_norm": 2.052164077758789,
       "learning_rate": 2.604166666666667e-05,
-      "loss": 4.2237,
+      "loss": 0.1178,
       "step": 100
     },
     {
       "epoch": 6.007291666666666,
-      "grad_norm": 2.455629587173462,
+      "grad_norm": 2.4891324043273926,
       "learning_rate": 2.8645833333333333e-05,
-      "loss": 4.2239,
+      "loss": 0.1106,
       "step": 110
     },
     {
       "epoch": 6.008333333333334,
-      "eval_accuracy": 0.010775862068965518,
-      "eval_loss": 4.2099738121032715,
-      "eval_runtime": 137.9002,
-      "eval_samples_per_second": 10.094,
-      "eval_steps_per_second": 0.16,
+      "eval_accuracy": 0.7407407407407407,
+      "eval_loss": 0.8108886480331421,
+      "eval_runtime": 55.2829,
+      "eval_samples_per_second": 4.884,
+      "eval_steps_per_second": 0.09,
       "step": 112
     },
     {
       "epoch": 7.004166666666666,
-      "grad_norm": 2.583024024963379,
+      "grad_norm": 2.5003340244293213,
       "learning_rate": 3.125e-05,
-      "loss": 4.2246,
+      "loss": 0.1346,
       "step": 120
     },
     {
       "epoch": 7.008333333333334,
-      "eval_accuracy": 0.014367816091954023,
-      "eval_loss": 4.211906909942627,
-      "eval_runtime": 131.9857,
-      "eval_samples_per_second": 10.547,
-      "eval_steps_per_second": 0.167,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.6663423180580139,
+      "eval_runtime": 59.5557,
+      "eval_samples_per_second": 4.534,
+      "eval_steps_per_second": 0.084,
       "step": 128
     },
     {
       "epoch": 8.001041666666667,
-      "grad_norm": 2.4764182567596436,
+      "grad_norm": 2.3909215927124023,
       "learning_rate": 3.385416666666667e-05,
-      "loss": 4.2207,
+      "loss": 0.1102,
       "step": 130
     },
     {
       "epoch": 8.00625,
-      "grad_norm": 2.181648015975952,
+      "grad_norm": 4.043548583984375,
       "learning_rate": 3.6458333333333336e-05,
-      "loss": 4.2131,
+      "loss": 0.1054,
       "step": 140
     },
     {
       "epoch": 8.008333333333333,
-      "eval_accuracy": 0.017241379310344827,
-      "eval_loss": 4.203038215637207,
-      "eval_runtime": 161.0561,
-      "eval_samples_per_second": 8.643,
-      "eval_steps_per_second": 0.137,
+      "eval_accuracy": 0.7851851851851852,
+      "eval_loss": 0.7931135892868042,
+      "eval_runtime": 51.4414,
+      "eval_samples_per_second": 5.249,
+      "eval_steps_per_second": 0.097,
       "step": 144
     },
     {
       "epoch": 9.003125,
-      "grad_norm": 2.136594533920288,
+      "grad_norm": 6.156685829162598,
       "learning_rate": 3.90625e-05,
-      "loss": 4.2064,
+      "loss": 0.0972,
       "step": 150
     },
     {
       "epoch": 9.008333333333333,
-      "grad_norm": 7.214341163635254,
+      "grad_norm": 26.1621036529541,
       "learning_rate": 4.166666666666667e-05,
-      "loss": 4.236,
+      "loss": 0.1302,
       "step": 160
     },
     {
       "epoch": 9.008333333333333,
-      "eval_accuracy": 0.01293103448275862,
-      "eval_loss": 4.197862148284912,
-      "eval_runtime": 121.1445,
-      "eval_samples_per_second": 11.49,
-      "eval_steps_per_second": 0.182,
+      "eval_accuracy": 0.7592592592592593,
+      "eval_loss": 0.8379598259925842,
+      "eval_runtime": 56.0501,
+      "eval_samples_per_second": 4.817,
+      "eval_steps_per_second": 0.089,
       "step": 160
     },
     {
       "epoch": 10.005208333333334,
-      "grad_norm": 2.4436376094818115,
+      "grad_norm": 3.0394575595855713,
       "learning_rate": 4.4270833333333337e-05,
-      "loss": 4.177,
+      "loss": 0.1201,
       "step": 170
     },
     {
       "epoch": 10.008333333333333,
-      "eval_accuracy": 0.032327586206896554,
-      "eval_loss": 4.132750511169434,
-      "eval_runtime": 132.3308,
-      "eval_samples_per_second": 10.519,
-      "eval_steps_per_second": 0.166,
+      "eval_accuracy": 0.7703703703703704,
+      "eval_loss": 0.7757737636566162,
+      "eval_runtime": 51.161,
+      "eval_samples_per_second": 5.277,
+      "eval_steps_per_second": 0.098,
       "step": 176
     },
     {
       "epoch": 11.002083333333333,
-      "grad_norm": 3.184831142425537,
+      "grad_norm": 3.2897756099700928,
       "learning_rate": 4.6875e-05,
-      "loss": 4.1437,
+      "loss": 0.1016,
       "step": 180
     },
     {
       "epoch": 11.007291666666667,
-      "grad_norm": 4.086268901824951,
+      "grad_norm": 4.202846527099609,
       "learning_rate": 4.947916666666667e-05,
-      "loss": 4.0656,
+      "loss": 0.0992,
       "step": 190
     },
     {
       "epoch": 11.008333333333333,
-      "eval_accuracy": 0.031609195402298854,
-      "eval_loss": 4.0372843742370605,
-      "eval_runtime": 128.8196,
-      "eval_samples_per_second": 10.806,
-      "eval_steps_per_second": 0.171,
+      "eval_accuracy": 0.725925925925926,
+      "eval_loss": 0.9271950125694275,
+      "eval_runtime": 56.717,
+      "eval_samples_per_second": 4.76,
+      "eval_steps_per_second": 0.088,
       "step": 192
     },
     {
       "epoch": 12.004166666666666,
-      "grad_norm": 4.314004421234131,
+      "grad_norm": 1.9054981470108032,
       "learning_rate": 4.976851851851852e-05,
-      "loss": 3.9339,
+      "loss": 0.11,
       "step": 200
     },
     {
       "epoch": 12.008333333333333,
-      "eval_accuracy": 0.05028735632183908,
-      "eval_loss": 3.884765386581421,
-      "eval_runtime": 129.9723,
-      "eval_samples_per_second": 10.71,
-      "eval_steps_per_second": 0.169,
+      "eval_accuracy": 0.7666666666666667,
+      "eval_loss": 0.8362606167793274,
+      "eval_runtime": 62.2624,
+      "eval_samples_per_second": 4.336,
+      "eval_steps_per_second": 0.08,
       "step": 208
     },
     {
       "epoch": 13.001041666666667,
-      "grad_norm": 4.494894027709961,
+      "grad_norm": 2.9541594982147217,
       "learning_rate": 4.947916666666667e-05,
-      "loss": 3.8696,
+      "loss": 0.1089,
       "step": 210
     },
     {
       "epoch": 13.00625,
-      "grad_norm": 3.506269931793213,
+      "grad_norm": 5.1113739013671875,
       "learning_rate": 4.9189814814814815e-05,
-      "loss": 3.7197,
+      "loss": 0.122,
       "step": 220
     },
     {
       "epoch": 13.008333333333333,
-      "eval_accuracy": 0.07830459770114942,
-      "eval_loss": 3.768052816390991,
-      "eval_runtime": 122.2061,
-      "eval_samples_per_second": 11.391,
-      "eval_steps_per_second": 0.18,
+      "eval_accuracy": 0.8111111111111111,
+      "eval_loss": 0.6285049319267273,
+      "eval_runtime": 58.9383,
+      "eval_samples_per_second": 4.581,
+      "eval_steps_per_second": 0.085,
       "step": 224
     },
     {
       "epoch": 14.003125,
-      "grad_norm": 3.8743553161621094,
+      "grad_norm": 5.306451797485352,
       "learning_rate": 4.8900462962962965e-05,
-      "loss": 3.6495,
+      "loss": 0.0947,
       "step": 230
     },
     {
       "epoch": 14.008333333333333,
-      "grad_norm": 13.633111000061035,
+      "grad_norm": 14.834625244140625,
       "learning_rate": 4.8611111111111115e-05,
-      "loss": 3.5657,
+      "loss": 0.1336,
       "step": 240
     },
     {
       "epoch": 14.008333333333333,
-      "eval_accuracy": 0.11853448275862069,
-      "eval_loss": 3.5790250301361084,
-      "eval_runtime": 131.0036,
-      "eval_samples_per_second": 10.626,
-      "eval_steps_per_second": 0.168,
+      "eval_accuracy": 0.8185185185185185,
+      "eval_loss": 0.6990402936935425,
+      "eval_runtime": 58.7688,
+      "eval_samples_per_second": 4.594,
+      "eval_steps_per_second": 0.085,
       "step": 240
     },
     {
       "epoch": 15.005208333333334,
-      "grad_norm": 3.9671919345855713,
+      "grad_norm": 2.0207409858703613,
       "learning_rate": 4.8321759259259265e-05,
-      "loss": 3.3108,
+      "loss": 0.0996,
       "step": 250
     },
     {
       "epoch": 15.008333333333333,
-      "eval_accuracy": 0.14655172413793102,
-      "eval_loss": 3.564199924468994,
-      "eval_runtime": 143.9145,
-      "eval_samples_per_second": 9.672,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.8037037037037037,
+      "eval_loss": 0.7356700897216797,
+      "eval_runtime": 57.999,
+      "eval_samples_per_second": 4.655,
+      "eval_steps_per_second": 0.086,
       "step": 256
     },
     {
       "epoch": 16.002083333333335,
-      "grad_norm": 4.458219528198242,
+      "grad_norm": 0.8878017067909241,
       "learning_rate": 4.803240740740741e-05,
-      "loss": 3.2608,
+      "loss": 0.0731,
       "step": 260
     },
     {
       "epoch": 16.007291666666667,
-      "grad_norm": 4.314128398895264,
+      "grad_norm": 4.644617080688477,
       "learning_rate": 4.774305555555556e-05,
-      "loss": 3.1687,
+      "loss": 0.0711,
       "step": 270
     },
     {
       "epoch": 16.008333333333333,
-      "eval_accuracy": 0.18318965517241378,
-      "eval_loss": 3.32100772857666,
-      "eval_runtime": 118.6776,
-      "eval_samples_per_second": 11.729,
-      "eval_steps_per_second": 0.185,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.662126898765564,
+      "eval_runtime": 59.7158,
+      "eval_samples_per_second": 4.521,
+      "eval_steps_per_second": 0.084,
       "step": 272
     },
     {
       "epoch": 17.004166666666666,
-      "grad_norm": 4.7279372215271,
+      "grad_norm": 1.8143174648284912,
       "learning_rate": 4.745370370370371e-05,
-      "loss": 3.1376,
+      "loss": 0.0839,
       "step": 280
     },
     {
       "epoch": 17.008333333333333,
-      "eval_accuracy": 0.22916666666666666,
-      "eval_loss": 3.194972276687622,
-      "eval_runtime": 131.4719,
-      "eval_samples_per_second": 10.588,
-      "eval_steps_per_second": 0.167,
+      "eval_accuracy": 0.7814814814814814,
+      "eval_loss": 0.7743860483169556,
+      "eval_runtime": 58.8674,
+      "eval_samples_per_second": 4.587,
+      "eval_steps_per_second": 0.085,
       "step": 288
     },
     {
       "epoch": 18.001041666666666,
-      "grad_norm": 6.386988162994385,
+      "grad_norm": 6.041594505310059,
       "learning_rate": 4.716435185185186e-05,
-      "loss": 3.0321,
+      "loss": 0.0641,
       "step": 290
     },
     {
       "epoch": 18.00625,
-      "grad_norm": 5.669564247131348,
+      "grad_norm": 3.4158430099487305,
       "learning_rate": 4.6875e-05,
-      "loss": 2.8366,
+      "loss": 0.0865,
       "step": 300
     },
     {
       "epoch": 18.008333333333333,
-      "eval_accuracy": 0.2492816091954023,
-      "eval_loss": 3.1269867420196533,
-      "eval_runtime": 113.3706,
-      "eval_samples_per_second": 12.278,
-      "eval_steps_per_second": 0.194,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.6456056833267212,
+      "eval_runtime": 59.2035,
+      "eval_samples_per_second": 4.561,
+      "eval_steps_per_second": 0.084,
       "step": 304
     },
     {
       "epoch": 19.003125,
-      "grad_norm": 6.121387004852295,
+      "grad_norm": 1.0697002410888672,
       "learning_rate": 4.658564814814815e-05,
-      "loss": 2.8072,
+      "loss": 0.0546,
       "step": 310
     },
     {
       "epoch": 19.008333333333333,
-      "grad_norm": 13.984885215759277,
+      "grad_norm": 11.687989234924316,
       "learning_rate": 4.62962962962963e-05,
-      "loss": 2.6811,
+      "loss": 0.0607,
       "step": 320
     },
     {
       "epoch": 19.008333333333333,
-      "eval_accuracy": 0.28951149425287354,
-      "eval_loss": 2.999732732772827,
-      "eval_runtime": 120.8139,
-      "eval_samples_per_second": 11.522,
-      "eval_steps_per_second": 0.182,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.727846622467041,
+      "eval_runtime": 55.4859,
+      "eval_samples_per_second": 4.866,
+      "eval_steps_per_second": 0.09,
       "step": 320
     },
     {
       "epoch": 20.005208333333332,
-      "grad_norm": 5.307132244110107,
+      "grad_norm": 3.3323593139648438,
       "learning_rate": 4.6006944444444444e-05,
-      "loss": 2.6104,
+      "loss": 0.0672,
       "step": 330
     },
     {
       "epoch": 20.008333333333333,
-      "eval_accuracy": 0.27729885057471265,
-      "eval_loss": 2.9776358604431152,
-      "eval_runtime": 144.4295,
-      "eval_samples_per_second": 9.638,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.8,
+      "eval_loss": 0.7862735390663147,
+      "eval_runtime": 55.9915,
+      "eval_samples_per_second": 4.822,
+      "eval_steps_per_second": 0.089,
       "step": 336
     },
     {
       "epoch": 21.002083333333335,
-      "grad_norm": 5.0186638832092285,
+      "grad_norm": 2.702950954437256,
       "learning_rate": 4.5717592592592594e-05,
-      "loss": 2.5321,
+      "loss": 0.0583,
       "step": 340
     },
     {
       "epoch": 21.007291666666667,
-      "grad_norm": 5.094930648803711,
+      "grad_norm": 0.8738414645195007,
       "learning_rate": 4.5428240740740744e-05,
-      "loss": 2.5156,
+      "loss": 0.0575,
       "step": 350
     },
     {
       "epoch": 21.008333333333333,
-      "eval_accuracy": 0.3125,
-      "eval_loss": 2.862489700317383,
-      "eval_runtime": 144.5549,
-      "eval_samples_per_second": 9.63,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.8185185185185185,
+      "eval_loss": 0.6789024472236633,
+      "eval_runtime": 58.2242,
+      "eval_samples_per_second": 4.637,
+      "eval_steps_per_second": 0.086,
       "step": 352
     },
     {
       "epoch": 22.004166666666666,
-      "grad_norm": 5.007638931274414,
+      "grad_norm": 0.7883135676383972,
       "learning_rate": 4.5138888888888894e-05,
-      "loss": 2.3804,
+      "loss": 0.0527,
       "step": 360
     },
     {
       "epoch": 22.008333333333333,
-      "eval_accuracy": 0.2974137931034483,
-      "eval_loss": 2.822291135787964,
-      "eval_runtime": 161.0637,
-      "eval_samples_per_second": 8.643,
-      "eval_steps_per_second": 0.137,
+      "eval_accuracy": 0.8148148148148148,
+      "eval_loss": 0.6200993657112122,
+      "eval_runtime": 60.0717,
+      "eval_samples_per_second": 4.495,
+      "eval_steps_per_second": 0.083,
       "step": 368
     },
     {
       "epoch": 23.001041666666666,
-      "grad_norm": 6.657947540283203,
+      "grad_norm": 6.070339202880859,
       "learning_rate": 4.484953703703704e-05,
-      "loss": 2.418,
+      "loss": 0.043,
       "step": 370
     },
     {
       "epoch": 23.00625,
-      "grad_norm": 5.21594762802124,
+      "grad_norm": 2.227738857269287,
       "learning_rate": 4.456018518518519e-05,
-      "loss": 2.2389,
+      "loss": 0.0856,
       "step": 380
     },
     {
       "epoch": 23.008333333333333,
-      "eval_accuracy": 0.34123563218390807,
-      "eval_loss": 2.7041523456573486,
-      "eval_runtime": 163.8868,
-      "eval_samples_per_second": 8.494,
-      "eval_steps_per_second": 0.134,
+      "eval_accuracy": 0.8,
+      "eval_loss": 0.6439046859741211,
+      "eval_runtime": 60.3917,
+      "eval_samples_per_second": 4.471,
+      "eval_steps_per_second": 0.083,
       "step": 384
     },
     {
       "epoch": 24.003125,
-      "grad_norm": 6.3385009765625,
+      "grad_norm": 4.7137556076049805,
       "learning_rate": 4.4270833333333337e-05,
-      "loss": 2.195,
+      "loss": 0.0627,
       "step": 390
     },
     {
       "epoch": 24.008333333333333,
-      "grad_norm": 14.863781929016113,
+      "grad_norm": 2.8616156578063965,
       "learning_rate": 4.3981481481481486e-05,
-      "loss": 2.1889,
+      "loss": 0.0621,
       "step": 400
     },
     {
       "epoch": 24.008333333333333,
-      "eval_accuracy": 0.3225574712643678,
-      "eval_loss": 2.6846468448638916,
-      "eval_runtime": 162.1045,
-      "eval_samples_per_second": 8.587,
-      "eval_steps_per_second": 0.136,
+      "eval_accuracy": 0.7703703703703704,
+      "eval_loss": 0.8606389760971069,
+      "eval_runtime": 58.2765,
+      "eval_samples_per_second": 4.633,
+      "eval_steps_per_second": 0.086,
       "step": 400
     },
     {
       "epoch": 25.005208333333332,
-      "grad_norm": 7.218128204345703,
+      "grad_norm": 2.4379124641418457,
       "learning_rate": 4.369212962962963e-05,
-      "loss": 1.9829,
+      "loss": 0.0725,
       "step": 410
     },
     {
       "epoch": 25.008333333333333,
-      "eval_accuracy": 0.35919540229885055,
-      "eval_loss": 2.6066086292266846,
-      "eval_runtime": 143.5991,
-      "eval_samples_per_second": 9.694,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.635921835899353,
+      "eval_runtime": 60.9701,
+      "eval_samples_per_second": 4.428,
+      "eval_steps_per_second": 0.082,
       "step": 416
     },
     {
       "epoch": 26.002083333333335,
-      "grad_norm": 5.6082763671875,
+      "grad_norm": 3.5918140411376953,
       "learning_rate": 4.340277777777778e-05,
-      "loss": 2.1112,
+      "loss": 0.0377,
       "step": 420
     },
     {
       "epoch": 26.007291666666667,
-      "grad_norm": 5.779495716094971,
+      "grad_norm": 5.705934047698975,
       "learning_rate": 4.311342592592593e-05,
-      "loss": 1.9466,
+      "loss": 0.0659,
       "step": 430
     },
     {
       "epoch": 26.008333333333333,
-      "eval_accuracy": 0.3642241379310345,
-      "eval_loss": 2.584472417831421,
-      "eval_runtime": 134.6836,
-      "eval_samples_per_second": 10.335,
-      "eval_steps_per_second": 0.163,
+      "eval_accuracy": 0.825925925925926,
+      "eval_loss": 0.651319146156311,
+      "eval_runtime": 52.9634,
+      "eval_samples_per_second": 5.098,
+      "eval_steps_per_second": 0.094,
       "step": 432
     },
     {
       "epoch": 27.004166666666666,
-      "grad_norm": 5.747377395629883,
+      "grad_norm": 2.7037317752838135,
       "learning_rate": 4.282407407407408e-05,
-      "loss": 1.8991,
+      "loss": 0.036,
       "step": 440
     },
     {
       "epoch": 27.008333333333333,
-      "eval_accuracy": 0.3922413793103448,
-      "eval_loss": 2.5150339603424072,
-      "eval_runtime": 160.9218,
-      "eval_samples_per_second": 8.65,
-      "eval_steps_per_second": 0.137,
+      "eval_accuracy": 0.8111111111111111,
+      "eval_loss": 0.6299635767936707,
+      "eval_runtime": 70.2621,
+      "eval_samples_per_second": 3.843,
+      "eval_steps_per_second": 0.071,
       "step": 448
     },
     {
       "epoch": 28.001041666666666,
-      "grad_norm": 5.8224639892578125,
+      "grad_norm": 2.0669760704040527,
       "learning_rate": 4.253472222222222e-05,
-      "loss": 1.9699,
+      "loss": 0.0386,
       "step": 450
     },
     {
       "epoch": 28.00625,
-      "grad_norm": 6.3729329109191895,
+      "grad_norm": 0.44117018580436707,
       "learning_rate": 4.224537037037037e-05,
-      "loss": 1.8629,
+      "loss": 0.0337,
       "step": 460
     },
     {
       "epoch": 28.008333333333333,
-      "eval_accuracy": 0.3994252873563218,
-      "eval_loss": 2.4958128929138184,
-      "eval_runtime": 133.0775,
-      "eval_samples_per_second": 10.46,
-      "eval_steps_per_second": 0.165,
+      "eval_accuracy": 0.8444444444444444,
+      "eval_loss": 0.6410818696022034,
+      "eval_runtime": 50.5516,
+      "eval_samples_per_second": 5.341,
+      "eval_steps_per_second": 0.099,
       "step": 464
     },
     {
       "epoch": 29.003125,
-      "grad_norm": 7.140518665313721,
+      "grad_norm": 1.7283190488815308,
       "learning_rate": 4.195601851851852e-05,
-      "loss": 1.8899,
+      "loss": 0.0291,
       "step": 470
     },
     {
       "epoch": 29.008333333333333,
-      "grad_norm": 27.132600784301758,
+      "grad_norm": 0.5234139561653137,
       "learning_rate": 4.166666666666667e-05,
-      "loss": 1.8563,
+      "loss": 0.0249,
       "step": 480
     },
     {
       "epoch": 29.008333333333333,
-      "eval_accuracy": 0.3994252873563218,
-      "eval_loss": 2.503603219985962,
-      "eval_runtime": 146.1023,
-      "eval_samples_per_second": 9.528,
-      "eval_steps_per_second": 0.151,
+      "eval_accuracy": 0.8592592592592593,
+      "eval_loss": 0.565715491771698,
+      "eval_runtime": 50.8232,
+      "eval_samples_per_second": 5.313,
+      "eval_steps_per_second": 0.098,
       "step": 480
     },
     {
       "epoch": 30.005208333333332,
-      "grad_norm": 6.655496120452881,
+      "grad_norm": 0.6814005374908447,
       "learning_rate": 4.1377314814814815e-05,
-      "loss": 1.832,
+      "loss": 0.0236,
       "step": 490
     },
     {
       "epoch": 30.008333333333333,
-      "eval_accuracy": 0.40373563218390807,
-      "eval_loss": 2.4212334156036377,
-      "eval_runtime": 144.6194,
-      "eval_samples_per_second": 9.625,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.8296296296296296,
+      "eval_loss": 0.5584858059883118,
+      "eval_runtime": 61.3005,
+      "eval_samples_per_second": 4.405,
+      "eval_steps_per_second": 0.082,
       "step": 496
     },
     {
       "epoch": 31.002083333333335,
-      "grad_norm": 7.301593780517578,
+      "grad_norm": 4.192360877990723,
       "learning_rate": 4.1087962962962965e-05,
-      "loss": 1.6729,
+      "loss": 0.0492,
       "step": 500
     },
     {
       "epoch": 31.007291666666667,
-      "grad_norm": 6.040745735168457,
+      "grad_norm": 6.4834885597229,
       "learning_rate": 4.0798611111111115e-05,
-      "loss": 1.7148,
+      "loss": 0.0488,
       "step": 510
     },
     {
       "epoch": 31.008333333333333,
-      "eval_accuracy": 0.42528735632183906,
-      "eval_loss": 2.389084815979004,
-      "eval_runtime": 141.2811,
-      "eval_samples_per_second": 9.853,
-      "eval_steps_per_second": 0.156,
+      "eval_accuracy": 0.8148148148148148,
+      "eval_loss": 0.6616935729980469,
+      "eval_runtime": 51.8099,
+      "eval_samples_per_second": 5.211,
+      "eval_steps_per_second": 0.097,
       "step": 512
     },
     {
       "epoch": 32.00416666666667,
-      "grad_norm": 6.346589088439941,
+      "grad_norm": 1.9468162059783936,
       "learning_rate": 4.0509259259259265e-05,
-      "loss": 1.6525,
+      "loss": 0.0327,
       "step": 520
     },
     {
       "epoch": 32.00833333333333,
-      "eval_accuracy": 0.4109195402298851,
-      "eval_loss": 2.3817079067230225,
-      "eval_runtime": 122.5427,
-      "eval_samples_per_second": 11.359,
-      "eval_steps_per_second": 0.18,
+      "eval_accuracy": 0.8407407407407408,
+      "eval_loss": 0.5680005550384521,
+      "eval_runtime": 58.0544,
+      "eval_samples_per_second": 4.651,
+      "eval_steps_per_second": 0.086,
       "step": 528
     },
     {
       "epoch": 33.001041666666666,
-      "grad_norm": 6.116299629211426,
+      "grad_norm": 0.5346788763999939,
       "learning_rate": 4.021990740740741e-05,
-      "loss": 1.6292,
+      "loss": 0.013,
       "step": 530
     },
     {
       "epoch": 33.00625,
-      "grad_norm": 6.249851703643799,
+      "grad_norm": 1.564922571182251,
       "learning_rate": 3.993055555555556e-05,
-      "loss": 1.6489,
+      "loss": 0.0367,
       "step": 540
     },
     {
       "epoch": 33.00833333333333,
-      "eval_accuracy": 0.4274425287356322,
-      "eval_loss": 2.3350534439086914,
-      "eval_runtime": 118.7762,
-      "eval_samples_per_second": 11.72,
-      "eval_steps_per_second": 0.185,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.7029944658279419,
+      "eval_runtime": 53.8788,
+      "eval_samples_per_second": 5.011,
+      "eval_steps_per_second": 0.093,
       "step": 544
     },
     {
       "epoch": 34.003125,
-      "grad_norm": 5.983939170837402,
+      "grad_norm": 4.385405540466309,
       "learning_rate": 3.964120370370371e-05,
-      "loss": 1.6495,
+      "loss": 0.0388,
       "step": 550
     },
     {
       "epoch": 34.00833333333333,
-      "grad_norm": 17.406435012817383,
+      "grad_norm": 0.3302398920059204,
       "learning_rate": 3.935185185185186e-05,
-      "loss": 1.6928,
+      "loss": 0.0226,
       "step": 560
     },
     {
       "epoch": 34.00833333333333,
-      "eval_accuracy": 0.4188218390804598,
-      "eval_loss": 2.3494503498077393,
-      "eval_runtime": 112.8904,
-      "eval_samples_per_second": 12.331,
-      "eval_steps_per_second": 0.195,
+      "eval_accuracy": 0.7592592592592593,
+      "eval_loss": 0.8866225481033325,
+      "eval_runtime": 50.498,
+      "eval_samples_per_second": 5.347,
+      "eval_steps_per_second": 0.099,
       "step": 560
     },
     {
       "epoch": 35.005208333333336,
-      "grad_norm": 6.939193248748779,
+      "grad_norm": 1.117612600326538,
       "learning_rate": 3.90625e-05,
-      "loss": 1.503,
+      "loss": 0.0277,
       "step": 570
     },
     {
       "epoch": 35.00833333333333,
-      "eval_accuracy": 0.4274425287356322,
-      "eval_loss": 2.2961111068725586,
-      "eval_runtime": 122.4307,
-      "eval_samples_per_second": 11.37,
-      "eval_steps_per_second": 0.18,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.8433952331542969,
+      "eval_runtime": 50.9828,
+      "eval_samples_per_second": 5.296,
+      "eval_steps_per_second": 0.098,
       "step": 576
     },
     {
       "epoch": 36.00208333333333,
-      "grad_norm": 8.047181129455566,
+      "grad_norm": 1.3228788375854492,
       "learning_rate": 3.877314814814815e-05,
-      "loss": 1.4577,
+      "loss": 0.0218,
       "step": 580
     },
     {
       "epoch": 36.00729166666667,
-      "grad_norm": 5.338744640350342,
+      "grad_norm": 0.6263003349304199,
       "learning_rate": 3.84837962962963e-05,
-      "loss": 1.5126,
+      "loss": 0.0136,
       "step": 590
     },
     {
       "epoch": 36.00833333333333,
-      "eval_accuracy": 0.4454022988505747,
-      "eval_loss": 2.2619619369506836,
-      "eval_runtime": 143.5341,
-      "eval_samples_per_second": 9.698,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.7777777777777778,
+      "eval_loss": 0.7817791700363159,
+      "eval_runtime": 52.59,
+      "eval_samples_per_second": 5.134,
+      "eval_steps_per_second": 0.095,
       "step": 592
     },
     {
       "epoch": 37.00416666666667,
-      "grad_norm": 6.184549808502197,
+      "grad_norm": 1.3871808052062988,
       "learning_rate": 3.8194444444444444e-05,
-      "loss": 1.4732,
+      "loss": 0.017,
       "step": 600
     },
     {
       "epoch": 37.00833333333333,
-      "eval_accuracy": 0.4367816091954023,
-      "eval_loss": 2.2635250091552734,
-      "eval_runtime": 145.8205,
-      "eval_samples_per_second": 9.546,
-      "eval_steps_per_second": 0.151,
+      "eval_accuracy": 0.7592592592592593,
+      "eval_loss": 0.7850788235664368,
+      "eval_runtime": 51.6164,
+      "eval_samples_per_second": 5.231,
+      "eval_steps_per_second": 0.097,
       "step": 608
     },
     {
       "epoch": 38.001041666666666,
-      "grad_norm": 6.274258136749268,
+      "grad_norm": 0.17293395102024078,
       "learning_rate": 3.7905092592592594e-05,
-      "loss": 1.52,
+      "loss": 0.0855,
       "step": 610
     },
     {
       "epoch": 38.00625,
-      "grad_norm": 6.8546576499938965,
+      "grad_norm": 2.0498270988464355,
       "learning_rate": 3.7615740740740744e-05,
-      "loss": 1.5553,
+      "loss": 0.0391,
       "step": 620
     },
     {
       "epoch": 38.00833333333333,
-      "eval_accuracy": 0.4489942528735632,
-      "eval_loss": 2.2326316833496094,
-      "eval_runtime": 121.0135,
-      "eval_samples_per_second": 11.503,
-      "eval_steps_per_second": 0.182,
+      "eval_accuracy": 0.7481481481481481,
+      "eval_loss": 1.0256072282791138,
+      "eval_runtime": 53.8376,
+      "eval_samples_per_second": 5.015,
+      "eval_steps_per_second": 0.093,
       "step": 624
     },
     {
       "epoch": 39.003125,
-      "grad_norm": 6.396101951599121,
+      "grad_norm": 6.026589393615723,
       "learning_rate": 3.7326388888888893e-05,
-      "loss": 1.4549,
+      "loss": 0.0373,
       "step": 630
     },
     {
       "epoch": 39.00833333333333,
-      "grad_norm": 22.021570205688477,
+      "grad_norm": 0.1762811839580536,
       "learning_rate": 3.7037037037037037e-05,
-      "loss": 1.7115,
+      "loss": 0.0211,
       "step": 640
     },
     {
       "epoch": 39.00833333333333,
-      "eval_accuracy": 0.44037356321839083,
-      "eval_loss": 2.2265872955322266,
-      "eval_runtime": 132.2101,
-      "eval_samples_per_second": 10.529,
-      "eval_steps_per_second": 0.166,
+      "eval_accuracy": 0.7592592592592593,
+      "eval_loss": 0.922510027885437,
+      "eval_runtime": 51.7154,
+      "eval_samples_per_second": 5.221,
+      "eval_steps_per_second": 0.097,
       "step": 640
     },
     {
       "epoch": 40.005208333333336,
-      "grad_norm": 5.9612627029418945,
+      "grad_norm": 3.7255866527557373,
       "learning_rate": 3.6747685185185186e-05,
-      "loss": 1.4851,
+      "loss": 0.0322,
       "step": 650
     },
     {
       "epoch": 40.00833333333333,
-      "eval_accuracy": 0.4274425287356322,
-      "eval_loss": 2.2690393924713135,
-      "eval_runtime": 133.5177,
-      "eval_samples_per_second": 10.426,
-      "eval_steps_per_second": 0.165,
+      "eval_accuracy": 0.7925925925925926,
+      "eval_loss": 0.7321712374687195,
+      "eval_runtime": 69.4978,
+      "eval_samples_per_second": 3.885,
+      "eval_steps_per_second": 0.072,
       "step": 656
     },
     {
       "epoch": 41.00208333333333,
-      "grad_norm": 5.4721550941467285,
+      "grad_norm": 0.14019830524921417,
       "learning_rate": 3.6458333333333336e-05,
-      "loss": 1.4883,
+      "loss": 0.0111,
       "step": 660
     },
     {
       "epoch": 41.00729166666667,
-      "grad_norm": 8.272603034973145,
+      "grad_norm": 0.7536612153053284,
       "learning_rate": 3.6168981481481486e-05,
-      "loss": 1.455,
+      "loss": 0.0203,
       "step": 670
     },
     {
       "epoch": 41.00833333333333,
-      "eval_accuracy": 0.45689655172413796,
-      "eval_loss": 2.192133665084839,
-      "eval_runtime": 134.8188,
-      "eval_samples_per_second": 10.325,
-      "eval_steps_per_second": 0.163,
+      "eval_accuracy": 0.7851851851851852,
+      "eval_loss": 0.7956416606903076,
+      "eval_runtime": 50.3091,
+      "eval_samples_per_second": 5.367,
+      "eval_steps_per_second": 0.099,
       "step": 672
     },
     {
       "epoch": 42.00416666666667,
-      "grad_norm": 6.0662407875061035,
+      "grad_norm": 1.30323326587677,
       "learning_rate": 3.587962962962963e-05,
-      "loss": 1.4827,
+      "loss": 0.0223,
       "step": 680
     },
     {
       "epoch": 42.00833333333333,
-      "eval_accuracy": 0.4504310344827586,
-      "eval_loss": 2.238664388656616,
-      "eval_runtime": 129.8736,
-      "eval_samples_per_second": 10.718,
-      "eval_steps_per_second": 0.169,
+      "eval_accuracy": 0.7703703703703704,
+      "eval_loss": 0.8495256304740906,
+      "eval_runtime": 54.6719,
+      "eval_samples_per_second": 4.939,
+      "eval_steps_per_second": 0.091,
       "step": 688
     },
     {
       "epoch": 43.001041666666666,
-      "grad_norm": 8.369169235229492,
+      "grad_norm": 0.34746453166007996,
       "learning_rate": 3.559027777777778e-05,
-      "loss": 1.2639,
+      "loss": 0.0085,
       "step": 690
     },
     {
       "epoch": 43.00625,
-      "grad_norm": 9.68371868133545,
+      "grad_norm": 1.755631685256958,
       "learning_rate": 3.530092592592593e-05,
-      "loss": 1.4839,
+      "loss": 0.0228,
       "step": 700
     },
     {
       "epoch": 43.00833333333333,
-      "eval_accuracy": 0.4432471264367816,
-      "eval_loss": 2.2020034790039062,
-      "eval_runtime": 146.2451,
-      "eval_samples_per_second": 9.518,
-      "eval_steps_per_second": 0.15,
+      "eval_accuracy": 0.825925925925926,
+      "eval_loss": 0.6640289425849915,
+      "eval_runtime": 55.9727,
+      "eval_samples_per_second": 4.824,
+      "eval_steps_per_second": 0.089,
       "step": 704
     },
     {
       "epoch": 44.003125,
-      "grad_norm": 6.168234825134277,
+      "grad_norm": 0.9141967296600342,
       "learning_rate": 3.501157407407408e-05,
-      "loss": 1.1762,
+      "loss": 0.0133,
       "step": 710
     },
     {
       "epoch": 44.00833333333333,
-      "grad_norm": 29.257152557373047,
+      "grad_norm": 0.3903932571411133,
       "learning_rate": 3.472222222222222e-05,
-      "loss": 1.2879,
+      "loss": 0.0115,
       "step": 720
     },
     {
       "epoch": 44.00833333333333,
-      "eval_accuracy": 0.4396551724137931,
-      "eval_loss": 2.195861339569092,
-      "eval_runtime": 118.2366,
-      "eval_samples_per_second": 11.773,
-      "eval_steps_per_second": 0.186,
+      "eval_accuracy": 0.7592592592592593,
+      "eval_loss": 0.9644998908042908,
+      "eval_runtime": 51.7453,
+      "eval_samples_per_second": 5.218,
+      "eval_steps_per_second": 0.097,
       "step": 720
     },
     {
       "epoch": 45.005208333333336,
-      "grad_norm": 7.243413925170898,
+      "grad_norm": 0.17332051694393158,
       "learning_rate": 3.443287037037037e-05,
-      "loss": 1.2722,
+      "loss": 0.0222,
       "step": 730
     },
     {
       "epoch": 45.00833333333333,
-      "eval_accuracy": 0.44396551724137934,
-      "eval_loss": 2.2157773971557617,
-      "eval_runtime": 131.0025,
-      "eval_samples_per_second": 10.626,
-      "eval_steps_per_second": 0.168,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.659512996673584,
+      "eval_runtime": 55.7661,
+      "eval_samples_per_second": 4.842,
+      "eval_steps_per_second": 0.09,
       "step": 736
     },
     {
       "epoch": 46.00208333333333,
-      "grad_norm": 6.851569652557373,
+      "grad_norm": 0.8124841451644897,
       "learning_rate": 3.414351851851852e-05,
-      "loss": 1.4204,
+      "loss": 0.0091,
       "step": 740
     },
     {
       "epoch": 46.00729166666667,
-      "grad_norm": 5.8133463859558105,
+      "grad_norm": 6.734714984893799,
       "learning_rate": 3.385416666666667e-05,
-      "loss": 1.2225,
+      "loss": 0.0165,
       "step": 750
     },
     {
       "epoch": 46.00833333333333,
-      "eval_accuracy": 0.46623563218390807,
-      "eval_loss": 2.1567935943603516,
-      "eval_runtime": 151.0401,
-      "eval_samples_per_second": 9.216,
-      "eval_steps_per_second": 0.146,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.7119552493095398,
+      "eval_runtime": 52.2301,
+      "eval_samples_per_second": 5.169,
+      "eval_steps_per_second": 0.096,
       "step": 752
     },
     {
       "epoch": 47.00416666666667,
-      "grad_norm": 5.665341377258301,
+      "grad_norm": 0.2893330752849579,
       "learning_rate": 3.3564814814814815e-05,
-      "loss": 1.1821,
+      "loss": 0.0165,
       "step": 760
     },
     {
       "epoch": 47.00833333333333,
-      "eval_accuracy": 0.47629310344827586,
-      "eval_loss": 2.1312315464019775,
-      "eval_runtime": 113.8029,
-      "eval_samples_per_second": 12.232,
-      "eval_steps_per_second": 0.193,
+      "eval_accuracy": 0.8,
+      "eval_loss": 0.8027238845825195,
+      "eval_runtime": 67.238,
+      "eval_samples_per_second": 4.016,
+      "eval_steps_per_second": 0.074,
       "step": 768
     },
     {
       "epoch": 48.001041666666666,
-      "grad_norm": 6.758544445037842,
+      "grad_norm": 1.3208953142166138,
       "learning_rate": 3.3275462962962965e-05,
-      "loss": 1.2554,
+      "loss": 0.0124,
       "step": 770
     },
     {
       "epoch": 48.00625,
-      "grad_norm": 7.730432510375977,
+      "grad_norm": 0.2127707153558731,
       "learning_rate": 3.2986111111111115e-05,
-      "loss": 1.2406,
+      "loss": 0.0166,
       "step": 780
     },
     {
       "epoch": 48.00833333333333,
-      "eval_accuracy": 0.47844827586206895,
-      "eval_loss": 2.1162068843841553,
-      "eval_runtime": 138.6268,
-      "eval_samples_per_second": 10.041,
-      "eval_steps_per_second": 0.159,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.848490297794342,
+      "eval_runtime": 67.7188,
+      "eval_samples_per_second": 3.987,
+      "eval_steps_per_second": 0.074,
       "step": 784
     },
     {
       "epoch": 49.003125,
-      "grad_norm": 5.654488563537598,
+      "grad_norm": 0.08924805372953415,
       "learning_rate": 3.2696759259259265e-05,
-      "loss": 1.2432,
+      "loss": 0.0106,
       "step": 790
     },
     {
       "epoch": 49.00833333333333,
-      "grad_norm": 21.777118682861328,
+      "grad_norm": 0.055565591901540756,
       "learning_rate": 3.240740740740741e-05,
-      "loss": 1.1717,
+      "loss": 0.0097,
       "step": 800
     },
     {
       "epoch": 49.00833333333333,
-      "eval_accuracy": 0.4755747126436782,
-      "eval_loss": 2.1368021965026855,
-      "eval_runtime": 124.2033,
-      "eval_samples_per_second": 11.207,
-      "eval_steps_per_second": 0.177,
+      "eval_accuracy": 0.7925925925925926,
+      "eval_loss": 0.850395143032074,
+      "eval_runtime": 49.8196,
+      "eval_samples_per_second": 5.42,
+      "eval_steps_per_second": 0.1,
       "step": 800
     },
     {
       "epoch": 50.005208333333336,
-      "grad_norm": 6.306783199310303,
+      "grad_norm": 2.7151601314544678,
       "learning_rate": 3.211805555555556e-05,
-      "loss": 1.2366,
+      "loss": 0.0257,
       "step": 810
     },
     {
       "epoch": 50.00833333333333,
-      "eval_accuracy": 0.4834770114942529,
-      "eval_loss": 2.1134183406829834,
-      "eval_runtime": 139.8423,
-      "eval_samples_per_second": 9.954,
-      "eval_steps_per_second": 0.157,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.7933987975120544,
+      "eval_runtime": 50.3254,
+      "eval_samples_per_second": 5.365,
+      "eval_steps_per_second": 0.099,
       "step": 816
     },
     {
       "epoch": 51.00208333333333,
-      "grad_norm": 6.758057117462158,
+      "grad_norm": 0.1394781917333603,
       "learning_rate": 3.182870370370371e-05,
-      "loss": 1.1925,
+      "loss": 0.0219,
       "step": 820
     },
     {
       "epoch": 51.00729166666667,
-      "grad_norm": 6.371450424194336,
+      "grad_norm": 2.1916935443878174,
       "learning_rate": 3.153935185185186e-05,
-      "loss": 1.2534,
+      "loss": 0.0172,
       "step": 830
     },
     {
       "epoch": 51.00833333333333,
-      "eval_accuracy": 0.4734195402298851,
-      "eval_loss": 2.096381664276123,
-      "eval_runtime": 153.8561,
-      "eval_samples_per_second": 9.047,
-      "eval_steps_per_second": 0.143,
+      "eval_accuracy": 0.8037037037037037,
+      "eval_loss": 0.7561771869659424,
+      "eval_runtime": 48.799,
+      "eval_samples_per_second": 5.533,
+      "eval_steps_per_second": 0.102,
       "step": 832
     },
     {
       "epoch": 52.00416666666667,
-      "grad_norm": 6.33159065246582,
+      "grad_norm": 0.6336559057235718,
       "learning_rate": 3.125e-05,
-      "loss": 1.2322,
+      "loss": 0.0064,
       "step": 840
     },
     {
       "epoch": 52.00833333333333,
-      "eval_accuracy": 0.4640804597701149,
-      "eval_loss": 2.150590658187866,
-      "eval_runtime": 115.0189,
-      "eval_samples_per_second": 12.102,
-      "eval_steps_per_second": 0.191,
+      "eval_accuracy": 0.8111111111111111,
+      "eval_loss": 0.709699273109436,
+      "eval_runtime": 59.7458,
+      "eval_samples_per_second": 4.519,
+      "eval_steps_per_second": 0.084,
       "step": 848
     },
     {
       "epoch": 53.001041666666666,
-      "grad_norm": 5.289525985717773,
+      "grad_norm": 0.23525160551071167,
       "learning_rate": 3.0960648148148144e-05,
-      "loss": 1.2281,
+      "loss": 0.0191,
       "step": 850
     },
     {
       "epoch": 53.00625,
-      "grad_norm": 6.158384799957275,
+      "grad_norm": 0.26842543482780457,
       "learning_rate": 3.06712962962963e-05,
-      "loss": 1.2742,
+      "loss": 0.0052,
       "step": 860
     },
     {
       "epoch": 53.00833333333333,
-      "eval_accuracy": 0.45905172413793105,
-      "eval_loss": 2.171858310699463,
-      "eval_runtime": 139.4117,
-      "eval_samples_per_second": 9.985,
-      "eval_steps_per_second": 0.158,
+      "eval_accuracy": 0.7962962962962963,
+      "eval_loss": 0.7537274956703186,
+      "eval_runtime": 50.2392,
+      "eval_samples_per_second": 5.374,
+      "eval_steps_per_second": 0.1,
       "step": 864
     },
     {
       "epoch": 54.003125,
-      "grad_norm": 4.942337989807129,
+      "grad_norm": 2.6570372581481934,
       "learning_rate": 3.0381944444444444e-05,
-      "loss": 1.1286,
+      "loss": 0.0154,
       "step": 870
     },
     {
       "epoch": 54.00833333333333,
-      "grad_norm": 20.070188522338867,
+      "grad_norm": 0.16351526975631714,
       "learning_rate": 3.0092592592592593e-05,
-      "loss": 1.132,
+      "loss": 0.012,
       "step": 880
     },
     {
       "epoch": 54.00833333333333,
-      "eval_accuracy": 0.46479885057471265,
-      "eval_loss": 2.1557226181030273,
-      "eval_runtime": 133.4523,
-      "eval_samples_per_second": 10.431,
-      "eval_steps_per_second": 0.165,
+      "eval_accuracy": 0.8074074074074075,
+      "eval_loss": 0.7386136054992676,
+      "eval_runtime": 58.982,
+      "eval_samples_per_second": 4.578,
+      "eval_steps_per_second": 0.085,
       "step": 880
     },
     {
       "epoch": 55.005208333333336,
-      "grad_norm": 7.356424808502197,
+      "grad_norm": 0.10845500230789185,
       "learning_rate": 2.980324074074074e-05,
-      "loss": 1.1306,
+      "loss": 0.0174,
       "step": 890
     },
     {
       "epoch": 55.00833333333333,
-      "eval_accuracy": 0.48778735632183906,
-      "eval_loss": 2.101032018661499,
-      "eval_runtime": 120.4781,
-      "eval_samples_per_second": 11.554,
-      "eval_steps_per_second": 0.183,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.6893912553787231,
+      "eval_runtime": 55.5333,
+      "eval_samples_per_second": 4.862,
+      "eval_steps_per_second": 0.09,
       "step": 896
     },
     {
       "epoch": 56.00208333333333,
-      "grad_norm": 6.045055389404297,
+      "grad_norm": 0.20301461219787598,
       "learning_rate": 2.951388888888889e-05,
-      "loss": 1.1428,
+      "loss": 0.0085,
       "step": 900
     },
     {
       "epoch": 56.00729166666667,
-      "grad_norm": 8.3392915725708,
+      "grad_norm": 0.4829808473587036,
       "learning_rate": 2.9224537037037036e-05,
-      "loss": 1.2719,
+      "loss": 0.0151,
       "step": 910
     },
     {
       "epoch": 56.00833333333333,
-      "eval_accuracy": 0.4619252873563218,
-      "eval_loss": 2.148237466812134,
-      "eval_runtime": 113.6281,
-      "eval_samples_per_second": 12.25,
-      "eval_steps_per_second": 0.194,
+      "eval_accuracy": 0.7666666666666667,
+      "eval_loss": 0.9359716773033142,
+      "eval_runtime": 51.4879,
+      "eval_samples_per_second": 5.244,
+      "eval_steps_per_second": 0.097,
       "step": 912
     },
     {
       "epoch": 57.00416666666667,
-      "grad_norm": 5.768066883087158,
+      "grad_norm": 1.294859766960144,
       "learning_rate": 2.8935185185185186e-05,
-      "loss": 1.1549,
+      "loss": 0.0081,
       "step": 920
     },
     {
       "epoch": 57.00833333333333,
-      "eval_accuracy": 0.4813218390804598,
-      "eval_loss": 2.0960781574249268,
-      "eval_runtime": 151.6984,
-      "eval_samples_per_second": 9.176,
-      "eval_steps_per_second": 0.145,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.7101936340332031,
+      "eval_runtime": 53.7781,
+      "eval_samples_per_second": 5.021,
+      "eval_steps_per_second": 0.093,
       "step": 928
     },
     {
       "epoch": 58.001041666666666,
-      "grad_norm": 7.034135818481445,
+      "grad_norm": 5.858212471008301,
       "learning_rate": 2.8645833333333333e-05,
-      "loss": 1.0771,
+      "loss": 0.0091,
       "step": 930
     },
     {
       "epoch": 58.00625,
-      "grad_norm": 7.099185466766357,
+      "grad_norm": 0.4849383533000946,
       "learning_rate": 2.8356481481481483e-05,
-      "loss": 1.1495,
+      "loss": 0.0142,
       "step": 940
     },
     {
       "epoch": 58.00833333333333,
-      "eval_accuracy": 0.47629310344827586,
-      "eval_loss": 2.124603509902954,
-      "eval_runtime": 127.3723,
-      "eval_samples_per_second": 10.929,
-      "eval_steps_per_second": 0.173,
+      "eval_accuracy": 0.8111111111111111,
+      "eval_loss": 0.7865731120109558,
+      "eval_runtime": 52.9255,
+      "eval_samples_per_second": 5.102,
+      "eval_steps_per_second": 0.094,
       "step": 944
     },
     {
       "epoch": 59.003125,
-      "grad_norm": 6.804399013519287,
+      "grad_norm": 0.24274876713752747,
       "learning_rate": 2.806712962962963e-05,
-      "loss": 1.1043,
+      "loss": 0.0129,
       "step": 950
     },
     {
       "epoch": 59.00833333333333,
-      "grad_norm": 31.837963104248047,
+      "grad_norm": 0.06653541326522827,
       "learning_rate": 2.777777777777778e-05,
-      "loss": 1.2539,
+      "loss": 0.0169,
       "step": 960
     },
     {
       "epoch": 59.00833333333333,
-      "eval_accuracy": 0.48060344827586204,
-      "eval_loss": 2.1117918491363525,
-      "eval_runtime": 142.6628,
-      "eval_samples_per_second": 9.757,
-      "eval_steps_per_second": 0.154,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6516271233558655,
+      "eval_runtime": 51.7732,
+      "eval_samples_per_second": 5.215,
+      "eval_steps_per_second": 0.097,
       "step": 960
     },
     {
       "epoch": 60.005208333333336,
-      "grad_norm": 7.992947578430176,
+      "grad_norm": 2.3222341537475586,
       "learning_rate": 2.7488425925925926e-05,
-      "loss": 1.1719,
+      "loss": 0.0149,
       "step": 970
     },
     {
       "epoch": 60.00833333333333,
-      "eval_accuracy": 0.49353448275862066,
-      "eval_loss": 2.0665571689605713,
-      "eval_runtime": 129.4806,
-      "eval_samples_per_second": 10.751,
-      "eval_steps_per_second": 0.17,
+      "eval_accuracy": 0.7555555555555555,
+      "eval_loss": 1.0039104223251343,
+      "eval_runtime": 47.1858,
+      "eval_samples_per_second": 5.722,
+      "eval_steps_per_second": 0.106,
       "step": 976
     },
     {
       "epoch": 61.00208333333333,
-      "grad_norm": 5.261826038360596,
+      "grad_norm": 0.18627400696277618,
       "learning_rate": 2.7199074074074076e-05,
-      "loss": 1.055,
+      "loss": 0.0126,
       "step": 980
     },
     {
       "epoch": 61.00729166666667,
-      "grad_norm": 8.344293594360352,
+      "grad_norm": 7.43377161026001,
       "learning_rate": 2.6909722222222222e-05,
-      "loss": 1.1108,
+      "loss": 0.0106,
       "step": 990
     },
     {
       "epoch": 61.00833333333333,
-      "eval_accuracy": 0.4834770114942529,
-      "eval_loss": 2.0630218982696533,
-      "eval_runtime": 111.5142,
-      "eval_samples_per_second": 12.483,
-      "eval_steps_per_second": 0.197,
+      "eval_accuracy": 0.8407407407407408,
+      "eval_loss": 0.6569637656211853,
+      "eval_runtime": 51.3789,
+      "eval_samples_per_second": 5.255,
+      "eval_steps_per_second": 0.097,
       "step": 992
     },
     {
       "epoch": 62.00416666666667,
-      "grad_norm": 7.500652313232422,
+      "grad_norm": 0.07728447020053864,
       "learning_rate": 2.6620370370370372e-05,
-      "loss": 1.0417,
+      "loss": 0.005,
       "step": 1000
     },
     {
       "epoch": 62.00833333333333,
-      "eval_accuracy": 0.4899425287356322,
-      "eval_loss": 2.0634570121765137,
-      "eval_runtime": 126.2555,
-      "eval_samples_per_second": 11.025,
-      "eval_steps_per_second": 0.174,
+      "eval_accuracy": 0.8037037037037037,
+      "eval_loss": 0.7251839637756348,
+      "eval_runtime": 52.0231,
+      "eval_samples_per_second": 5.19,
+      "eval_steps_per_second": 0.096,
       "step": 1008
     },
     {
       "epoch": 63.001041666666666,
-      "grad_norm": 6.283971786499023,
+      "grad_norm": 0.05448053032159805,
       "learning_rate": 2.633101851851852e-05,
-      "loss": 1.2428,
+      "loss": 0.0153,
       "step": 1010
     },
     {
       "epoch": 63.00625,
-      "grad_norm": 7.511735439300537,
+      "grad_norm": 0.268702894449234,
       "learning_rate": 2.604166666666667e-05,
-      "loss": 1.1755,
+      "loss": 0.0115,
       "step": 1020
     },
     {
       "epoch": 63.00833333333333,
-      "eval_accuracy": 0.47198275862068967,
-      "eval_loss": 2.0862152576446533,
-      "eval_runtime": 124.1187,
-      "eval_samples_per_second": 11.215,
-      "eval_steps_per_second": 0.177,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.6913211941719055,
+      "eval_runtime": 50.5967,
+      "eval_samples_per_second": 5.336,
+      "eval_steps_per_second": 0.099,
       "step": 1024
     },
     {
       "epoch": 64.003125,
-      "grad_norm": 6.025021076202393,
+      "grad_norm": 5.630356788635254,
       "learning_rate": 2.5752314814814815e-05,
-      "loss": 1.0898,
+      "loss": 0.007,
       "step": 1030
     },
     {
       "epoch": 64.00833333333334,
-      "grad_norm": 23.160959243774414,
+      "grad_norm": 0.07044278085231781,
       "learning_rate": 2.5462962962962965e-05,
-      "loss": 1.0512,
+      "loss": 0.0059,
       "step": 1040
     },
     {
       "epoch": 64.00833333333334,
-      "eval_accuracy": 0.48778735632183906,
-      "eval_loss": 2.073040723800659,
-      "eval_runtime": 128.3013,
-      "eval_samples_per_second": 10.849,
-      "eval_steps_per_second": 0.171,
+      "eval_accuracy": 0.8481481481481481,
+      "eval_loss": 0.6858277320861816,
+      "eval_runtime": 53.2273,
+      "eval_samples_per_second": 5.073,
+      "eval_steps_per_second": 0.094,
       "step": 1040
     },
     {
       "epoch": 65.00520833333333,
-      "grad_norm": 5.617940902709961,
+      "grad_norm": 0.04406805336475372,
       "learning_rate": 2.517361111111111e-05,
-      "loss": 0.9824,
+      "loss": 0.0225,
       "step": 1050
     },
     {
       "epoch": 65.00833333333334,
-      "eval_accuracy": 0.4906609195402299,
-      "eval_loss": 2.070878744125366,
-      "eval_runtime": 122.5737,
-      "eval_samples_per_second": 11.356,
-      "eval_steps_per_second": 0.179,
+      "eval_accuracy": 0.8148148148148148,
+      "eval_loss": 0.7341809272766113,
+      "eval_runtime": 53.5884,
+      "eval_samples_per_second": 5.038,
+      "eval_steps_per_second": 0.093,
       "step": 1056
     },
     {
       "epoch": 66.00208333333333,
-      "grad_norm": 5.892145156860352,
+      "grad_norm": 0.5853094458580017,
       "learning_rate": 2.488425925925926e-05,
-      "loss": 1.1867,
+      "loss": 0.0225,
       "step": 1060
     },
     {
       "epoch": 66.00729166666666,
-      "grad_norm": 7.249321937561035,
+      "grad_norm": 1.6443512439727783,
       "learning_rate": 2.4594907407407408e-05,
-      "loss": 1.0924,
+      "loss": 0.0151,
       "step": 1070
     },
     {
       "epoch": 66.00833333333334,
-      "eval_accuracy": 0.46120689655172414,
-      "eval_loss": 2.16375732421875,
-      "eval_runtime": 110.676,
-      "eval_samples_per_second": 12.577,
-      "eval_steps_per_second": 0.199,
+      "eval_accuracy": 0.825925925925926,
+      "eval_loss": 0.6859686374664307,
+      "eval_runtime": 53.9253,
+      "eval_samples_per_second": 5.007,
+      "eval_steps_per_second": 0.093,
       "step": 1072
     },
     {
       "epoch": 67.00416666666666,
-      "grad_norm": 6.243124485015869,
+      "grad_norm": 0.05437076464295387,
       "learning_rate": 2.4305555555555558e-05,
-      "loss": 1.1027,
+      "loss": 0.0098,
       "step": 1080
     },
     {
       "epoch": 67.00833333333334,
-      "eval_accuracy": 0.47772988505747127,
-      "eval_loss": 2.0572071075439453,
-      "eval_runtime": 123.131,
-      "eval_samples_per_second": 11.305,
-      "eval_steps_per_second": 0.179,
+      "eval_accuracy": 0.8296296296296296,
+      "eval_loss": 0.7040747404098511,
+      "eval_runtime": 53.9398,
+      "eval_samples_per_second": 5.006,
+      "eval_steps_per_second": 0.093,
       "step": 1088
     },
     {
       "epoch": 68.00104166666667,
-      "grad_norm": 7.192434787750244,
+      "grad_norm": 0.15694469213485718,
       "learning_rate": 2.4016203703703704e-05,
-      "loss": 0.942,
+      "loss": 0.0078,
       "step": 1090
     },
     {
       "epoch": 68.00625,
-      "grad_norm": 5.061285018920898,
+      "grad_norm": 2.5937747955322266,
       "learning_rate": 2.3726851851851854e-05,
-      "loss": 1.0956,
+      "loss": 0.0097,
       "step": 1100
     },
     {
       "epoch": 68.00833333333334,
-      "eval_accuracy": 0.4892241379310345,
-      "eval_loss": 2.050154685974121,
-      "eval_runtime": 159.2133,
-      "eval_samples_per_second": 8.743,
-      "eval_steps_per_second": 0.138,
+      "eval_accuracy": 0.8185185185185185,
+      "eval_loss": 0.7320748567581177,
+      "eval_runtime": 53.8385,
+      "eval_samples_per_second": 5.015,
+      "eval_steps_per_second": 0.093,
       "step": 1104
     },
     {
       "epoch": 69.003125,
-      "grad_norm": 6.748392581939697,
+      "grad_norm": 0.08000241965055466,
       "learning_rate": 2.34375e-05,
-      "loss": 1.0495,
+      "loss": 0.0068,
       "step": 1110
     },
     {
       "epoch": 69.00833333333334,
-      "grad_norm": 21.487567901611328,
+      "grad_norm": 0.09789244085550308,
       "learning_rate": 2.314814814814815e-05,
-      "loss": 0.8823,
+      "loss": 0.014,
       "step": 1120
     },
     {
       "epoch": 69.00833333333334,
-      "eval_accuracy": 0.4755747126436782,
-      "eval_loss": 2.112837076187134,
-      "eval_runtime": 115.1946,
-      "eval_samples_per_second": 12.084,
-      "eval_steps_per_second": 0.191,
+      "eval_accuracy": 0.8481481481481481,
+      "eval_loss": 0.6250659227371216,
+      "eval_runtime": 50.0971,
+      "eval_samples_per_second": 5.39,
+      "eval_steps_per_second": 0.1,
       "step": 1120
     },
     {
       "epoch": 70.00520833333333,
-      "grad_norm": 6.899831295013428,
+      "grad_norm": 1.5017173290252686,
       "learning_rate": 2.2858796296296297e-05,
-      "loss": 1.0344,
+      "loss": 0.0252,
       "step": 1130
     },
     {
       "epoch": 70.00833333333334,
-      "eval_accuracy": 0.47270114942528735,
-      "eval_loss": 2.095017433166504,
-      "eval_runtime": 147.4741,
-      "eval_samples_per_second": 9.439,
-      "eval_steps_per_second": 0.149,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6770647764205933,
+      "eval_runtime": 66.7356,
+      "eval_samples_per_second": 4.046,
+      "eval_steps_per_second": 0.075,
       "step": 1136
     },
     {
       "epoch": 71.00208333333333,
-      "grad_norm": 6.141778469085693,
+      "grad_norm": 1.2484806776046753,
       "learning_rate": 2.2569444444444447e-05,
-      "loss": 1.0074,
+      "loss": 0.0054,
       "step": 1140
     },
     {
       "epoch": 71.00729166666666,
-      "grad_norm": 5.4406352043151855,
+      "grad_norm": 0.040970832109451294,
       "learning_rate": 2.2280092592592593e-05,
-      "loss": 1.0887,
+      "loss": 0.0052,
       "step": 1150
     },
     {
       "epoch": 71.00833333333334,
-      "eval_accuracy": 0.4942528735632184,
-      "eval_loss": 2.0542972087860107,
-      "eval_runtime": 125.512,
-      "eval_samples_per_second": 11.091,
-      "eval_steps_per_second": 0.175,
+      "eval_accuracy": 0.8,
+      "eval_loss": 0.7527381181716919,
+      "eval_runtime": 49.5137,
+      "eval_samples_per_second": 5.453,
+      "eval_steps_per_second": 0.101,
       "step": 1152
     },
     {
       "epoch": 72.00416666666666,
-      "grad_norm": 6.820794582366943,
+      "grad_norm": 0.05043178051710129,
       "learning_rate": 2.1990740740740743e-05,
-      "loss": 1.0763,
+      "loss": 0.0189,
       "step": 1160
     },
     {
       "epoch": 72.00833333333334,
-      "eval_accuracy": 0.4906609195402299,
-      "eval_loss": 2.053474187850952,
-      "eval_runtime": 138.7554,
-      "eval_samples_per_second": 10.032,
-      "eval_steps_per_second": 0.159,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.6935626268386841,
+      "eval_runtime": 51.8732,
+      "eval_samples_per_second": 5.205,
+      "eval_steps_per_second": 0.096,
       "step": 1168
     },
     {
       "epoch": 73.00104166666667,
-      "grad_norm": 6.077042102813721,
+      "grad_norm": 0.08731618523597717,
       "learning_rate": 2.170138888888889e-05,
-      "loss": 0.9623,
+      "loss": 0.0045,
       "step": 1170
     },
     {
       "epoch": 73.00625,
-      "grad_norm": 8.70025634765625,
+      "grad_norm": 0.6820685267448425,
       "learning_rate": 2.141203703703704e-05,
-      "loss": 0.9652,
+      "loss": 0.0038,
       "step": 1180
     },
     {
       "epoch": 73.00833333333334,
-      "eval_accuracy": 0.5,
-      "eval_loss": 2.028047561645508,
-      "eval_runtime": 126.5448,
-      "eval_samples_per_second": 11.0,
-      "eval_steps_per_second": 0.174,
+      "eval_accuracy": 0.8296296296296296,
+      "eval_loss": 0.6540727019309998,
+      "eval_runtime": 50.7592,
+      "eval_samples_per_second": 5.319,
+      "eval_steps_per_second": 0.099,
       "step": 1184
     },
     {
       "epoch": 74.003125,
-      "grad_norm": 7.789737701416016,
+      "grad_norm": 0.055838149040937424,
       "learning_rate": 2.1122685185185186e-05,
-      "loss": 0.9284,
+      "loss": 0.0128,
       "step": 1190
     },
     {
       "epoch": 74.00833333333334,
-      "grad_norm": 15.859330177307129,
+      "grad_norm": 0.033164795488119125,
       "learning_rate": 2.0833333333333336e-05,
-      "loss": 1.0445,
+      "loss": 0.0027,
       "step": 1200
     },
     {
       "epoch": 74.00833333333334,
-      "eval_accuracy": 0.48204022988505746,
-      "eval_loss": 2.055063486099243,
-      "eval_runtime": 143.3562,
-      "eval_samples_per_second": 9.71,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.8074074074074075,
+      "eval_loss": 0.7257384061813354,
+      "eval_runtime": 50.8461,
+      "eval_samples_per_second": 5.31,
+      "eval_steps_per_second": 0.098,
       "step": 1200
     },
     {
       "epoch": 75.00520833333333,
-      "grad_norm": 6.045502662658691,
+      "grad_norm": 0.0417243167757988,
       "learning_rate": 2.0543981481481483e-05,
-      "loss": 0.9844,
+      "loss": 0.0028,
       "step": 1210
     },
     {
       "epoch": 75.00833333333334,
-      "eval_accuracy": 0.5043103448275862,
-      "eval_loss": 2.051361083984375,
-      "eval_runtime": 128.397,
-      "eval_samples_per_second": 10.841,
-      "eval_steps_per_second": 0.171,
+      "eval_accuracy": 0.8185185185185185,
+      "eval_loss": 0.6686450242996216,
+      "eval_runtime": 57.1325,
+      "eval_samples_per_second": 4.726,
+      "eval_steps_per_second": 0.088,
       "step": 1216
     },
     {
       "epoch": 76.00208333333333,
-      "grad_norm": 4.970855712890625,
+      "grad_norm": 0.04405437409877777,
       "learning_rate": 2.0254629629629632e-05,
-      "loss": 1.1088,
+      "loss": 0.0035,
       "step": 1220
     },
     {
       "epoch": 76.00729166666666,
-      "grad_norm": 7.561370849609375,
+      "grad_norm": 0.9214801788330078,
       "learning_rate": 1.996527777777778e-05,
-      "loss": 1.0809,
+      "loss": 0.0034,
       "step": 1230
     },
     {
       "epoch": 76.00833333333334,
-      "eval_accuracy": 0.5021551724137931,
-      "eval_loss": 2.05523419380188,
-      "eval_runtime": 129.1447,
-      "eval_samples_per_second": 10.779,
-      "eval_steps_per_second": 0.17,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6239407658576965,
+      "eval_runtime": 57.7463,
+      "eval_samples_per_second": 4.676,
+      "eval_steps_per_second": 0.087,
       "step": 1232
     },
     {
       "epoch": 77.00416666666666,
-      "grad_norm": 6.377460479736328,
+      "grad_norm": 0.12569206953048706,
       "learning_rate": 1.967592592592593e-05,
-      "loss": 1.1158,
+      "loss": 0.0111,
       "step": 1240
     },
     {
       "epoch": 77.00833333333334,
-      "eval_accuracy": 0.48778735632183906,
-      "eval_loss": 2.04689621925354,
-      "eval_runtime": 145.6931,
-      "eval_samples_per_second": 9.554,
-      "eval_steps_per_second": 0.151,
+      "eval_accuracy": 0.7925925925925926,
+      "eval_loss": 0.771929919719696,
+      "eval_runtime": 57.2385,
+      "eval_samples_per_second": 4.717,
+      "eval_steps_per_second": 0.087,
       "step": 1248
     },
     {
       "epoch": 78.00104166666667,
-      "grad_norm": 7.5645904541015625,
+      "grad_norm": 0.08424292504787445,
       "learning_rate": 1.9386574074074075e-05,
-      "loss": 0.9078,
+      "loss": 0.0059,
       "step": 1250
     },
     {
       "epoch": 78.00625,
-      "grad_norm": 6.4087419509887695,
+      "grad_norm": 0.084795281291008,
       "learning_rate": 1.9097222222222222e-05,
-      "loss": 0.9017,
+      "loss": 0.009,
       "step": 1260
     },
     {
       "epoch": 78.00833333333334,
-      "eval_accuracy": 0.4906609195402299,
-      "eval_loss": 2.0516154766082764,
-      "eval_runtime": 121.723,
-      "eval_samples_per_second": 11.436,
-      "eval_steps_per_second": 0.181,
+      "eval_accuracy": 0.8185185185185185,
+      "eval_loss": 0.6882457733154297,
+      "eval_runtime": 58.6356,
+      "eval_samples_per_second": 4.605,
+      "eval_steps_per_second": 0.085,
       "step": 1264
     },
     {
       "epoch": 79.003125,
-      "grad_norm": 5.661813735961914,
+      "grad_norm": 1.1351326704025269,
       "learning_rate": 1.8807870370370372e-05,
-      "loss": 0.8761,
+      "loss": 0.0044,
       "step": 1270
     },
     {
       "epoch": 79.00833333333334,
-      "grad_norm": 33.69865798950195,
+      "grad_norm": 0.1627674102783203,
       "learning_rate": 1.8518518518518518e-05,
-      "loss": 1.0449,
+      "loss": 0.0038,
       "step": 1280
     },
     {
       "epoch": 79.00833333333334,
-      "eval_accuracy": 0.4863505747126437,
-      "eval_loss": 2.0770106315612793,
-      "eval_runtime": 129.9823,
-      "eval_samples_per_second": 10.709,
-      "eval_steps_per_second": 0.169,
+      "eval_accuracy": 0.8222222222222222,
+      "eval_loss": 0.7039524912834167,
+      "eval_runtime": 54.0543,
+      "eval_samples_per_second": 4.995,
+      "eval_steps_per_second": 0.092,
       "step": 1280
     },
     {
       "epoch": 80.00520833333333,
-      "grad_norm": 5.260113716125488,
+      "grad_norm": 0.3010275363922119,
       "learning_rate": 1.8229166666666668e-05,
-      "loss": 1.0167,
+      "loss": 0.005,
       "step": 1290
     },
     {
       "epoch": 80.00833333333334,
-      "eval_accuracy": 0.4942528735632184,
-      "eval_loss": 2.0373690128326416,
-      "eval_runtime": 120.0265,
-      "eval_samples_per_second": 11.597,
-      "eval_steps_per_second": 0.183,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6955244541168213,
+      "eval_runtime": 50.75,
+      "eval_samples_per_second": 5.32,
+      "eval_steps_per_second": 0.099,
       "step": 1296
     },
     {
       "epoch": 81.00208333333333,
-      "grad_norm": 6.28135347366333,
+      "grad_norm": 0.06365453451871872,
       "learning_rate": 1.7939814814814815e-05,
-      "loss": 1.0278,
+      "loss": 0.0024,
       "step": 1300
     },
     {
       "epoch": 81.00729166666666,
-      "grad_norm": 7.153203010559082,
+      "grad_norm": 0.12171023339033127,
       "learning_rate": 1.7650462962962965e-05,
-      "loss": 0.975,
+      "loss": 0.003,
       "step": 1310
     },
     {
       "epoch": 81.00833333333334,
-      "eval_accuracy": 0.4942528735632184,
-      "eval_loss": 2.0630548000335693,
-      "eval_runtime": 127.397,
-      "eval_samples_per_second": 10.926,
-      "eval_steps_per_second": 0.173,
+      "eval_accuracy": 0.8481481481481481,
+      "eval_loss": 0.679672122001648,
+      "eval_runtime": 55.0864,
+      "eval_samples_per_second": 4.901,
+      "eval_steps_per_second": 0.091,
       "step": 1312
     },
     {
       "epoch": 82.00416666666666,
-      "grad_norm": 6.697967052459717,
+      "grad_norm": 1.2039811611175537,
       "learning_rate": 1.736111111111111e-05,
-      "loss": 0.9285,
+      "loss": 0.0035,
       "step": 1320
     },
     {
       "epoch": 82.00833333333334,
-      "eval_accuracy": 0.4870689655172414,
-      "eval_loss": 2.049909830093384,
-      "eval_runtime": 143.6575,
-      "eval_samples_per_second": 9.69,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.65481036901474,
+      "eval_runtime": 51.9003,
+      "eval_samples_per_second": 5.202,
+      "eval_steps_per_second": 0.096,
       "step": 1328
     },
     {
       "epoch": 83.00104166666667,
-      "grad_norm": 7.098538875579834,
+      "grad_norm": 0.020734239369630814,
       "learning_rate": 1.707175925925926e-05,
-      "loss": 0.8193,
+      "loss": 0.004,
       "step": 1330
     },
     {
       "epoch": 83.00625,
-      "grad_norm": 5.9787468910217285,
+      "grad_norm": 0.0892987847328186,
       "learning_rate": 1.6782407407407408e-05,
-      "loss": 0.9762,
+      "loss": 0.0029,
       "step": 1340
     },
     {
       "epoch": 83.00833333333334,
-      "eval_accuracy": 0.4964080459770115,
-      "eval_loss": 2.061843156814575,
-      "eval_runtime": 107.3354,
-      "eval_samples_per_second": 12.969,
-      "eval_steps_per_second": 0.205,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6407437920570374,
+      "eval_runtime": 52.8462,
+      "eval_samples_per_second": 5.109,
+      "eval_steps_per_second": 0.095,
       "step": 1344
     },
     {
       "epoch": 84.003125,
-      "grad_norm": 6.269992351531982,
+      "grad_norm": 0.3291865885257721,
       "learning_rate": 1.6493055555555557e-05,
-      "loss": 1.0743,
+      "loss": 0.0037,
       "step": 1350
     },
     {
       "epoch": 84.00833333333334,
-      "grad_norm": 30.140321731567383,
+      "grad_norm": 0.07655324041843414,
       "learning_rate": 1.6203703703703704e-05,
-      "loss": 0.9454,
+      "loss": 0.0131,
       "step": 1360
     },
     {
       "epoch": 84.00833333333334,
-      "eval_accuracy": 0.4992816091954023,
-      "eval_loss": 2.046207904815674,
-      "eval_runtime": 143.1149,
-      "eval_samples_per_second": 9.726,
-      "eval_steps_per_second": 0.154,
+      "eval_accuracy": 0.8407407407407408,
+      "eval_loss": 0.6151522994041443,
+      "eval_runtime": 54.4088,
+      "eval_samples_per_second": 4.962,
+      "eval_steps_per_second": 0.092,
       "step": 1360
     },
     {
       "epoch": 85.00520833333333,
-      "grad_norm": 6.801133632659912,
+      "grad_norm": 0.04456999897956848,
       "learning_rate": 1.5914351851851854e-05,
-      "loss": 0.8665,
+      "loss": 0.0026,
       "step": 1370
     },
     {
       "epoch": 85.00833333333334,
-      "eval_accuracy": 0.4892241379310345,
-      "eval_loss": 2.076526165008545,
-      "eval_runtime": 132.1822,
-      "eval_samples_per_second": 10.531,
-      "eval_steps_per_second": 0.166,
+      "eval_accuracy": 0.8444444444444444,
+      "eval_loss": 0.5862600803375244,
+      "eval_runtime": 63.1334,
+      "eval_samples_per_second": 4.277,
+      "eval_steps_per_second": 0.079,
       "step": 1376
     },
     {
       "epoch": 86.00208333333333,
-      "grad_norm": 6.922851085662842,
+      "grad_norm": 0.07492371648550034,
       "learning_rate": 1.5625e-05,
-      "loss": 0.8844,
+      "loss": 0.0103,
       "step": 1380
     },
     {
       "epoch": 86.00729166666666,
-      "grad_norm": 6.79546594619751,
+      "grad_norm": 0.19561482965946198,
       "learning_rate": 1.533564814814815e-05,
-      "loss": 0.9202,
+      "loss": 0.0048,
       "step": 1390
     },
     {
       "epoch": 86.00833333333334,
-      "eval_accuracy": 0.4949712643678161,
-      "eval_loss": 2.0512619018554688,
-      "eval_runtime": 117.1467,
-      "eval_samples_per_second": 11.883,
-      "eval_steps_per_second": 0.188,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.6047504544258118,
+      "eval_runtime": 60.9766,
+      "eval_samples_per_second": 4.428,
+      "eval_steps_per_second": 0.082,
       "step": 1392
     },
     {
       "epoch": 87.00416666666666,
-      "grad_norm": 5.933279514312744,
+      "grad_norm": 0.02398071438074112,
       "learning_rate": 1.5046296296296297e-05,
-      "loss": 0.8186,
+      "loss": 0.0032,
       "step": 1400
     },
     {
       "epoch": 87.00833333333334,
-      "eval_accuracy": 0.5093390804597702,
-      "eval_loss": 2.0254082679748535,
-      "eval_runtime": 139.5548,
-      "eval_samples_per_second": 9.975,
-      "eval_steps_per_second": 0.158,
+      "eval_accuracy": 0.8481481481481481,
+      "eval_loss": 0.606376588344574,
+      "eval_runtime": 61.0061,
+      "eval_samples_per_second": 4.426,
+      "eval_steps_per_second": 0.082,
       "step": 1408
     },
     {
       "epoch": 88.00104166666667,
-      "grad_norm": 7.079472541809082,
+      "grad_norm": 0.024840401485562325,
       "learning_rate": 1.4756944444444445e-05,
-      "loss": 0.8471,
+      "loss": 0.0021,
       "step": 1410
     },
     {
       "epoch": 88.00625,
-      "grad_norm": 5.367373466491699,
+      "grad_norm": 0.32311511039733887,
       "learning_rate": 1.4467592592592593e-05,
-      "loss": 0.8659,
+      "loss": 0.0067,
       "step": 1420
     },
     {
       "epoch": 88.00833333333334,
-      "eval_accuracy": 0.4791666666666667,
-      "eval_loss": 2.1059958934783936,
-      "eval_runtime": 109.122,
-      "eval_samples_per_second": 12.756,
-      "eval_steps_per_second": 0.202,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6492321491241455,
+      "eval_runtime": 59.5983,
+      "eval_samples_per_second": 4.53,
+      "eval_steps_per_second": 0.084,
       "step": 1424
     },
     {
       "epoch": 89.003125,
-      "grad_norm": 4.5717926025390625,
+      "grad_norm": 0.032252054661512375,
       "learning_rate": 1.4178240740740741e-05,
-      "loss": 0.8132,
+      "loss": 0.0062,
       "step": 1430
     },
     {
       "epoch": 89.00833333333334,
-      "grad_norm": 11.70888900756836,
+      "grad_norm": 0.06694841384887695,
       "learning_rate": 1.388888888888889e-05,
-      "loss": 0.8789,
+      "loss": 0.0077,
       "step": 1440
     },
     {
       "epoch": 89.00833333333334,
-      "eval_accuracy": 0.4964080459770115,
-      "eval_loss": 2.029611110687256,
-      "eval_runtime": 117.991,
-      "eval_samples_per_second": 11.798,
-      "eval_steps_per_second": 0.186,
+      "eval_accuracy": 0.7851851851851852,
+      "eval_loss": 0.751960813999176,
+      "eval_runtime": 58.3708,
+      "eval_samples_per_second": 4.626,
+      "eval_steps_per_second": 0.086,
       "step": 1440
     },
     {
       "epoch": 90.00520833333333,
-      "grad_norm": 6.696922779083252,
+      "grad_norm": 0.08605653792619705,
       "learning_rate": 1.3599537037037038e-05,
-      "loss": 0.8592,
+      "loss": 0.012,
       "step": 1450
     },
     {
       "epoch": 90.00833333333334,
-      "eval_accuracy": 0.4849137931034483,
-      "eval_loss": 2.0757148265838623,
-      "eval_runtime": 125.7116,
-      "eval_samples_per_second": 11.073,
-      "eval_steps_per_second": 0.175,
+      "eval_accuracy": 0.8037037037037037,
+      "eval_loss": 0.7661921977996826,
+      "eval_runtime": 58.3671,
+      "eval_samples_per_second": 4.626,
+      "eval_steps_per_second": 0.086,
       "step": 1456
     },
     {
       "epoch": 91.00208333333333,
-      "grad_norm": 6.2329535484313965,
+      "grad_norm": 0.025169596076011658,
       "learning_rate": 1.3310185185185186e-05,
-      "loss": 0.8472,
+      "loss": 0.0053,
       "step": 1460
     },
     {
       "epoch": 91.00729166666666,
-      "grad_norm": 7.030002117156982,
+      "grad_norm": 0.278448224067688,
       "learning_rate": 1.3020833333333334e-05,
-      "loss": 0.8093,
+      "loss": 0.0092,
       "step": 1470
     },
     {
       "epoch": 91.00833333333334,
-      "eval_accuracy": 0.4985632183908046,
-      "eval_loss": 2.0289297103881836,
-      "eval_runtime": 143.6382,
-      "eval_samples_per_second": 9.691,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.8074074074074075,
+      "eval_loss": 0.7106200456619263,
+      "eval_runtime": 62.5207,
+      "eval_samples_per_second": 4.319,
+      "eval_steps_per_second": 0.08,
       "step": 1472
     },
     {
       "epoch": 92.00416666666666,
-      "grad_norm": 5.606525897979736,
+      "grad_norm": 0.038209304213523865,
       "learning_rate": 1.2731481481481482e-05,
-      "loss": 0.9074,
+      "loss": 0.0034,
       "step": 1480
     },
     {
       "epoch": 92.00833333333334,
-      "eval_accuracy": 0.4920977011494253,
-      "eval_loss": 2.0539023876190186,
-      "eval_runtime": 145.5404,
-      "eval_samples_per_second": 9.564,
-      "eval_steps_per_second": 0.151,
+      "eval_accuracy": 0.8111111111111111,
+      "eval_loss": 0.7589225769042969,
+      "eval_runtime": 64.0017,
+      "eval_samples_per_second": 4.219,
+      "eval_steps_per_second": 0.078,
       "step": 1488
     },
     {
       "epoch": 93.00104166666667,
-      "grad_norm": 7.5830793380737305,
+      "grad_norm": 0.06385576725006104,
       "learning_rate": 1.244212962962963e-05,
-      "loss": 0.82,
+      "loss": 0.0167,
       "step": 1490
     },
     {
       "epoch": 93.00625,
-      "grad_norm": 7.022946357727051,
+      "grad_norm": 0.3713780641555786,
       "learning_rate": 1.2152777777777779e-05,
-      "loss": 0.82,
+      "loss": 0.0042,
       "step": 1500
     },
     {
       "epoch": 93.00833333333334,
-      "eval_accuracy": 0.5014367816091954,
-      "eval_loss": 2.0480799674987793,
-      "eval_runtime": 126.661,
-      "eval_samples_per_second": 10.99,
-      "eval_steps_per_second": 0.174,
+      "eval_accuracy": 0.8296296296296296,
+      "eval_loss": 0.6381981372833252,
+      "eval_runtime": 66.7638,
+      "eval_samples_per_second": 4.044,
+      "eval_steps_per_second": 0.075,
       "step": 1504
     },
     {
       "epoch": 94.003125,
-      "grad_norm": 8.179780960083008,
+      "grad_norm": 0.015576478093862534,
       "learning_rate": 1.1863425925925927e-05,
-      "loss": 1.0252,
+      "loss": 0.0444,
       "step": 1510
     },
     {
       "epoch": 94.00833333333334,
-      "grad_norm": 18.23526954650879,
+      "grad_norm": 0.049867402762174606,
       "learning_rate": 1.1574074074074075e-05,
-      "loss": 0.8318,
+      "loss": 0.0053,
       "step": 1520
     },
     {
       "epoch": 94.00833333333334,
-      "eval_accuracy": 0.5021551724137931,
-      "eval_loss": 2.030928134918213,
-      "eval_runtime": 144.5617,
-      "eval_samples_per_second": 9.629,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.615300714969635,
+      "eval_runtime": 58.6328,
+      "eval_samples_per_second": 4.605,
+      "eval_steps_per_second": 0.085,
       "step": 1520
     },
     {
       "epoch": 95.00520833333333,
-      "grad_norm": 5.861065864562988,
+      "grad_norm": 1.0250205993652344,
       "learning_rate": 1.1284722222222223e-05,
-      "loss": 0.8337,
+      "loss": 0.0038,
       "step": 1530
     },
     {
       "epoch": 95.00833333333334,
-      "eval_accuracy": 0.5050287356321839,
-      "eval_loss": 2.0334632396698,
-      "eval_runtime": 153.2179,
-      "eval_samples_per_second": 9.085,
-      "eval_steps_per_second": 0.144,
+      "eval_accuracy": 0.837037037037037,
+      "eval_loss": 0.6227499842643738,
+      "eval_runtime": 58.3792,
+      "eval_samples_per_second": 4.625,
+      "eval_steps_per_second": 0.086,
       "step": 1536
     },
     {
       "epoch": 96.00208333333333,
-      "grad_norm": 6.104484558105469,
+      "grad_norm": 0.10840343683958054,
       "learning_rate": 1.0995370370370372e-05,
-      "loss": 0.8918,
+      "loss": 0.0024,
       "step": 1540
     },
     {
       "epoch": 96.00729166666666,
-      "grad_norm": 6.614218711853027,
+      "grad_norm": 0.014415646903216839,
       "learning_rate": 1.070601851851852e-05,
-      "loss": 0.9089,
+      "loss": 0.002,
       "step": 1550
     },
     {
       "epoch": 96.00833333333334,
-      "eval_accuracy": 0.5021551724137931,
-      "eval_loss": 2.045605421066284,
-      "eval_runtime": 123.198,
-      "eval_samples_per_second": 11.299,
-      "eval_steps_per_second": 0.179,
+      "eval_accuracy": 0.8407407407407408,
+      "eval_loss": 0.6424070000648499,
+      "eval_runtime": 60.0655,
+      "eval_samples_per_second": 4.495,
+      "eval_steps_per_second": 0.083,
       "step": 1552
     },
     {
       "epoch": 97.00416666666666,
-      "grad_norm": 6.799188137054443,
+      "grad_norm": 0.03282202407717705,
       "learning_rate": 1.0416666666666668e-05,
-      "loss": 0.8189,
+      "loss": 0.0063,
       "step": 1560
     },
     {
       "epoch": 97.00833333333334,
-      "eval_accuracy": 0.4985632183908046,
-      "eval_loss": 2.0107243061065674,
-      "eval_runtime": 132.087,
-      "eval_samples_per_second": 10.539,
-      "eval_steps_per_second": 0.167,
+      "eval_accuracy": 0.8481481481481481,
+      "eval_loss": 0.62154620885849,
+      "eval_runtime": 63.3885,
+      "eval_samples_per_second": 4.259,
+      "eval_steps_per_second": 0.079,
       "step": 1568
     },
     {
       "epoch": 98.00104166666667,
-      "grad_norm": 6.34768533706665,
+      "grad_norm": 0.07932916283607483,
       "learning_rate": 1.0127314814814816e-05,
-      "loss": 0.8,
+      "loss": 0.0118,
       "step": 1570
     },
     {
       "epoch": 98.00625,
-      "grad_norm": 6.501392364501953,
+      "grad_norm": 0.07543420791625977,
       "learning_rate": 9.837962962962964e-06,
-      "loss": 0.7603,
+      "loss": 0.0021,
       "step": 1580
     },
     {
       "epoch": 98.00833333333334,
-      "eval_accuracy": 0.507183908045977,
-      "eval_loss": 2.014706611633301,
-      "eval_runtime": 107.2213,
-      "eval_samples_per_second": 12.982,
-      "eval_steps_per_second": 0.205,
+      "eval_accuracy": 0.8333333333333334,
+      "eval_loss": 0.635489821434021,
+      "eval_runtime": 61.475,
+      "eval_samples_per_second": 4.392,
+      "eval_steps_per_second": 0.081,
       "step": 1584
     },
     {
       "epoch": 99.003125,
-      "grad_norm": 4.626859664916992,
+      "grad_norm": 0.028682081028819084,
       "learning_rate": 9.548611111111111e-06,
-      "loss": 0.7984,
+      "loss": 0.0145,
       "step": 1590
     },
     {
       "epoch": 99.00833333333334,
-      "grad_norm": 19.75746726989746,
+      "grad_norm": 0.02590627782046795,
       "learning_rate": 9.259259259259259e-06,
-      "loss": 0.9197,
+      "loss": 0.0022,
       "step": 1600
     },
     {
       "epoch": 99.00833333333334,
-      "eval_accuracy": 0.5014367816091954,
-      "eval_loss": 2.043818712234497,
-      "eval_runtime": 144.3126,
-      "eval_samples_per_second": 9.646,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.8407407407407408,
+      "eval_loss": 0.6141006946563721,
+      "eval_runtime": 61.305,
+      "eval_samples_per_second": 4.404,
+      "eval_steps_per_second": 0.082,
       "step": 1600
     },
     {
       "epoch": 100.00520833333333,
-      "grad_norm": 5.76575231552124,
+      "grad_norm": 0.039108309894800186,
       "learning_rate": 8.969907407407407e-06,
-      "loss": 0.8021,
+      "loss": 0.002,
       "step": 1610
     },
     {
       "epoch": 100.00833333333334,
-      "eval_accuracy": 0.5,
-      "eval_loss": 2.0310871601104736,
-      "eval_runtime": 124.2032,
-      "eval_samples_per_second": 11.207,
-      "eval_steps_per_second": 0.177,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.5681954622268677,
+      "eval_runtime": 59.6149,
+      "eval_samples_per_second": 4.529,
+      "eval_steps_per_second": 0.084,
       "step": 1616
     },
     {
       "epoch": 101.00208333333333,
-      "grad_norm": 7.446949005126953,
+      "grad_norm": 0.013332217000424862,
       "learning_rate": 8.680555555555556e-06,
-      "loss": 0.8517,
+      "loss": 0.0061,
       "step": 1620
     },
     {
       "epoch": 101.00729166666666,
-      "grad_norm": 7.860858917236328,
+      "grad_norm": 0.16490715742111206,
       "learning_rate": 8.391203703703704e-06,
-      "loss": 0.7474,
+      "loss": 0.0058,
       "step": 1630
     },
     {
       "epoch": 101.00833333333334,
-      "eval_accuracy": 0.5007183908045977,
-      "eval_loss": 2.0348401069641113,
-      "eval_runtime": 122.1603,
-      "eval_samples_per_second": 11.395,
-      "eval_steps_per_second": 0.18,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.5804283618927002,
+      "eval_runtime": 60.5778,
+      "eval_samples_per_second": 4.457,
+      "eval_steps_per_second": 0.083,
       "step": 1632
     },
     {
       "epoch": 102.00416666666666,
-      "grad_norm": 6.173117160797119,
+      "grad_norm": 0.020627934485673904,
       "learning_rate": 8.101851851851852e-06,
-      "loss": 0.9423,
+      "loss": 0.0027,
       "step": 1640
     },
     {
       "epoch": 102.00833333333334,
-      "eval_accuracy": 0.5007183908045977,
-      "eval_loss": 2.0177173614501953,
-      "eval_runtime": 127.9201,
-      "eval_samples_per_second": 10.882,
-      "eval_steps_per_second": 0.172,
+      "eval_accuracy": 0.8555555555555555,
+      "eval_loss": 0.5723864436149597,
+      "eval_runtime": 58.8413,
+      "eval_samples_per_second": 4.589,
+      "eval_steps_per_second": 0.085,
       "step": 1648
     },
     {
       "epoch": 103.00104166666667,
-      "grad_norm": 7.340484142303467,
+      "grad_norm": 0.0398273728787899,
       "learning_rate": 7.8125e-06,
-      "loss": 0.8699,
+      "loss": 0.0018,
       "step": 1650
     },
     {
       "epoch": 103.00625,
-      "grad_norm": 7.385336875915527,
+      "grad_norm": 0.7983164191246033,
       "learning_rate": 7.523148148148148e-06,
-      "loss": 0.8135,
+      "loss": 0.0026,
       "step": 1660
     },
     {
       "epoch": 103.00833333333334,
-      "eval_accuracy": 0.5028735632183908,
-      "eval_loss": 2.0140013694763184,
-      "eval_runtime": 147.4404,
-      "eval_samples_per_second": 9.441,
-      "eval_steps_per_second": 0.149,
+      "eval_accuracy": 0.8629629629629629,
+      "eval_loss": 0.5556555986404419,
+      "eval_runtime": 60.0508,
+      "eval_samples_per_second": 4.496,
+      "eval_steps_per_second": 0.083,
       "step": 1664
     },
     {
       "epoch": 104.003125,
-      "grad_norm": 5.808743000030518,
+      "grad_norm": 0.02057529427111149,
       "learning_rate": 7.2337962962962966e-06,
-      "loss": 0.8795,
+      "loss": 0.013,
       "step": 1670
     },
     {
       "epoch": 104.00833333333334,
-      "grad_norm": 20.817060470581055,
+      "grad_norm": 0.023396655917167664,
       "learning_rate": 6.944444444444445e-06,
-      "loss": 0.8244,
+      "loss": 0.0016,
       "step": 1680
     },
     {
       "epoch": 104.00833333333334,
-      "eval_accuracy": 0.4985632183908046,
-      "eval_loss": 2.012415885925293,
-      "eval_runtime": 137.531,
-      "eval_samples_per_second": 10.121,
-      "eval_steps_per_second": 0.16,
+      "eval_accuracy": 0.8592592592592593,
+      "eval_loss": 0.5465064644813538,
+      "eval_runtime": 57.6357,
+      "eval_samples_per_second": 4.685,
+      "eval_steps_per_second": 0.087,
       "step": 1680
     },
     {
       "epoch": 105.00520833333333,
-      "grad_norm": 7.121223449707031,
+      "grad_norm": 0.014929450117051601,
       "learning_rate": 6.655092592592593e-06,
-      "loss": 0.8446,
+      "loss": 0.0018,
       "step": 1690
     },
     {
       "epoch": 105.00833333333334,
-      "eval_accuracy": 0.5064655172413793,
-      "eval_loss": 2.0021567344665527,
-      "eval_runtime": 128.7884,
-      "eval_samples_per_second": 10.808,
-      "eval_steps_per_second": 0.171,
+      "eval_accuracy": 0.8629629629629629,
+      "eval_loss": 0.5636276602745056,
+      "eval_runtime": 58.0116,
+      "eval_samples_per_second": 4.654,
+      "eval_steps_per_second": 0.086,
       "step": 1696
     },
     {
       "epoch": 106.00208333333333,
-      "grad_norm": 6.345254898071289,
+      "grad_norm": 0.01618959940969944,
       "learning_rate": 6.365740740740741e-06,
-      "loss": 0.7981,
+      "loss": 0.0017,
       "step": 1700
     },
     {
       "epoch": 106.00729166666666,
-      "grad_norm": 8.024818420410156,
+      "grad_norm": 0.024775177240371704,
       "learning_rate": 6.076388888888889e-06,
-      "loss": 0.7965,
+      "loss": 0.0022,
       "step": 1710
     },
     {
       "epoch": 106.00833333333334,
-      "eval_accuracy": 0.5107758620689655,
-      "eval_loss": 1.9956977367401123,
-      "eval_runtime": 119.9052,
-      "eval_samples_per_second": 11.609,
-      "eval_steps_per_second": 0.183,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.5932139158248901,
+      "eval_runtime": 58.104,
+      "eval_samples_per_second": 4.647,
+      "eval_steps_per_second": 0.086,
       "step": 1712
     },
     {
       "epoch": 107.00416666666666,
-      "grad_norm": 5.641129016876221,
+      "grad_norm": 0.010166914202272892,
       "learning_rate": 5.787037037037038e-06,
-      "loss": 0.8256,
+      "loss": 0.0018,
       "step": 1720
     },
     {
       "epoch": 107.00833333333334,
-      "eval_accuracy": 0.5107758620689655,
-      "eval_loss": 1.999496579170227,
-      "eval_runtime": 134.9022,
-      "eval_samples_per_second": 10.319,
-      "eval_steps_per_second": 0.163,
+      "eval_accuracy": 0.8592592592592593,
+      "eval_loss": 0.588383138179779,
+      "eval_runtime": 58.8949,
+      "eval_samples_per_second": 4.584,
+      "eval_steps_per_second": 0.085,
       "step": 1728
     },
     {
       "epoch": 108.00104166666667,
-      "grad_norm": 6.308507919311523,
+      "grad_norm": 0.2191196233034134,
       "learning_rate": 5.497685185185186e-06,
-      "loss": 0.7253,
+      "loss": 0.0018,
       "step": 1730
     },
     {
       "epoch": 108.00625,
-      "grad_norm": 6.363261699676514,
+      "grad_norm": 0.272781103849411,
       "learning_rate": 5.208333333333334e-06,
-      "loss": 0.8448,
+      "loss": 0.0018,
       "step": 1740
     },
     {
       "epoch": 108.00833333333334,
-      "eval_accuracy": 0.5093390804597702,
-      "eval_loss": 2.005643844604492,
-      "eval_runtime": 143.6016,
-      "eval_samples_per_second": 9.693,
-      "eval_steps_per_second": 0.153,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.5960416197776794,
+      "eval_runtime": 60.5719,
+      "eval_samples_per_second": 4.458,
+      "eval_steps_per_second": 0.083,
       "step": 1744
     },
     {
       "epoch": 109.003125,
-      "grad_norm": 4.125205993652344,
+      "grad_norm": 0.026597756892442703,
       "learning_rate": 4.918981481481482e-06,
-      "loss": 0.8608,
+      "loss": 0.0017,
       "step": 1750
     },
     {
       "epoch": 109.00833333333334,
-      "grad_norm": 1.9473612308502197,
+      "grad_norm": 0.01299307867884636,
       "learning_rate": 4.6296296296296296e-06,
-      "loss": 0.7144,
+      "loss": 0.0041,
       "step": 1760
     },
     {
       "epoch": 109.00833333333334,
-      "eval_accuracy": 0.507183908045977,
-      "eval_loss": 2.0083651542663574,
-      "eval_runtime": 169.1186,
-      "eval_samples_per_second": 8.231,
-      "eval_steps_per_second": 0.13,
+      "eval_accuracy": 0.8555555555555555,
+      "eval_loss": 0.5983728170394897,
+      "eval_runtime": 61.079,
+      "eval_samples_per_second": 4.421,
+      "eval_steps_per_second": 0.082,
       "step": 1760
     },
     {
       "epoch": 110.00520833333333,
-      "grad_norm": 7.592528343200684,
+      "grad_norm": 0.04651355370879173,
       "learning_rate": 4.340277777777778e-06,
-      "loss": 0.7869,
+      "loss": 0.0019,
       "step": 1770
     },
     {
       "epoch": 110.00833333333334,
-      "eval_accuracy": 0.5114942528735632,
-      "eval_loss": 1.9966795444488525,
-      "eval_runtime": 122.5349,
-      "eval_samples_per_second": 11.36,
-      "eval_steps_per_second": 0.18,
+      "eval_accuracy": 0.8518518518518519,
+      "eval_loss": 0.6015256643295288,
+      "eval_runtime": 61.0967,
+      "eval_samples_per_second": 4.419,
+      "eval_steps_per_second": 0.082,
       "step": 1776
     },
     {
       "epoch": 111.00208333333333,
-      "grad_norm": 6.077642440795898,
+      "grad_norm": 1.950492262840271,
       "learning_rate": 4.050925925925926e-06,
-      "loss": 0.924,
+      "loss": 0.0026,
       "step": 1780
     },
     {
       "epoch": 111.00729166666666,
-      "grad_norm": 7.033000946044922,
+      "grad_norm": 0.014990141615271568,
       "learning_rate": 3.761574074074074e-06,
-      "loss": 0.8149,
+      "loss": 0.0031,
       "step": 1790
     },
     {
       "epoch": 111.00833333333334,
-      "eval_accuracy": 0.5114942528735632,
-      "eval_loss": 1.997259259223938,
-      "eval_runtime": 131.0162,
-      "eval_samples_per_second": 10.625,
-      "eval_steps_per_second": 0.168,
+      "eval_accuracy": 0.8592592592592593,
+      "eval_loss": 0.5940619111061096,
+      "eval_runtime": 60.2381,
+      "eval_samples_per_second": 4.482,
+      "eval_steps_per_second": 0.083,
       "step": 1792
     },
     {
       "epoch": 112.00416666666666,
-      "grad_norm": 6.99604606628418,
+      "grad_norm": 0.0172797292470932,
       "learning_rate": 3.4722222222222224e-06,
-      "loss": 0.7896,
+      "loss": 0.0056,
       "step": 1800
     },
     {
       "epoch": 112.00833333333334,
-      "eval_accuracy": 0.5122126436781609,
-      "eval_loss": 2.0014069080352783,
-      "eval_runtime": 135.8894,
-      "eval_samples_per_second": 10.244,
-      "eval_steps_per_second": 0.162,
+      "eval_accuracy": 0.8592592592592593,
+      "eval_loss": 0.5957003235816956,
+      "eval_runtime": 57.2533,
+      "eval_samples_per_second": 4.716,
+      "eval_steps_per_second": 0.087,
       "step": 1808
     },
     {
       "epoch": 113.00104166666667,
-      "grad_norm": 5.9290690422058105,
+      "grad_norm": 0.4851263463497162,
       "learning_rate": 3.1828703703703706e-06,
-      "loss": 0.6788,
+      "loss": 0.0019,
       "step": 1810
     },
     {
       "epoch": 113.00625,
-      "grad_norm": 5.406201362609863,
+      "grad_norm": 0.010226945392787457,
       "learning_rate": 2.893518518518519e-06,
-      "loss": 0.8189,
+      "loss": 0.0014,
       "step": 1820
     },
     {
       "epoch": 113.00833333333334,
-      "eval_accuracy": 0.514367816091954,
-      "eval_loss": 1.998863697052002,
-      "eval_runtime": 120.7041,
-      "eval_samples_per_second": 11.532,
-      "eval_steps_per_second": 0.182,
+      "eval_accuracy": 0.8592592592592593,
+      "eval_loss": 0.6007441878318787,
+      "eval_runtime": 62.7358,
+      "eval_samples_per_second": 4.304,
+      "eval_steps_per_second": 0.08,
       "step": 1824
     },
     {
       "epoch": 114.003125,
-      "grad_norm": 7.0976881980896,
+      "grad_norm": 0.022903593257069588,
       "learning_rate": 2.604166666666667e-06,
-      "loss": 0.7602,
+      "loss": 0.0041,
       "step": 1830
     },
     {
       "epoch": 114.00833333333334,
-      "grad_norm": 16.11831283569336,
+      "grad_norm": 1.950323462486267,
       "learning_rate": 2.3148148148148148e-06,
-      "loss": 0.6775,
+      "loss": 0.0145,
       "step": 1840
     },
     {
       "epoch": 114.00833333333334,
-      "eval_accuracy": 0.5122126436781609,
-      "eval_loss": 1.9956809282302856,
-      "eval_runtime": 143.0208,
-      "eval_samples_per_second": 9.733,
-      "eval_steps_per_second": 0.154,
+      "eval_accuracy": 0.8444444444444444,
+      "eval_loss": 0.6138308048248291,
+      "eval_runtime": 61.2348,
+      "eval_samples_per_second": 4.409,
+      "eval_steps_per_second": 0.082,
       "step": 1840
     },
     {
       "epoch": 115.00520833333333,
-      "grad_norm": 6.661690711975098,
+      "grad_norm": 0.021297235041856766,
       "learning_rate": 2.025462962962963e-06,
-      "loss": 0.8642,
+      "loss": 0.002,
       "step": 1850
     },
     {
       "epoch": 115.00833333333334,
-      "eval_accuracy": 0.5100574712643678,
-      "eval_loss": 2.000101089477539,
-      "eval_runtime": 116.9876,
-      "eval_samples_per_second": 11.899,
-      "eval_steps_per_second": 0.188,
+      "eval_accuracy": 0.8407407407407408,
+      "eval_loss": 0.620459794998169,
+      "eval_runtime": 61.1627,
+      "eval_samples_per_second": 4.414,
+      "eval_steps_per_second": 0.082,
       "step": 1856
     },
     {
       "epoch": 116.00208333333333,
-      "grad_norm": 5.262021064758301,
+      "grad_norm": 0.015099190175533295,
       "learning_rate": 1.7361111111111112e-06,
-      "loss": 0.7406,
+      "loss": 0.0017,
       "step": 1860
     },
     {
       "epoch": 116.00729166666666,
-      "grad_norm": 4.934671401977539,
+      "grad_norm": 4.604526996612549,
       "learning_rate": 1.4467592592592594e-06,
-      "loss": 0.7308,
+      "loss": 0.0046,
       "step": 1870
     },
     {
       "epoch": 116.00833333333334,
-      "eval_accuracy": 0.5086206896551724,
-      "eval_loss": 1.9895482063293457,
-      "eval_runtime": 135.8417,
-      "eval_samples_per_second": 10.247,
-      "eval_steps_per_second": 0.162,
+      "eval_accuracy": 0.8444444444444444,
+      "eval_loss": 0.6194358468055725,
+      "eval_runtime": 63.3221,
+      "eval_samples_per_second": 4.264,
+      "eval_steps_per_second": 0.079,
       "step": 1872
     },
     {
       "epoch": 117.00416666666666,
-      "grad_norm": 5.465080261230469,
+      "grad_norm": 0.0836496651172638,
       "learning_rate": 1.1574074074074074e-06,
-      "loss": 0.8616,
+      "loss": 0.0018,
       "step": 1880
     },
     {
       "epoch": 117.00833333333334,
-      "eval_accuracy": 0.5079022988505747,
-      "eval_loss": 1.985381007194519,
-      "eval_runtime": 117.3279,
-      "eval_samples_per_second": 11.864,
-      "eval_steps_per_second": 0.188,
+      "eval_accuracy": 0.8444444444444444,
+      "eval_loss": 0.618888795375824,
+      "eval_runtime": 60.3515,
+      "eval_samples_per_second": 4.474,
+      "eval_steps_per_second": 0.083,
       "step": 1888
     },
     {
       "epoch": 118.00104166666667,
-      "grad_norm": 6.937384128570557,
+      "grad_norm": 0.01113935001194477,
       "learning_rate": 8.680555555555556e-07,
-      "loss": 0.7361,
+      "loss": 0.0069,
       "step": 1890
     },
     {
       "epoch": 118.00625,
-      "grad_norm": 5.723898887634277,
+      "grad_norm": 0.05385176092386246,
       "learning_rate": 5.787037037037037e-07,
-      "loss": 0.7763,
+      "loss": 0.0023,
       "step": 1900
     },
     {
       "epoch": 118.00833333333334,
-      "eval_accuracy": 0.507183908045977,
-      "eval_loss": 1.9895660877227783,
-      "eval_runtime": 147.9074,
-      "eval_samples_per_second": 9.411,
-      "eval_steps_per_second": 0.149,
+      "eval_accuracy": 0.8444444444444444,
+      "eval_loss": 0.6390697956085205,
+      "eval_runtime": 64.997,
+      "eval_samples_per_second": 4.154,
+      "eval_steps_per_second": 0.077,
       "step": 1904
     },
     {
       "epoch": 119.003125,
-      "grad_norm": 6.339710712432861,
+      "grad_norm": 0.012112889438867569,
       "learning_rate": 2.8935185185185185e-07,
-      "loss": 0.7345,
+      "loss": 0.0027,
       "step": 1910
     },
     {
       "epoch": 119.00833333333334,
-      "grad_norm": 20.4032039642334,
+      "grad_norm": 0.05198315158486366,
       "learning_rate": 0.0,
-      "loss": 0.8009,
+      "loss": 0.0021,
       "step": 1920
     },
     {
       "epoch": 119.00833333333334,
-      "eval_accuracy": 0.507183908045977,
-      "eval_loss": 1.9902666807174683,
-      "eval_runtime": 138.8377,
-      "eval_samples_per_second": 10.026,
-      "eval_steps_per_second": 0.158,
+      "eval_accuracy": 0.8481481481481481,
+      "eval_loss": 0.622653603553772,
+      "eval_runtime": 63.8793,
+      "eval_samples_per_second": 4.227,
+      "eval_steps_per_second": 0.078,
       "step": 1920
     },
     {
       "epoch": 119.00833333333334,
       "step": 1920,
       "total_flos": 1.4452753827235627e+20,
-      "train_loss": 1.618524906039238,
-      "train_runtime": 28327.5233,
-      "train_samples_per_second": 4.338,
-      "train_steps_per_second": 0.068
+      "train_loss": 0.034724208731010245,
+      "train_runtime": 33430.1897,
+      "train_samples_per_second": 3.676,
+      "train_steps_per_second": 0.057
     },
     {
       "epoch": 119.00833333333334,
-      "eval_accuracy": 0.514367816091954,
-      "eval_loss": 1.998863697052002,
-      "eval_runtime": 145.2117,
-      "eval_samples_per_second": 9.586,
-      "eval_steps_per_second": 0.152,
+      "eval_accuracy": 0.8629629629629629,
+      "eval_loss": 0.5552316308021545,
+      "eval_runtime": 65.2796,
+      "eval_samples_per_second": 4.136,
+      "eval_steps_per_second": 0.077,
       "step": 1920
     },
     {
       "epoch": 119.00833333333334,
-      "eval_accuracy": 0.514367816091954,
-      "eval_loss": 1.998863697052002,
-      "eval_runtime": 148.8403,
-      "eval_samples_per_second": 9.352,
-      "eval_steps_per_second": 0.148,
+      "eval_accuracy": 0.8629629629629629,
+      "eval_loss": 0.557027280330658,
+      "eval_runtime": 58.3366,
+      "eval_samples_per_second": 4.628,
+      "eval_steps_per_second": 0.086,
       "step": 1920
     }
   ],