diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,9 +1,9 @@
 {
-  "best_metric": 0.4903908967971802,
-  "best_model_checkpoint": "./beans_outputs/checkpoint-621",
-  "epoch": 50.0,
+  "best_metric": 0.4896911084651947,
+  "best_model_checkpoint": "./beans_outputs/checkpoint-1495",
+  "epoch": 200.0,
   "eval_steps": 500,
-  "global_step": 1150,
+  "global_step": 4600,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -1264,19 +1264,3784 @@
       "step": 1150
     },
     {
-      "epoch": 50.0,
-      "step": 1150,
-      "total_flos": 2.72467378584576e+17,
-      "train_loss": 0.4791690407628598,
-      "train_runtime": 1616.3187,
-      "train_samples_per_second": 44.391,
-      "train_steps_per_second": 0.711
+      "epoch": 50.43478260869565,
+      "grad_norm": 2.7469470500946045,
+      "learning_rate": 1.4956521739130436e-05,
+      "loss": 0.4877,
+      "step": 1160
+    },
+    {
+      "epoch": 50.869565217391305,
+      "grad_norm": 2.5254504680633545,
+      "learning_rate": 1.491304347826087e-05,
+      "loss": 0.4366,
+      "step": 1170
+    },
+    {
+      "epoch": 51.0,
+      "eval_accuracy": 0.7755905511811023,
+      "eval_loss": 0.497986376285553,
+      "eval_runtime": 3.9676,
+      "eval_samples_per_second": 64.019,
+      "eval_steps_per_second": 0.504,
+      "step": 1173
+    },
+    {
+      "epoch": 51.30434782608695,
+      "grad_norm": 2.032457113265991,
+      "learning_rate": 1.4869565217391306e-05,
+      "loss": 0.4663,
+      "step": 1180
+    },
+    {
+      "epoch": 51.73913043478261,
+      "grad_norm": 2.802882432937622,
+      "learning_rate": 1.4826086956521741e-05,
+      "loss": 0.4467,
+      "step": 1190
+    },
+    {
+      "epoch": 52.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.49471431970596313,
+      "eval_runtime": 4.8494,
+      "eval_samples_per_second": 52.378,
+      "eval_steps_per_second": 0.412,
+      "step": 1196
+    },
+    {
+      "epoch": 52.17391304347826,
+      "grad_norm": 2.2359466552734375,
+      "learning_rate": 1.4782608695652174e-05,
+      "loss": 0.4424,
+      "step": 1200
+    },
+    {
+      "epoch": 52.608695652173914,
+      "grad_norm": 3.219308376312256,
+      "learning_rate": 1.473913043478261e-05,
+      "loss": 0.4797,
+      "step": 1210
+    },
+    {
+      "epoch": 53.0,
+      "eval_accuracy": 0.7755905511811023,
+      "eval_loss": 0.4950390160083771,
+      "eval_runtime": 3.8997,
+      "eval_samples_per_second": 65.133,
+      "eval_steps_per_second": 0.513,
+      "step": 1219
+    },
+    {
+      "epoch": 53.04347826086956,
+      "grad_norm": 2.6939969062805176,
+      "learning_rate": 1.4695652173913045e-05,
+      "loss": 0.4256,
+      "step": 1220
+    },
+    {
+      "epoch": 53.47826086956522,
+      "grad_norm": 2.6343085765838623,
+      "learning_rate": 1.465217391304348e-05,
+      "loss": 0.4192,
+      "step": 1230
+    },
+    {
+      "epoch": 53.91304347826087,
+      "grad_norm": 10.655885696411133,
+      "learning_rate": 1.4608695652173915e-05,
+      "loss": 0.4544,
+      "step": 1240
+    },
+    {
+      "epoch": 54.0,
+      "eval_accuracy": 0.7716535433070866,
+      "eval_loss": 0.49978330731391907,
+      "eval_runtime": 3.7833,
+      "eval_samples_per_second": 67.137,
+      "eval_steps_per_second": 0.529,
+      "step": 1242
+    },
+    {
+      "epoch": 54.34782608695652,
+      "grad_norm": 4.838284969329834,
+      "learning_rate": 1.456521739130435e-05,
+      "loss": 0.4361,
+      "step": 1250
+    },
+    {
+      "epoch": 54.78260869565217,
+      "grad_norm": 3.1171820163726807,
+      "learning_rate": 1.4521739130434785e-05,
+      "loss": 0.4466,
+      "step": 1260
+    },
+    {
+      "epoch": 55.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.49803978204727173,
+      "eval_runtime": 3.7735,
+      "eval_samples_per_second": 67.311,
+      "eval_steps_per_second": 0.53,
+      "step": 1265
+    },
+    {
+      "epoch": 55.21739130434783,
+      "grad_norm": 2.5744450092315674,
+      "learning_rate": 1.447826086956522e-05,
+      "loss": 0.4511,
+      "step": 1270
+    },
+    {
+      "epoch": 55.65217391304348,
+      "grad_norm": 7.211576461791992,
+      "learning_rate": 1.4434782608695654e-05,
+      "loss": 0.4599,
+      "step": 1280
+    },
+    {
+      "epoch": 56.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.4962967336177826,
+      "eval_runtime": 4.8613,
+      "eval_samples_per_second": 52.249,
+      "eval_steps_per_second": 0.411,
+      "step": 1288
+    },
+    {
+      "epoch": 56.08695652173913,
+      "grad_norm": 1.932460069656372,
+      "learning_rate": 1.4391304347826087e-05,
+      "loss": 0.4168,
+      "step": 1290
+    },
+    {
+      "epoch": 56.52173913043478,
+      "grad_norm": 5.841196537017822,
+      "learning_rate": 1.4347826086956522e-05,
+      "loss": 0.4622,
+      "step": 1300
+    },
+    {
+      "epoch": 56.95652173913044,
+      "grad_norm": 1.948188066482544,
+      "learning_rate": 1.4304347826086957e-05,
+      "loss": 0.4458,
+      "step": 1310
+    },
+    {
+      "epoch": 57.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.49557480216026306,
+      "eval_runtime": 3.6949,
+      "eval_samples_per_second": 68.744,
+      "eval_steps_per_second": 0.541,
+      "step": 1311
+    },
+    {
+      "epoch": 57.391304347826086,
+      "grad_norm": 4.304020881652832,
+      "learning_rate": 1.4260869565217392e-05,
+      "loss": 0.4378,
+      "step": 1320
+    },
+    {
+      "epoch": 57.82608695652174,
+      "grad_norm": 2.710130453109741,
+      "learning_rate": 1.4217391304347828e-05,
+      "loss": 0.4296,
+      "step": 1330
+    },
+    {
+      "epoch": 58.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.49939388036727905,
+      "eval_runtime": 3.7356,
+      "eval_samples_per_second": 67.995,
+      "eval_steps_per_second": 0.535,
+      "step": 1334
+    },
+    {
+      "epoch": 58.26086956521739,
+      "grad_norm": 3.730140209197998,
+      "learning_rate": 1.4173913043478263e-05,
+      "loss": 0.4664,
+      "step": 1340
+    },
+    {
+      "epoch": 58.69565217391305,
+      "grad_norm": 9.71405029296875,
+      "learning_rate": 1.4130434782608698e-05,
+      "loss": 0.4415,
+      "step": 1350
+    },
+    {
+      "epoch": 59.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.4997561573982239,
+      "eval_runtime": 4.9859,
+      "eval_samples_per_second": 50.943,
+      "eval_steps_per_second": 0.401,
+      "step": 1357
+    },
+    {
+      "epoch": 59.130434782608695,
+      "grad_norm": 2.7752935886383057,
+      "learning_rate": 1.4086956521739133e-05,
+      "loss": 0.4436,
+      "step": 1360
+    },
+    {
+      "epoch": 59.56521739130435,
+      "grad_norm": 4.0491251945495605,
+      "learning_rate": 1.4043478260869568e-05,
+      "loss": 0.4442,
+      "step": 1370
+    },
+    {
+      "epoch": 60.0,
+      "grad_norm": 3.6015145778656006,
+      "learning_rate": 1.4e-05,
+      "loss": 0.4036,
+      "step": 1380
+    },
+    {
+      "epoch": 60.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.4996122717857361,
+      "eval_runtime": 3.8039,
+      "eval_samples_per_second": 66.774,
+      "eval_steps_per_second": 0.526,
+      "step": 1380
+    },
+    {
+      "epoch": 60.43478260869565,
+      "grad_norm": 2.5297908782958984,
+      "learning_rate": 1.3956521739130435e-05,
+      "loss": 0.4364,
+      "step": 1390
+    },
+    {
+      "epoch": 60.869565217391305,
+      "grad_norm": 2.8682429790496826,
+      "learning_rate": 1.391304347826087e-05,
+      "loss": 0.4406,
+      "step": 1400
+    },
+    {
+      "epoch": 61.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5022182464599609,
+      "eval_runtime": 3.7333,
+      "eval_samples_per_second": 68.036,
+      "eval_steps_per_second": 0.536,
+      "step": 1403
+    },
+    {
+      "epoch": 61.30434782608695,
+      "grad_norm": 3.3014872074127197,
+      "learning_rate": 1.3869565217391305e-05,
+      "loss": 0.4346,
+      "step": 1410
+    },
+    {
+      "epoch": 61.73913043478261,
+      "grad_norm": 3.4654860496520996,
+      "learning_rate": 1.382608695652174e-05,
+      "loss": 0.4235,
+      "step": 1420
+    },
+    {
+      "epoch": 62.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5018435120582581,
+      "eval_runtime": 4.9797,
+      "eval_samples_per_second": 51.007,
+      "eval_steps_per_second": 0.402,
+      "step": 1426
+    },
+    {
+      "epoch": 62.17391304347826,
+      "grad_norm": 6.609365940093994,
+      "learning_rate": 1.3782608695652175e-05,
+      "loss": 0.4884,
+      "step": 1430
+    },
+    {
+      "epoch": 62.608695652173914,
+      "grad_norm": 3.425076484680176,
+      "learning_rate": 1.373913043478261e-05,
+      "loss": 0.4492,
+      "step": 1440
+    },
+    {
+      "epoch": 63.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.4963783323764801,
+      "eval_runtime": 3.7677,
+      "eval_samples_per_second": 67.416,
+      "eval_steps_per_second": 0.531,
+      "step": 1449
+    },
+    {
+      "epoch": 63.04347826086956,
+      "grad_norm": 4.069096088409424,
+      "learning_rate": 1.3695652173913046e-05,
+      "loss": 0.4119,
+      "step": 1450
+    },
+    {
+      "epoch": 63.47826086956522,
+      "grad_norm": 2.3584377765655518,
+      "learning_rate": 1.3652173913043479e-05,
+      "loss": 0.4356,
+      "step": 1460
+    },
+    {
+      "epoch": 63.91304347826087,
+      "grad_norm": 12.776151657104492,
+      "learning_rate": 1.3608695652173913e-05,
+      "loss": 0.4065,
+      "step": 1470
+    },
+    {
+      "epoch": 64.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.4952879250049591,
+      "eval_runtime": 3.7751,
+      "eval_samples_per_second": 67.284,
+      "eval_steps_per_second": 0.53,
+      "step": 1472
+    },
+    {
+      "epoch": 64.34782608695652,
+      "grad_norm": 2.501909017562866,
+      "learning_rate": 1.3565217391304348e-05,
+      "loss": 0.4356,
+      "step": 1480
+    },
+    {
+      "epoch": 64.78260869565217,
+      "grad_norm": 2.650075674057007,
+      "learning_rate": 1.3521739130434783e-05,
+      "loss": 0.4474,
+      "step": 1490
+    },
+    {
+      "epoch": 65.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.4896911084651947,
+      "eval_runtime": 5.1494,
+      "eval_samples_per_second": 49.326,
+      "eval_steps_per_second": 0.388,
+      "step": 1495
+    },
+    {
+      "epoch": 65.21739130434783,
+      "grad_norm": 4.465973854064941,
+      "learning_rate": 1.3478260869565218e-05,
+      "loss": 0.4246,
+      "step": 1500
+    },
+    {
+      "epoch": 65.65217391304348,
+      "grad_norm": 2.057035207748413,
+      "learning_rate": 1.3434782608695653e-05,
+      "loss": 0.4605,
+      "step": 1510
+    },
+    {
+      "epoch": 66.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.5039426684379578,
+      "eval_runtime": 3.7424,
+      "eval_samples_per_second": 67.871,
+      "eval_steps_per_second": 0.534,
+      "step": 1518
+    },
+    {
+      "epoch": 66.08695652173913,
+      "grad_norm": 3.064012050628662,
+      "learning_rate": 1.3391304347826088e-05,
+      "loss": 0.4157,
+      "step": 1520
+    },
+    {
+      "epoch": 66.52173913043478,
+      "grad_norm": 3.7584011554718018,
+      "learning_rate": 1.3347826086956523e-05,
+      "loss": 0.4553,
+      "step": 1530
+    },
+    {
+      "epoch": 66.95652173913044,
+      "grad_norm": 2.945054054260254,
+      "learning_rate": 1.3304347826086958e-05,
+      "loss": 0.436,
+      "step": 1540
+    },
+    {
+      "epoch": 67.0,
+      "eval_accuracy": 0.7755905511811023,
+      "eval_loss": 0.5024412274360657,
+      "eval_runtime": 3.7621,
+      "eval_samples_per_second": 67.516,
+      "eval_steps_per_second": 0.532,
+      "step": 1541
+    },
+    {
+      "epoch": 67.3913043478261,
+      "grad_norm": 3.1257166862487793,
+      "learning_rate": 1.3260869565217392e-05,
+      "loss": 0.4173,
+      "step": 1550
+    },
+    {
+      "epoch": 67.82608695652173,
+      "grad_norm": 5.225259304046631,
+      "learning_rate": 1.3217391304347827e-05,
+      "loss": 0.4746,
+      "step": 1560
+    },
+    {
+      "epoch": 68.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5006521344184875,
+      "eval_runtime": 5.0155,
+      "eval_samples_per_second": 50.643,
+      "eval_steps_per_second": 0.399,
+      "step": 1564
+    },
+    {
+      "epoch": 68.26086956521739,
+      "grad_norm": 3.3438003063201904,
+      "learning_rate": 1.3173913043478262e-05,
+      "loss": 0.3957,
+      "step": 1570
+    },
+    {
+      "epoch": 68.69565217391305,
+      "grad_norm": 2.6640641689300537,
+      "learning_rate": 1.3130434782608697e-05,
+      "loss": 0.4555,
+      "step": 1580
+    },
+    {
+      "epoch": 69.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5053796172142029,
+      "eval_runtime": 3.7207,
+      "eval_samples_per_second": 68.266,
+      "eval_steps_per_second": 0.538,
+      "step": 1587
+    },
+    {
+      "epoch": 69.1304347826087,
+      "grad_norm": 6.726771831512451,
+      "learning_rate": 1.308695652173913e-05,
+      "loss": 0.4322,
+      "step": 1590
+    },
+    {
+      "epoch": 69.56521739130434,
+      "grad_norm": 3.231029748916626,
+      "learning_rate": 1.3043478260869566e-05,
+      "loss": 0.3949,
+      "step": 1600
+    },
+    {
+      "epoch": 70.0,
+      "grad_norm": 6.560612201690674,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.433,
+      "step": 1610
+    },
+    {
+      "epoch": 70.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.49738696217536926,
+      "eval_runtime": 3.716,
+      "eval_samples_per_second": 68.353,
+      "eval_steps_per_second": 0.538,
+      "step": 1610
+    },
+    {
+      "epoch": 70.43478260869566,
+      "grad_norm": 2.6361474990844727,
+      "learning_rate": 1.2956521739130436e-05,
+      "loss": 0.4158,
+      "step": 1620
+    },
+    {
+      "epoch": 70.8695652173913,
+      "grad_norm": 2.7182960510253906,
+      "learning_rate": 1.2913043478260871e-05,
+      "loss": 0.4503,
+      "step": 1630
+    },
+    {
+      "epoch": 71.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.509588897228241,
+      "eval_runtime": 5.0488,
+      "eval_samples_per_second": 50.309,
+      "eval_steps_per_second": 0.396,
+      "step": 1633
+    },
+    {
+      "epoch": 71.30434782608695,
+      "grad_norm": 1.938330888748169,
+      "learning_rate": 1.2869565217391305e-05,
+      "loss": 0.4285,
+      "step": 1640
+    },
+    {
+      "epoch": 71.73913043478261,
+      "grad_norm": 3.1797378063201904,
+      "learning_rate": 1.282608695652174e-05,
+      "loss": 0.4424,
+      "step": 1650
+    },
+    {
+      "epoch": 72.0,
+      "eval_accuracy": 0.7755905511811023,
+      "eval_loss": 0.5040333867073059,
+      "eval_runtime": 3.7598,
+      "eval_samples_per_second": 67.557,
+      "eval_steps_per_second": 0.532,
+      "step": 1656
+    },
+    {
+      "epoch": 72.17391304347827,
+      "grad_norm": 3.028841257095337,
+      "learning_rate": 1.2782608695652175e-05,
+      "loss": 0.4278,
+      "step": 1660
+    },
+    {
+      "epoch": 72.6086956521739,
+      "grad_norm": 3.0137178897857666,
+      "learning_rate": 1.273913043478261e-05,
+      "loss": 0.4331,
+      "step": 1670
+    },
+    {
+      "epoch": 73.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.505591869354248,
+      "eval_runtime": 3.7202,
+      "eval_samples_per_second": 68.277,
+      "eval_steps_per_second": 0.538,
+      "step": 1679
+    },
+    {
+      "epoch": 73.04347826086956,
+      "grad_norm": 3.519934892654419,
+      "learning_rate": 1.2695652173913045e-05,
+      "loss": 0.4523,
+      "step": 1680
+    },
+    {
+      "epoch": 73.47826086956522,
+      "grad_norm": 2.6839394569396973,
+      "learning_rate": 1.265217391304348e-05,
+      "loss": 0.4143,
+      "step": 1690
+    },
+    {
+      "epoch": 73.91304347826087,
+      "grad_norm": 4.223355770111084,
+      "learning_rate": 1.2608695652173915e-05,
+      "loss": 0.4263,
+      "step": 1700
+    },
+    {
+      "epoch": 74.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5025500059127808,
+      "eval_runtime": 5.0354,
+      "eval_samples_per_second": 50.442,
+      "eval_steps_per_second": 0.397,
+      "step": 1702
+    },
+    {
+      "epoch": 74.34782608695652,
+      "grad_norm": 2.633610248565674,
+      "learning_rate": 1.2565217391304349e-05,
+      "loss": 0.4451,
+      "step": 1710
+    },
+    {
+      "epoch": 74.78260869565217,
+      "grad_norm": 4.227041721343994,
+      "learning_rate": 1.2521739130434784e-05,
+      "loss": 0.4305,
+      "step": 1720
+    },
+    {
+      "epoch": 75.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5032832026481628,
+      "eval_runtime": 3.7074,
+      "eval_samples_per_second": 68.512,
+      "eval_steps_per_second": 0.539,
+      "step": 1725
+    },
+    {
+      "epoch": 75.21739130434783,
+      "grad_norm": 3.885732412338257,
+      "learning_rate": 1.2478260869565217e-05,
+      "loss": 0.4177,
+      "step": 1730
+    },
+    {
+      "epoch": 75.65217391304348,
+      "grad_norm": 6.669870853424072,
+      "learning_rate": 1.2434782608695652e-05,
+      "loss": 0.4271,
+      "step": 1740
+    },
+    {
+      "epoch": 76.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5014809966087341,
+      "eval_runtime": 3.6911,
+      "eval_samples_per_second": 68.814,
+      "eval_steps_per_second": 0.542,
+      "step": 1748
+    },
+    {
+      "epoch": 76.08695652173913,
+      "grad_norm": 1.820388913154602,
+      "learning_rate": 1.2391304347826088e-05,
+      "loss": 0.4457,
+      "step": 1750
+    },
+    {
+      "epoch": 76.52173913043478,
+      "grad_norm": 2.142805337905884,
+      "learning_rate": 1.2347826086956523e-05,
+      "loss": 0.3962,
+      "step": 1760
+    },
+    {
+      "epoch": 76.95652173913044,
+      "grad_norm": 3.5151073932647705,
+      "learning_rate": 1.2304347826086958e-05,
+      "loss": 0.4635,
+      "step": 1770
+    },
+    {
+      "epoch": 77.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.49884113669395447,
+      "eval_runtime": 4.3354,
+      "eval_samples_per_second": 58.587,
+      "eval_steps_per_second": 0.461,
+      "step": 1771
+    },
+    {
+      "epoch": 77.3913043478261,
+      "grad_norm": 3.867955207824707,
+      "learning_rate": 1.2260869565217393e-05,
+      "loss": 0.4616,
+      "step": 1780
+    },
+    {
+      "epoch": 77.82608695652173,
+      "grad_norm": 2.6050870418548584,
+      "learning_rate": 1.2217391304347828e-05,
+      "loss": 0.4212,
+      "step": 1790
+    },
+    {
+      "epoch": 78.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.4993511438369751,
+      "eval_runtime": 4.267,
+      "eval_samples_per_second": 59.527,
+      "eval_steps_per_second": 0.469,
+      "step": 1794
+    },
+    {
+      "epoch": 78.26086956521739,
+      "grad_norm": 2.1961538791656494,
+      "learning_rate": 1.2173913043478263e-05,
+      "loss": 0.4191,
+      "step": 1800
+    },
+    {
+      "epoch": 78.69565217391305,
+      "grad_norm": 6.02454948425293,
+      "learning_rate": 1.2130434782608698e-05,
+      "loss": 0.4154,
+      "step": 1810
+    },
+    {
+      "epoch": 79.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5044043660163879,
+      "eval_runtime": 3.7036,
+      "eval_samples_per_second": 68.581,
+      "eval_steps_per_second": 0.54,
+      "step": 1817
+    },
+    {
+      "epoch": 79.1304347826087,
+      "grad_norm": 2.1048858165740967,
+      "learning_rate": 1.208695652173913e-05,
+      "loss": 0.4196,
+      "step": 1820
+    },
+    {
+      "epoch": 79.56521739130434,
+      "grad_norm": 2.8622193336486816,
+      "learning_rate": 1.2043478260869565e-05,
+      "loss": 0.4314,
+      "step": 1830
+    },
+    {
+      "epoch": 80.0,
+      "grad_norm": 6.1558427810668945,
+      "learning_rate": 1.2e-05,
+      "loss": 0.4288,
+      "step": 1840
+    },
+    {
+      "epoch": 80.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5033003687858582,
+      "eval_runtime": 3.7575,
+      "eval_samples_per_second": 67.598,
+      "eval_steps_per_second": 0.532,
+      "step": 1840
+    },
+    {
+      "epoch": 80.43478260869566,
+      "grad_norm": 3.3254945278167725,
+      "learning_rate": 1.1956521739130435e-05,
+      "loss": 0.4297,
+      "step": 1850
+    },
+    {
+      "epoch": 80.8695652173913,
+      "grad_norm": 2.2818620204925537,
+      "learning_rate": 1.191304347826087e-05,
+      "loss": 0.4211,
+      "step": 1860
+    },
+    {
+      "epoch": 81.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5050157904624939,
+      "eval_runtime": 5.0113,
+      "eval_samples_per_second": 50.685,
+      "eval_steps_per_second": 0.399,
+      "step": 1863
+    },
+    {
+      "epoch": 81.30434782608695,
+      "grad_norm": 4.174459934234619,
+      "learning_rate": 1.1869565217391306e-05,
+      "loss": 0.4229,
+      "step": 1870
+    },
+    {
+      "epoch": 81.73913043478261,
+      "grad_norm": 2.87514066696167,
+      "learning_rate": 1.182608695652174e-05,
+      "loss": 0.4022,
+      "step": 1880
+    },
+    {
+      "epoch": 82.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5021248459815979,
+      "eval_runtime": 3.7629,
+      "eval_samples_per_second": 67.5,
+      "eval_steps_per_second": 0.531,
+      "step": 1886
+    },
+    {
+      "epoch": 82.17391304347827,
+      "grad_norm": 5.307149410247803,
+      "learning_rate": 1.1782608695652176e-05,
+      "loss": 0.4564,
+      "step": 1890
+    },
+    {
+      "epoch": 82.6086956521739,
+      "grad_norm": 4.411511421203613,
+      "learning_rate": 1.1739130434782611e-05,
+      "loss": 0.4477,
+      "step": 1900
+    },
+    {
+      "epoch": 83.0,
+      "eval_accuracy": 0.7755905511811023,
+      "eval_loss": 0.509568452835083,
+      "eval_runtime": 5.3605,
+      "eval_samples_per_second": 47.384,
+      "eval_steps_per_second": 0.373,
+      "step": 1909
+    },
+    {
+      "epoch": 83.04347826086956,
+      "grad_norm": 2.478482246398926,
+      "learning_rate": 1.1695652173913043e-05,
+      "loss": 0.4118,
+      "step": 1910
+    },
+    {
+      "epoch": 83.47826086956522,
+      "grad_norm": 2.000185012817383,
+      "learning_rate": 1.1652173913043478e-05,
+      "loss": 0.4486,
+      "step": 1920
+    },
+    {
+      "epoch": 83.91304347826087,
+      "grad_norm": 4.231175422668457,
+      "learning_rate": 1.1608695652173913e-05,
+      "loss": 0.4091,
+      "step": 1930
+    },
+    {
+      "epoch": 84.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.501672625541687,
+      "eval_runtime": 4.6714,
+      "eval_samples_per_second": 54.374,
+      "eval_steps_per_second": 0.428,
+      "step": 1932
+    },
+    {
+      "epoch": 84.34782608695652,
+      "grad_norm": 8.062799453735352,
+      "learning_rate": 1.1565217391304348e-05,
+      "loss": 0.4108,
+      "step": 1940
+    },
+    {
+      "epoch": 84.78260869565217,
+      "grad_norm": 3.525912046432495,
+      "learning_rate": 1.1521739130434783e-05,
+      "loss": 0.4284,
+      "step": 1950
+    },
+    {
+      "epoch": 85.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.5094006061553955,
+      "eval_runtime": 3.7057,
+      "eval_samples_per_second": 68.544,
+      "eval_steps_per_second": 0.54,
+      "step": 1955
+    },
+    {
+      "epoch": 85.21739130434783,
+      "grad_norm": 2.8294172286987305,
+      "learning_rate": 1.1478260869565218e-05,
+      "loss": 0.4341,
+      "step": 1960
+    },
+    {
+      "epoch": 85.65217391304348,
+      "grad_norm": 2.6164603233337402,
+      "learning_rate": 1.1434782608695654e-05,
+      "loss": 0.4317,
+      "step": 1970
+    },
+    {
+      "epoch": 86.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5055702328681946,
+      "eval_runtime": 3.78,
+      "eval_samples_per_second": 67.195,
+      "eval_steps_per_second": 0.529,
+      "step": 1978
+    },
+    {
+      "epoch": 86.08695652173913,
+      "grad_norm": 5.29531717300415,
+      "learning_rate": 1.1391304347826089e-05,
+      "loss": 0.3842,
+      "step": 1980
+    },
+    {
+      "epoch": 86.52173913043478,
+      "grad_norm": 3.8016159534454346,
+      "learning_rate": 1.1347826086956524e-05,
+      "loss": 0.4294,
+      "step": 1990
+    },
+    {
+      "epoch": 86.95652173913044,
+      "grad_norm": 2.229055643081665,
+      "learning_rate": 1.1304347826086957e-05,
+      "loss": 0.4011,
+      "step": 2000
+    },
+    {
+      "epoch": 87.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.4991566836833954,
+      "eval_runtime": 4.9616,
+      "eval_samples_per_second": 51.194,
+      "eval_steps_per_second": 0.403,
+      "step": 2001
+    },
+    {
+      "epoch": 87.3913043478261,
+      "grad_norm": 4.449975490570068,
+      "learning_rate": 1.1260869565217392e-05,
+      "loss": 0.4413,
+      "step": 2010
+    },
+    {
+      "epoch": 87.82608695652173,
+      "grad_norm": 3.4843342304229736,
+      "learning_rate": 1.1217391304347827e-05,
+      "loss": 0.4043,
+      "step": 2020
+    },
+    {
+      "epoch": 88.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5106358528137207,
+      "eval_runtime": 3.759,
+      "eval_samples_per_second": 67.571,
+      "eval_steps_per_second": 0.532,
+      "step": 2024
+    },
+    {
+      "epoch": 88.26086956521739,
+      "grad_norm": 3.2311477661132812,
+      "learning_rate": 1.1173913043478261e-05,
+      "loss": 0.4127,
+      "step": 2030
+    },
+    {
+      "epoch": 88.69565217391305,
+      "grad_norm": 3.511033058166504,
+      "learning_rate": 1.1130434782608696e-05,
+      "loss": 0.4233,
+      "step": 2040
+    },
+    {
+      "epoch": 89.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5082967877388,
+      "eval_runtime": 3.7279,
+      "eval_samples_per_second": 68.135,
+      "eval_steps_per_second": 0.536,
+      "step": 2047
+    },
+    {
+      "epoch": 89.1304347826087,
+      "grad_norm": 3.1737847328186035,
+      "learning_rate": 1.1086956521739131e-05,
+      "loss": 0.4449,
+      "step": 2050
+    },
+    {
+      "epoch": 89.56521739130434,
+      "grad_norm": 3.3332552909851074,
+      "learning_rate": 1.1043478260869566e-05,
+      "loss": 0.4148,
+      "step": 2060
+    },
+    {
+      "epoch": 90.0,
+      "grad_norm": 5.011209487915039,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.4383,
+      "step": 2070
+    },
+    {
+      "epoch": 90.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5015798211097717,
+      "eval_runtime": 5.0126,
+      "eval_samples_per_second": 50.672,
+      "eval_steps_per_second": 0.399,
+      "step": 2070
+    },
+    {
+      "epoch": 90.43478260869566,
+      "grad_norm": 2.4368808269500732,
+      "learning_rate": 1.0956521739130435e-05,
+      "loss": 0.4133,
+      "step": 2080
+    },
+    {
+      "epoch": 90.8695652173913,
+      "grad_norm": 5.885110378265381,
+      "learning_rate": 1.091304347826087e-05,
+      "loss": 0.4328,
+      "step": 2090
+    },
+    {
+      "epoch": 91.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5062097311019897,
+      "eval_runtime": 3.7299,
+      "eval_samples_per_second": 68.098,
+      "eval_steps_per_second": 0.536,
+      "step": 2093
+    },
+    {
+      "epoch": 91.30434782608695,
+      "grad_norm": 2.2072901725769043,
+      "learning_rate": 1.0869565217391305e-05,
+      "loss": 0.4137,
+      "step": 2100
+    },
+    {
+      "epoch": 91.73913043478261,
+      "grad_norm": 1.9848076105117798,
+      "learning_rate": 1.082608695652174e-05,
+      "loss": 0.3978,
+      "step": 2110
+    },
+    {
+      "epoch": 92.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5026075839996338,
+      "eval_runtime": 3.7759,
+      "eval_samples_per_second": 67.268,
+      "eval_steps_per_second": 0.53,
+      "step": 2116
+    },
+    {
+      "epoch": 92.17391304347827,
+      "grad_norm": 3.738398313522339,
+      "learning_rate": 1.0782608695652175e-05,
+      "loss": 0.4459,
+      "step": 2120
+    },
+    {
+      "epoch": 92.6086956521739,
+      "grad_norm": 3.0096168518066406,
+      "learning_rate": 1.073913043478261e-05,
+      "loss": 0.4052,
+      "step": 2130
+    },
+    {
+      "epoch": 93.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.49642127752304077,
+      "eval_runtime": 5.0236,
+      "eval_samples_per_second": 50.562,
+      "eval_steps_per_second": 0.398,
+      "step": 2139
+    },
+    {
+      "epoch": 93.04347826086956,
+      "grad_norm": 3.7452170848846436,
+      "learning_rate": 1.0695652173913046e-05,
+      "loss": 0.4205,
+      "step": 2140
+    },
+    {
+      "epoch": 93.47826086956522,
+      "grad_norm": 3.8985049724578857,
+      "learning_rate": 1.0652173913043479e-05,
+      "loss": 0.4171,
+      "step": 2150
+    },
+    {
+      "epoch": 93.91304347826087,
+      "grad_norm": 2.283020496368408,
+      "learning_rate": 1.0608695652173914e-05,
+      "loss": 0.3938,
+      "step": 2160
+    },
+    {
+      "epoch": 94.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5036487579345703,
+      "eval_runtime": 3.7298,
+      "eval_samples_per_second": 68.101,
+      "eval_steps_per_second": 0.536,
+      "step": 2162
+    },
+    {
+      "epoch": 94.34782608695652,
+      "grad_norm": 7.054046630859375,
+      "learning_rate": 1.0565217391304348e-05,
+      "loss": 0.4336,
+      "step": 2170
+    },
+    {
+      "epoch": 94.78260869565217,
+      "grad_norm": 3.131002902984619,
+      "learning_rate": 1.0521739130434783e-05,
+      "loss": 0.393,
+      "step": 2180
+    },
+    {
+      "epoch": 95.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5102458596229553,
+      "eval_runtime": 3.6839,
+      "eval_samples_per_second": 68.949,
+      "eval_steps_per_second": 0.543,
+      "step": 2185
+    },
+    {
+      "epoch": 95.21739130434783,
+      "grad_norm": 2.4622268676757812,
+      "learning_rate": 1.0478260869565218e-05,
+      "loss": 0.3997,
+      "step": 2190
+    },
+    {
+      "epoch": 95.65217391304348,
+      "grad_norm": 3.815375566482544,
+      "learning_rate": 1.0434782608695653e-05,
+      "loss": 0.4294,
+      "step": 2200
+    },
+    {
+      "epoch": 96.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5002910494804382,
+      "eval_runtime": 4.8997,
+      "eval_samples_per_second": 51.84,
+      "eval_steps_per_second": 0.408,
+      "step": 2208
+    },
+    {
+      "epoch": 96.08695652173913,
+      "grad_norm": 8.787290573120117,
+      "learning_rate": 1.0391304347826088e-05,
+      "loss": 0.4155,
+      "step": 2210
+    },
+    {
+      "epoch": 96.52173913043478,
+      "grad_norm": 2.8499906063079834,
+      "learning_rate": 1.0347826086956523e-05,
+      "loss": 0.4095,
+      "step": 2220
+    },
+    {
+      "epoch": 96.95652173913044,
+      "grad_norm": 6.26355504989624,
+      "learning_rate": 1.0304347826086958e-05,
+      "loss": 0.4122,
+      "step": 2230
+    },
+    {
+      "epoch": 97.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5013226270675659,
+      "eval_runtime": 3.7744,
+      "eval_samples_per_second": 67.296,
+      "eval_steps_per_second": 0.53,
+      "step": 2231
+    },
+    {
+      "epoch": 97.3913043478261,
+      "grad_norm": 3.257772445678711,
+      "learning_rate": 1.0260869565217393e-05,
+      "loss": 0.3522,
+      "step": 2240
+    },
+    {
+      "epoch": 97.82608695652173,
+      "grad_norm": 2.788611888885498,
+      "learning_rate": 1.0217391304347829e-05,
+      "loss": 0.4207,
+      "step": 2250
+    },
+    {
+      "epoch": 98.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.507587730884552,
+      "eval_runtime": 3.8534,
+      "eval_samples_per_second": 65.915,
+      "eval_steps_per_second": 0.519,
+      "step": 2254
+    },
+    {
+      "epoch": 98.26086956521739,
+      "grad_norm": 2.974043846130371,
+      "learning_rate": 1.017391304347826e-05,
+      "loss": 0.4352,
+      "step": 2260
+    },
+    {
+      "epoch": 98.69565217391305,
+      "grad_norm": 3.231869697570801,
+      "learning_rate": 1.0130434782608695e-05,
+      "loss": 0.4127,
+      "step": 2270
+    },
+    {
+      "epoch": 99.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.503979504108429,
+      "eval_runtime": 5.0998,
+      "eval_samples_per_second": 49.806,
+      "eval_steps_per_second": 0.392,
+      "step": 2277
+    },
+    {
+      "epoch": 99.1304347826087,
+      "grad_norm": 2.597999095916748,
+      "learning_rate": 1.008695652173913e-05,
+      "loss": 0.3888,
+      "step": 2280
+    },
+    {
+      "epoch": 99.56521739130434,
+      "grad_norm": 4.4219889640808105,
+      "learning_rate": 1.0043478260869566e-05,
+      "loss": 0.3921,
+      "step": 2290
+    },
+    {
+      "epoch": 100.0,
+      "grad_norm": 4.641758441925049,
+      "learning_rate": 1e-05,
+      "loss": 0.441,
+      "step": 2300
+    },
+    {
+      "epoch": 100.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.502194881439209,
+      "eval_runtime": 3.7077,
+      "eval_samples_per_second": 68.506,
+      "eval_steps_per_second": 0.539,
+      "step": 2300
+    },
+    {
+      "epoch": 100.43478260869566,
+      "grad_norm": 2.948529005050659,
+      "learning_rate": 9.956521739130436e-06,
+      "loss": 0.4324,
+      "step": 2310
+    },
+    {
+      "epoch": 100.8695652173913,
+      "grad_norm": 2.4855594635009766,
+      "learning_rate": 9.913043478260871e-06,
+      "loss": 0.3938,
+      "step": 2320
+    },
+    {
+      "epoch": 101.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.4974897503852844,
+      "eval_runtime": 3.7364,
+      "eval_samples_per_second": 67.98,
+      "eval_steps_per_second": 0.535,
+      "step": 2323
+    },
+    {
+      "epoch": 101.30434782608695,
+      "grad_norm": 4.753269195556641,
+      "learning_rate": 9.869565217391304e-06,
+      "loss": 0.3918,
+      "step": 2330
+    },
+    {
+      "epoch": 101.73913043478261,
+      "grad_norm": 5.000470161437988,
+      "learning_rate": 9.82608695652174e-06,
+      "loss": 0.4109,
+      "step": 2340
+    },
+    {
+      "epoch": 102.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5018798112869263,
+      "eval_runtime": 4.8425,
+      "eval_samples_per_second": 52.452,
+      "eval_steps_per_second": 0.413,
+      "step": 2346
+    },
+    {
+      "epoch": 102.17391304347827,
+      "grad_norm": 2.8584697246551514,
+      "learning_rate": 9.782608695652175e-06,
+      "loss": 0.4199,
+      "step": 2350
+    },
+    {
+      "epoch": 102.6086956521739,
+      "grad_norm": 2.773083448410034,
+      "learning_rate": 9.73913043478261e-06,
+      "loss": 0.4299,
+      "step": 2360
+    },
+    {
+      "epoch": 103.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5060404539108276,
+      "eval_runtime": 3.7179,
+      "eval_samples_per_second": 68.318,
+      "eval_steps_per_second": 0.538,
+      "step": 2369
+    },
+    {
+      "epoch": 103.04347826086956,
+      "grad_norm": 1.847158670425415,
+      "learning_rate": 9.695652173913043e-06,
+      "loss": 0.3834,
+      "step": 2370
+    },
+    {
+      "epoch": 103.47826086956522,
+      "grad_norm": 4.114128112792969,
+      "learning_rate": 9.652173913043478e-06,
+      "loss": 0.4061,
+      "step": 2380
+    },
+    {
+      "epoch": 103.91304347826087,
+      "grad_norm": 5.080406665802002,
+      "learning_rate": 9.608695652173914e-06,
+      "loss": 0.4148,
+      "step": 2390
+    },
+    {
+      "epoch": 104.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5038026571273804,
+      "eval_runtime": 3.7535,
+      "eval_samples_per_second": 67.671,
+      "eval_steps_per_second": 0.533,
+      "step": 2392
+    },
+    {
+      "epoch": 104.34782608695652,
+      "grad_norm": 3.291896104812622,
+      "learning_rate": 9.565217391304349e-06,
+      "loss": 0.4272,
+      "step": 2400
+    },
+    {
+      "epoch": 104.78260869565217,
+      "grad_norm": 2.7959041595458984,
+      "learning_rate": 9.521739130434784e-06,
+      "loss": 0.4179,
+      "step": 2410
+    },
+    {
+      "epoch": 105.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5064316391944885,
+      "eval_runtime": 4.8627,
+      "eval_samples_per_second": 52.235,
+      "eval_steps_per_second": 0.411,
+      "step": 2415
+    },
+    {
+      "epoch": 105.21739130434783,
+      "grad_norm": 5.880518913269043,
+      "learning_rate": 9.478260869565217e-06,
+      "loss": 0.4155,
+      "step": 2420
+    },
+    {
+      "epoch": 105.65217391304348,
+      "grad_norm": 2.2435200214385986,
+      "learning_rate": 9.434782608695652e-06,
+      "loss": 0.4352,
+      "step": 2430
+    },
+    {
+      "epoch": 106.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5059410929679871,
+      "eval_runtime": 3.7149,
+      "eval_samples_per_second": 68.373,
+      "eval_steps_per_second": 0.538,
+      "step": 2438
+    },
+    {
+      "epoch": 106.08695652173913,
+      "grad_norm": 3.1865811347961426,
+      "learning_rate": 9.391304347826087e-06,
+      "loss": 0.3997,
+      "step": 2440
+    },
+    {
+      "epoch": 106.52173913043478,
+      "grad_norm": 4.0479936599731445,
+      "learning_rate": 9.347826086956523e-06,
+      "loss": 0.401,
+      "step": 2450
+    },
+    {
+      "epoch": 106.95652173913044,
+      "grad_norm": 2.87663197517395,
+      "learning_rate": 9.304347826086956e-06,
+      "loss": 0.4027,
+      "step": 2460
+    },
+    {
+      "epoch": 107.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5025486350059509,
+      "eval_runtime": 3.7614,
+      "eval_samples_per_second": 67.528,
+      "eval_steps_per_second": 0.532,
+      "step": 2461
+    },
+    {
+      "epoch": 107.3913043478261,
+      "grad_norm": 2.630986452102661,
+      "learning_rate": 9.260869565217391e-06,
+      "loss": 0.3828,
+      "step": 2470
+    },
+    {
+      "epoch": 107.82608695652173,
+      "grad_norm": 2.9700822830200195,
+      "learning_rate": 9.217391304347826e-06,
+      "loss": 0.4002,
+      "step": 2480
+    },
+    {
+      "epoch": 108.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5020495653152466,
+      "eval_runtime": 4.6331,
+      "eval_samples_per_second": 54.823,
+      "eval_steps_per_second": 0.432,
+      "step": 2484
+    },
+    {
+      "epoch": 108.26086956521739,
+      "grad_norm": 4.361221790313721,
+      "learning_rate": 9.173913043478261e-06,
+      "loss": 0.405,
+      "step": 2490
+    },
+    {
+      "epoch": 108.69565217391305,
+      "grad_norm": 2.9328296184539795,
+      "learning_rate": 9.130434782608697e-06,
+      "loss": 0.3988,
+      "step": 2500
+    },
+    {
+      "epoch": 109.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5063354969024658,
+      "eval_runtime": 3.8012,
+      "eval_samples_per_second": 66.821,
+      "eval_steps_per_second": 0.526,
+      "step": 2507
+    },
+    {
+      "epoch": 109.1304347826087,
+      "grad_norm": 2.3236513137817383,
+      "learning_rate": 9.086956521739132e-06,
+      "loss": 0.3894,
+      "step": 2510
+    },
+    {
+      "epoch": 109.56521739130434,
+      "grad_norm": 3.4379804134368896,
+      "learning_rate": 9.043478260869565e-06,
+      "loss": 0.4023,
+      "step": 2520
+    },
+    {
+      "epoch": 110.0,
+      "grad_norm": 4.300137042999268,
+      "learning_rate": 9e-06,
+      "loss": 0.4095,
+      "step": 2530
+    },
+    {
+      "epoch": 110.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5034452676773071,
+      "eval_runtime": 3.7021,
+      "eval_samples_per_second": 68.61,
+      "eval_steps_per_second": 0.54,
+      "step": 2530
+    },
+    {
+      "epoch": 110.43478260869566,
+      "grad_norm": 2.190524101257324,
+      "learning_rate": 8.956521739130435e-06,
+      "loss": 0.4072,
+      "step": 2540
+    },
+    {
+      "epoch": 110.8695652173913,
+      "grad_norm": 2.2291879653930664,
+      "learning_rate": 8.91304347826087e-06,
+      "loss": 0.4001,
+      "step": 2550
+    },
+    {
+      "epoch": 111.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.505436360836029,
+      "eval_runtime": 4.2919,
+      "eval_samples_per_second": 59.182,
+      "eval_steps_per_second": 0.466,
+      "step": 2553
+    },
+    {
+      "epoch": 111.30434782608695,
+      "grad_norm": 3.1182541847229004,
+      "learning_rate": 8.869565217391306e-06,
+      "loss": 0.3904,
+      "step": 2560
+    },
+    {
+      "epoch": 111.73913043478261,
+      "grad_norm": 3.8375625610351562,
+      "learning_rate": 8.82608695652174e-06,
+      "loss": 0.4201,
+      "step": 2570
+    },
+    {
+      "epoch": 112.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5076125860214233,
+      "eval_runtime": 4.2691,
+      "eval_samples_per_second": 59.497,
+      "eval_steps_per_second": 0.468,
+      "step": 2576
+    },
+    {
+      "epoch": 112.17391304347827,
+      "grad_norm": 2.4231808185577393,
+      "learning_rate": 8.782608695652174e-06,
+      "loss": 0.3925,
+      "step": 2580
+    },
+    {
+      "epoch": 112.6086956521739,
+      "grad_norm": 4.854309558868408,
+      "learning_rate": 8.73913043478261e-06,
+      "loss": 0.4134,
+      "step": 2590
+    },
+    {
+      "epoch": 113.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5069688558578491,
+      "eval_runtime": 3.7367,
+      "eval_samples_per_second": 67.974,
+      "eval_steps_per_second": 0.535,
+      "step": 2599
+    },
+    {
+      "epoch": 113.04347826086956,
+      "grad_norm": 4.327704429626465,
+      "learning_rate": 8.695652173913044e-06,
+      "loss": 0.3959,
+      "step": 2600
+    },
+    {
+      "epoch": 113.47826086956522,
+      "grad_norm": 2.8718910217285156,
+      "learning_rate": 8.65217391304348e-06,
+      "loss": 0.3806,
+      "step": 2610
+    },
+    {
+      "epoch": 113.91304347826087,
+      "grad_norm": 5.400497913360596,
+      "learning_rate": 8.608695652173915e-06,
+      "loss": 0.3614,
+      "step": 2620
+    },
+    {
+      "epoch": 114.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5032684206962585,
+      "eval_runtime": 4.4717,
+      "eval_samples_per_second": 56.802,
+      "eval_steps_per_second": 0.447,
+      "step": 2622
+    },
+    {
+      "epoch": 114.34782608695652,
+      "grad_norm": 2.7276597023010254,
+      "learning_rate": 8.56521739130435e-06,
+      "loss": 0.3956,
+      "step": 2630
+    },
+    {
+      "epoch": 114.78260869565217,
+      "grad_norm": 3.339860200881958,
+      "learning_rate": 8.521739130434783e-06,
+      "loss": 0.3928,
+      "step": 2640
+    },
+    {
+      "epoch": 115.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5042973160743713,
+      "eval_runtime": 4.1216,
+      "eval_samples_per_second": 61.627,
+      "eval_steps_per_second": 0.485,
+      "step": 2645
+    },
+    {
+      "epoch": 115.21739130434783,
+      "grad_norm": 2.435579538345337,
+      "learning_rate": 8.478260869565218e-06,
+      "loss": 0.4149,
+      "step": 2650
+    },
+    {
+      "epoch": 115.65217391304348,
+      "grad_norm": 3.9001612663269043,
+      "learning_rate": 8.434782608695653e-06,
+      "loss": 0.435,
+      "step": 2660
+    },
+    {
+      "epoch": 116.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.4998602271080017,
+      "eval_runtime": 3.7703,
+      "eval_samples_per_second": 67.369,
+      "eval_steps_per_second": 0.53,
+      "step": 2668
+    },
+    {
+      "epoch": 116.08695652173913,
+      "grad_norm": 4.031954288482666,
+      "learning_rate": 8.391304347826089e-06,
+      "loss": 0.3575,
+      "step": 2670
+    },
+    {
+      "epoch": 116.52173913043478,
+      "grad_norm": 3.1172120571136475,
+      "learning_rate": 8.347826086956522e-06,
+      "loss": 0.4062,
+      "step": 2680
+    },
+    {
+      "epoch": 116.95652173913044,
+      "grad_norm": 2.6061761379241943,
+      "learning_rate": 8.304347826086957e-06,
+      "loss": 0.4162,
+      "step": 2690
+    },
+    {
+      "epoch": 117.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5132189393043518,
+      "eval_runtime": 4.4279,
+      "eval_samples_per_second": 57.363,
+      "eval_steps_per_second": 0.452,
+      "step": 2691
+    },
+    {
+      "epoch": 117.3913043478261,
+      "grad_norm": 1.7457960844039917,
+      "learning_rate": 8.260869565217392e-06,
+      "loss": 0.3887,
+      "step": 2700
+    },
+    {
+      "epoch": 117.82608695652173,
+      "grad_norm": 5.013397216796875,
+      "learning_rate": 8.217391304347827e-06,
+      "loss": 0.4078,
+      "step": 2710
+    },
+    {
+      "epoch": 118.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.5088200569152832,
+      "eval_runtime": 4.1897,
+      "eval_samples_per_second": 60.625,
+      "eval_steps_per_second": 0.477,
+      "step": 2714
+    },
+    {
+      "epoch": 118.26086956521739,
+      "grad_norm": 3.4758872985839844,
+      "learning_rate": 8.173913043478263e-06,
+      "loss": 0.4251,
+      "step": 2720
+    },
+    {
+      "epoch": 118.69565217391305,
+      "grad_norm": 1.8225319385528564,
+      "learning_rate": 8.130434782608696e-06,
+      "loss": 0.4025,
+      "step": 2730
+    },
+    {
+      "epoch": 119.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.507527768611908,
+      "eval_runtime": 3.721,
+      "eval_samples_per_second": 68.261,
+      "eval_steps_per_second": 0.537,
+      "step": 2737
+    },
+    {
+      "epoch": 119.1304347826087,
+      "grad_norm": 4.636626720428467,
+      "learning_rate": 8.086956521739131e-06,
+      "loss": 0.4024,
+      "step": 2740
+    },
+    {
+      "epoch": 119.56521739130434,
+      "grad_norm": 2.249758720397949,
+      "learning_rate": 8.043478260869566e-06,
+      "loss": 0.3917,
+      "step": 2750
+    },
+    {
+      "epoch": 120.0,
+      "grad_norm": 6.408204555511475,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.4096,
+      "step": 2760
+    },
+    {
+      "epoch": 120.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.502310574054718,
+      "eval_runtime": 4.469,
+      "eval_samples_per_second": 56.836,
+      "eval_steps_per_second": 0.448,
+      "step": 2760
+    },
+    {
+      "epoch": 120.43478260869566,
+      "grad_norm": 2.495302200317383,
+      "learning_rate": 7.956521739130435e-06,
+      "loss": 0.3791,
+      "step": 2770
+    },
+    {
+      "epoch": 120.8695652173913,
+      "grad_norm": 2.840449571609497,
+      "learning_rate": 7.91304347826087e-06,
+      "loss": 0.3879,
+      "step": 2780
+    },
+    {
+      "epoch": 121.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5062641501426697,
+      "eval_runtime": 3.9315,
+      "eval_samples_per_second": 64.606,
+      "eval_steps_per_second": 0.509,
+      "step": 2783
+    },
+    {
+      "epoch": 121.30434782608695,
+      "grad_norm": 4.82555627822876,
+      "learning_rate": 7.869565217391305e-06,
+      "loss": 0.4232,
+      "step": 2790
+    },
+    {
+      "epoch": 121.73913043478261,
+      "grad_norm": 3.220736503601074,
+      "learning_rate": 7.82608695652174e-06,
+      "loss": 0.4033,
+      "step": 2800
+    },
+    {
+      "epoch": 122.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5001329183578491,
+      "eval_runtime": 3.7903,
+      "eval_samples_per_second": 67.012,
+      "eval_steps_per_second": 0.528,
+      "step": 2806
+    },
+    {
+      "epoch": 122.17391304347827,
+      "grad_norm": 4.516547203063965,
+      "learning_rate": 7.782608695652174e-06,
+      "loss": 0.4144,
+      "step": 2810
+    },
+    {
+      "epoch": 122.6086956521739,
+      "grad_norm": 2.559272289276123,
+      "learning_rate": 7.739130434782609e-06,
+      "loss": 0.3927,
+      "step": 2820
+    },
+    {
+      "epoch": 123.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.5087068676948547,
+      "eval_runtime": 4.404,
+      "eval_samples_per_second": 57.675,
+      "eval_steps_per_second": 0.454,
+      "step": 2829
+    },
+    {
+      "epoch": 123.04347826086956,
+      "grad_norm": 3.344332695007324,
+      "learning_rate": 7.695652173913044e-06,
+      "loss": 0.4016,
+      "step": 2830
+    },
+    {
+      "epoch": 123.47826086956522,
+      "grad_norm": 2.610856533050537,
+      "learning_rate": 7.652173913043479e-06,
+      "loss": 0.3925,
+      "step": 2840
+    },
+    {
+      "epoch": 123.91304347826087,
+      "grad_norm": 3.501596689224243,
+      "learning_rate": 7.608695652173914e-06,
+      "loss": 0.3803,
+      "step": 2850
+    },
+    {
+      "epoch": 124.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5149940848350525,
+      "eval_runtime": 4.2503,
+      "eval_samples_per_second": 59.76,
+      "eval_steps_per_second": 0.471,
+      "step": 2852
+    },
+    {
+      "epoch": 124.34782608695652,
+      "grad_norm": 4.040353298187256,
+      "learning_rate": 7.565217391304348e-06,
+      "loss": 0.4101,
+      "step": 2860
+    },
+    {
+      "epoch": 124.78260869565217,
+      "grad_norm": 3.1806752681732178,
+      "learning_rate": 7.5217391304347835e-06,
+      "loss": 0.4248,
+      "step": 2870
+    },
+    {
+      "epoch": 125.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.515027642250061,
+      "eval_runtime": 3.7006,
+      "eval_samples_per_second": 68.638,
+      "eval_steps_per_second": 0.54,
+      "step": 2875
+    },
+    {
+      "epoch": 125.21739130434783,
+      "grad_norm": 2.976123332977295,
+      "learning_rate": 7.478260869565218e-06,
+      "loss": 0.3806,
+      "step": 2880
+    },
+    {
+      "epoch": 125.65217391304348,
+      "grad_norm": 4.0399250984191895,
+      "learning_rate": 7.434782608695653e-06,
+      "loss": 0.3874,
+      "step": 2890
+    },
+    {
+      "epoch": 126.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5157892107963562,
+      "eval_runtime": 3.8292,
+      "eval_samples_per_second": 66.332,
+      "eval_steps_per_second": 0.522,
+      "step": 2898
+    },
+    {
+      "epoch": 126.08695652173913,
+      "grad_norm": 2.8186984062194824,
+      "learning_rate": 7.391304347826087e-06,
+      "loss": 0.4068,
+      "step": 2900
+    },
+    {
+      "epoch": 126.52173913043478,
+      "grad_norm": 1.7811031341552734,
+      "learning_rate": 7.347826086956522e-06,
+      "loss": 0.4188,
+      "step": 2910
+    },
+    {
+      "epoch": 126.95652173913044,
+      "grad_norm": 2.591479539871216,
+      "learning_rate": 7.304347826086957e-06,
+      "loss": 0.3646,
+      "step": 2920
+    },
+    {
+      "epoch": 127.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.4979710578918457,
+      "eval_runtime": 4.9476,
+      "eval_samples_per_second": 51.338,
+      "eval_steps_per_second": 0.404,
+      "step": 2921
+    },
+    {
+      "epoch": 127.3913043478261,
+      "grad_norm": 3.097064733505249,
+      "learning_rate": 7.2608695652173925e-06,
+      "loss": 0.3809,
+      "step": 2930
+    },
+    {
+      "epoch": 127.82608695652173,
+      "grad_norm": 4.5358805656433105,
+      "learning_rate": 7.217391304347827e-06,
+      "loss": 0.4115,
+      "step": 2940
+    },
+    {
+      "epoch": 128.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.507692813873291,
+      "eval_runtime": 3.7086,
+      "eval_samples_per_second": 68.489,
+      "eval_steps_per_second": 0.539,
+      "step": 2944
+    },
+    {
+      "epoch": 128.2608695652174,
+      "grad_norm": 4.192093372344971,
+      "learning_rate": 7.173913043478261e-06,
+      "loss": 0.3931,
+      "step": 2950
+    },
+    {
+      "epoch": 128.69565217391303,
+      "grad_norm": 2.4763779640197754,
+      "learning_rate": 7.130434782608696e-06,
+      "loss": 0.385,
+      "step": 2960
+    },
+    {
+      "epoch": 129.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5153175592422485,
+      "eval_runtime": 3.726,
+      "eval_samples_per_second": 68.17,
+      "eval_steps_per_second": 0.537,
+      "step": 2967
+    },
+    {
+      "epoch": 129.1304347826087,
+      "grad_norm": 2.906510353088379,
+      "learning_rate": 7.086956521739131e-06,
+      "loss": 0.4009,
+      "step": 2970
+    },
+    {
+      "epoch": 129.56521739130434,
+      "grad_norm": 5.497567653656006,
+      "learning_rate": 7.0434782608695665e-06,
+      "loss": 0.4091,
+      "step": 2980
+    },
+    {
+      "epoch": 130.0,
+      "grad_norm": 4.277368068695068,
+      "learning_rate": 7e-06,
+      "loss": 0.4064,
+      "step": 2990
+    },
+    {
+      "epoch": 130.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.511443018913269,
+      "eval_runtime": 5.0723,
+      "eval_samples_per_second": 50.076,
+      "eval_steps_per_second": 0.394,
+      "step": 2990
+    },
+    {
+      "epoch": 130.43478260869566,
+      "grad_norm": 2.3368613719940186,
+      "learning_rate": 6.956521739130435e-06,
+      "loss": 0.3762,
+      "step": 3000
+    },
+    {
+      "epoch": 130.8695652173913,
+      "grad_norm": 2.983280897140503,
+      "learning_rate": 6.91304347826087e-06,
+      "loss": 0.4168,
+      "step": 3010
+    },
+    {
+      "epoch": 131.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5056832432746887,
+      "eval_runtime": 3.7472,
+      "eval_samples_per_second": 67.785,
+      "eval_steps_per_second": 0.534,
+      "step": 3013
+    },
+    {
+      "epoch": 131.30434782608697,
+      "grad_norm": 5.6472978591918945,
+      "learning_rate": 6.869565217391305e-06,
+      "loss": 0.3454,
+      "step": 3020
+    },
+    {
+      "epoch": 131.7391304347826,
+      "grad_norm": 2.710934638977051,
+      "learning_rate": 6.8260869565217395e-06,
+      "loss": 0.4319,
+      "step": 3030
+    },
+    {
+      "epoch": 132.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5041180849075317,
+      "eval_runtime": 3.7165,
+      "eval_samples_per_second": 68.344,
+      "eval_steps_per_second": 0.538,
+      "step": 3036
+    },
+    {
+      "epoch": 132.17391304347825,
+      "grad_norm": 2.8998305797576904,
+      "learning_rate": 6.782608695652174e-06,
+      "loss": 0.3769,
+      "step": 3040
+    },
+    {
+      "epoch": 132.6086956521739,
+      "grad_norm": 3.503068208694458,
+      "learning_rate": 6.739130434782609e-06,
+      "loss": 0.4234,
+      "step": 3050
+    },
+    {
+      "epoch": 133.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5119389891624451,
+      "eval_runtime": 4.8197,
+      "eval_samples_per_second": 52.701,
+      "eval_steps_per_second": 0.415,
+      "step": 3059
+    },
+    {
+      "epoch": 133.04347826086956,
+      "grad_norm": 2.628817319869995,
+      "learning_rate": 6.695652173913044e-06,
+      "loss": 0.3984,
+      "step": 3060
+    },
+    {
+      "epoch": 133.47826086956522,
+      "grad_norm": 3.1060750484466553,
+      "learning_rate": 6.652173913043479e-06,
+      "loss": 0.4147,
+      "step": 3070
+    },
+    {
+      "epoch": 133.91304347826087,
+      "grad_norm": 6.7668328285217285,
+      "learning_rate": 6.6086956521739135e-06,
+      "loss": 0.3721,
+      "step": 3080
+    },
+    {
+      "epoch": 134.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.51175457239151,
+      "eval_runtime": 3.7909,
+      "eval_samples_per_second": 67.003,
+      "eval_steps_per_second": 0.528,
+      "step": 3082
+    },
+    {
+      "epoch": 134.34782608695653,
+      "grad_norm": 6.763729572296143,
+      "learning_rate": 6.565217391304349e-06,
+      "loss": 0.386,
+      "step": 3090
+    },
+    {
+      "epoch": 134.7826086956522,
+      "grad_norm": 4.876804828643799,
+      "learning_rate": 6.521739130434783e-06,
+      "loss": 0.3709,
+      "step": 3100
+    },
+    {
+      "epoch": 135.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5078221559524536,
+      "eval_runtime": 3.7684,
+      "eval_samples_per_second": 67.402,
+      "eval_steps_per_second": 0.531,
+      "step": 3105
+    },
+    {
+      "epoch": 135.2173913043478,
+      "grad_norm": 3.7445313930511475,
+      "learning_rate": 6.478260869565218e-06,
+      "loss": 0.3592,
+      "step": 3110
+    },
+    {
+      "epoch": 135.65217391304347,
+      "grad_norm": 5.715231418609619,
+      "learning_rate": 6.434782608695652e-06,
+      "loss": 0.4149,
+      "step": 3120
+    },
+    {
+      "epoch": 136.0,
+      "eval_accuracy": 0.7795275590551181,
+      "eval_loss": 0.5163589715957642,
+      "eval_runtime": 4.6746,
+      "eval_samples_per_second": 54.336,
+      "eval_steps_per_second": 0.428,
+      "step": 3128
+    },
+    {
+      "epoch": 136.08695652173913,
+      "grad_norm": 3.3850629329681396,
+      "learning_rate": 6.391304347826087e-06,
+      "loss": 0.3681,
+      "step": 3130
+    },
+    {
+      "epoch": 136.52173913043478,
+      "grad_norm": 5.502380847930908,
+      "learning_rate": 6.3478260869565225e-06,
+      "loss": 0.3629,
+      "step": 3140
+    },
+    {
+      "epoch": 136.95652173913044,
+      "grad_norm": 4.158088684082031,
+      "learning_rate": 6.304347826086958e-06,
+      "loss": 0.416,
+      "step": 3150
+    },
+    {
+      "epoch": 137.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5123007297515869,
+      "eval_runtime": 3.789,
+      "eval_samples_per_second": 67.036,
+      "eval_steps_per_second": 0.528,
+      "step": 3151
+    },
+    {
+      "epoch": 137.3913043478261,
+      "grad_norm": 2.241478681564331,
+      "learning_rate": 6.260869565217392e-06,
+      "loss": 0.4089,
+      "step": 3160
+    },
+    {
+      "epoch": 137.82608695652175,
+      "grad_norm": 4.336514472961426,
+      "learning_rate": 6.217391304347826e-06,
+      "loss": 0.406,
+      "step": 3170
+    },
+    {
+      "epoch": 138.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5115824937820435,
+      "eval_runtime": 3.7195,
+      "eval_samples_per_second": 68.288,
+      "eval_steps_per_second": 0.538,
+      "step": 3174
+    },
+    {
+      "epoch": 138.2608695652174,
+      "grad_norm": 2.154179334640503,
+      "learning_rate": 6.173913043478261e-06,
+      "loss": 0.4018,
+      "step": 3180
+    },
+    {
+      "epoch": 138.69565217391303,
+      "grad_norm": 3.2215845584869385,
+      "learning_rate": 6.1304347826086965e-06,
+      "loss": 0.3613,
+      "step": 3190
+    },
+    {
+      "epoch": 139.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5169662237167358,
+      "eval_runtime": 4.4593,
+      "eval_samples_per_second": 56.959,
+      "eval_steps_per_second": 0.448,
+      "step": 3197
+    },
+    {
+      "epoch": 139.1304347826087,
+      "grad_norm": 2.800915241241455,
+      "learning_rate": 6.086956521739132e-06,
+      "loss": 0.3863,
+      "step": 3200
+    },
+    {
+      "epoch": 139.56521739130434,
+      "grad_norm": 7.433578014373779,
+      "learning_rate": 6.043478260869565e-06,
+      "loss": 0.4278,
+      "step": 3210
+    },
+    {
+      "epoch": 140.0,
+      "grad_norm": 3.887300968170166,
+      "learning_rate": 6e-06,
+      "loss": 0.3786,
+      "step": 3220
+    },
+    {
+      "epoch": 140.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5098868608474731,
+      "eval_runtime": 4.1343,
+      "eval_samples_per_second": 61.437,
+      "eval_steps_per_second": 0.484,
+      "step": 3220
+    },
+    {
+      "epoch": 140.43478260869566,
+      "grad_norm": 3.3379013538360596,
+      "learning_rate": 5.956521739130435e-06,
+      "loss": 0.405,
+      "step": 3230
+    },
+    {
+      "epoch": 140.8695652173913,
+      "grad_norm": 3.2763419151306152,
+      "learning_rate": 5.91304347826087e-06,
+      "loss": 0.3976,
+      "step": 3240
+    },
+    {
+      "epoch": 141.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5111474394798279,
+      "eval_runtime": 3.8029,
+      "eval_samples_per_second": 66.792,
+      "eval_steps_per_second": 0.526,
+      "step": 3243
+    },
+    {
+      "epoch": 141.30434782608697,
+      "grad_norm": 3.1908023357391357,
+      "learning_rate": 5.8695652173913055e-06,
+      "loss": 0.3856,
+      "step": 3250
+    },
+    {
+      "epoch": 141.7391304347826,
+      "grad_norm": 3.875778913497925,
+      "learning_rate": 5.826086956521739e-06,
+      "loss": 0.371,
+      "step": 3260
+    },
+    {
+      "epoch": 142.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5081124901771545,
+      "eval_runtime": 4.5605,
+      "eval_samples_per_second": 55.696,
+      "eval_steps_per_second": 0.439,
+      "step": 3266
+    },
+    {
+      "epoch": 142.17391304347825,
+      "grad_norm": 2.925506353378296,
+      "learning_rate": 5.782608695652174e-06,
+      "loss": 0.4169,
+      "step": 3270
+    },
+    {
+      "epoch": 142.6086956521739,
+      "grad_norm": 9.266388893127441,
+      "learning_rate": 5.739130434782609e-06,
+      "loss": 0.4056,
+      "step": 3280
+    },
+    {
+      "epoch": 143.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5098369717597961,
+      "eval_runtime": 3.9928,
+      "eval_samples_per_second": 63.615,
+      "eval_steps_per_second": 0.501,
+      "step": 3289
+    },
+    {
+      "epoch": 143.04347826086956,
+      "grad_norm": 25.856365203857422,
+      "learning_rate": 5.695652173913044e-06,
+      "loss": 0.3757,
+      "step": 3290
+    },
+    {
+      "epoch": 143.47826086956522,
+      "grad_norm": 2.937258720397949,
+      "learning_rate": 5.652173913043479e-06,
+      "loss": 0.3745,
+      "step": 3300
+    },
+    {
+      "epoch": 143.91304347826087,
+      "grad_norm": 3.236806631088257,
+      "learning_rate": 5.608695652173914e-06,
+      "loss": 0.4214,
+      "step": 3310
+    },
+    {
+      "epoch": 144.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5085259675979614,
+      "eval_runtime": 3.7621,
+      "eval_samples_per_second": 67.516,
+      "eval_steps_per_second": 0.532,
+      "step": 3312
+    },
+    {
+      "epoch": 144.34782608695653,
+      "grad_norm": 3.6454241275787354,
+      "learning_rate": 5.565217391304348e-06,
+      "loss": 0.3659,
+      "step": 3320
+    },
+    {
+      "epoch": 144.7826086956522,
+      "grad_norm": 3.4510464668273926,
+      "learning_rate": 5.521739130434783e-06,
+      "loss": 0.3832,
+      "step": 3330
+    },
+    {
+      "epoch": 145.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5084368586540222,
+      "eval_runtime": 4.2307,
+      "eval_samples_per_second": 60.037,
+      "eval_steps_per_second": 0.473,
+      "step": 3335
+    },
+    {
+      "epoch": 145.2173913043478,
+      "grad_norm": 2.4478542804718018,
+      "learning_rate": 5.478260869565217e-06,
+      "loss": 0.3522,
+      "step": 3340
+    },
+    {
+      "epoch": 145.65217391304347,
+      "grad_norm": 4.097745895385742,
+      "learning_rate": 5.4347826086956525e-06,
+      "loss": 0.3762,
+      "step": 3350
+    },
+    {
+      "epoch": 146.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5060733556747437,
+      "eval_runtime": 4.203,
+      "eval_samples_per_second": 60.433,
+      "eval_steps_per_second": 0.476,
+      "step": 3358
+    },
+    {
+      "epoch": 146.08695652173913,
+      "grad_norm": 3.384960651397705,
+      "learning_rate": 5.391304347826088e-06,
+      "loss": 0.3761,
+      "step": 3360
+    },
+    {
+      "epoch": 146.52173913043478,
+      "grad_norm": 2.909395217895508,
+      "learning_rate": 5.347826086956523e-06,
+      "loss": 0.3902,
+      "step": 3370
+    },
+    {
+      "epoch": 146.95652173913044,
+      "grad_norm": 2.538163900375366,
+      "learning_rate": 5.304347826086957e-06,
+      "loss": 0.4118,
+      "step": 3380
+    },
+    {
+      "epoch": 147.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5111083984375,
+      "eval_runtime": 3.8334,
+      "eval_samples_per_second": 66.26,
+      "eval_steps_per_second": 0.522,
+      "step": 3381
+    },
+    {
+      "epoch": 147.3913043478261,
+      "grad_norm": 2.9644970893859863,
+      "learning_rate": 5.260869565217391e-06,
+      "loss": 0.3802,
+      "step": 3390
+    },
+    {
+      "epoch": 147.82608695652175,
+      "grad_norm": 3.0972464084625244,
+      "learning_rate": 5.2173913043478265e-06,
+      "loss": 0.3866,
+      "step": 3400
+    },
+    {
+      "epoch": 148.0,
+      "eval_accuracy": 0.8070866141732284,
+      "eval_loss": 0.5092455148696899,
+      "eval_runtime": 5.6174,
+      "eval_samples_per_second": 45.216,
+      "eval_steps_per_second": 0.356,
+      "step": 3404
+    },
+    {
+      "epoch": 148.2608695652174,
+      "grad_norm": 1.625214695930481,
+      "learning_rate": 5.173913043478262e-06,
+      "loss": 0.3584,
+      "step": 3410
+    },
+    {
+      "epoch": 148.69565217391303,
+      "grad_norm": 15.01403522491455,
+      "learning_rate": 5.130434782608697e-06,
+      "loss": 0.3869,
+      "step": 3420
+    },
+    {
+      "epoch": 149.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.512187659740448,
+      "eval_runtime": 3.8515,
+      "eval_samples_per_second": 65.948,
+      "eval_steps_per_second": 0.519,
+      "step": 3427
+    },
+    {
+      "epoch": 149.1304347826087,
+      "grad_norm": 1.9776344299316406,
+      "learning_rate": 5.08695652173913e-06,
+      "loss": 0.3921,
+      "step": 3430
+    },
+    {
+      "epoch": 149.56521739130434,
+      "grad_norm": 2.336129665374756,
+      "learning_rate": 5.043478260869565e-06,
+      "loss": 0.4048,
+      "step": 3440
+    },
+    {
+      "epoch": 150.0,
+      "grad_norm": 3.6398816108703613,
+      "learning_rate": 5e-06,
+      "loss": 0.3734,
+      "step": 3450
+    },
+    {
+      "epoch": 150.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5116916298866272,
+      "eval_runtime": 3.7475,
+      "eval_samples_per_second": 67.779,
+      "eval_steps_per_second": 0.534,
+      "step": 3450
+    },
+    {
+      "epoch": 150.43478260869566,
+      "grad_norm": 2.299021005630493,
+      "learning_rate": 4.9565217391304355e-06,
+      "loss": 0.3734,
+      "step": 3460
+    },
+    {
+      "epoch": 150.8695652173913,
+      "grad_norm": 3.107494831085205,
+      "learning_rate": 4.91304347826087e-06,
+      "loss": 0.4061,
+      "step": 3470
+    },
+    {
+      "epoch": 151.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5094764232635498,
+      "eval_runtime": 4.4075,
+      "eval_samples_per_second": 57.629,
+      "eval_steps_per_second": 0.454,
+      "step": 3473
+    },
+    {
+      "epoch": 151.30434782608697,
+      "grad_norm": 2.319066286087036,
+      "learning_rate": 4.869565217391305e-06,
+      "loss": 0.3681,
+      "step": 3480
+    },
+    {
+      "epoch": 151.7391304347826,
+      "grad_norm": 2.7603538036346436,
+      "learning_rate": 4.826086956521739e-06,
+      "loss": 0.3705,
+      "step": 3490
+    },
+    {
+      "epoch": 152.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5171404480934143,
+      "eval_runtime": 4.362,
+      "eval_samples_per_second": 58.23,
+      "eval_steps_per_second": 0.459,
+      "step": 3496
+    },
+    {
+      "epoch": 152.17391304347825,
+      "grad_norm": 2.0375826358795166,
+      "learning_rate": 4.782608695652174e-06,
+      "loss": 0.3882,
+      "step": 3500
+    },
+    {
+      "epoch": 152.6086956521739,
+      "grad_norm": 2.8498833179473877,
+      "learning_rate": 4.739130434782609e-06,
+      "loss": 0.3873,
+      "step": 3510
+    },
+    {
+      "epoch": 153.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5179200768470764,
+      "eval_runtime": 3.7588,
+      "eval_samples_per_second": 67.575,
+      "eval_steps_per_second": 0.532,
+      "step": 3519
+    },
+    {
+      "epoch": 153.04347826086956,
+      "grad_norm": 2.707977533340454,
+      "learning_rate": 4.695652173913044e-06,
+      "loss": 0.3979,
+      "step": 3520
+    },
+    {
+      "epoch": 153.47826086956522,
+      "grad_norm": 3.5183486938476562,
+      "learning_rate": 4.652173913043478e-06,
+      "loss": 0.4025,
+      "step": 3530
+    },
+    {
+      "epoch": 153.91304347826087,
+      "grad_norm": 2.90291166305542,
+      "learning_rate": 4.608695652173913e-06,
+      "loss": 0.3927,
+      "step": 3540
+    },
+    {
+      "epoch": 154.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5117496252059937,
+      "eval_runtime": 3.7541,
+      "eval_samples_per_second": 67.659,
+      "eval_steps_per_second": 0.533,
+      "step": 3542
+    },
+    {
+      "epoch": 154.34782608695653,
+      "grad_norm": 4.005958080291748,
+      "learning_rate": 4.565217391304348e-06,
+      "loss": 0.4011,
+      "step": 3550
+    },
+    {
+      "epoch": 154.7826086956522,
+      "grad_norm": 2.469202995300293,
+      "learning_rate": 4.5217391304347826e-06,
+      "loss": 0.3807,
+      "step": 3560
+    },
+    {
+      "epoch": 155.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5133464932441711,
+      "eval_runtime": 5.8154,
+      "eval_samples_per_second": 43.677,
+      "eval_steps_per_second": 0.344,
+      "step": 3565
+    },
+    {
+      "epoch": 155.2173913043478,
+      "grad_norm": 3.2248237133026123,
+      "learning_rate": 4.478260869565218e-06,
+      "loss": 0.4498,
+      "step": 3570
+    },
+    {
+      "epoch": 155.65217391304347,
+      "grad_norm": 3.463270425796509,
+      "learning_rate": 4.434782608695653e-06,
+      "loss": 0.3761,
+      "step": 3580
+    },
+    {
+      "epoch": 156.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5140319466590881,
+      "eval_runtime": 3.7668,
+      "eval_samples_per_second": 67.432,
+      "eval_steps_per_second": 0.531,
+      "step": 3588
+    },
+    {
+      "epoch": 156.08695652173913,
+      "grad_norm": 3.640611171722412,
+      "learning_rate": 4.391304347826087e-06,
+      "loss": 0.3609,
+      "step": 3590
+    },
+    {
+      "epoch": 156.52173913043478,
+      "grad_norm": 4.198793888092041,
+      "learning_rate": 4.347826086956522e-06,
+      "loss": 0.3984,
+      "step": 3600
+    },
+    {
+      "epoch": 156.95652173913044,
+      "grad_norm": 2.9035775661468506,
+      "learning_rate": 4.304347826086957e-06,
+      "loss": 0.3964,
+      "step": 3610
+    },
+    {
+      "epoch": 157.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5117691159248352,
+      "eval_runtime": 3.7832,
+      "eval_samples_per_second": 67.138,
+      "eval_steps_per_second": 0.529,
+      "step": 3611
+    },
+    {
+      "epoch": 157.3913043478261,
+      "grad_norm": 5.13762092590332,
+      "learning_rate": 4.260869565217392e-06,
+      "loss": 0.3818,
+      "step": 3620
+    },
+    {
+      "epoch": 157.82608695652175,
+      "grad_norm": 8.948963165283203,
+      "learning_rate": 4.217391304347827e-06,
+      "loss": 0.39,
+      "step": 3630
+    },
+    {
+      "epoch": 158.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5122236609458923,
+      "eval_runtime": 4.6309,
+      "eval_samples_per_second": 54.849,
+      "eval_steps_per_second": 0.432,
+      "step": 3634
+    },
+    {
+      "epoch": 158.2608695652174,
+      "grad_norm": 2.4759654998779297,
+      "learning_rate": 4.173913043478261e-06,
+      "loss": 0.3784,
+      "step": 3640
+    },
+    {
+      "epoch": 158.69565217391303,
+      "grad_norm": 2.407663106918335,
+      "learning_rate": 4.130434782608696e-06,
+      "loss": 0.3943,
+      "step": 3650
+    },
+    {
+      "epoch": 159.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5125917196273804,
+      "eval_runtime": 3.7278,
+      "eval_samples_per_second": 68.138,
+      "eval_steps_per_second": 0.537,
+      "step": 3657
+    },
+    {
+      "epoch": 159.1304347826087,
+      "grad_norm": 2.2464840412139893,
+      "learning_rate": 4.086956521739131e-06,
+      "loss": 0.3675,
+      "step": 3660
+    },
+    {
+      "epoch": 159.56521739130434,
+      "grad_norm": 3.0186944007873535,
+      "learning_rate": 4.0434782608695655e-06,
+      "loss": 0.355,
+      "step": 3670
+    },
+    {
+      "epoch": 160.0,
+      "grad_norm": 9.606362342834473,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.3417,
+      "step": 3680
+    },
+    {
+      "epoch": 160.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5096677541732788,
+      "eval_runtime": 3.7505,
+      "eval_samples_per_second": 67.724,
+      "eval_steps_per_second": 0.533,
+      "step": 3680
+    },
+    {
+      "epoch": 160.43478260869566,
+      "grad_norm": 3.155024766921997,
+      "learning_rate": 3.956521739130435e-06,
+      "loss": 0.3951,
+      "step": 3690
+    },
+    {
+      "epoch": 160.8695652173913,
+      "grad_norm": 2.3195645809173584,
+      "learning_rate": 3.91304347826087e-06,
+      "loss": 0.3996,
+      "step": 3700
+    },
+    {
+      "epoch": 161.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5048008561134338,
+      "eval_runtime": 4.9463,
+      "eval_samples_per_second": 51.351,
+      "eval_steps_per_second": 0.404,
+      "step": 3703
+    },
+    {
+      "epoch": 161.30434782608697,
+      "grad_norm": 16.818618774414062,
+      "learning_rate": 3.869565217391304e-06,
+      "loss": 0.3613,
+      "step": 3710
+    },
+    {
+      "epoch": 161.7391304347826,
+      "grad_norm": 5.290389060974121,
+      "learning_rate": 3.8260869565217395e-06,
+      "loss": 0.4,
+      "step": 3720
+    },
+    {
+      "epoch": 162.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5148473978042603,
+      "eval_runtime": 3.7348,
+      "eval_samples_per_second": 68.008,
+      "eval_steps_per_second": 0.535,
+      "step": 3726
+    },
+    {
+      "epoch": 162.17391304347825,
+      "grad_norm": 4.7519330978393555,
+      "learning_rate": 3.782608695652174e-06,
+      "loss": 0.3983,
+      "step": 3730
+    },
+    {
+      "epoch": 162.6086956521739,
+      "grad_norm": 2.433164358139038,
+      "learning_rate": 3.739130434782609e-06,
+      "loss": 0.4051,
+      "step": 3740
+    },
+    {
+      "epoch": 163.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5150399804115295,
+      "eval_runtime": 3.7013,
+      "eval_samples_per_second": 68.625,
+      "eval_steps_per_second": 0.54,
+      "step": 3749
+    },
+    {
+      "epoch": 163.04347826086956,
+      "grad_norm": 2.870962381362915,
+      "learning_rate": 3.6956521739130436e-06,
+      "loss": 0.3903,
+      "step": 3750
+    },
+    {
+      "epoch": 163.47826086956522,
+      "grad_norm": 3.3795669078826904,
+      "learning_rate": 3.6521739130434787e-06,
+      "loss": 0.3981,
+      "step": 3760
+    },
+    {
+      "epoch": 163.91304347826087,
+      "grad_norm": 4.447073936462402,
+      "learning_rate": 3.6086956521739134e-06,
+      "loss": 0.3973,
+      "step": 3770
+    },
+    {
+      "epoch": 164.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5036624073982239,
+      "eval_runtime": 4.8343,
+      "eval_samples_per_second": 52.541,
+      "eval_steps_per_second": 0.414,
+      "step": 3772
+    },
+    {
+      "epoch": 164.34782608695653,
+      "grad_norm": 2.5403716564178467,
+      "learning_rate": 3.565217391304348e-06,
+      "loss": 0.3586,
+      "step": 3780
+    },
+    {
+      "epoch": 164.7826086956522,
+      "grad_norm": 2.5216853618621826,
+      "learning_rate": 3.5217391304347832e-06,
+      "loss": 0.3963,
+      "step": 3790
+    },
+    {
+      "epoch": 165.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5048288702964783,
+      "eval_runtime": 3.7404,
+      "eval_samples_per_second": 67.907,
+      "eval_steps_per_second": 0.535,
+      "step": 3795
+    },
+    {
+      "epoch": 165.2173913043478,
+      "grad_norm": 3.382376194000244,
+      "learning_rate": 3.4782608695652175e-06,
+      "loss": 0.4012,
+      "step": 3800
+    },
+    {
+      "epoch": 165.65217391304347,
+      "grad_norm": 3.0021872520446777,
+      "learning_rate": 3.4347826086956526e-06,
+      "loss": 0.3568,
+      "step": 3810
+    },
+    {
+      "epoch": 166.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5167564749717712,
+      "eval_runtime": 3.6895,
+      "eval_samples_per_second": 68.845,
+      "eval_steps_per_second": 0.542,
+      "step": 3818
+    },
+    {
+      "epoch": 166.08695652173913,
+      "grad_norm": 4.209798812866211,
+      "learning_rate": 3.391304347826087e-06,
+      "loss": 0.4217,
+      "step": 3820
+    },
+    {
+      "epoch": 166.52173913043478,
+      "grad_norm": 2.3605332374572754,
+      "learning_rate": 3.347826086956522e-06,
+      "loss": 0.3897,
+      "step": 3830
+    },
+    {
+      "epoch": 166.95652173913044,
+      "grad_norm": 7.9494733810424805,
+      "learning_rate": 3.3043478260869567e-06,
+      "loss": 0.3995,
+      "step": 3840
+    },
+    {
+      "epoch": 167.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5096150636672974,
+      "eval_runtime": 4.9956,
+      "eval_samples_per_second": 50.845,
+      "eval_steps_per_second": 0.4,
+      "step": 3841
+    },
+    {
+      "epoch": 167.3913043478261,
+      "grad_norm": 3.431043863296509,
+      "learning_rate": 3.2608695652173914e-06,
+      "loss": 0.3765,
+      "step": 3850
+    },
+    {
+      "epoch": 167.82608695652175,
+      "grad_norm": 3.4384922981262207,
+      "learning_rate": 3.217391304347826e-06,
+      "loss": 0.3628,
+      "step": 3860
+    },
+    {
+      "epoch": 168.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5101594924926758,
+      "eval_runtime": 3.7705,
+      "eval_samples_per_second": 67.365,
+      "eval_steps_per_second": 0.53,
+      "step": 3864
+    },
+    {
+      "epoch": 168.2608695652174,
+      "grad_norm": 8.502880096435547,
+      "learning_rate": 3.1739130434782613e-06,
+      "loss": 0.3857,
+      "step": 3870
+    },
+    {
+      "epoch": 168.69565217391303,
+      "grad_norm": 2.5634241104125977,
+      "learning_rate": 3.130434782608696e-06,
+      "loss": 0.3836,
+      "step": 3880
+    },
+    {
+      "epoch": 169.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5133307576179504,
+      "eval_runtime": 3.7532,
+      "eval_samples_per_second": 67.676,
+      "eval_steps_per_second": 0.533,
+      "step": 3887
+    },
+    {
+      "epoch": 169.1304347826087,
+      "grad_norm": 3.617677927017212,
+      "learning_rate": 3.0869565217391307e-06,
+      "loss": 0.4251,
+      "step": 3890
+    },
+    {
+      "epoch": 169.56521739130434,
+      "grad_norm": 3.9091439247131348,
+      "learning_rate": 3.043478260869566e-06,
+      "loss": 0.3747,
+      "step": 3900
+    },
+    {
+      "epoch": 170.0,
+      "grad_norm": 12.626005172729492,
+      "learning_rate": 3e-06,
+      "loss": 0.3646,
+      "step": 3910
+    },
+    {
+      "epoch": 170.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5099019408226013,
+      "eval_runtime": 4.85,
+      "eval_samples_per_second": 52.372,
+      "eval_steps_per_second": 0.412,
+      "step": 3910
+    },
+    {
+      "epoch": 170.43478260869566,
+      "grad_norm": 5.3712263107299805,
+      "learning_rate": 2.956521739130435e-06,
+      "loss": 0.3732,
+      "step": 3920
+    },
+    {
+      "epoch": 170.8695652173913,
+      "grad_norm": 2.489645481109619,
+      "learning_rate": 2.9130434782608695e-06,
+      "loss": 0.3789,
+      "step": 3930
+    },
+    {
+      "epoch": 171.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5151440501213074,
+      "eval_runtime": 3.7265,
+      "eval_samples_per_second": 68.161,
+      "eval_steps_per_second": 0.537,
+      "step": 3933
+    },
+    {
+      "epoch": 171.30434782608697,
+      "grad_norm": 60.279747009277344,
+      "learning_rate": 2.8695652173913046e-06,
+      "loss": 0.3372,
+      "step": 3940
+    },
+    {
+      "epoch": 171.7391304347826,
+      "grad_norm": 5.177385330200195,
+      "learning_rate": 2.8260869565217393e-06,
+      "loss": 0.3832,
+      "step": 3950
+    },
+    {
+      "epoch": 172.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5148643255233765,
+      "eval_runtime": 3.7835,
+      "eval_samples_per_second": 67.134,
+      "eval_steps_per_second": 0.529,
+      "step": 3956
+    },
+    {
+      "epoch": 172.17391304347825,
+      "grad_norm": 2.5841851234436035,
+      "learning_rate": 2.782608695652174e-06,
+      "loss": 0.405,
+      "step": 3960
+    },
+    {
+      "epoch": 172.6086956521739,
+      "grad_norm": 2.6472222805023193,
+      "learning_rate": 2.7391304347826087e-06,
+      "loss": 0.3476,
+      "step": 3970
+    },
+    {
+      "epoch": 173.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5178123116493225,
+      "eval_runtime": 5.0055,
+      "eval_samples_per_second": 50.745,
+      "eval_steps_per_second": 0.4,
+      "step": 3979
+    },
+    {
+      "epoch": 173.04347826086956,
+      "grad_norm": 2.3995625972747803,
+      "learning_rate": 2.695652173913044e-06,
+      "loss": 0.4347,
+      "step": 3980
+    },
+    {
+      "epoch": 173.47826086956522,
+      "grad_norm": 4.958439826965332,
+      "learning_rate": 2.6521739130434785e-06,
+      "loss": 0.3886,
+      "step": 3990
+    },
+    {
+      "epoch": 173.91304347826087,
+      "grad_norm": 4.661713600158691,
+      "learning_rate": 2.6086956521739132e-06,
+      "loss": 0.3806,
+      "step": 4000
+    },
+    {
+      "epoch": 174.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5080812573432922,
+      "eval_runtime": 3.6838,
+      "eval_samples_per_second": 68.951,
+      "eval_steps_per_second": 0.543,
+      "step": 4002
+    },
+    {
+      "epoch": 174.34782608695653,
+      "grad_norm": 2.979862928390503,
+      "learning_rate": 2.5652173913043484e-06,
+      "loss": 0.3429,
+      "step": 4010
+    },
+    {
+      "epoch": 174.7826086956522,
+      "grad_norm": 1.8571139574050903,
+      "learning_rate": 2.5217391304347826e-06,
+      "loss": 0.4053,
+      "step": 4020
+    },
+    {
+      "epoch": 175.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5099707245826721,
+      "eval_runtime": 3.7194,
+      "eval_samples_per_second": 68.291,
+      "eval_steps_per_second": 0.538,
+      "step": 4025
+    },
+    {
+      "epoch": 175.2173913043478,
+      "grad_norm": 2.364047050476074,
+      "learning_rate": 2.4782608695652178e-06,
+      "loss": 0.3774,
+      "step": 4030
+    },
+    {
+      "epoch": 175.65217391304347,
+      "grad_norm": 4.220658779144287,
+      "learning_rate": 2.4347826086956525e-06,
+      "loss": 0.3986,
+      "step": 4040
+    },
+    {
+      "epoch": 176.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5189133286476135,
+      "eval_runtime": 5.0478,
+      "eval_samples_per_second": 50.319,
+      "eval_steps_per_second": 0.396,
+      "step": 4048
+    },
+    {
+      "epoch": 176.08695652173913,
+      "grad_norm": 2.9689295291900635,
+      "learning_rate": 2.391304347826087e-06,
+      "loss": 0.4225,
+      "step": 4050
+    },
+    {
+      "epoch": 176.52173913043478,
+      "grad_norm": 3.78476881980896,
+      "learning_rate": 2.347826086956522e-06,
+      "loss": 0.3798,
+      "step": 4060
+    },
+    {
+      "epoch": 176.95652173913044,
+      "grad_norm": 2.3258774280548096,
+      "learning_rate": 2.3043478260869566e-06,
+      "loss": 0.3827,
+      "step": 4070
+    },
+    {
+      "epoch": 177.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5128843784332275,
+      "eval_runtime": 3.7539,
+      "eval_samples_per_second": 67.663,
+      "eval_steps_per_second": 0.533,
+      "step": 4071
+    },
+    {
+      "epoch": 177.3913043478261,
+      "grad_norm": 2.329585313796997,
+      "learning_rate": 2.2608695652173913e-06,
+      "loss": 0.329,
+      "step": 4080
+    },
+    {
+      "epoch": 177.82608695652175,
+      "grad_norm": 3.0889029502868652,
+      "learning_rate": 2.2173913043478264e-06,
+      "loss": 0.3892,
+      "step": 4090
+    },
+    {
+      "epoch": 178.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5099364519119263,
+      "eval_runtime": 3.764,
+      "eval_samples_per_second": 67.482,
+      "eval_steps_per_second": 0.531,
+      "step": 4094
+    },
+    {
+      "epoch": 178.2608695652174,
+      "grad_norm": 3.167226791381836,
+      "learning_rate": 2.173913043478261e-06,
+      "loss": 0.3801,
+      "step": 4100
+    },
+    {
+      "epoch": 178.69565217391303,
+      "grad_norm": 2.857957601547241,
+      "learning_rate": 2.130434782608696e-06,
+      "loss": 0.3955,
+      "step": 4110
+    },
+    {
+      "epoch": 179.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5212357640266418,
+      "eval_runtime": 4.8308,
+      "eval_samples_per_second": 52.579,
+      "eval_steps_per_second": 0.414,
+      "step": 4117
+    },
+    {
+      "epoch": 179.1304347826087,
+      "grad_norm": 8.153979301452637,
+      "learning_rate": 2.0869565217391305e-06,
+      "loss": 0.4062,
+      "step": 4120
+    },
+    {
+      "epoch": 179.56521739130434,
+      "grad_norm": 3.2647910118103027,
+      "learning_rate": 2.0434782608695656e-06,
+      "loss": 0.3603,
+      "step": 4130
+    },
+    {
+      "epoch": 180.0,
+      "grad_norm": 4.87031364440918,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.4077,
+      "step": 4140
+    },
+    {
+      "epoch": 180.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5101702213287354,
+      "eval_runtime": 3.8052,
+      "eval_samples_per_second": 66.75,
+      "eval_steps_per_second": 0.526,
+      "step": 4140
+    },
+    {
+      "epoch": 180.43478260869566,
+      "grad_norm": 3.3625569343566895,
+      "learning_rate": 1.956521739130435e-06,
+      "loss": 0.3881,
+      "step": 4150
+    },
+    {
+      "epoch": 180.8695652173913,
+      "grad_norm": 3.717646360397339,
+      "learning_rate": 1.9130434782608697e-06,
+      "loss": 0.3579,
+      "step": 4160
+    },
+    {
+      "epoch": 181.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5099858641624451,
+      "eval_runtime": 3.707,
+      "eval_samples_per_second": 68.52,
+      "eval_steps_per_second": 0.54,
+      "step": 4163
+    },
+    {
+      "epoch": 181.30434782608697,
+      "grad_norm": 2.5178964138031006,
+      "learning_rate": 1.8695652173913044e-06,
+      "loss": 0.3828,
+      "step": 4170
+    },
+    {
+      "epoch": 181.7391304347826,
+      "grad_norm": 3.244948625564575,
+      "learning_rate": 1.8260869565217394e-06,
+      "loss": 0.3666,
+      "step": 4180
+    },
+    {
+      "epoch": 182.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5247715711593628,
+      "eval_runtime": 4.2228,
+      "eval_samples_per_second": 60.149,
+      "eval_steps_per_second": 0.474,
+      "step": 4186
+    },
+    {
+      "epoch": 182.17391304347825,
+      "grad_norm": 3.418851613998413,
+      "learning_rate": 1.782608695652174e-06,
+      "loss": 0.4,
+      "step": 4190
+    },
+    {
+      "epoch": 182.6086956521739,
+      "grad_norm": 2.247349262237549,
+      "learning_rate": 1.7391304347826088e-06,
+      "loss": 0.3746,
+      "step": 4200
+    },
+    {
+      "epoch": 183.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5220462083816528,
+      "eval_runtime": 4.3605,
+      "eval_samples_per_second": 58.25,
+      "eval_steps_per_second": 0.459,
+      "step": 4209
+    },
+    {
+      "epoch": 183.04347826086956,
+      "grad_norm": 5.591789245605469,
+      "learning_rate": 1.6956521739130435e-06,
+      "loss": 0.3971,
+      "step": 4210
+    },
+    {
+      "epoch": 183.47826086956522,
+      "grad_norm": 2.8663575649261475,
+      "learning_rate": 1.6521739130434784e-06,
+      "loss": 0.3516,
+      "step": 4220
+    },
+    {
+      "epoch": 183.91304347826087,
+      "grad_norm": 5.791408061981201,
+      "learning_rate": 1.608695652173913e-06,
+      "loss": 0.3867,
+      "step": 4230
+    },
+    {
+      "epoch": 184.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5172824859619141,
+      "eval_runtime": 3.8331,
+      "eval_samples_per_second": 66.265,
+      "eval_steps_per_second": 0.522,
+      "step": 4232
+    },
+    {
+      "epoch": 184.34782608695653,
+      "grad_norm": 3.3605191707611084,
+      "learning_rate": 1.565217391304348e-06,
+      "loss": 0.3911,
+      "step": 4240
+    },
+    {
+      "epoch": 184.7826086956522,
+      "grad_norm": 3.4683103561401367,
+      "learning_rate": 1.521739130434783e-06,
+      "loss": 0.4024,
+      "step": 4250
+    },
+    {
+      "epoch": 185.0,
+      "eval_accuracy": 0.7874015748031497,
+      "eval_loss": 0.5248106122016907,
+      "eval_runtime": 4.5705,
+      "eval_samples_per_second": 55.574,
+      "eval_steps_per_second": 0.438,
+      "step": 4255
+    },
+    {
+      "epoch": 185.2173913043478,
+      "grad_norm": 4.495180130004883,
+      "learning_rate": 1.4782608695652176e-06,
+      "loss": 0.3931,
+      "step": 4260
+    },
+    {
+      "epoch": 185.65217391304347,
+      "grad_norm": 4.51051139831543,
+      "learning_rate": 1.4347826086956523e-06,
+      "loss": 0.4014,
+      "step": 4270
+    },
+    {
+      "epoch": 186.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5084752440452576,
+      "eval_runtime": 4.1594,
+      "eval_samples_per_second": 61.066,
+      "eval_steps_per_second": 0.481,
+      "step": 4278
+    },
+    {
+      "epoch": 186.08695652173913,
+      "grad_norm": 6.847979545593262,
+      "learning_rate": 1.391304347826087e-06,
+      "loss": 0.3887,
+      "step": 4280
+    },
+    {
+      "epoch": 186.52173913043478,
+      "grad_norm": 8.414494514465332,
+      "learning_rate": 1.347826086956522e-06,
+      "loss": 0.3876,
+      "step": 4290
+    },
+    {
+      "epoch": 186.95652173913044,
+      "grad_norm": 2.0459609031677246,
+      "learning_rate": 1.3043478260869566e-06,
+      "loss": 0.3445,
+      "step": 4300
+    },
+    {
+      "epoch": 187.0,
+      "eval_accuracy": 0.8031496062992126,
+      "eval_loss": 0.5136986970901489,
+      "eval_runtime": 3.7104,
+      "eval_samples_per_second": 68.456,
+      "eval_steps_per_second": 0.539,
+      "step": 4301
+    },
+    {
+      "epoch": 187.3913043478261,
+      "grad_norm": 2.7707877159118652,
+      "learning_rate": 1.2608695652173913e-06,
+      "loss": 0.4067,
+      "step": 4310
+    },
+    {
+      "epoch": 187.82608695652175,
+      "grad_norm": 2.2277884483337402,
+      "learning_rate": 1.2173913043478262e-06,
+      "loss": 0.382,
+      "step": 4320
+    },
+    {
+      "epoch": 188.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.521314799785614,
+      "eval_runtime": 4.1528,
+      "eval_samples_per_second": 61.164,
+      "eval_steps_per_second": 0.482,
+      "step": 4324
+    },
+    {
+      "epoch": 188.2608695652174,
+      "grad_norm": 4.299314498901367,
+      "learning_rate": 1.173913043478261e-06,
+      "loss": 0.3717,
+      "step": 4330
+    },
+    {
+      "epoch": 188.69565217391303,
+      "grad_norm": 2.479510545730591,
+      "learning_rate": 1.1304347826086956e-06,
+      "loss": 0.3673,
+      "step": 4340
+    },
+    {
+      "epoch": 189.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5241702795028687,
+      "eval_runtime": 4.1853,
+      "eval_samples_per_second": 60.689,
+      "eval_steps_per_second": 0.478,
+      "step": 4347
+    },
+    {
+      "epoch": 189.1304347826087,
+      "grad_norm": 3.9942944049835205,
+      "learning_rate": 1.0869565217391306e-06,
+      "loss": 0.4158,
+      "step": 4350
+    },
+    {
+      "epoch": 189.56521739130434,
+      "grad_norm": 2.8651175498962402,
+      "learning_rate": 1.0434782608695653e-06,
+      "loss": 0.3919,
+      "step": 4360
+    },
+    {
+      "epoch": 190.0,
+      "grad_norm": 3.1065168380737305,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.3631,
+      "step": 4370
+    },
+    {
+      "epoch": 190.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5146118402481079,
+      "eval_runtime": 3.7356,
+      "eval_samples_per_second": 67.995,
+      "eval_steps_per_second": 0.535,
+      "step": 4370
+    },
+    {
+      "epoch": 190.43478260869566,
+      "grad_norm": 3.8796093463897705,
+      "learning_rate": 9.565217391304349e-07,
+      "loss": 0.3893,
+      "step": 4380
+    },
+    {
+      "epoch": 190.8695652173913,
+      "grad_norm": 3.2894842624664307,
+      "learning_rate": 9.130434782608697e-07,
+      "loss": 0.393,
+      "step": 4390
+    },
+    {
+      "epoch": 191.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5097819566726685,
+      "eval_runtime": 3.7404,
+      "eval_samples_per_second": 67.908,
+      "eval_steps_per_second": 0.535,
+      "step": 4393
+    },
+    {
+      "epoch": 191.30434782608697,
+      "grad_norm": 2.4112348556518555,
+      "learning_rate": 8.695652173913044e-07,
+      "loss": 0.4037,
+      "step": 4400
+    },
+    {
+      "epoch": 191.7391304347826,
+      "grad_norm": 2.4510791301727295,
+      "learning_rate": 8.260869565217392e-07,
+      "loss": 0.3806,
+      "step": 4410
+    },
+    {
+      "epoch": 192.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5133717656135559,
+      "eval_runtime": 5.016,
+      "eval_samples_per_second": 50.638,
+      "eval_steps_per_second": 0.399,
+      "step": 4416
+    },
+    {
+      "epoch": 192.17391304347825,
+      "grad_norm": 3.1017332077026367,
+      "learning_rate": 7.82608695652174e-07,
+      "loss": 0.3598,
+      "step": 4420
+    },
+    {
+      "epoch": 192.6086956521739,
+      "grad_norm": 3.5164568424224854,
+      "learning_rate": 7.391304347826088e-07,
+      "loss": 0.3789,
+      "step": 4430
+    },
+    {
+      "epoch": 193.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5127285718917847,
+      "eval_runtime": 3.736,
+      "eval_samples_per_second": 67.988,
+      "eval_steps_per_second": 0.535,
+      "step": 4439
+    },
+    {
+      "epoch": 193.04347826086956,
+      "grad_norm": 86.44344329833984,
+      "learning_rate": 6.956521739130435e-07,
+      "loss": 0.3858,
+      "step": 4440
+    },
+    {
+      "epoch": 193.47826086956522,
+      "grad_norm": 2.892185688018799,
+      "learning_rate": 6.521739130434783e-07,
+      "loss": 0.3894,
+      "step": 4450
+    },
+    {
+      "epoch": 193.91304347826087,
+      "grad_norm": 2.0254733562469482,
+      "learning_rate": 6.086956521739131e-07,
+      "loss": 0.3717,
+      "step": 4460
+    },
+    {
+      "epoch": 194.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5183544158935547,
+      "eval_runtime": 3.7197,
+      "eval_samples_per_second": 68.285,
+      "eval_steps_per_second": 0.538,
+      "step": 4462
+    },
+    {
+      "epoch": 194.34782608695653,
+      "grad_norm": 4.124297618865967,
+      "learning_rate": 5.652173913043478e-07,
+      "loss": 0.4098,
+      "step": 4470
+    },
+    {
+      "epoch": 194.7826086956522,
+      "grad_norm": 4.1497955322265625,
+      "learning_rate": 5.217391304347826e-07,
+      "loss": 0.361,
+      "step": 4480
+    },
+    {
+      "epoch": 195.0,
+      "eval_accuracy": 0.7834645669291339,
+      "eval_loss": 0.5185708999633789,
+      "eval_runtime": 4.9741,
+      "eval_samples_per_second": 51.064,
+      "eval_steps_per_second": 0.402,
+      "step": 4485
+    },
+    {
+      "epoch": 195.2173913043478,
+      "grad_norm": 11.268845558166504,
+      "learning_rate": 4.782608695652174e-07,
+      "loss": 0.3786,
+      "step": 4490
+    },
+    {
+      "epoch": 195.65217391304347,
+      "grad_norm": 3.9937920570373535,
+      "learning_rate": 4.347826086956522e-07,
+      "loss": 0.3722,
+      "step": 4500
+    },
+    {
+      "epoch": 196.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5107359886169434,
+      "eval_runtime": 3.7506,
+      "eval_samples_per_second": 67.723,
+      "eval_steps_per_second": 0.533,
+      "step": 4508
+    },
+    {
+      "epoch": 196.08695652173913,
+      "grad_norm": 2.869596004486084,
+      "learning_rate": 3.91304347826087e-07,
+      "loss": 0.3985,
+      "step": 4510
+    },
+    {
+      "epoch": 196.52173913043478,
+      "grad_norm": 6.21280574798584,
+      "learning_rate": 3.4782608695652175e-07,
+      "loss": 0.4019,
+      "step": 4520
+    },
+    {
+      "epoch": 196.95652173913044,
+      "grad_norm": 2.2324206829071045,
+      "learning_rate": 3.0434782608695656e-07,
+      "loss": 0.3551,
+      "step": 4530
+    },
+    {
+      "epoch": 197.0,
+      "eval_accuracy": 0.7952755905511811,
+      "eval_loss": 0.5174936056137085,
+      "eval_runtime": 3.6975,
+      "eval_samples_per_second": 68.695,
+      "eval_steps_per_second": 0.541,
+      "step": 4531
+    },
+    {
+      "epoch": 197.3913043478261,
+      "grad_norm": 2.6415905952453613,
+      "learning_rate": 2.608695652173913e-07,
+      "loss": 0.3919,
+      "step": 4540
+    },
+    {
+      "epoch": 197.82608695652175,
+      "grad_norm": 5.146513938903809,
+      "learning_rate": 2.173913043478261e-07,
+      "loss": 0.3649,
+      "step": 4550
+    },
+    {
+      "epoch": 198.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5135703682899475,
+      "eval_runtime": 4.9875,
+      "eval_samples_per_second": 50.928,
+      "eval_steps_per_second": 0.401,
+      "step": 4554
+    },
+    {
+      "epoch": 198.2608695652174,
+      "grad_norm": 3.1943461894989014,
+      "learning_rate": 1.7391304347826088e-07,
+      "loss": 0.3763,
+      "step": 4560
+    },
+    {
+      "epoch": 198.69565217391303,
+      "grad_norm": 2.8955743312835693,
+      "learning_rate": 1.3043478260869566e-07,
+      "loss": 0.3749,
+      "step": 4570
+    },
+    {
+      "epoch": 199.0,
+      "eval_accuracy": 0.7913385826771654,
+      "eval_loss": 0.5192672610282898,
+      "eval_runtime": 3.6944,
+      "eval_samples_per_second": 68.753,
+      "eval_steps_per_second": 0.541,
+      "step": 4577
+    },
+    {
+      "epoch": 199.1304347826087,
+      "grad_norm": 12.166488647460938,
+      "learning_rate": 8.695652173913044e-08,
+      "loss": 0.3869,
+      "step": 4580
+    },
+    {
+      "epoch": 199.56521739130434,
+      "grad_norm": 2.9687561988830566,
+      "learning_rate": 4.347826086956522e-08,
+      "loss": 0.3926,
+      "step": 4590
+    },
+    {
+      "epoch": 200.0,
+      "grad_norm": 4.834624290466309,
+      "learning_rate": 0.0,
+      "loss": 0.3782,
+      "step": 4600
+    },
+    {
+      "epoch": 200.0,
+      "eval_accuracy": 0.7992125984251969,
+      "eval_loss": 0.5181651711463928,
+      "eval_runtime": 3.7789,
+      "eval_samples_per_second": 67.216,
+      "eval_steps_per_second": 0.529,
+      "step": 4600
+    },
+    {
+      "epoch": 200.0,
+      "step": 4600,
+      "total_flos": 1.089869514338304e+18,
+      "train_loss": 0.30267460563908455,
+      "train_runtime": 4787.5341,
+      "train_samples_per_second": 59.947,
+      "train_steps_per_second": 0.961
     }
   ],
   "logging_steps": 10,
-  "max_steps": 1150,
+  "max_steps": 4600,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 50,
+  "num_train_epochs": 200,
   "save_steps": 500,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -1290,7 +5055,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 2.72467378584576e+17,
+  "total_flos": 1.089869514338304e+18,
   "train_batch_size": 64,
   "trial_name": null,
   "trial_params": null