diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4162 @@
+{
+  "best_metric": 0.44,
+  "best_model_checkpoint": "5c_2/checkpoint-2209",
+  "epoch": 97.00891304347826,
+  "eval_steps": 500,
+  "global_step": 4600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.002173913043478261,
+      "grad_norm": 3.132235288619995,
+      "learning_rate": 2.173913043478261e-07,
+      "loss": 1.3788,
+      "step": 10
+    },
+    {
+      "epoch": 0.004347826086956522,
+      "grad_norm": 5.476442337036133,
+      "learning_rate": 4.347826086956522e-07,
+      "loss": 1.3742,
+      "step": 20
+    },
+    {
+      "epoch": 0.006521739130434782,
+      "grad_norm": 4.188530921936035,
+      "learning_rate": 6.521739130434783e-07,
+      "loss": 1.3419,
+      "step": 30
+    },
+    {
+      "epoch": 0.008695652173913044,
+      "grad_norm": 4.2109761238098145,
+      "learning_rate": 8.695652173913044e-07,
+      "loss": 1.3192,
+      "step": 40
+    },
+    {
+      "epoch": 0.010217391304347826,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.3455283641815186,
+      "eval_runtime": 8.2639,
+      "eval_samples_per_second": 3.025,
+      "eval_steps_per_second": 0.605,
+      "step": 47
+    },
+    {
+      "epoch": 1.0006521739130434,
+      "grad_norm": 7.003238677978516,
+      "learning_rate": 1.0869565217391306e-06,
+      "loss": 1.2342,
+      "step": 50
+    },
+    {
+      "epoch": 1.0028260869565218,
+      "grad_norm": 4.680798053741455,
+      "learning_rate": 1.3043478260869566e-06,
+      "loss": 1.2008,
+      "step": 60
+    },
+    {
+      "epoch": 1.005,
+      "grad_norm": 12.677494049072266,
+      "learning_rate": 1.521739130434783e-06,
+      "loss": 1.0899,
+      "step": 70
+    },
+    {
+      "epoch": 1.0071739130434783,
+      "grad_norm": 15.23046875,
+      "learning_rate": 1.7391304347826088e-06,
+      "loss": 1.0554,
+      "step": 80
+    },
+    {
+      "epoch": 1.0093478260869566,
+      "grad_norm": 11.614619255065918,
+      "learning_rate": 1.956521739130435e-06,
+      "loss": 0.8974,
+      "step": 90
+    },
+    {
+      "epoch": 1.0102173913043477,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.4305167198181152,
+      "eval_runtime": 7.4693,
+      "eval_samples_per_second": 3.347,
+      "eval_steps_per_second": 0.669,
+      "step": 94
+    },
+    {
+      "epoch": 2.001304347826087,
+      "grad_norm": 11.64830493927002,
+      "learning_rate": 2.173913043478261e-06,
+      "loss": 1.0618,
+      "step": 100
+    },
+    {
+      "epoch": 2.0034782608695654,
+      "grad_norm": 6.811112403869629,
+      "learning_rate": 2.391304347826087e-06,
+      "loss": 0.8375,
+      "step": 110
+    },
+    {
+      "epoch": 2.0056521739130435,
+      "grad_norm": 14.380611419677734,
+      "learning_rate": 2.6086956521739132e-06,
+      "loss": 0.9417,
+      "step": 120
+    },
+    {
+      "epoch": 2.0078260869565216,
+      "grad_norm": 9.205263137817383,
+      "learning_rate": 2.8260869565217393e-06,
+      "loss": 0.8511,
+      "step": 130
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 11.153182029724121,
+      "learning_rate": 3.043478260869566e-06,
+      "loss": 0.9209,
+      "step": 140
+    },
+    {
+      "epoch": 2.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.5682109594345093,
+      "eval_runtime": 7.537,
+      "eval_samples_per_second": 3.317,
+      "eval_steps_per_second": 0.663,
+      "step": 141
+    },
+    {
+      "epoch": 3.0019565217391304,
+      "grad_norm": 11.33986759185791,
+      "learning_rate": 3.2608695652173914e-06,
+      "loss": 1.0064,
+      "step": 150
+    },
+    {
+      "epoch": 3.0041304347826085,
+      "grad_norm": 6.213976860046387,
+      "learning_rate": 3.4782608695652175e-06,
+      "loss": 0.808,
+      "step": 160
+    },
+    {
+      "epoch": 3.006304347826087,
+      "grad_norm": 9.90012264251709,
+      "learning_rate": 3.6956521739130436e-06,
+      "loss": 1.201,
+      "step": 170
+    },
+    {
+      "epoch": 3.0084782608695653,
+      "grad_norm": 11.080544471740723,
+      "learning_rate": 3.91304347826087e-06,
+      "loss": 0.7915,
+      "step": 180
+    },
+    {
+      "epoch": 3.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.6475093364715576,
+      "eval_runtime": 7.4739,
+      "eval_samples_per_second": 3.345,
+      "eval_steps_per_second": 0.669,
+      "step": 188
+    },
+    {
+      "epoch": 4.0004347826086954,
+      "grad_norm": 13.799714088439941,
+      "learning_rate": 4.130434782608696e-06,
+      "loss": 1.0211,
+      "step": 190
+    },
+    {
+      "epoch": 4.002608695652174,
+      "grad_norm": 6.624669551849365,
+      "learning_rate": 4.347826086956522e-06,
+      "loss": 0.7803,
+      "step": 200
+    },
+    {
+      "epoch": 4.004782608695653,
+      "grad_norm": 8.803510665893555,
+      "learning_rate": 4.565217391304348e-06,
+      "loss": 0.7924,
+      "step": 210
+    },
+    {
+      "epoch": 4.006956521739131,
+      "grad_norm": 6.722883701324463,
+      "learning_rate": 4.782608695652174e-06,
+      "loss": 0.7971,
+      "step": 220
+    },
+    {
+      "epoch": 4.009130434782609,
+      "grad_norm": 8.884916305541992,
+      "learning_rate": 5e-06,
+      "loss": 0.9571,
+      "step": 230
+    },
+    {
+      "epoch": 4.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.4177318811416626,
+      "eval_runtime": 6.4192,
+      "eval_samples_per_second": 3.895,
+      "eval_steps_per_second": 0.779,
+      "step": 235
+    },
+    {
+      "epoch": 5.001086956521739,
+      "grad_norm": 6.163194179534912,
+      "learning_rate": 5.2173913043478265e-06,
+      "loss": 1.0597,
+      "step": 240
+    },
+    {
+      "epoch": 5.003260869565217,
+      "grad_norm": 6.332143306732178,
+      "learning_rate": 5.4347826086956525e-06,
+      "loss": 1.0605,
+      "step": 250
+    },
+    {
+      "epoch": 5.005434782608695,
+      "grad_norm": 7.887217998504639,
+      "learning_rate": 5.652173913043479e-06,
+      "loss": 0.8734,
+      "step": 260
+    },
+    {
+      "epoch": 5.0076086956521735,
+      "grad_norm": 7.5437164306640625,
+      "learning_rate": 5.8695652173913055e-06,
+      "loss": 0.7325,
+      "step": 270
+    },
+    {
+      "epoch": 5.0097826086956525,
+      "grad_norm": 8.990574836730957,
+      "learning_rate": 6.086956521739132e-06,
+      "loss": 0.9085,
+      "step": 280
+    },
+    {
+      "epoch": 5.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.4975625276565552,
+      "eval_runtime": 7.0969,
+      "eval_samples_per_second": 3.523,
+      "eval_steps_per_second": 0.705,
+      "step": 282
+    },
+    {
+      "epoch": 6.001739130434783,
+      "grad_norm": 9.910781860351562,
+      "learning_rate": 6.304347826086958e-06,
+      "loss": 0.8632,
+      "step": 290
+    },
+    {
+      "epoch": 6.003913043478261,
+      "grad_norm": 17.275583267211914,
+      "learning_rate": 6.521739130434783e-06,
+      "loss": 0.9834,
+      "step": 300
+    },
+    {
+      "epoch": 6.006086956521739,
+      "grad_norm": 7.485463619232178,
+      "learning_rate": 6.739130434782609e-06,
+      "loss": 0.9132,
+      "step": 310
+    },
+    {
+      "epoch": 6.008260869565217,
+      "grad_norm": 4.864223480224609,
+      "learning_rate": 6.956521739130435e-06,
+      "loss": 0.8815,
+      "step": 320
+    },
+    {
+      "epoch": 6.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.6249030828475952,
+      "eval_runtime": 7.0312,
+      "eval_samples_per_second": 3.556,
+      "eval_steps_per_second": 0.711,
+      "step": 329
+    },
+    {
+      "epoch": 7.000217391304348,
+      "grad_norm": 9.021862983703613,
+      "learning_rate": 7.173913043478261e-06,
+      "loss": 0.8582,
+      "step": 330
+    },
+    {
+      "epoch": 7.002391304347826,
+      "grad_norm": 7.927940368652344,
+      "learning_rate": 7.391304347826087e-06,
+      "loss": 0.6438,
+      "step": 340
+    },
+    {
+      "epoch": 7.004565217391304,
+      "grad_norm": 17.572071075439453,
+      "learning_rate": 7.608695652173914e-06,
+      "loss": 1.1519,
+      "step": 350
+    },
+    {
+      "epoch": 7.006739130434783,
+      "grad_norm": 6.535980224609375,
+      "learning_rate": 7.82608695652174e-06,
+      "loss": 0.9535,
+      "step": 360
+    },
+    {
+      "epoch": 7.008913043478261,
+      "grad_norm": 8.023582458496094,
+      "learning_rate": 8.043478260869566e-06,
+      "loss": 0.9245,
+      "step": 370
+    },
+    {
+      "epoch": 7.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.7362802028656006,
+      "eval_runtime": 7.0278,
+      "eval_samples_per_second": 3.557,
+      "eval_steps_per_second": 0.711,
+      "step": 376
+    },
+    {
+      "epoch": 8.000869565217391,
+      "grad_norm": 8.489022254943848,
+      "learning_rate": 8.260869565217392e-06,
+      "loss": 0.7537,
+      "step": 380
+    },
+    {
+      "epoch": 8.003043478260869,
+      "grad_norm": 7.774221420288086,
+      "learning_rate": 8.478260869565218e-06,
+      "loss": 0.8413,
+      "step": 390
+    },
+    {
+      "epoch": 8.005217391304347,
+      "grad_norm": 6.500988960266113,
+      "learning_rate": 8.695652173913044e-06,
+      "loss": 1.0033,
+      "step": 400
+    },
+    {
+      "epoch": 8.007391304347825,
+      "grad_norm": 5.181256294250488,
+      "learning_rate": 8.91304347826087e-06,
+      "loss": 0.9294,
+      "step": 410
+    },
+    {
+      "epoch": 8.009565217391305,
+      "grad_norm": 20.103971481323242,
+      "learning_rate": 9.130434782608697e-06,
+      "loss": 0.9179,
+      "step": 420
+    },
+    {
+      "epoch": 8.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.4074734449386597,
+      "eval_runtime": 7.3683,
+      "eval_samples_per_second": 3.393,
+      "eval_steps_per_second": 0.679,
+      "step": 423
+    },
+    {
+      "epoch": 9.001521739130435,
+      "grad_norm": 6.049601078033447,
+      "learning_rate": 9.347826086956523e-06,
+      "loss": 0.6285,
+      "step": 430
+    },
+    {
+      "epoch": 9.003695652173914,
+      "grad_norm": 6.180543422698975,
+      "learning_rate": 9.565217391304349e-06,
+      "loss": 0.807,
+      "step": 440
+    },
+    {
+      "epoch": 9.005869565217392,
+      "grad_norm": 7.220910549163818,
+      "learning_rate": 9.782608695652175e-06,
+      "loss": 0.8798,
+      "step": 450
+    },
+    {
+      "epoch": 9.00804347826087,
+      "grad_norm": 12.359790802001953,
+      "learning_rate": 1e-05,
+      "loss": 0.9162,
+      "step": 460
+    },
+    {
+      "epoch": 9.010217391304348,
+      "grad_norm": 27.027353286743164,
+      "learning_rate": 9.97584541062802e-06,
+      "loss": 1.0738,
+      "step": 470
+    },
+    {
+      "epoch": 9.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.953555941581726,
+      "eval_runtime": 7.4282,
+      "eval_samples_per_second": 3.366,
+      "eval_steps_per_second": 0.673,
+      "step": 470
+    },
+    {
+      "epoch": 10.002173913043478,
+      "grad_norm": 9.702156066894531,
+      "learning_rate": 9.95169082125604e-06,
+      "loss": 0.7061,
+      "step": 480
+    },
+    {
+      "epoch": 10.004347826086956,
+      "grad_norm": 5.7458014488220215,
+      "learning_rate": 9.927536231884058e-06,
+      "loss": 0.7154,
+      "step": 490
+    },
+    {
+      "epoch": 10.006521739130434,
+      "grad_norm": 10.728633880615234,
+      "learning_rate": 9.903381642512077e-06,
+      "loss": 0.8171,
+      "step": 500
+    },
+    {
+      "epoch": 10.008695652173913,
+      "grad_norm": 2.1833744049072266,
+      "learning_rate": 9.879227053140097e-06,
+      "loss": 0.5865,
+      "step": 510
+    },
+    {
+      "epoch": 10.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.6996039152145386,
+      "eval_runtime": 7.4442,
+      "eval_samples_per_second": 3.358,
+      "eval_steps_per_second": 0.672,
+      "step": 517
+    },
+    {
+      "epoch": 11.000652173913043,
+      "grad_norm": 11.105751991271973,
+      "learning_rate": 9.855072463768118e-06,
+      "loss": 1.2835,
+      "step": 520
+    },
+    {
+      "epoch": 11.00282608695652,
+      "grad_norm": 30.184377670288086,
+      "learning_rate": 9.830917874396136e-06,
+      "loss": 0.924,
+      "step": 530
+    },
+    {
+      "epoch": 11.005,
+      "grad_norm": 8.80075740814209,
+      "learning_rate": 9.806763285024155e-06,
+      "loss": 1.0119,
+      "step": 540
+    },
+    {
+      "epoch": 11.007173913043479,
+      "grad_norm": 20.036352157592773,
+      "learning_rate": 9.782608695652175e-06,
+      "loss": 0.8836,
+      "step": 550
+    },
+    {
+      "epoch": 11.009347826086957,
+      "grad_norm": 1.3486090898513794,
+      "learning_rate": 9.758454106280194e-06,
+      "loss": 0.4752,
+      "step": 560
+    },
+    {
+      "epoch": 11.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 2.1711320877075195,
+      "eval_runtime": 6.2994,
+      "eval_samples_per_second": 3.969,
+      "eval_steps_per_second": 0.794,
+      "step": 564
+    },
+    {
+      "epoch": 12.001304347826087,
+      "grad_norm": 10.300837516784668,
+      "learning_rate": 9.734299516908214e-06,
+      "loss": 0.9957,
+      "step": 570
+    },
+    {
+      "epoch": 12.003478260869565,
+      "grad_norm": 6.451653003692627,
+      "learning_rate": 9.710144927536233e-06,
+      "loss": 0.7474,
+      "step": 580
+    },
+    {
+      "epoch": 12.005652173913044,
+      "grad_norm": 13.948019981384277,
+      "learning_rate": 9.685990338164251e-06,
+      "loss": 0.6747,
+      "step": 590
+    },
+    {
+      "epoch": 12.007826086956522,
+      "grad_norm": 17.124176025390625,
+      "learning_rate": 9.66183574879227e-06,
+      "loss": 0.7575,
+      "step": 600
+    },
+    {
+      "epoch": 12.01,
+      "grad_norm": 15.972012519836426,
+      "learning_rate": 9.63768115942029e-06,
+      "loss": 0.7409,
+      "step": 610
+    },
+    {
+      "epoch": 12.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 1.7252395153045654,
+      "eval_runtime": 6.6006,
+      "eval_samples_per_second": 3.788,
+      "eval_steps_per_second": 0.758,
+      "step": 611
+    },
+    {
+      "epoch": 13.00195652173913,
+      "grad_norm": 7.705373764038086,
+      "learning_rate": 9.61352657004831e-06,
+      "loss": 0.5027,
+      "step": 620
+    },
+    {
+      "epoch": 13.004130434782608,
+      "grad_norm": 16.61420249938965,
+      "learning_rate": 9.58937198067633e-06,
+      "loss": 0.7508,
+      "step": 630
+    },
+    {
+      "epoch": 13.006304347826086,
+      "grad_norm": 10.502535820007324,
+      "learning_rate": 9.565217391304349e-06,
+      "loss": 0.7997,
+      "step": 640
+    },
+    {
+      "epoch": 13.008478260869564,
+      "grad_norm": 12.413015365600586,
+      "learning_rate": 9.541062801932368e-06,
+      "loss": 0.7534,
+      "step": 650
+    },
+    {
+      "epoch": 13.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 1.9987900257110596,
+      "eval_runtime": 7.0116,
+      "eval_samples_per_second": 3.565,
+      "eval_steps_per_second": 0.713,
+      "step": 658
+    },
+    {
+      "epoch": 14.000434782608696,
+      "grad_norm": 1.7208877801895142,
+      "learning_rate": 9.516908212560388e-06,
+      "loss": 0.6252,
+      "step": 660
+    },
+    {
+      "epoch": 14.002608695652174,
+      "grad_norm": 1.498140811920166,
+      "learning_rate": 9.492753623188407e-06,
+      "loss": 0.7554,
+      "step": 670
+    },
+    {
+      "epoch": 14.004782608695653,
+      "grad_norm": 39.054325103759766,
+      "learning_rate": 9.468599033816425e-06,
+      "loss": 0.5343,
+      "step": 680
+    },
+    {
+      "epoch": 14.00695652173913,
+      "grad_norm": 10.334756851196289,
+      "learning_rate": 9.444444444444445e-06,
+      "loss": 0.855,
+      "step": 690
+    },
+    {
+      "epoch": 14.009130434782609,
+      "grad_norm": 16.51199722290039,
+      "learning_rate": 9.420289855072464e-06,
+      "loss": 0.6109,
+      "step": 700
+    },
+    {
+      "epoch": 14.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 1.7449324131011963,
+      "eval_runtime": 7.5709,
+      "eval_samples_per_second": 3.302,
+      "eval_steps_per_second": 0.66,
+      "step": 705
+    },
+    {
+      "epoch": 15.001086956521739,
+      "grad_norm": 32.96943664550781,
+      "learning_rate": 9.396135265700484e-06,
+      "loss": 0.9145,
+      "step": 710
+    },
+    {
+      "epoch": 15.003260869565217,
+      "grad_norm": 16.733196258544922,
+      "learning_rate": 9.371980676328503e-06,
+      "loss": 0.7627,
+      "step": 720
+    },
+    {
+      "epoch": 15.005434782608695,
+      "grad_norm": 14.321381568908691,
+      "learning_rate": 9.347826086956523e-06,
+      "loss": 0.7481,
+      "step": 730
+    },
+    {
+      "epoch": 15.007608695652173,
+      "grad_norm": 15.458819389343262,
+      "learning_rate": 9.323671497584542e-06,
+      "loss": 0.6374,
+      "step": 740
+    },
+    {
+      "epoch": 15.009782608695652,
+      "grad_norm": 0.3697209358215332,
+      "learning_rate": 9.299516908212562e-06,
+      "loss": 0.4217,
+      "step": 750
+    },
+    {
+      "epoch": 15.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 2.9983742237091064,
+      "eval_runtime": 7.0335,
+      "eval_samples_per_second": 3.554,
+      "eval_steps_per_second": 0.711,
+      "step": 752
+    },
+    {
+      "epoch": 16.001739130434782,
+      "grad_norm": 1.2474554777145386,
+      "learning_rate": 9.275362318840581e-06,
+      "loss": 1.1848,
+      "step": 760
+    },
+    {
+      "epoch": 16.00391304347826,
+      "grad_norm": 10.79831600189209,
+      "learning_rate": 9.251207729468599e-06,
+      "loss": 0.6795,
+      "step": 770
+    },
+    {
+      "epoch": 16.006086956521738,
+      "grad_norm": 20.671566009521484,
+      "learning_rate": 9.227053140096618e-06,
+      "loss": 0.5105,
+      "step": 780
+    },
+    {
+      "epoch": 16.008260869565216,
+      "grad_norm": 2.542451858520508,
+      "learning_rate": 9.202898550724638e-06,
+      "loss": 0.8409,
+      "step": 790
+    },
+    {
+      "epoch": 16.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 1.6708961725234985,
+      "eval_runtime": 7.0051,
+      "eval_samples_per_second": 3.569,
+      "eval_steps_per_second": 0.714,
+      "step": 799
+    },
+    {
+      "epoch": 17.000217391304346,
+      "grad_norm": 11.911920547485352,
+      "learning_rate": 9.178743961352658e-06,
+      "loss": 0.6834,
+      "step": 800
+    },
+    {
+      "epoch": 17.002391304347825,
+      "grad_norm": 18.490251541137695,
+      "learning_rate": 9.154589371980677e-06,
+      "loss": 0.6909,
+      "step": 810
+    },
+    {
+      "epoch": 17.004565217391303,
+      "grad_norm": 13.711281776428223,
+      "learning_rate": 9.130434782608697e-06,
+      "loss": 0.6639,
+      "step": 820
+    },
+    {
+      "epoch": 17.006739130434784,
+      "grad_norm": 14.244696617126465,
+      "learning_rate": 9.106280193236716e-06,
+      "loss": 0.7085,
+      "step": 830
+    },
+    {
+      "epoch": 17.008913043478262,
+      "grad_norm": 11.212364196777344,
+      "learning_rate": 9.082125603864736e-06,
+      "loss": 0.6114,
+      "step": 840
+    },
+    {
+      "epoch": 17.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 1.9013595581054688,
+      "eval_runtime": 7.253,
+      "eval_samples_per_second": 3.447,
+      "eval_steps_per_second": 0.689,
+      "step": 846
+    },
+    {
+      "epoch": 18.000869565217393,
+      "grad_norm": 0.3804018795490265,
+      "learning_rate": 9.057971014492755e-06,
+      "loss": 0.3704,
+      "step": 850
+    },
+    {
+      "epoch": 18.00304347826087,
+      "grad_norm": 47.33468246459961,
+      "learning_rate": 9.033816425120775e-06,
+      "loss": 0.7076,
+      "step": 860
+    },
+    {
+      "epoch": 18.00521739130435,
+      "grad_norm": 27.488746643066406,
+      "learning_rate": 9.009661835748792e-06,
+      "loss": 0.4679,
+      "step": 870
+    },
+    {
+      "epoch": 18.007391304347827,
+      "grad_norm": 20.485361099243164,
+      "learning_rate": 8.985507246376812e-06,
+      "loss": 0.6203,
+      "step": 880
+    },
+    {
+      "epoch": 18.009565217391305,
+      "grad_norm": 11.710765838623047,
+      "learning_rate": 8.961352657004831e-06,
+      "loss": 0.6806,
+      "step": 890
+    },
+    {
+      "epoch": 18.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 1.876340627670288,
+      "eval_runtime": 7.0078,
+      "eval_samples_per_second": 3.567,
+      "eval_steps_per_second": 0.713,
+      "step": 893
+    },
+    {
+      "epoch": 19.001521739130435,
+      "grad_norm": 13.085403442382812,
+      "learning_rate": 8.937198067632851e-06,
+      "loss": 0.5103,
+      "step": 900
+    },
+    {
+      "epoch": 19.003695652173914,
+      "grad_norm": 18.02910614013672,
+      "learning_rate": 8.91304347826087e-06,
+      "loss": 0.3566,
+      "step": 910
+    },
+    {
+      "epoch": 19.00586956521739,
+      "grad_norm": 13.15645980834961,
+      "learning_rate": 8.888888888888888e-06,
+      "loss": 0.4992,
+      "step": 920
+    },
+    {
+      "epoch": 19.00804347826087,
+      "grad_norm": 22.144899368286133,
+      "learning_rate": 8.86473429951691e-06,
+      "loss": 0.4577,
+      "step": 930
+    },
+    {
+      "epoch": 19.010217391304348,
+      "grad_norm": 43.63472366333008,
+      "learning_rate": 8.840579710144929e-06,
+      "loss": 0.5359,
+      "step": 940
+    },
+    {
+      "epoch": 19.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 2.1816036701202393,
+      "eval_runtime": 7.0001,
+      "eval_samples_per_second": 3.571,
+      "eval_steps_per_second": 0.714,
+      "step": 940
+    },
+    {
+      "epoch": 20.002173913043478,
+      "grad_norm": 17.214570999145508,
+      "learning_rate": 8.816425120772949e-06,
+      "loss": 0.4247,
+      "step": 950
+    },
+    {
+      "epoch": 20.004347826086956,
+      "grad_norm": 12.690553665161133,
+      "learning_rate": 8.792270531400966e-06,
+      "loss": 0.3248,
+      "step": 960
+    },
+    {
+      "epoch": 20.006521739130434,
+      "grad_norm": 61.57204055786133,
+      "learning_rate": 8.768115942028986e-06,
+      "loss": 0.5981,
+      "step": 970
+    },
+    {
+      "epoch": 20.008695652173913,
+      "grad_norm": 13.446467399597168,
+      "learning_rate": 8.743961352657005e-06,
+      "loss": 0.5989,
+      "step": 980
+    },
+    {
+      "epoch": 20.010217391304348,
+      "eval_accuracy": 0.32,
+      "eval_loss": 1.7238571643829346,
+      "eval_runtime": 7.0733,
+      "eval_samples_per_second": 3.534,
+      "eval_steps_per_second": 0.707,
+      "step": 987
+    },
+    {
+      "epoch": 21.000652173913043,
+      "grad_norm": 4.935718536376953,
+      "learning_rate": 8.719806763285025e-06,
+      "loss": 0.3831,
+      "step": 990
+    },
+    {
+      "epoch": 21.00282608695652,
+      "grad_norm": 16.678447723388672,
+      "learning_rate": 8.695652173913044e-06,
+      "loss": 0.2781,
+      "step": 1000
+    },
+    {
+      "epoch": 21.005,
+      "grad_norm": 25.6951847076416,
+      "learning_rate": 8.671497584541064e-06,
+      "loss": 0.518,
+      "step": 1010
+    },
+    {
+      "epoch": 21.007173913043477,
+      "grad_norm": 14.607623100280762,
+      "learning_rate": 8.647342995169082e-06,
+      "loss": 0.4972,
+      "step": 1020
+    },
+    {
+      "epoch": 21.009347826086955,
+      "grad_norm": 41.22285461425781,
+      "learning_rate": 8.623188405797103e-06,
+      "loss": 0.4145,
+      "step": 1030
+    },
+    {
+      "epoch": 21.010217391304348,
+      "eval_accuracy": 0.32,
+      "eval_loss": 2.245953321456909,
+      "eval_runtime": 7.0547,
+      "eval_samples_per_second": 3.544,
+      "eval_steps_per_second": 0.709,
+      "step": 1034
+    },
+    {
+      "epoch": 22.001304347826085,
+      "grad_norm": 21.144567489624023,
+      "learning_rate": 8.599033816425122e-06,
+      "loss": 0.3487,
+      "step": 1040
+    },
+    {
+      "epoch": 22.003478260869564,
+      "grad_norm": 21.607402801513672,
+      "learning_rate": 8.57487922705314e-06,
+      "loss": 0.4854,
+      "step": 1050
+    },
+    {
+      "epoch": 22.00565217391304,
+      "grad_norm": 47.5101432800293,
+      "learning_rate": 8.55072463768116e-06,
+      "loss": 0.3654,
+      "step": 1060
+    },
+    {
+      "epoch": 22.007826086956523,
+      "grad_norm": 36.82216262817383,
+      "learning_rate": 8.52657004830918e-06,
+      "loss": 0.4297,
+      "step": 1070
+    },
+    {
+      "epoch": 22.01,
+      "grad_norm": 18.2718448638916,
+      "learning_rate": 8.502415458937199e-06,
+      "loss": 0.3841,
+      "step": 1080
+    },
+    {
+      "epoch": 22.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 2.4508769512176514,
+      "eval_runtime": 7.0615,
+      "eval_samples_per_second": 3.54,
+      "eval_steps_per_second": 0.708,
+      "step": 1081
+    },
+    {
+      "epoch": 23.00195652173913,
+      "grad_norm": 129.92491149902344,
+      "learning_rate": 8.478260869565218e-06,
+      "loss": 0.2186,
+      "step": 1090
+    },
+    {
+      "epoch": 23.00413043478261,
+      "grad_norm": 17.69215202331543,
+      "learning_rate": 8.454106280193238e-06,
+      "loss": 0.57,
+      "step": 1100
+    },
+    {
+      "epoch": 23.006304347826088,
+      "grad_norm": 32.97102737426758,
+      "learning_rate": 8.429951690821256e-06,
+      "loss": 0.6273,
+      "step": 1110
+    },
+    {
+      "epoch": 23.008478260869566,
+      "grad_norm": 13.1940336227417,
+      "learning_rate": 8.405797101449275e-06,
+      "loss": 0.4356,
+      "step": 1120
+    },
+    {
+      "epoch": 23.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 2.112544298171997,
+      "eval_runtime": 7.006,
+      "eval_samples_per_second": 3.568,
+      "eval_steps_per_second": 0.714,
+      "step": 1128
+    },
+    {
+      "epoch": 24.000434782608696,
+      "grad_norm": 33.830413818359375,
+      "learning_rate": 8.381642512077295e-06,
+      "loss": 0.4494,
+      "step": 1130
+    },
+    {
+      "epoch": 24.002608695652174,
+      "grad_norm": 9.115934371948242,
+      "learning_rate": 8.357487922705316e-06,
+      "loss": 0.2399,
+      "step": 1140
+    },
+    {
+      "epoch": 24.004782608695653,
+      "grad_norm": 9.865811347961426,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.4098,
+      "step": 1150
+    },
+    {
+      "epoch": 24.00695652173913,
+      "grad_norm": 58.77852249145508,
+      "learning_rate": 8.309178743961353e-06,
+      "loss": 0.4483,
+      "step": 1160
+    },
+    {
+      "epoch": 24.00913043478261,
+      "grad_norm": 2.725935220718384,
+      "learning_rate": 8.285024154589373e-06,
+      "loss": 0.2509,
+      "step": 1170
+    },
+    {
+      "epoch": 24.010217391304348,
+      "eval_accuracy": 0.32,
+      "eval_loss": 2.6513030529022217,
+      "eval_runtime": 7.1137,
+      "eval_samples_per_second": 3.514,
+      "eval_steps_per_second": 0.703,
+      "step": 1175
+    },
+    {
+      "epoch": 25.00108695652174,
+      "grad_norm": 47.503807067871094,
+      "learning_rate": 8.260869565217392e-06,
+      "loss": 0.3978,
+      "step": 1180
+    },
+    {
+      "epoch": 25.003260869565217,
+      "grad_norm": 3.1732006072998047,
+      "learning_rate": 8.236714975845412e-06,
+      "loss": 0.2354,
+      "step": 1190
+    },
+    {
+      "epoch": 25.005434782608695,
+      "grad_norm": 1.1716434955596924,
+      "learning_rate": 8.212560386473431e-06,
+      "loss": 0.2235,
+      "step": 1200
+    },
+    {
+      "epoch": 25.007608695652173,
+      "grad_norm": 21.524194717407227,
+      "learning_rate": 8.188405797101449e-06,
+      "loss": 0.4126,
+      "step": 1210
+    },
+    {
+      "epoch": 25.00978260869565,
+      "grad_norm": 20.019350051879883,
+      "learning_rate": 8.164251207729469e-06,
+      "loss": 0.4963,
+      "step": 1220
+    },
+    {
+      "epoch": 25.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 2.8018980026245117,
+      "eval_runtime": 7.2482,
+      "eval_samples_per_second": 3.449,
+      "eval_steps_per_second": 0.69,
+      "step": 1222
+    },
+    {
+      "epoch": 26.001739130434782,
+      "grad_norm": 36.1866340637207,
+      "learning_rate": 8.140096618357488e-06,
+      "loss": 0.5625,
+      "step": 1230
+    },
+    {
+      "epoch": 26.00391304347826,
+      "grad_norm": 48.510860443115234,
+      "learning_rate": 8.115942028985508e-06,
+      "loss": 0.2733,
+      "step": 1240
+    },
+    {
+      "epoch": 26.006086956521738,
+      "grad_norm": 35.73210144042969,
+      "learning_rate": 8.091787439613527e-06,
+      "loss": 0.2478,
+      "step": 1250
+    },
+    {
+      "epoch": 26.008260869565216,
+      "grad_norm": 18.470474243164062,
+      "learning_rate": 8.067632850241547e-06,
+      "loss": 0.1915,
+      "step": 1260
+    },
+    {
+      "epoch": 26.010217391304348,
+      "eval_accuracy": 0.32,
+      "eval_loss": 2.4637062549591064,
+      "eval_runtime": 7.6909,
+      "eval_samples_per_second": 3.251,
+      "eval_steps_per_second": 0.65,
+      "step": 1269
+    },
+    {
+      "epoch": 27.000217391304346,
+      "grad_norm": 36.4515495300293,
+      "learning_rate": 8.043478260869566e-06,
+      "loss": 0.5219,
+      "step": 1270
+    },
+    {
+      "epoch": 27.002391304347825,
+      "grad_norm": 40.779380798339844,
+      "learning_rate": 8.019323671497586e-06,
+      "loss": 0.3625,
+      "step": 1280
+    },
+    {
+      "epoch": 27.004565217391303,
+      "grad_norm": 11.368120193481445,
+      "learning_rate": 7.995169082125605e-06,
+      "loss": 0.1531,
+      "step": 1290
+    },
+    {
+      "epoch": 27.006739130434784,
+      "grad_norm": 14.41855239868164,
+      "learning_rate": 7.971014492753623e-06,
+      "loss": 0.3014,
+      "step": 1300
+    },
+    {
+      "epoch": 27.008913043478262,
+      "grad_norm": 0.03663391247391701,
+      "learning_rate": 7.946859903381643e-06,
+      "loss": 0.1269,
+      "step": 1310
+    },
+    {
+      "epoch": 27.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 2.8957390785217285,
+      "eval_runtime": 6.9966,
+      "eval_samples_per_second": 3.573,
+      "eval_steps_per_second": 0.715,
+      "step": 1316
+    },
+    {
+      "epoch": 28.000869565217393,
+      "grad_norm": 21.387739181518555,
+      "learning_rate": 7.922705314009662e-06,
+      "loss": 0.5231,
+      "step": 1320
+    },
+    {
+      "epoch": 28.00304347826087,
+      "grad_norm": 13.809257507324219,
+      "learning_rate": 7.898550724637682e-06,
+      "loss": 0.1749,
+      "step": 1330
+    },
+    {
+      "epoch": 28.00521739130435,
+      "grad_norm": 58.56370162963867,
+      "learning_rate": 7.874396135265701e-06,
+      "loss": 0.2982,
+      "step": 1340
+    },
+    {
+      "epoch": 28.007391304347827,
+      "grad_norm": 2.6313860416412354,
+      "learning_rate": 7.85024154589372e-06,
+      "loss": 0.1082,
+      "step": 1350
+    },
+    {
+      "epoch": 28.009565217391305,
+      "grad_norm": 80.68356323242188,
+      "learning_rate": 7.82608695652174e-06,
+      "loss": 0.3599,
+      "step": 1360
+    },
+    {
+      "epoch": 28.010217391304348,
+      "eval_accuracy": 0.36,
+      "eval_loss": 2.585341215133667,
+      "eval_runtime": 7.4662,
+      "eval_samples_per_second": 3.348,
+      "eval_steps_per_second": 0.67,
+      "step": 1363
+    },
+    {
+      "epoch": 29.001521739130435,
+      "grad_norm": 2.553622007369995,
+      "learning_rate": 7.80193236714976e-06,
+      "loss": 0.3539,
+      "step": 1370
+    },
+    {
+      "epoch": 29.003695652173914,
+      "grad_norm": 3.36917781829834,
+      "learning_rate": 7.77777777777778e-06,
+      "loss": 0.4175,
+      "step": 1380
+    },
+    {
+      "epoch": 29.00586956521739,
+      "grad_norm": 2.5340640544891357,
+      "learning_rate": 7.753623188405797e-06,
+      "loss": 0.0719,
+      "step": 1390
+    },
+    {
+      "epoch": 29.00804347826087,
+      "grad_norm": 8.34886646270752,
+      "learning_rate": 7.729468599033817e-06,
+      "loss": 0.3319,
+      "step": 1400
+    },
+    {
+      "epoch": 29.010217391304348,
+      "grad_norm": 24.99821662902832,
+      "learning_rate": 7.705314009661836e-06,
+      "loss": 0.399,
+      "step": 1410
+    },
+    {
+      "epoch": 29.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 3.363311529159546,
+      "eval_runtime": 7.4383,
+      "eval_samples_per_second": 3.361,
+      "eval_steps_per_second": 0.672,
+      "step": 1410
+    },
+    {
+      "epoch": 30.002173913043478,
+      "grad_norm": 3.317831039428711,
+      "learning_rate": 7.681159420289856e-06,
+      "loss": 0.2035,
+      "step": 1420
+    },
+    {
+      "epoch": 30.004347826086956,
+      "grad_norm": 1.5610860586166382,
+      "learning_rate": 7.657004830917875e-06,
+      "loss": 0.2635,
+      "step": 1430
+    },
+    {
+      "epoch": 30.006521739130434,
+      "grad_norm": 0.9607851505279541,
+      "learning_rate": 7.632850241545895e-06,
+      "loss": 0.1763,
+      "step": 1440
+    },
+    {
+      "epoch": 30.008695652173913,
+      "grad_norm": 55.51839065551758,
+      "learning_rate": 7.608695652173914e-06,
+      "loss": 0.205,
+      "step": 1450
+    },
+    {
+      "epoch": 30.010217391304348,
+      "eval_accuracy": 0.32,
+      "eval_loss": 3.0276477336883545,
+      "eval_runtime": 7.0639,
+      "eval_samples_per_second": 3.539,
+      "eval_steps_per_second": 0.708,
+      "step": 1457
+    },
+    {
+      "epoch": 31.000652173913043,
+      "grad_norm": 30.79387664794922,
+      "learning_rate": 7.584541062801934e-06,
+      "loss": 0.229,
+      "step": 1460
+    },
+    {
+      "epoch": 31.00282608695652,
+      "grad_norm": 0.7464058995246887,
+      "learning_rate": 7.560386473429952e-06,
+      "loss": 0.1409,
+      "step": 1470
+    },
+    {
+      "epoch": 31.005,
+      "grad_norm": 50.541080474853516,
+      "learning_rate": 7.536231884057972e-06,
+      "loss": 0.2637,
+      "step": 1480
+    },
+    {
+      "epoch": 31.007173913043477,
+      "grad_norm": 0.211073100566864,
+      "learning_rate": 7.512077294685991e-06,
+      "loss": 0.0746,
+      "step": 1490
+    },
+    {
+      "epoch": 31.009347826086955,
+      "grad_norm": 23.4665584564209,
+      "learning_rate": 7.48792270531401e-06,
+      "loss": 0.0945,
+      "step": 1500
+    },
+    {
+      "epoch": 31.010217391304348,
+      "eval_accuracy": 0.4,
+      "eval_loss": 3.396014451980591,
+      "eval_runtime": 7.0227,
+      "eval_samples_per_second": 3.56,
+      "eval_steps_per_second": 0.712,
+      "step": 1504
+    },
+    {
+      "epoch": 32.001304347826085,
+      "grad_norm": 0.1441574990749359,
+      "learning_rate": 7.4637681159420295e-06,
+      "loss": 0.2904,
+      "step": 1510
+    },
+    {
+      "epoch": 32.003478260869564,
+      "grad_norm": 0.24322232604026794,
+      "learning_rate": 7.439613526570049e-06,
+      "loss": 0.0857,
+      "step": 1520
+    },
+    {
+      "epoch": 32.00565217391304,
+      "grad_norm": 60.05118942260742,
+      "learning_rate": 7.415458937198068e-06,
+      "loss": 0.2013,
+      "step": 1530
+    },
+    {
+      "epoch": 32.00782608695652,
+      "grad_norm": 1.1909397840499878,
+      "learning_rate": 7.391304347826087e-06,
+      "loss": 0.0717,
+      "step": 1540
+    },
+    {
+      "epoch": 32.01,
+      "grad_norm": 111.34762573242188,
+      "learning_rate": 7.3671497584541075e-06,
+      "loss": 0.3376,
+      "step": 1550
+    },
+    {
+      "epoch": 32.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 3.0445356369018555,
+      "eval_runtime": 7.0055,
+      "eval_samples_per_second": 3.569,
+      "eval_steps_per_second": 0.714,
+      "step": 1551
+    },
+    {
+      "epoch": 33.00195652173913,
+      "grad_norm": 0.30092719197273254,
+      "learning_rate": 7.342995169082127e-06,
+      "loss": 0.1294,
+      "step": 1560
+    },
+    {
+      "epoch": 33.00413043478261,
+      "grad_norm": 0.05873465910553932,
+      "learning_rate": 7.318840579710146e-06,
+      "loss": 0.2487,
+      "step": 1570
+    },
+    {
+      "epoch": 33.00630434782609,
+      "grad_norm": 9.693734169006348,
+      "learning_rate": 7.294685990338165e-06,
+      "loss": 0.2849,
+      "step": 1580
+    },
+    {
+      "epoch": 33.008478260869566,
+      "grad_norm": 89.94793701171875,
+      "learning_rate": 7.270531400966184e-06,
+      "loss": 0.2407,
+      "step": 1590
+    },
+    {
+      "epoch": 33.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 2.8461215496063232,
+      "eval_runtime": 7.0733,
+      "eval_samples_per_second": 3.534,
+      "eval_steps_per_second": 0.707,
+      "step": 1598
+    },
+    {
+      "epoch": 34.00043478260869,
+      "grad_norm": 4.439624309539795,
+      "learning_rate": 7.246376811594203e-06,
+      "loss": 0.1256,
+      "step": 1600
+    },
+    {
+      "epoch": 34.00260869565217,
+      "grad_norm": 7.371945381164551,
+      "learning_rate": 7.222222222222223e-06,
+      "loss": 0.1617,
+      "step": 1610
+    },
+    {
+      "epoch": 34.00478260869565,
+      "grad_norm": 1.0538614988327026,
+      "learning_rate": 7.1980676328502416e-06,
+      "loss": 0.1513,
+      "step": 1620
+    },
+    {
+      "epoch": 34.00695652173913,
+      "grad_norm": 0.032363422214984894,
+      "learning_rate": 7.173913043478261e-06,
+      "loss": 0.1773,
+      "step": 1630
+    },
+    {
+      "epoch": 34.009130434782605,
+      "grad_norm": 14.874954223632812,
+      "learning_rate": 7.149758454106281e-06,
+      "loss": 0.1653,
+      "step": 1640
+    },
+    {
+      "epoch": 34.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 3.173734188079834,
+      "eval_runtime": 7.0852,
+      "eval_samples_per_second": 3.528,
+      "eval_steps_per_second": 0.706,
+      "step": 1645
+    },
+    {
+      "epoch": 35.00108695652174,
+      "grad_norm": 16.16836166381836,
+      "learning_rate": 7.125603864734299e-06,
+      "loss": 0.2382,
+      "step": 1650
+    },
+    {
+      "epoch": 35.00326086956522,
+      "grad_norm": 104.3167953491211,
+      "learning_rate": 7.10144927536232e-06,
+      "loss": 0.2295,
+      "step": 1660
+    },
+    {
+      "epoch": 35.005434782608695,
+      "grad_norm": 0.13320963084697723,
+      "learning_rate": 7.077294685990339e-06,
+      "loss": 0.2243,
+      "step": 1670
+    },
+    {
+      "epoch": 35.00760869565217,
+      "grad_norm": 41.24268341064453,
+      "learning_rate": 7.053140096618359e-06,
+      "loss": 0.1473,
+      "step": 1680
+    },
+    {
+      "epoch": 35.00978260869565,
+      "grad_norm": 0.0757836177945137,
+      "learning_rate": 7.028985507246377e-06,
+      "loss": 0.187,
+      "step": 1690
+    },
+    {
+      "epoch": 35.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 3.564199209213257,
+      "eval_runtime": 7.0337,
+      "eval_samples_per_second": 3.554,
+      "eval_steps_per_second": 0.711,
+      "step": 1692
+    },
+    {
+      "epoch": 36.001739130434785,
+      "grad_norm": 23.14582061767578,
+      "learning_rate": 7.004830917874397e-06,
+      "loss": 0.1639,
+      "step": 1700
+    },
+    {
+      "epoch": 36.00391304347826,
+      "grad_norm": 2.1773264408111572,
+      "learning_rate": 6.980676328502416e-06,
+      "loss": 0.0897,
+      "step": 1710
+    },
+    {
+      "epoch": 36.00608695652174,
+      "grad_norm": 0.5130124688148499,
+      "learning_rate": 6.956521739130435e-06,
+      "loss": 0.0447,
+      "step": 1720
+    },
+    {
+      "epoch": 36.00826086956522,
+      "grad_norm": 0.04473530501127243,
+      "learning_rate": 6.9323671497584545e-06,
+      "loss": 0.2339,
+      "step": 1730
+    },
+    {
+      "epoch": 36.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 3.6020455360412598,
+      "eval_runtime": 7.0891,
+      "eval_samples_per_second": 3.527,
+      "eval_steps_per_second": 0.705,
+      "step": 1739
+    },
+    {
+      "epoch": 37.000217391304346,
+      "grad_norm": 1.4224348068237305,
+      "learning_rate": 6.908212560386473e-06,
+      "loss": 0.0398,
+      "step": 1740
+    },
+    {
+      "epoch": 37.002391304347825,
+      "grad_norm": 141.16114807128906,
+      "learning_rate": 6.884057971014493e-06,
+      "loss": 0.1587,
+      "step": 1750
+    },
+    {
+      "epoch": 37.0045652173913,
+      "grad_norm": 0.31278282403945923,
+      "learning_rate": 6.859903381642513e-06,
+      "loss": 0.1382,
+      "step": 1760
+    },
+    {
+      "epoch": 37.00673913043478,
+      "grad_norm": 0.7922609448432922,
+      "learning_rate": 6.835748792270533e-06,
+      "loss": 0.0916,
+      "step": 1770
+    },
+    {
+      "epoch": 37.00891304347826,
+      "grad_norm": 10.504807472229004,
+      "learning_rate": 6.811594202898551e-06,
+      "loss": 0.1097,
+      "step": 1780
+    },
+    {
+      "epoch": 37.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 3.563100814819336,
+      "eval_runtime": 7.2679,
+      "eval_samples_per_second": 3.44,
+      "eval_steps_per_second": 0.688,
+      "step": 1786
+    },
+    {
+      "epoch": 38.00086956521739,
+      "grad_norm": 0.9079655408859253,
+      "learning_rate": 6.787439613526571e-06,
+      "loss": 0.0146,
+      "step": 1790
+    },
+    {
+      "epoch": 38.00304347826087,
+      "grad_norm": 2.608656167984009,
+      "learning_rate": 6.76328502415459e-06,
+      "loss": 0.1952,
+      "step": 1800
+    },
+    {
+      "epoch": 38.00521739130435,
+      "grad_norm": 0.17260250449180603,
+      "learning_rate": 6.739130434782609e-06,
+      "loss": 0.1533,
+      "step": 1810
+    },
+    {
+      "epoch": 38.00739130434783,
+      "grad_norm": 90.72824096679688,
+      "learning_rate": 6.7149758454106285e-06,
+      "loss": 0.1522,
+      "step": 1820
+    },
+    {
+      "epoch": 38.009565217391305,
+      "grad_norm": 0.20508749783039093,
+      "learning_rate": 6.690821256038648e-06,
+      "loss": 0.2859,
+      "step": 1830
+    },
+    {
+      "epoch": 38.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 3.6048107147216797,
+      "eval_runtime": 7.0008,
+      "eval_samples_per_second": 3.571,
+      "eval_steps_per_second": 0.714,
+      "step": 1833
+    },
+    {
+      "epoch": 39.00152173913043,
+      "grad_norm": 0.336029589176178,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.0454,
+      "step": 1840
+    },
+    {
+      "epoch": 39.00369565217391,
+      "grad_norm": 71.1222915649414,
+      "learning_rate": 6.642512077294686e-06,
+      "loss": 0.0504,
+      "step": 1850
+    },
+    {
+      "epoch": 39.00586956521739,
+      "grad_norm": 0.13150547444820404,
+      "learning_rate": 6.6183574879227065e-06,
+      "loss": 0.065,
+      "step": 1860
+    },
+    {
+      "epoch": 39.008043478260866,
+      "grad_norm": 0.2860020399093628,
+      "learning_rate": 6.594202898550725e-06,
+      "loss": 0.1275,
+      "step": 1870
+    },
+    {
+      "epoch": 39.010217391304344,
+      "grad_norm": 0.018725862726569176,
+      "learning_rate": 6.570048309178745e-06,
+      "loss": 0.0123,
+      "step": 1880
+    },
+    {
+      "epoch": 39.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.202207088470459,
+      "eval_runtime": 7.0352,
+      "eval_samples_per_second": 3.554,
+      "eval_steps_per_second": 0.711,
+      "step": 1880
+    },
+    {
+      "epoch": 40.00217391304348,
+      "grad_norm": 0.2596185803413391,
+      "learning_rate": 6.545893719806764e-06,
+      "loss": 0.0094,
+      "step": 1890
+    },
+    {
+      "epoch": 40.004347826086956,
+      "grad_norm": 29.194820404052734,
+      "learning_rate": 6.521739130434783e-06,
+      "loss": 0.1313,
+      "step": 1900
+    },
+    {
+      "epoch": 40.006521739130434,
+      "grad_norm": 0.1385490745306015,
+      "learning_rate": 6.497584541062802e-06,
+      "loss": 0.1186,
+      "step": 1910
+    },
+    {
+      "epoch": 40.00869565217391,
+      "grad_norm": 0.030303888022899628,
+      "learning_rate": 6.473429951690822e-06,
+      "loss": 0.0062,
+      "step": 1920
+    },
+    {
+      "epoch": 40.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.256433486938477,
+      "eval_runtime": 7.0964,
+      "eval_samples_per_second": 3.523,
+      "eval_steps_per_second": 0.705,
+      "step": 1927
+    },
+    {
+      "epoch": 41.000652173913046,
+      "grad_norm": 0.23668155074119568,
+      "learning_rate": 6.449275362318841e-06,
+      "loss": 0.0072,
+      "step": 1930
+    },
+    {
+      "epoch": 41.002826086956524,
+      "grad_norm": 0.04657760262489319,
+      "learning_rate": 6.42512077294686e-06,
+      "loss": 0.1138,
+      "step": 1940
+    },
+    {
+      "epoch": 41.005,
+      "grad_norm": 0.11511340737342834,
+      "learning_rate": 6.40096618357488e-06,
+      "loss": 0.0256,
+      "step": 1950
+    },
+    {
+      "epoch": 41.00717391304348,
+      "grad_norm": 0.1367231011390686,
+      "learning_rate": 6.376811594202898e-06,
+      "loss": 0.0074,
+      "step": 1960
+    },
+    {
+      "epoch": 41.00934782608696,
+      "grad_norm": 4.625364780426025,
+      "learning_rate": 6.352657004830919e-06,
+      "loss": 0.031,
+      "step": 1970
+    },
+    {
+      "epoch": 41.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.046452045440674,
+      "eval_runtime": 7.054,
+      "eval_samples_per_second": 3.544,
+      "eval_steps_per_second": 0.709,
+      "step": 1974
+    },
+    {
+      "epoch": 42.001304347826085,
+      "grad_norm": 1.0009948015213013,
+      "learning_rate": 6.328502415458938e-06,
+      "loss": 0.0154,
+      "step": 1980
+    },
+    {
+      "epoch": 42.003478260869564,
+      "grad_norm": 0.06303862482309341,
+      "learning_rate": 6.304347826086958e-06,
+      "loss": 0.0264,
+      "step": 1990
+    },
+    {
+      "epoch": 42.00565217391304,
+      "grad_norm": 0.541151225566864,
+      "learning_rate": 6.280193236714976e-06,
+      "loss": 0.1562,
+      "step": 2000
+    },
+    {
+      "epoch": 42.00782608695652,
+      "grad_norm": 0.02539551630616188,
+      "learning_rate": 6.256038647342996e-06,
+      "loss": 0.2316,
+      "step": 2010
+    },
+    {
+      "epoch": 42.01,
+      "grad_norm": 148.5316925048828,
+      "learning_rate": 6.2318840579710145e-06,
+      "loss": 0.1045,
+      "step": 2020
+    },
+    {
+      "epoch": 42.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 3.537886142730713,
+      "eval_runtime": 6.6427,
+      "eval_samples_per_second": 3.764,
+      "eval_steps_per_second": 0.753,
+      "step": 2021
+    },
+    {
+      "epoch": 43.00195652173913,
+      "grad_norm": 0.08496909588575363,
+      "learning_rate": 6.207729468599034e-06,
+      "loss": 0.0055,
+      "step": 2030
+    },
+    {
+      "epoch": 43.00413043478261,
+      "grad_norm": 1.4194090366363525,
+      "learning_rate": 6.1835748792270535e-06,
+      "loss": 0.0936,
+      "step": 2040
+    },
+    {
+      "epoch": 43.00630434782609,
+      "grad_norm": 0.04825194552540779,
+      "learning_rate": 6.159420289855072e-06,
+      "loss": 0.2127,
+      "step": 2050
+    },
+    {
+      "epoch": 43.008478260869566,
+      "grad_norm": 0.02581864595413208,
+      "learning_rate": 6.135265700483092e-06,
+      "loss": 0.0025,
+      "step": 2060
+    },
+    {
+      "epoch": 43.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.188045501708984,
+      "eval_runtime": 6.6377,
+      "eval_samples_per_second": 3.766,
+      "eval_steps_per_second": 0.753,
+      "step": 2068
+    },
+    {
+      "epoch": 44.00043478260869,
+      "grad_norm": 0.600941002368927,
+      "learning_rate": 6.111111111111112e-06,
+      "loss": 0.0245,
+      "step": 2070
+    },
+    {
+      "epoch": 44.00260869565217,
+      "grad_norm": 0.13209854066371918,
+      "learning_rate": 6.086956521739132e-06,
+      "loss": 0.0106,
+      "step": 2080
+    },
+    {
+      "epoch": 44.00478260869565,
+      "grad_norm": 0.04085930436849594,
+      "learning_rate": 6.06280193236715e-06,
+      "loss": 0.0021,
+      "step": 2090
+    },
+    {
+      "epoch": 44.00695652173913,
+      "grad_norm": 4.8549909591674805,
+      "learning_rate": 6.03864734299517e-06,
+      "loss": 0.0027,
+      "step": 2100
+    },
+    {
+      "epoch": 44.009130434782605,
+      "grad_norm": 0.02438848279416561,
+      "learning_rate": 6.014492753623189e-06,
+      "loss": 0.2103,
+      "step": 2110
+    },
+    {
+      "epoch": 44.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.448570251464844,
+      "eval_runtime": 7.0519,
+      "eval_samples_per_second": 3.545,
+      "eval_steps_per_second": 0.709,
+      "step": 2115
+    },
+    {
+      "epoch": 45.00108695652174,
+      "grad_norm": 0.0407257117331028,
+      "learning_rate": 5.990338164251208e-06,
+      "loss": 0.0899,
+      "step": 2120
+    },
+    {
+      "epoch": 45.00326086956522,
+      "grad_norm": 48.495948791503906,
+      "learning_rate": 5.9661835748792275e-06,
+      "loss": 0.0148,
+      "step": 2130
+    },
+    {
+      "epoch": 45.005434782608695,
+      "grad_norm": 84.51445007324219,
+      "learning_rate": 5.942028985507247e-06,
+      "loss": 0.0836,
+      "step": 2140
+    },
+    {
+      "epoch": 45.00760869565217,
+      "grad_norm": 0.004853805061429739,
+      "learning_rate": 5.917874396135266e-06,
+      "loss": 0.0195,
+      "step": 2150
+    },
+    {
+      "epoch": 45.00978260869565,
+      "grad_norm": 0.0302343238145113,
+      "learning_rate": 5.893719806763285e-06,
+      "loss": 0.0035,
+      "step": 2160
+    },
+    {
+      "epoch": 45.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 3.788334846496582,
+      "eval_runtime": 7.0511,
+      "eval_samples_per_second": 3.546,
+      "eval_steps_per_second": 0.709,
+      "step": 2162
+    },
+    {
+      "epoch": 46.001739130434785,
+      "grad_norm": 0.03520100191235542,
+      "learning_rate": 5.8695652173913055e-06,
+      "loss": 0.1397,
+      "step": 2170
+    },
+    {
+      "epoch": 46.00391304347826,
+      "grad_norm": 0.018664751201868057,
+      "learning_rate": 5.845410628019324e-06,
+      "loss": 0.0112,
+      "step": 2180
+    },
+    {
+      "epoch": 46.00608695652174,
+      "grad_norm": 0.010453383438289165,
+      "learning_rate": 5.821256038647344e-06,
+      "loss": 0.1068,
+      "step": 2190
+    },
+    {
+      "epoch": 46.00826086956522,
+      "grad_norm": 32.75054168701172,
+      "learning_rate": 5.797101449275363e-06,
+      "loss": 0.0117,
+      "step": 2200
+    },
+    {
+      "epoch": 46.010217391304344,
+      "eval_accuracy": 0.44,
+      "eval_loss": 3.826441764831543,
+      "eval_runtime": 6.6745,
+      "eval_samples_per_second": 3.746,
+      "eval_steps_per_second": 0.749,
+      "step": 2209
+    },
+    {
+      "epoch": 47.000217391304346,
+      "grad_norm": 0.019730685278773308,
+      "learning_rate": 5.772946859903382e-06,
+      "loss": 0.0026,
+      "step": 2210
+    },
+    {
+      "epoch": 47.002391304347825,
+      "grad_norm": 5.27133846282959,
+      "learning_rate": 5.748792270531401e-06,
+      "loss": 0.0235,
+      "step": 2220
+    },
+    {
+      "epoch": 47.0045652173913,
+      "grad_norm": 0.3507259488105774,
+      "learning_rate": 5.724637681159421e-06,
+      "loss": 0.1714,
+      "step": 2230
+    },
+    {
+      "epoch": 47.00673913043478,
+      "grad_norm": 0.11976295709609985,
+      "learning_rate": 5.70048309178744e-06,
+      "loss": 0.1462,
+      "step": 2240
+    },
+    {
+      "epoch": 47.00891304347826,
+      "grad_norm": 0.026138843968510628,
+      "learning_rate": 5.676328502415459e-06,
+      "loss": 0.0027,
+      "step": 2250
+    },
+    {
+      "epoch": 47.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.2370924949646,
+      "eval_runtime": 6.4755,
+      "eval_samples_per_second": 3.861,
+      "eval_steps_per_second": 0.772,
+      "step": 2256
+    },
+    {
+      "epoch": 48.00086956521739,
+      "grad_norm": 0.014283141121268272,
+      "learning_rate": 5.652173913043479e-06,
+      "loss": 0.0042,
+      "step": 2260
+    },
+    {
+      "epoch": 48.00304347826087,
+      "grad_norm": 0.07310983538627625,
+      "learning_rate": 5.628019323671497e-06,
+      "loss": 0.096,
+      "step": 2270
+    },
+    {
+      "epoch": 48.00521739130435,
+      "grad_norm": 0.008005861192941666,
+      "learning_rate": 5.603864734299518e-06,
+      "loss": 0.0053,
+      "step": 2280
+    },
+    {
+      "epoch": 48.00739130434783,
+      "grad_norm": 0.013425138778984547,
+      "learning_rate": 5.579710144927537e-06,
+      "loss": 0.1492,
+      "step": 2290
+    },
+    {
+      "epoch": 48.009565217391305,
+      "grad_norm": 0.003499469254165888,
+      "learning_rate": 5.555555555555557e-06,
+      "loss": 0.0174,
+      "step": 2300
+    },
+    {
+      "epoch": 48.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.045090675354004,
+      "eval_runtime": 6.6952,
+      "eval_samples_per_second": 3.734,
+      "eval_steps_per_second": 0.747,
+      "step": 2303
+    },
+    {
+      "epoch": 49.00152173913043,
+      "grad_norm": 0.2776978611946106,
+      "learning_rate": 5.531400966183575e-06,
+      "loss": 0.002,
+      "step": 2310
+    },
+    {
+      "epoch": 49.00369565217391,
+      "grad_norm": 0.0137174716219306,
+      "learning_rate": 5.507246376811595e-06,
+      "loss": 0.0032,
+      "step": 2320
+    },
+    {
+      "epoch": 49.00586956521739,
+      "grad_norm": 0.03487164154648781,
+      "learning_rate": 5.4830917874396135e-06,
+      "loss": 0.0018,
+      "step": 2330
+    },
+    {
+      "epoch": 49.008043478260866,
+      "grad_norm": 0.06460028141736984,
+      "learning_rate": 5.458937198067633e-06,
+      "loss": 0.0107,
+      "step": 2340
+    },
+    {
+      "epoch": 49.010217391304344,
+      "grad_norm": 0.02553311362862587,
+      "learning_rate": 5.4347826086956525e-06,
+      "loss": 0.0199,
+      "step": 2350
+    },
+    {
+      "epoch": 49.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.099611282348633,
+      "eval_runtime": 6.8822,
+      "eval_samples_per_second": 3.633,
+      "eval_steps_per_second": 0.727,
+      "step": 2350
+    },
+    {
+      "epoch": 50.00217391304348,
+      "grad_norm": 106.11552429199219,
+      "learning_rate": 5.410628019323671e-06,
+      "loss": 0.0885,
+      "step": 2360
+    },
+    {
+      "epoch": 50.004347826086956,
+      "grad_norm": 23.03952407836914,
+      "learning_rate": 5.386473429951691e-06,
+      "loss": 0.0202,
+      "step": 2370
+    },
+    {
+      "epoch": 50.006521739130434,
+      "grad_norm": 0.009521090425550938,
+      "learning_rate": 5.362318840579711e-06,
+      "loss": 0.0009,
+      "step": 2380
+    },
+    {
+      "epoch": 50.00869565217391,
+      "grad_norm": 0.050981443375349045,
+      "learning_rate": 5.338164251207731e-06,
+      "loss": 0.0082,
+      "step": 2390
+    },
+    {
+      "epoch": 50.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.568161964416504,
+      "eval_runtime": 7.0154,
+      "eval_samples_per_second": 3.564,
+      "eval_steps_per_second": 0.713,
+      "step": 2397
+    },
+    {
+      "epoch": 51.000652173913046,
+      "grad_norm": 0.006439452059566975,
+      "learning_rate": 5.314009661835749e-06,
+      "loss": 0.0692,
+      "step": 2400
+    },
+    {
+      "epoch": 51.002826086956524,
+      "grad_norm": 0.01826869137585163,
+      "learning_rate": 5.289855072463769e-06,
+      "loss": 0.0169,
+      "step": 2410
+    },
+    {
+      "epoch": 51.005,
+      "grad_norm": 0.23446208238601685,
+      "learning_rate": 5.265700483091788e-06,
+      "loss": 0.076,
+      "step": 2420
+    },
+    {
+      "epoch": 51.00717391304348,
+      "grad_norm": 0.11446990817785263,
+      "learning_rate": 5.241545893719807e-06,
+      "loss": 0.0743,
+      "step": 2430
+    },
+    {
+      "epoch": 51.00934782608696,
+      "grad_norm": 148.3246612548828,
+      "learning_rate": 5.2173913043478265e-06,
+      "loss": 0.0186,
+      "step": 2440
+    },
+    {
+      "epoch": 51.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.003583908081055,
+      "eval_runtime": 7.0506,
+      "eval_samples_per_second": 3.546,
+      "eval_steps_per_second": 0.709,
+      "step": 2444
+    },
+    {
+      "epoch": 52.001304347826085,
+      "grad_norm": 0.0546116940677166,
+      "learning_rate": 5.193236714975846e-06,
+      "loss": 0.1326,
+      "step": 2450
+    },
+    {
+      "epoch": 52.003478260869564,
+      "grad_norm": 0.06657025218009949,
+      "learning_rate": 5.169082125603865e-06,
+      "loss": 0.1174,
+      "step": 2460
+    },
+    {
+      "epoch": 52.00565217391304,
+      "grad_norm": 0.014060840010643005,
+      "learning_rate": 5.144927536231884e-06,
+      "loss": 0.0006,
+      "step": 2470
+    },
+    {
+      "epoch": 52.00782608695652,
+      "grad_norm": 0.6909917593002319,
+      "learning_rate": 5.1207729468599045e-06,
+      "loss": 0.0021,
+      "step": 2480
+    },
+    {
+      "epoch": 52.01,
+      "grad_norm": 0.07262730598449707,
+      "learning_rate": 5.096618357487923e-06,
+      "loss": 0.1483,
+      "step": 2490
+    },
+    {
+      "epoch": 52.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 3.8019299507141113,
+      "eval_runtime": 6.9894,
+      "eval_samples_per_second": 3.577,
+      "eval_steps_per_second": 0.715,
+      "step": 2491
+    },
+    {
+      "epoch": 53.00195652173913,
+      "grad_norm": 0.2312413901090622,
+      "learning_rate": 5.072463768115943e-06,
+      "loss": 0.0322,
+      "step": 2500
+    },
+    {
+      "epoch": 53.00413043478261,
+      "grad_norm": 70.85338592529297,
+      "learning_rate": 5.048309178743962e-06,
+      "loss": 0.0395,
+      "step": 2510
+    },
+    {
+      "epoch": 53.00630434782609,
+      "grad_norm": 0.04813679680228233,
+      "learning_rate": 5.024154589371981e-06,
+      "loss": 0.0029,
+      "step": 2520
+    },
+    {
+      "epoch": 53.008478260869566,
+      "grad_norm": 0.16456535458564758,
+      "learning_rate": 5e-06,
+      "loss": 0.1276,
+      "step": 2530
+    },
+    {
+      "epoch": 53.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 3.9253478050231934,
+      "eval_runtime": 6.6132,
+      "eval_samples_per_second": 3.78,
+      "eval_steps_per_second": 0.756,
+      "step": 2538
+    },
+    {
+      "epoch": 54.00043478260869,
+      "grad_norm": 0.041296545416116714,
+      "learning_rate": 4.97584541062802e-06,
+      "loss": 0.0045,
+      "step": 2540
+    },
+    {
+      "epoch": 54.00260869565217,
+      "grad_norm": 0.21262913942337036,
+      "learning_rate": 4.951690821256039e-06,
+      "loss": 0.1289,
+      "step": 2550
+    },
+    {
+      "epoch": 54.00478260869565,
+      "grad_norm": 0.010958844795823097,
+      "learning_rate": 4.927536231884059e-06,
+      "loss": 0.0014,
+      "step": 2560
+    },
+    {
+      "epoch": 54.00695652173913,
+      "grad_norm": 0.20751166343688965,
+      "learning_rate": 4.903381642512078e-06,
+      "loss": 0.0046,
+      "step": 2570
+    },
+    {
+      "epoch": 54.009130434782605,
+      "grad_norm": 0.5311007499694824,
+      "learning_rate": 4.879227053140097e-06,
+      "loss": 0.0601,
+      "step": 2580
+    },
+    {
+      "epoch": 54.010217391304344,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.504699230194092,
+      "eval_runtime": 6.6327,
+      "eval_samples_per_second": 3.769,
+      "eval_steps_per_second": 0.754,
+      "step": 2585
+    },
+    {
+      "epoch": 55.00108695652174,
+      "grad_norm": 0.01531294360756874,
+      "learning_rate": 4.855072463768117e-06,
+      "loss": 0.0067,
+      "step": 2590
+    },
+    {
+      "epoch": 55.00326086956522,
+      "grad_norm": 0.008121052756905556,
+      "learning_rate": 4.830917874396135e-06,
+      "loss": 0.0083,
+      "step": 2600
+    },
+    {
+      "epoch": 55.005434782608695,
+      "grad_norm": 0.031669892370700836,
+      "learning_rate": 4.806763285024155e-06,
+      "loss": 0.0282,
+      "step": 2610
+    },
+    {
+      "epoch": 55.00760869565217,
+      "grad_norm": 0.07723254710435867,
+      "learning_rate": 4.782608695652174e-06,
+      "loss": 0.1465,
+      "step": 2620
+    },
+    {
+      "epoch": 55.00978260869565,
+      "grad_norm": 0.016228657215833664,
+      "learning_rate": 4.758454106280194e-06,
+      "loss": 0.0027,
+      "step": 2630
+    },
+    {
+      "epoch": 55.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.57468318939209,
+      "eval_runtime": 6.9851,
+      "eval_samples_per_second": 3.579,
+      "eval_steps_per_second": 0.716,
+      "step": 2632
+    },
+    {
+      "epoch": 56.001739130434785,
+      "grad_norm": 7.166647911071777,
+      "learning_rate": 4.7342995169082125e-06,
+      "loss": 0.0249,
+      "step": 2640
+    },
+    {
+      "epoch": 56.00391304347826,
+      "grad_norm": 83.3346939086914,
+      "learning_rate": 4.710144927536232e-06,
+      "loss": 0.1309,
+      "step": 2650
+    },
+    {
+      "epoch": 56.00608695652174,
+      "grad_norm": 6.305859565734863,
+      "learning_rate": 4.6859903381642516e-06,
+      "loss": 0.127,
+      "step": 2660
+    },
+    {
+      "epoch": 56.00826086956522,
+      "grad_norm": 18.75379753112793,
+      "learning_rate": 4.661835748792271e-06,
+      "loss": 0.0055,
+      "step": 2670
+    },
+    {
+      "epoch": 56.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.236307144165039,
+      "eval_runtime": 7.0843,
+      "eval_samples_per_second": 3.529,
+      "eval_steps_per_second": 0.706,
+      "step": 2679
+    },
+    {
+      "epoch": 57.000217391304346,
+      "grad_norm": 0.015196056105196476,
+      "learning_rate": 4.637681159420291e-06,
+      "loss": 0.0011,
+      "step": 2680
+    },
+    {
+      "epoch": 57.002391304347825,
+      "grad_norm": 116.98242950439453,
+      "learning_rate": 4.613526570048309e-06,
+      "loss": 0.0195,
+      "step": 2690
+    },
+    {
+      "epoch": 57.0045652173913,
+      "grad_norm": 0.06870768219232559,
+      "learning_rate": 4.589371980676329e-06,
+      "loss": 0.007,
+      "step": 2700
+    },
+    {
+      "epoch": 57.00673913043478,
+      "grad_norm": 0.006763842422515154,
+      "learning_rate": 4.565217391304348e-06,
+      "loss": 0.0006,
+      "step": 2710
+    },
+    {
+      "epoch": 57.00891304347826,
+      "grad_norm": 1.417715311050415,
+      "learning_rate": 4.541062801932368e-06,
+      "loss": 0.0338,
+      "step": 2720
+    },
+    {
+      "epoch": 57.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.3327741622924805,
+      "eval_runtime": 7.0625,
+      "eval_samples_per_second": 3.54,
+      "eval_steps_per_second": 0.708,
+      "step": 2726
+    },
+    {
+      "epoch": 58.00086956521739,
+      "grad_norm": 0.10849138349294662,
+      "learning_rate": 4.516908212560387e-06,
+      "loss": 0.1213,
+      "step": 2730
+    },
+    {
+      "epoch": 58.00304347826087,
+      "grad_norm": 2.3681507110595703,
+      "learning_rate": 4.492753623188406e-06,
+      "loss": 0.0028,
+      "step": 2740
+    },
+    {
+      "epoch": 58.00521739130435,
+      "grad_norm": 184.0817413330078,
+      "learning_rate": 4.4685990338164255e-06,
+      "loss": 0.0685,
+      "step": 2750
+    },
+    {
+      "epoch": 58.00739130434783,
+      "grad_norm": 0.029288914054632187,
+      "learning_rate": 4.444444444444444e-06,
+      "loss": 0.1392,
+      "step": 2760
+    },
+    {
+      "epoch": 58.009565217391305,
+      "grad_norm": 0.048003844916820526,
+      "learning_rate": 4.4202898550724645e-06,
+      "loss": 0.0005,
+      "step": 2770
+    },
+    {
+      "epoch": 58.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.589659214019775,
+      "eval_runtime": 7.0912,
+      "eval_samples_per_second": 3.526,
+      "eval_steps_per_second": 0.705,
+      "step": 2773
+    },
+    {
+      "epoch": 59.00152173913043,
+      "grad_norm": 0.055417075753211975,
+      "learning_rate": 4.396135265700483e-06,
+      "loss": 0.0018,
+      "step": 2780
+    },
+    {
+      "epoch": 59.00369565217391,
+      "grad_norm": 0.017945546656847,
+      "learning_rate": 4.371980676328503e-06,
+      "loss": 0.0234,
+      "step": 2790
+    },
+    {
+      "epoch": 59.00586956521739,
+      "grad_norm": 0.037890467792749405,
+      "learning_rate": 4.347826086956522e-06,
+      "loss": 0.0026,
+      "step": 2800
+    },
+    {
+      "epoch": 59.008043478260866,
+      "grad_norm": 0.01206418126821518,
+      "learning_rate": 4.323671497584541e-06,
+      "loss": 0.0009,
+      "step": 2810
+    },
+    {
+      "epoch": 59.010217391304344,
+      "grad_norm": 0.021620292216539383,
+      "learning_rate": 4.299516908212561e-06,
+      "loss": 0.0489,
+      "step": 2820
+    },
+    {
+      "epoch": 59.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.741189002990723,
+      "eval_runtime": 7.0633,
+      "eval_samples_per_second": 3.539,
+      "eval_steps_per_second": 0.708,
+      "step": 2820
+    },
+    {
+      "epoch": 60.00217391304348,
+      "grad_norm": 1.386226773262024,
+      "learning_rate": 4.27536231884058e-06,
+      "loss": 0.0062,
+      "step": 2830
+    },
+    {
+      "epoch": 60.004347826086956,
+      "grad_norm": 0.0061995419673621655,
+      "learning_rate": 4.251207729468599e-06,
+      "loss": 0.0006,
+      "step": 2840
+    },
+    {
+      "epoch": 60.006521739130434,
+      "grad_norm": 0.011240825988352299,
+      "learning_rate": 4.227053140096619e-06,
+      "loss": 0.0025,
+      "step": 2850
+    },
+    {
+      "epoch": 60.00869565217391,
+      "grad_norm": 0.048650238662958145,
+      "learning_rate": 4.202898550724638e-06,
+      "loss": 0.11,
+      "step": 2860
+    },
+    {
+      "epoch": 60.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.799144744873047,
+      "eval_runtime": 7.2157,
+      "eval_samples_per_second": 3.465,
+      "eval_steps_per_second": 0.693,
+      "step": 2867
+    },
+    {
+      "epoch": 61.000652173913046,
+      "grad_norm": 1.3014930486679077,
+      "learning_rate": 4.178743961352658e-06,
+      "loss": 0.0009,
+      "step": 2870
+    },
+    {
+      "epoch": 61.002826086956524,
+      "grad_norm": 0.007794217672199011,
+      "learning_rate": 4.154589371980677e-06,
+      "loss": 0.1914,
+      "step": 2880
+    },
+    {
+      "epoch": 61.005,
+      "grad_norm": 0.20265735685825348,
+      "learning_rate": 4.130434782608696e-06,
+      "loss": 0.0039,
+      "step": 2890
+    },
+    {
+      "epoch": 61.00717391304348,
+      "grad_norm": 0.044721413403749466,
+      "learning_rate": 4.106280193236716e-06,
+      "loss": 0.037,
+      "step": 2900
+    },
+    {
+      "epoch": 61.00934782608696,
+      "grad_norm": 0.03816564008593559,
+      "learning_rate": 4.082125603864734e-06,
+      "loss": 0.0006,
+      "step": 2910
+    },
+    {
+      "epoch": 61.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.825016021728516,
+      "eval_runtime": 6.9846,
+      "eval_samples_per_second": 3.579,
+      "eval_steps_per_second": 0.716,
+      "step": 2914
+    },
+    {
+      "epoch": 62.001304347826085,
+      "grad_norm": 0.010720742866396904,
+      "learning_rate": 4.057971014492754e-06,
+      "loss": 0.0004,
+      "step": 2920
+    },
+    {
+      "epoch": 62.003478260869564,
+      "grad_norm": 0.023944050073623657,
+      "learning_rate": 4.033816425120773e-06,
+      "loss": 0.0003,
+      "step": 2930
+    },
+    {
+      "epoch": 62.00565217391304,
+      "grad_norm": 0.03520776703953743,
+      "learning_rate": 4.009661835748793e-06,
+      "loss": 0.0406,
+      "step": 2940
+    },
+    {
+      "epoch": 62.00782608695652,
+      "grad_norm": 1.1226750612258911,
+      "learning_rate": 3.9855072463768115e-06,
+      "loss": 0.2626,
+      "step": 2950
+    },
+    {
+      "epoch": 62.01,
+      "grad_norm": 0.09314927458763123,
+      "learning_rate": 3.961352657004831e-06,
+      "loss": 0.0008,
+      "step": 2960
+    },
+    {
+      "epoch": 62.010217391304344,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.756737232208252,
+      "eval_runtime": 7.0215,
+      "eval_samples_per_second": 3.561,
+      "eval_steps_per_second": 0.712,
+      "step": 2961
+    },
+    {
+      "epoch": 63.00195652173913,
+      "grad_norm": 0.03667507320642471,
+      "learning_rate": 3.9371980676328506e-06,
+      "loss": 0.0008,
+      "step": 2970
+    },
+    {
+      "epoch": 63.00413043478261,
+      "grad_norm": 6.374688148498535,
+      "learning_rate": 3.91304347826087e-06,
+      "loss": 0.0011,
+      "step": 2980
+    },
+    {
+      "epoch": 63.00630434782609,
+      "grad_norm": 0.003910560626536608,
+      "learning_rate": 3.88888888888889e-06,
+      "loss": 0.0026,
+      "step": 2990
+    },
+    {
+      "epoch": 63.008478260869566,
+      "grad_norm": 0.005920832511037588,
+      "learning_rate": 3.864734299516908e-06,
+      "loss": 0.0004,
+      "step": 3000
+    },
+    {
+      "epoch": 63.010217391304344,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.486661911010742,
+      "eval_runtime": 6.9991,
+      "eval_samples_per_second": 3.572,
+      "eval_steps_per_second": 0.714,
+      "step": 3008
+    },
+    {
+      "epoch": 64.00043478260869,
+      "grad_norm": 0.026754826307296753,
+      "learning_rate": 3.840579710144928e-06,
+      "loss": 0.0005,
+      "step": 3010
+    },
+    {
+      "epoch": 64.00260869565217,
+      "grad_norm": 0.03842983394861221,
+      "learning_rate": 3.816425120772947e-06,
+      "loss": 0.0625,
+      "step": 3020
+    },
+    {
+      "epoch": 64.00478260869565,
+      "grad_norm": 0.01191799622029066,
+      "learning_rate": 3.792270531400967e-06,
+      "loss": 0.0003,
+      "step": 3030
+    },
+    {
+      "epoch": 64.00695652173913,
+      "grad_norm": 54.559207916259766,
+      "learning_rate": 3.768115942028986e-06,
+      "loss": 0.0043,
+      "step": 3040
+    },
+    {
+      "epoch": 64.0091304347826,
+      "grad_norm": 0.028167851269245148,
+      "learning_rate": 3.743961352657005e-06,
+      "loss": 0.0877,
+      "step": 3050
+    },
+    {
+      "epoch": 64.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.8179755210876465,
+      "eval_runtime": 7.1302,
+      "eval_samples_per_second": 3.506,
+      "eval_steps_per_second": 0.701,
+      "step": 3055
+    },
+    {
+      "epoch": 65.00108695652175,
+      "grad_norm": 0.015800459310412407,
+      "learning_rate": 3.7198067632850245e-06,
+      "loss": 0.0016,
+      "step": 3060
+    },
+    {
+      "epoch": 65.00326086956522,
+      "grad_norm": 0.020454557612538338,
+      "learning_rate": 3.6956521739130436e-06,
+      "loss": 0.1618,
+      "step": 3070
+    },
+    {
+      "epoch": 65.0054347826087,
+      "grad_norm": 0.07312223315238953,
+      "learning_rate": 3.6714975845410635e-06,
+      "loss": 0.0005,
+      "step": 3080
+    },
+    {
+      "epoch": 65.00760869565218,
+      "grad_norm": 5.006317138671875,
+      "learning_rate": 3.6473429951690826e-06,
+      "loss": 0.004,
+      "step": 3090
+    },
+    {
+      "epoch": 65.00978260869566,
+      "grad_norm": 0.7734459042549133,
+      "learning_rate": 3.6231884057971017e-06,
+      "loss": 0.0009,
+      "step": 3100
+    },
+    {
+      "epoch": 65.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.320906162261963,
+      "eval_runtime": 6.6281,
+      "eval_samples_per_second": 3.772,
+      "eval_steps_per_second": 0.754,
+      "step": 3102
+    },
+    {
+      "epoch": 66.00173913043479,
+      "grad_norm": 0.1025114506483078,
+      "learning_rate": 3.5990338164251208e-06,
+      "loss": 0.0005,
+      "step": 3110
+    },
+    {
+      "epoch": 66.00391304347826,
+      "grad_norm": 0.027743550017476082,
+      "learning_rate": 3.5748792270531403e-06,
+      "loss": 0.0005,
+      "step": 3120
+    },
+    {
+      "epoch": 66.00608695652174,
+      "grad_norm": 0.003466652473434806,
+      "learning_rate": 3.55072463768116e-06,
+      "loss": 0.0003,
+      "step": 3130
+    },
+    {
+      "epoch": 66.00826086956522,
+      "grad_norm": 0.00631672004237771,
+      "learning_rate": 3.5265700483091793e-06,
+      "loss": 0.0004,
+      "step": 3140
+    },
+    {
+      "epoch": 66.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.37298583984375,
+      "eval_runtime": 6.6068,
+      "eval_samples_per_second": 3.784,
+      "eval_steps_per_second": 0.757,
+      "step": 3149
+    },
+    {
+      "epoch": 67.00021739130435,
+      "grad_norm": 0.002905324101448059,
+      "learning_rate": 3.5024154589371984e-06,
+      "loss": 0.0009,
+      "step": 3150
+    },
+    {
+      "epoch": 67.00239130434782,
+      "grad_norm": 3.351219654083252,
+      "learning_rate": 3.4782608695652175e-06,
+      "loss": 0.0025,
+      "step": 3160
+    },
+    {
+      "epoch": 67.0045652173913,
+      "grad_norm": 0.009294352494180202,
+      "learning_rate": 3.4541062801932366e-06,
+      "loss": 0.0013,
+      "step": 3170
+    },
+    {
+      "epoch": 67.00673913043478,
+      "grad_norm": 0.06682444363832474,
+      "learning_rate": 3.4299516908212565e-06,
+      "loss": 0.1511,
+      "step": 3180
+    },
+    {
+      "epoch": 67.00891304347826,
+      "grad_norm": 0.029500190168619156,
+      "learning_rate": 3.4057971014492756e-06,
+      "loss": 0.0005,
+      "step": 3190
+    },
+    {
+      "epoch": 67.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.057285308837891,
+      "eval_runtime": 7.0563,
+      "eval_samples_per_second": 3.543,
+      "eval_steps_per_second": 0.709,
+      "step": 3196
+    },
+    {
+      "epoch": 68.00086956521739,
+      "grad_norm": 0.03612295538187027,
+      "learning_rate": 3.381642512077295e-06,
+      "loss": 0.0134,
+      "step": 3200
+    },
+    {
+      "epoch": 68.00304347826086,
+      "grad_norm": 0.020889172330498695,
+      "learning_rate": 3.3574879227053142e-06,
+      "loss": 0.0087,
+      "step": 3210
+    },
+    {
+      "epoch": 68.00521739130434,
+      "grad_norm": 0.01311064325273037,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 0.0013,
+      "step": 3220
+    },
+    {
+      "epoch": 68.00739130434782,
+      "grad_norm": 8.516646385192871,
+      "learning_rate": 3.3091787439613533e-06,
+      "loss": 0.0844,
+      "step": 3230
+    },
+    {
+      "epoch": 68.0095652173913,
+      "grad_norm": 89.14869689941406,
+      "learning_rate": 3.2850241545893724e-06,
+      "loss": 0.0288,
+      "step": 3240
+    },
+    {
+      "epoch": 68.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 3.7278401851654053,
+      "eval_runtime": 7.0444,
+      "eval_samples_per_second": 3.549,
+      "eval_steps_per_second": 0.71,
+      "step": 3243
+    },
+    {
+      "epoch": 69.00152173913044,
+      "grad_norm": 170.97743225097656,
+      "learning_rate": 3.2608695652173914e-06,
+      "loss": 0.1804,
+      "step": 3250
+    },
+    {
+      "epoch": 69.00369565217392,
+      "grad_norm": 0.00593158695846796,
+      "learning_rate": 3.236714975845411e-06,
+      "loss": 0.0004,
+      "step": 3260
+    },
+    {
+      "epoch": 69.0058695652174,
+      "grad_norm": 0.007613878231495619,
+      "learning_rate": 3.21256038647343e-06,
+      "loss": 0.0013,
+      "step": 3270
+    },
+    {
+      "epoch": 69.00804347826087,
+      "grad_norm": 0.07744476944208145,
+      "learning_rate": 3.188405797101449e-06,
+      "loss": 0.0005,
+      "step": 3280
+    },
+    {
+      "epoch": 69.01021739130435,
+      "grad_norm": 0.0068251099437475204,
+      "learning_rate": 3.164251207729469e-06,
+      "loss": 0.0014,
+      "step": 3290
+    },
+    {
+      "epoch": 69.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.968076705932617,
+      "eval_runtime": 6.6127,
+      "eval_samples_per_second": 3.781,
+      "eval_steps_per_second": 0.756,
+      "step": 3290
+    },
+    {
+      "epoch": 70.00217391304348,
+      "grad_norm": 0.017752664163708687,
+      "learning_rate": 3.140096618357488e-06,
+      "loss": 0.0817,
+      "step": 3300
+    },
+    {
+      "epoch": 70.00434782608696,
+      "grad_norm": 0.059641920030117035,
+      "learning_rate": 3.1159420289855073e-06,
+      "loss": 0.0003,
+      "step": 3310
+    },
+    {
+      "epoch": 70.00652173913043,
+      "grad_norm": 0.010968188755214214,
+      "learning_rate": 3.0917874396135268e-06,
+      "loss": 0.035,
+      "step": 3320
+    },
+    {
+      "epoch": 70.00869565217391,
+      "grad_norm": 0.007694718427956104,
+      "learning_rate": 3.067632850241546e-06,
+      "loss": 0.0002,
+      "step": 3330
+    },
+    {
+      "epoch": 70.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.852163314819336,
+      "eval_runtime": 6.6266,
+      "eval_samples_per_second": 3.773,
+      "eval_steps_per_second": 0.755,
+      "step": 3337
+    },
+    {
+      "epoch": 71.00065217391304,
+      "grad_norm": 0.00220879097469151,
+      "learning_rate": 3.043478260869566e-06,
+      "loss": 0.0014,
+      "step": 3340
+    },
+    {
+      "epoch": 71.00282608695652,
+      "grad_norm": 0.0067679788917303085,
+      "learning_rate": 3.019323671497585e-06,
+      "loss": 0.0026,
+      "step": 3350
+    },
+    {
+      "epoch": 71.005,
+      "grad_norm": 0.010681414045393467,
+      "learning_rate": 2.995169082125604e-06,
+      "loss": 0.0004,
+      "step": 3360
+    },
+    {
+      "epoch": 71.00717391304347,
+      "grad_norm": 0.005109126213937998,
+      "learning_rate": 2.9710144927536235e-06,
+      "loss": 0.0309,
+      "step": 3370
+    },
+    {
+      "epoch": 71.00934782608695,
+      "grad_norm": 0.01897754892706871,
+      "learning_rate": 2.9468599033816426e-06,
+      "loss": 0.0009,
+      "step": 3380
+    },
+    {
+      "epoch": 71.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.947040557861328,
+      "eval_runtime": 6.9998,
+      "eval_samples_per_second": 3.572,
+      "eval_steps_per_second": 0.714,
+      "step": 3384
+    },
+    {
+      "epoch": 72.00130434782609,
+      "grad_norm": 0.009784792549908161,
+      "learning_rate": 2.922705314009662e-06,
+      "loss": 0.0004,
+      "step": 3390
+    },
+    {
+      "epoch": 72.00347826086957,
+      "grad_norm": 0.10001827031373978,
+      "learning_rate": 2.8985507246376816e-06,
+      "loss": 0.0003,
+      "step": 3400
+    },
+    {
+      "epoch": 72.00565217391305,
+      "grad_norm": 0.024172531440854073,
+      "learning_rate": 2.8743961352657007e-06,
+      "loss": 0.0006,
+      "step": 3410
+    },
+    {
+      "epoch": 72.00782608695653,
+      "grad_norm": 0.019321994855999947,
+      "learning_rate": 2.85024154589372e-06,
+      "loss": 0.0042,
+      "step": 3420
+    },
+    {
+      "epoch": 72.01,
+      "grad_norm": 0.017360197380185127,
+      "learning_rate": 2.8260869565217393e-06,
+      "loss": 0.0004,
+      "step": 3430
+    },
+    {
+      "epoch": 72.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.870550632476807,
+      "eval_runtime": 6.9868,
+      "eval_samples_per_second": 3.578,
+      "eval_steps_per_second": 0.716,
+      "step": 3431
+    },
+    {
+      "epoch": 73.00195652173913,
+      "grad_norm": 0.18712444603443146,
+      "learning_rate": 2.801932367149759e-06,
+      "loss": 0.0004,
+      "step": 3440
+    },
+    {
+      "epoch": 73.00413043478261,
+      "grad_norm": 0.00983288325369358,
+      "learning_rate": 2.7777777777777783e-06,
+      "loss": 0.0137,
+      "step": 3450
+    },
+    {
+      "epoch": 73.00630434782609,
+      "grad_norm": 0.022100107744336128,
+      "learning_rate": 2.7536231884057974e-06,
+      "loss": 0.0007,
+      "step": 3460
+    },
+    {
+      "epoch": 73.00847826086957,
+      "grad_norm": 0.08435442298650742,
+      "learning_rate": 2.7294685990338165e-06,
+      "loss": 0.0016,
+      "step": 3470
+    },
+    {
+      "epoch": 73.01021739130435,
+      "eval_accuracy": 0.32,
+      "eval_loss": 4.878505229949951,
+      "eval_runtime": 6.7027,
+      "eval_samples_per_second": 3.73,
+      "eval_steps_per_second": 0.746,
+      "step": 3478
+    },
+    {
+      "epoch": 74.00043478260869,
+      "grad_norm": 0.010630447417497635,
+      "learning_rate": 2.7053140096618356e-06,
+      "loss": 0.0002,
+      "step": 3480
+    },
+    {
+      "epoch": 74.00260869565217,
+      "grad_norm": 149.18994140625,
+      "learning_rate": 2.6811594202898555e-06,
+      "loss": 0.166,
+      "step": 3490
+    },
+    {
+      "epoch": 74.00478260869565,
+      "grad_norm": 0.005039245821535587,
+      "learning_rate": 2.6570048309178746e-06,
+      "loss": 0.0004,
+      "step": 3500
+    },
+    {
+      "epoch": 74.00695652173913,
+      "grad_norm": 0.01800839975476265,
+      "learning_rate": 2.632850241545894e-06,
+      "loss": 0.0004,
+      "step": 3510
+    },
+    {
+      "epoch": 74.0091304347826,
+      "grad_norm": 0.013507782481610775,
+      "learning_rate": 2.6086956521739132e-06,
+      "loss": 0.0003,
+      "step": 3520
+    },
+    {
+      "epoch": 74.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.998003959655762,
+      "eval_runtime": 6.6642,
+      "eval_samples_per_second": 3.751,
+      "eval_steps_per_second": 0.75,
+      "step": 3525
+    },
+    {
+      "epoch": 75.00108695652175,
+      "grad_norm": 0.018505793064832687,
+      "learning_rate": 2.5845410628019323e-06,
+      "loss": 0.0005,
+      "step": 3530
+    },
+    {
+      "epoch": 75.00326086956522,
+      "grad_norm": 0.037868183106184006,
+      "learning_rate": 2.5603864734299523e-06,
+      "loss": 0.0004,
+      "step": 3540
+    },
+    {
+      "epoch": 75.0054347826087,
+      "grad_norm": 0.004564860835671425,
+      "learning_rate": 2.5362318840579714e-06,
+      "loss": 0.0924,
+      "step": 3550
+    },
+    {
+      "epoch": 75.00760869565218,
+      "grad_norm": 0.01369723305106163,
+      "learning_rate": 2.5120772946859904e-06,
+      "loss": 0.0003,
+      "step": 3560
+    },
+    {
+      "epoch": 75.00978260869566,
+      "grad_norm": 0.017182549461722374,
+      "learning_rate": 2.48792270531401e-06,
+      "loss": 0.0003,
+      "step": 3570
+    },
+    {
+      "epoch": 75.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.72802209854126,
+      "eval_runtime": 7.0691,
+      "eval_samples_per_second": 3.537,
+      "eval_steps_per_second": 0.707,
+      "step": 3572
+    },
+    {
+      "epoch": 76.00173913043479,
+      "grad_norm": 0.006578261498361826,
+      "learning_rate": 2.4637681159420295e-06,
+      "loss": 0.0017,
+      "step": 3580
+    },
+    {
+      "epoch": 76.00391304347826,
+      "grad_norm": 0.031133858487010002,
+      "learning_rate": 2.4396135265700486e-06,
+      "loss": 0.0038,
+      "step": 3590
+    },
+    {
+      "epoch": 76.00608695652174,
+      "grad_norm": 0.0023055756464600563,
+      "learning_rate": 2.4154589371980677e-06,
+      "loss": 0.0002,
+      "step": 3600
+    },
+    {
+      "epoch": 76.00826086956522,
+      "grad_norm": 0.016801701858639717,
+      "learning_rate": 2.391304347826087e-06,
+      "loss": 0.0003,
+      "step": 3610
+    },
+    {
+      "epoch": 76.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 5.080868721008301,
+      "eval_runtime": 6.2886,
+      "eval_samples_per_second": 3.975,
+      "eval_steps_per_second": 0.795,
+      "step": 3619
+    },
+    {
+      "epoch": 77.00021739130435,
+      "grad_norm": 64.6976318359375,
+      "learning_rate": 2.3671497584541063e-06,
+      "loss": 0.0063,
+      "step": 3620
+    },
+    {
+      "epoch": 77.00239130434782,
+      "grad_norm": 0.6849684119224548,
+      "learning_rate": 2.3429951690821258e-06,
+      "loss": 0.0089,
+      "step": 3630
+    },
+    {
+      "epoch": 77.0045652173913,
+      "grad_norm": 0.018064072355628014,
+      "learning_rate": 2.3188405797101453e-06,
+      "loss": 0.0266,
+      "step": 3640
+    },
+    {
+      "epoch": 77.00673913043478,
+      "grad_norm": 0.07143070548772812,
+      "learning_rate": 2.2946859903381644e-06,
+      "loss": 0.0169,
+      "step": 3650
+    },
+    {
+      "epoch": 77.00891304347826,
+      "grad_norm": 0.04154340177774429,
+      "learning_rate": 2.270531400966184e-06,
+      "loss": 0.0005,
+      "step": 3660
+    },
+    {
+      "epoch": 77.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.811776638031006,
+      "eval_runtime": 6.2616,
+      "eval_samples_per_second": 3.993,
+      "eval_steps_per_second": 0.799,
+      "step": 3666
+    },
+    {
+      "epoch": 78.00086956521739,
+      "grad_norm": 0.016634808853268623,
+      "learning_rate": 2.246376811594203e-06,
+      "loss": 0.0002,
+      "step": 3670
+    },
+    {
+      "epoch": 78.00304347826086,
+      "grad_norm": 0.006576497573405504,
+      "learning_rate": 2.222222222222222e-06,
+      "loss": 0.0003,
+      "step": 3680
+    },
+    {
+      "epoch": 78.00521739130434,
+      "grad_norm": 0.0033253647852689028,
+      "learning_rate": 2.1980676328502416e-06,
+      "loss": 0.0003,
+      "step": 3690
+    },
+    {
+      "epoch": 78.00739130434782,
+      "grad_norm": 0.3980427384376526,
+      "learning_rate": 2.173913043478261e-06,
+      "loss": 0.0003,
+      "step": 3700
+    },
+    {
+      "epoch": 78.0095652173913,
+      "grad_norm": 0.0023840973153710365,
+      "learning_rate": 2.1497584541062806e-06,
+      "loss": 0.0003,
+      "step": 3710
+    },
+    {
+      "epoch": 78.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.743868350982666,
+      "eval_runtime": 6.5926,
+      "eval_samples_per_second": 3.792,
+      "eval_steps_per_second": 0.758,
+      "step": 3713
+    },
+    {
+      "epoch": 79.00152173913044,
+      "grad_norm": 0.004045234993100166,
+      "learning_rate": 2.1256038647342997e-06,
+      "loss": 0.0002,
+      "step": 3720
+    },
+    {
+      "epoch": 79.00369565217392,
+      "grad_norm": 0.019196512177586555,
+      "learning_rate": 2.101449275362319e-06,
+      "loss": 0.0003,
+      "step": 3730
+    },
+    {
+      "epoch": 79.0058695652174,
+      "grad_norm": 0.024642478674650192,
+      "learning_rate": 2.0772946859903383e-06,
+      "loss": 0.0004,
+      "step": 3740
+    },
+    {
+      "epoch": 79.00804347826087,
+      "grad_norm": 0.06350940465927124,
+      "learning_rate": 2.053140096618358e-06,
+      "loss": 0.0029,
+      "step": 3750
+    },
+    {
+      "epoch": 79.01021739130435,
+      "grad_norm": 0.003602983197197318,
+      "learning_rate": 2.028985507246377e-06,
+      "loss": 0.0003,
+      "step": 3760
+    },
+    {
+      "epoch": 79.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.9702677726745605,
+      "eval_runtime": 6.6584,
+      "eval_samples_per_second": 3.755,
+      "eval_steps_per_second": 0.751,
+      "step": 3760
+    },
+    {
+      "epoch": 80.00217391304348,
+      "grad_norm": 4.903309345245361,
+      "learning_rate": 2.0048309178743964e-06,
+      "loss": 0.0099,
+      "step": 3770
+    },
+    {
+      "epoch": 80.00434782608696,
+      "grad_norm": 0.004683238919824362,
+      "learning_rate": 1.9806763285024155e-06,
+      "loss": 0.0004,
+      "step": 3780
+    },
+    {
+      "epoch": 80.00652173913043,
+      "grad_norm": 0.04039665311574936,
+      "learning_rate": 1.956521739130435e-06,
+      "loss": 0.0021,
+      "step": 3790
+    },
+    {
+      "epoch": 80.00869565217391,
+      "grad_norm": 0.008333718404173851,
+      "learning_rate": 1.932367149758454e-06,
+      "loss": 0.0004,
+      "step": 3800
+    },
+    {
+      "epoch": 80.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.565735340118408,
+      "eval_runtime": 6.693,
+      "eval_samples_per_second": 3.735,
+      "eval_steps_per_second": 0.747,
+      "step": 3807
+    },
+    {
+      "epoch": 81.00065217391304,
+      "grad_norm": 0.00347050535492599,
+      "learning_rate": 1.9082125603864736e-06,
+      "loss": 0.0085,
+      "step": 3810
+    },
+    {
+      "epoch": 81.00282608695652,
+      "grad_norm": 0.005219504237174988,
+      "learning_rate": 1.884057971014493e-06,
+      "loss": 0.0003,
+      "step": 3820
+    },
+    {
+      "epoch": 81.005,
+      "grad_norm": 0.053393807262182236,
+      "learning_rate": 1.8599033816425122e-06,
+      "loss": 0.0003,
+      "step": 3830
+    },
+    {
+      "epoch": 81.00717391304347,
+      "grad_norm": 0.061141543090343475,
+      "learning_rate": 1.8357487922705318e-06,
+      "loss": 0.0188,
+      "step": 3840
+    },
+    {
+      "epoch": 81.00934782608695,
+      "grad_norm": 0.15630470216274261,
+      "learning_rate": 1.8115942028985508e-06,
+      "loss": 0.0004,
+      "step": 3850
+    },
+    {
+      "epoch": 81.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.508398532867432,
+      "eval_runtime": 6.6677,
+      "eval_samples_per_second": 3.749,
+      "eval_steps_per_second": 0.75,
+      "step": 3854
+    },
+    {
+      "epoch": 82.00130434782609,
+      "grad_norm": 0.010117909871041775,
+      "learning_rate": 1.7874396135265702e-06,
+      "loss": 0.0006,
+      "step": 3860
+    },
+    {
+      "epoch": 82.00347826086957,
+      "grad_norm": 0.007725616451352835,
+      "learning_rate": 1.7632850241545897e-06,
+      "loss": 0.0002,
+      "step": 3870
+    },
+    {
+      "epoch": 82.00565217391305,
+      "grad_norm": 0.009056983515620232,
+      "learning_rate": 1.7391304347826088e-06,
+      "loss": 0.0162,
+      "step": 3880
+    },
+    {
+      "epoch": 82.00782608695653,
+      "grad_norm": 0.004536367487162352,
+      "learning_rate": 1.7149758454106283e-06,
+      "loss": 0.0002,
+      "step": 3890
+    },
+    {
+      "epoch": 82.01,
+      "grad_norm": 0.0063750832341611385,
+      "learning_rate": 1.6908212560386476e-06,
+      "loss": 0.1261,
+      "step": 3900
+    },
+    {
+      "epoch": 82.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.88843297958374,
+      "eval_runtime": 7.2147,
+      "eval_samples_per_second": 3.465,
+      "eval_steps_per_second": 0.693,
+      "step": 3901
+    },
+    {
+      "epoch": 83.00195652173913,
+      "grad_norm": 0.006049334071576595,
+      "learning_rate": 1.6666666666666667e-06,
+      "loss": 0.0002,
+      "step": 3910
+    },
+    {
+      "epoch": 83.00413043478261,
+      "grad_norm": 0.03713194280862808,
+      "learning_rate": 1.6425120772946862e-06,
+      "loss": 0.0002,
+      "step": 3920
+    },
+    {
+      "epoch": 83.00630434782609,
+      "grad_norm": 0.00830780528485775,
+      "learning_rate": 1.6183574879227055e-06,
+      "loss": 0.0005,
+      "step": 3930
+    },
+    {
+      "epoch": 83.00847826086957,
+      "grad_norm": 0.015072002075612545,
+      "learning_rate": 1.5942028985507246e-06,
+      "loss": 0.0002,
+      "step": 3940
+    },
+    {
+      "epoch": 83.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.8646416664123535,
+      "eval_runtime": 6.611,
+      "eval_samples_per_second": 3.782,
+      "eval_steps_per_second": 0.756,
+      "step": 3948
+    },
+    {
+      "epoch": 84.00043478260869,
+      "grad_norm": 0.01025027222931385,
+      "learning_rate": 1.570048309178744e-06,
+      "loss": 0.0004,
+      "step": 3950
+    },
+    {
+      "epoch": 84.00260869565217,
+      "grad_norm": 0.026901597157120705,
+      "learning_rate": 1.5458937198067634e-06,
+      "loss": 0.0014,
+      "step": 3960
+    },
+    {
+      "epoch": 84.00478260869565,
+      "grad_norm": 0.009291891939938068,
+      "learning_rate": 1.521739130434783e-06,
+      "loss": 0.0011,
+      "step": 3970
+    },
+    {
+      "epoch": 84.00695652173913,
+      "grad_norm": 0.005257267504930496,
+      "learning_rate": 1.497584541062802e-06,
+      "loss": 0.0003,
+      "step": 3980
+    },
+    {
+      "epoch": 84.0091304347826,
+      "grad_norm": 0.002406481420621276,
+      "learning_rate": 1.4734299516908213e-06,
+      "loss": 0.0002,
+      "step": 3990
+    },
+    {
+      "epoch": 84.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.822509765625,
+      "eval_runtime": 6.2159,
+      "eval_samples_per_second": 4.022,
+      "eval_steps_per_second": 0.804,
+      "step": 3995
+    },
+    {
+      "epoch": 85.00108695652175,
+      "grad_norm": 0.0024325798731297255,
+      "learning_rate": 1.4492753623188408e-06,
+      "loss": 0.0003,
+      "step": 4000
+    },
+    {
+      "epoch": 85.00326086956522,
+      "grad_norm": 0.016618061810731888,
+      "learning_rate": 1.42512077294686e-06,
+      "loss": 0.0937,
+      "step": 4010
+    },
+    {
+      "epoch": 85.0054347826087,
+      "grad_norm": 0.0053650373592972755,
+      "learning_rate": 1.4009661835748794e-06,
+      "loss": 0.0003,
+      "step": 4020
+    },
+    {
+      "epoch": 85.00760869565218,
+      "grad_norm": 0.17295417189598083,
+      "learning_rate": 1.3768115942028987e-06,
+      "loss": 0.0002,
+      "step": 4030
+    },
+    {
+      "epoch": 85.00978260869566,
+      "grad_norm": 0.005120884161442518,
+      "learning_rate": 1.3526570048309178e-06,
+      "loss": 0.0003,
+      "step": 4040
+    },
+    {
+      "epoch": 85.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.720521926879883,
+      "eval_runtime": 6.6545,
+      "eval_samples_per_second": 3.757,
+      "eval_steps_per_second": 0.751,
+      "step": 4042
+    },
+    {
+      "epoch": 86.00173913043479,
+      "grad_norm": 0.002048624213784933,
+      "learning_rate": 1.3285024154589373e-06,
+      "loss": 0.0723,
+      "step": 4050
+    },
+    {
+      "epoch": 86.00391304347826,
+      "grad_norm": 39.07980728149414,
+      "learning_rate": 1.3043478260869566e-06,
+      "loss": 0.0039,
+      "step": 4060
+    },
+    {
+      "epoch": 86.00608695652174,
+      "grad_norm": 0.0075403256341814995,
+      "learning_rate": 1.2801932367149761e-06,
+      "loss": 0.0002,
+      "step": 4070
+    },
+    {
+      "epoch": 86.00826086956522,
+      "grad_norm": 0.010615762323141098,
+      "learning_rate": 1.2560386473429952e-06,
+      "loss": 0.0008,
+      "step": 4080
+    },
+    {
+      "epoch": 86.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.78880500793457,
+      "eval_runtime": 7.0889,
+      "eval_samples_per_second": 3.527,
+      "eval_steps_per_second": 0.705,
+      "step": 4089
+    },
+    {
+      "epoch": 87.00021739130435,
+      "grad_norm": 0.002574126934632659,
+      "learning_rate": 1.2318840579710147e-06,
+      "loss": 0.0004,
+      "step": 4090
+    },
+    {
+      "epoch": 87.00239130434782,
+      "grad_norm": 0.08530817180871964,
+      "learning_rate": 1.2077294685990338e-06,
+      "loss": 0.0388,
+      "step": 4100
+    },
+    {
+      "epoch": 87.0045652173913,
+      "grad_norm": 0.002941340673714876,
+      "learning_rate": 1.1835748792270531e-06,
+      "loss": 0.0004,
+      "step": 4110
+    },
+    {
+      "epoch": 87.00673913043478,
+      "grad_norm": 0.0027568840887397528,
+      "learning_rate": 1.1594202898550726e-06,
+      "loss": 0.0013,
+      "step": 4120
+    },
+    {
+      "epoch": 87.00891304347826,
+      "grad_norm": 0.4342869222164154,
+      "learning_rate": 1.135265700483092e-06,
+      "loss": 0.0004,
+      "step": 4130
+    },
+    {
+      "epoch": 87.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.85061502456665,
+      "eval_runtime": 6.6233,
+      "eval_samples_per_second": 3.775,
+      "eval_steps_per_second": 0.755,
+      "step": 4136
+    },
+    {
+      "epoch": 88.00086956521739,
+      "grad_norm": 0.01025103498250246,
+      "learning_rate": 1.111111111111111e-06,
+      "loss": 0.0002,
+      "step": 4140
+    },
+    {
+      "epoch": 88.00304347826086,
+      "grad_norm": 0.002221912145614624,
+      "learning_rate": 1.0869565217391306e-06,
+      "loss": 0.0001,
+      "step": 4150
+    },
+    {
+      "epoch": 88.00521739130434,
+      "grad_norm": 0.008240282535552979,
+      "learning_rate": 1.0628019323671499e-06,
+      "loss": 0.0003,
+      "step": 4160
+    },
+    {
+      "epoch": 88.00739130434782,
+      "grad_norm": 0.0045162090100348,
+      "learning_rate": 1.0386473429951692e-06,
+      "loss": 0.0002,
+      "step": 4170
+    },
+    {
+      "epoch": 88.0095652173913,
+      "grad_norm": 0.006467215716838837,
+      "learning_rate": 1.0144927536231885e-06,
+      "loss": 0.0004,
+      "step": 4180
+    },
+    {
+      "epoch": 88.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.816511631011963,
+      "eval_runtime": 7.0613,
+      "eval_samples_per_second": 3.54,
+      "eval_steps_per_second": 0.708,
+      "step": 4183
+    },
+    {
+      "epoch": 89.00152173913044,
+      "grad_norm": 0.005814757663756609,
+      "learning_rate": 9.903381642512078e-07,
+      "loss": 0.0002,
+      "step": 4190
+    },
+    {
+      "epoch": 89.00369565217392,
+      "grad_norm": 0.013174736872315407,
+      "learning_rate": 9.66183574879227e-07,
+      "loss": 0.0002,
+      "step": 4200
+    },
+    {
+      "epoch": 89.0058695652174,
+      "grad_norm": 0.022736093029379845,
+      "learning_rate": 9.420289855072465e-07,
+      "loss": 0.0012,
+      "step": 4210
+    },
+    {
+      "epoch": 89.00804347826087,
+      "grad_norm": 0.004482312127947807,
+      "learning_rate": 9.178743961352659e-07,
+      "loss": 0.0002,
+      "step": 4220
+    },
+    {
+      "epoch": 89.01021739130435,
+      "grad_norm": 0.004232966806739569,
+      "learning_rate": 8.937198067632851e-07,
+      "loss": 0.0006,
+      "step": 4230
+    },
+    {
+      "epoch": 89.01021739130435,
+      "eval_accuracy": 0.44,
+      "eval_loss": 4.686453819274902,
+      "eval_runtime": 7.0586,
+      "eval_samples_per_second": 3.542,
+      "eval_steps_per_second": 0.708,
+      "step": 4230
+    },
+    {
+      "epoch": 90.00217391304348,
+      "grad_norm": 0.002764122560620308,
+      "learning_rate": 8.695652173913044e-07,
+      "loss": 0.0005,
+      "step": 4240
+    },
+    {
+      "epoch": 90.00434782608696,
+      "grad_norm": 0.012936513870954514,
+      "learning_rate": 8.454106280193238e-07,
+      "loss": 0.0004,
+      "step": 4250
+    },
+    {
+      "epoch": 90.00652173913043,
+      "grad_norm": 0.006648113485425711,
+      "learning_rate": 8.212560386473431e-07,
+      "loss": 0.0002,
+      "step": 4260
+    },
+    {
+      "epoch": 90.00869565217391,
+      "grad_norm": 0.01485658623278141,
+      "learning_rate": 7.971014492753623e-07,
+      "loss": 0.0002,
+      "step": 4270
+    },
+    {
+      "epoch": 90.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.6192193031311035,
+      "eval_runtime": 7.0684,
+      "eval_samples_per_second": 3.537,
+      "eval_steps_per_second": 0.707,
+      "step": 4277
+    },
+    {
+      "epoch": 91.00065217391304,
+      "grad_norm": 0.013752800412476063,
+      "learning_rate": 7.729468599033817e-07,
+      "loss": 0.0002,
+      "step": 4280
+    },
+    {
+      "epoch": 91.00282608695652,
+      "grad_norm": 0.0028934464789927006,
+      "learning_rate": 7.48792270531401e-07,
+      "loss": 0.0002,
+      "step": 4290
+    },
+    {
+      "epoch": 91.005,
+      "grad_norm": 0.011021451093256474,
+      "learning_rate": 7.246376811594204e-07,
+      "loss": 0.0002,
+      "step": 4300
+    },
+    {
+      "epoch": 91.00717391304347,
+      "grad_norm": 0.056588321924209595,
+      "learning_rate": 7.004830917874397e-07,
+      "loss": 0.0075,
+      "step": 4310
+    },
+    {
+      "epoch": 91.00934782608695,
+      "grad_norm": 0.004861388821154833,
+      "learning_rate": 6.763285024154589e-07,
+      "loss": 0.0002,
+      "step": 4320
+    },
+    {
+      "epoch": 91.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.64890193939209,
+      "eval_runtime": 6.6242,
+      "eval_samples_per_second": 3.774,
+      "eval_steps_per_second": 0.755,
+      "step": 4324
+    },
+    {
+      "epoch": 92.00130434782609,
+      "grad_norm": 0.03238126263022423,
+      "learning_rate": 6.521739130434783e-07,
+      "loss": 0.0002,
+      "step": 4330
+    },
+    {
+      "epoch": 92.00347826086957,
+      "grad_norm": 0.006313295103609562,
+      "learning_rate": 6.280193236714976e-07,
+      "loss": 0.0002,
+      "step": 4340
+    },
+    {
+      "epoch": 92.00565217391305,
+      "grad_norm": 0.004186474252492189,
+      "learning_rate": 6.038647342995169e-07,
+      "loss": 0.0003,
+      "step": 4350
+    },
+    {
+      "epoch": 92.00782608695653,
+      "grad_norm": 0.00376342236995697,
+      "learning_rate": 5.797101449275363e-07,
+      "loss": 0.0002,
+      "step": 4360
+    },
+    {
+      "epoch": 92.01,
+      "grad_norm": 0.003788443747907877,
+      "learning_rate": 5.555555555555555e-07,
+      "loss": 0.0005,
+      "step": 4370
+    },
+    {
+      "epoch": 92.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.707370281219482,
+      "eval_runtime": 6.6095,
+      "eval_samples_per_second": 3.782,
+      "eval_steps_per_second": 0.756,
+      "step": 4371
+    },
+    {
+      "epoch": 93.00195652173913,
+      "grad_norm": 0.022369282320141792,
+      "learning_rate": 5.314009661835749e-07,
+      "loss": 0.0002,
+      "step": 4380
+    },
+    {
+      "epoch": 93.00413043478261,
+      "grad_norm": 0.012701062485575676,
+      "learning_rate": 5.072463768115942e-07,
+      "loss": 0.0002,
+      "step": 4390
+    },
+    {
+      "epoch": 93.00630434782609,
+      "grad_norm": 0.007889249362051487,
+      "learning_rate": 4.830917874396135e-07,
+      "loss": 0.0002,
+      "step": 4400
+    },
+    {
+      "epoch": 93.00847826086957,
+      "grad_norm": 0.0016512467991560698,
+      "learning_rate": 4.5893719806763294e-07,
+      "loss": 0.0002,
+      "step": 4410
+    },
+    {
+      "epoch": 93.01021739130435,
+      "eval_accuracy": 0.4,
+      "eval_loss": 4.692601680755615,
+      "eval_runtime": 6.6552,
+      "eval_samples_per_second": 3.756,
+      "eval_steps_per_second": 0.751,
+      "step": 4418
+    },
+    {
+      "epoch": 94.00043478260869,
+      "grad_norm": 0.0019178985385224223,
+      "learning_rate": 4.347826086956522e-07,
+      "loss": 0.0064,
+      "step": 4420
+    },
+    {
+      "epoch": 94.00260869565217,
+      "grad_norm": 0.002023448934778571,
+      "learning_rate": 4.1062801932367154e-07,
+      "loss": 0.0004,
+      "step": 4430
+    },
+    {
+      "epoch": 94.00478260869565,
+      "grad_norm": 0.003523347433656454,
+      "learning_rate": 3.8647342995169085e-07,
+      "loss": 0.0005,
+      "step": 4440
+    },
+    {
+      "epoch": 94.00695652173913,
+      "grad_norm": 0.007735758554190397,
+      "learning_rate": 3.623188405797102e-07,
+      "loss": 0.0096,
+      "step": 4450
+    },
+    {
+      "epoch": 94.0091304347826,
+      "grad_norm": 0.01277866493910551,
+      "learning_rate": 3.3816425120772945e-07,
+      "loss": 0.0012,
+      "step": 4460
+    },
+    {
+      "epoch": 94.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.728867053985596,
+      "eval_runtime": 6.615,
+      "eval_samples_per_second": 3.779,
+      "eval_steps_per_second": 0.756,
+      "step": 4465
+    },
+    {
+      "epoch": 95.00108695652175,
+      "grad_norm": 0.012651154771447182,
+      "learning_rate": 3.140096618357488e-07,
+      "loss": 0.0002,
+      "step": 4470
+    },
+    {
+      "epoch": 95.00326086956522,
+      "grad_norm": 0.005703654605895281,
+      "learning_rate": 2.8985507246376816e-07,
+      "loss": 0.0002,
+      "step": 4480
+    },
+    {
+      "epoch": 95.0054347826087,
+      "grad_norm": 0.001877317321486771,
+      "learning_rate": 2.6570048309178746e-07,
+      "loss": 0.0003,
+      "step": 4490
+    },
+    {
+      "epoch": 95.00760869565218,
+      "grad_norm": 0.0025666586589068174,
+      "learning_rate": 2.4154589371980677e-07,
+      "loss": 0.0002,
+      "step": 4500
+    },
+    {
+      "epoch": 95.00978260869566,
+      "grad_norm": 0.02972640097141266,
+      "learning_rate": 2.173913043478261e-07,
+      "loss": 0.0002,
+      "step": 4510
+    },
+    {
+      "epoch": 95.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.750493049621582,
+      "eval_runtime": 6.723,
+      "eval_samples_per_second": 3.719,
+      "eval_steps_per_second": 0.744,
+      "step": 4512
+    },
+    {
+      "epoch": 96.00173913043479,
+      "grad_norm": 0.0015502575552091002,
+      "learning_rate": 1.9323671497584542e-07,
+      "loss": 0.0002,
+      "step": 4520
+    },
+    {
+      "epoch": 96.00391304347826,
+      "grad_norm": 0.014091862365603447,
+      "learning_rate": 1.6908212560386473e-07,
+      "loss": 0.0003,
+      "step": 4530
+    },
+    {
+      "epoch": 96.00608695652174,
+      "grad_norm": 0.006117947865277529,
+      "learning_rate": 1.4492753623188408e-07,
+      "loss": 0.0001,
+      "step": 4540
+    },
+    {
+      "epoch": 96.00826086956522,
+      "grad_norm": 0.017958518117666245,
+      "learning_rate": 1.2077294685990338e-07,
+      "loss": 0.0002,
+      "step": 4550
+    },
+    {
+      "epoch": 96.01021739130435,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.749838352203369,
+      "eval_runtime": 6.8845,
+      "eval_samples_per_second": 3.631,
+      "eval_steps_per_second": 0.726,
+      "step": 4559
+    },
+    {
+      "epoch": 97.00021739130435,
+      "grad_norm": 0.011473325081169605,
+      "learning_rate": 9.661835748792271e-08,
+      "loss": 0.0002,
+      "step": 4560
+    },
+    {
+      "epoch": 97.00239130434782,
+      "grad_norm": 0.01986193284392357,
+      "learning_rate": 7.246376811594204e-08,
+      "loss": 0.0002,
+      "step": 4570
+    },
+    {
+      "epoch": 97.0045652173913,
+      "grad_norm": 0.005100281443446875,
+      "learning_rate": 4.8309178743961356e-08,
+      "loss": 0.001,
+      "step": 4580
+    },
+    {
+      "epoch": 97.00673913043478,
+      "grad_norm": 0.01051047258079052,
+      "learning_rate": 2.4154589371980678e-08,
+      "loss": 0.0002,
+      "step": 4590
+    },
+    {
+      "epoch": 97.00891304347826,
+      "grad_norm": 0.02039133943617344,
+      "learning_rate": 0.0,
+      "loss": 0.0002,
+      "step": 4600
+    },
+    {
+      "epoch": 97.00891304347826,
+      "eval_accuracy": 0.36,
+      "eval_loss": 4.752323150634766,
+      "eval_runtime": 8.9762,
+      "eval_samples_per_second": 2.785,
+      "eval_steps_per_second": 0.557,
+      "step": 4600
+    },
+    {
+      "epoch": 97.00891304347826,
+      "step": 4600,
+      "total_flos": 1.0056835417566177e+20,
+      "train_loss": 0.23909281312113465,
+      "train_runtime": 13016.5129,
+      "train_samples_per_second": 1.767,
+      "train_steps_per_second": 0.353
+    },
+    {
+      "epoch": 97.00891304347826,
+      "eval_accuracy": 0.44,
+      "eval_loss": 3.826442003250122,
+      "eval_runtime": 7.0086,
+      "eval_samples_per_second": 3.567,
+      "eval_steps_per_second": 0.713,
+      "step": 4600
+    },
+    {
+      "epoch": 97.00891304347826,
+      "eval_accuracy": 0.44,
+      "eval_loss": 3.826441764831543,
+      "eval_runtime": 7.003,
+      "eval_samples_per_second": 3.57,
+      "eval_steps_per_second": 0.714,
+      "step": 4600
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4600,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0056835417566177e+20,
+  "train_batch_size": 5,
+  "trial_name": null,
+  "trial_params": null
+}