diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7193 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.7266530334014996,
+  "eval_steps": 50,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0027266530334014998,
+      "grad_norm": 81.0,
+      "learning_rate": 4.997267759562842e-05,
+      "loss": 3.1321,
+      "step": 1
+    },
+    {
+      "epoch": 0.0054533060668029995,
+      "grad_norm": 10.125,
+      "learning_rate": 4.994535519125683e-05,
+      "loss": 1.2099,
+      "step": 2
+    },
+    {
+      "epoch": 0.0081799591002045,
+      "grad_norm": 4.84375,
+      "learning_rate": 4.991803278688525e-05,
+      "loss": 1.0262,
+      "step": 3
+    },
+    {
+      "epoch": 0.010906612133605999,
+      "grad_norm": 3.828125,
+      "learning_rate": 4.989071038251366e-05,
+      "loss": 1.0011,
+      "step": 4
+    },
+    {
+      "epoch": 0.013633265167007498,
+      "grad_norm": 2.953125,
+      "learning_rate": 4.986338797814208e-05,
+      "loss": 1.0045,
+      "step": 5
+    },
+    {
+      "epoch": 0.016359918200409,
+      "grad_norm": 3.78125,
+      "learning_rate": 4.9836065573770496e-05,
+      "loss": 0.9017,
+      "step": 6
+    },
+    {
+      "epoch": 0.0190865712338105,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.9808743169398906e-05,
+      "loss": 1.0102,
+      "step": 7
+    },
+    {
+      "epoch": 0.021813224267211998,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.978142076502732e-05,
+      "loss": 0.8594,
+      "step": 8
+    },
+    {
+      "epoch": 0.024539877300613498,
+      "grad_norm": 3.28125,
+      "learning_rate": 4.975409836065574e-05,
+      "loss": 0.9831,
+      "step": 9
+    },
+    {
+      "epoch": 0.027266530334014997,
+      "grad_norm": 3.171875,
+      "learning_rate": 4.9726775956284156e-05,
+      "loss": 1.0326,
+      "step": 10
+    },
+    {
+      "epoch": 0.029993183367416496,
+      "grad_norm": 2.5,
+      "learning_rate": 4.969945355191257e-05,
+      "loss": 0.8798,
+      "step": 11
+    },
+    {
+      "epoch": 0.032719836400818,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.967213114754098e-05,
+      "loss": 0.8995,
+      "step": 12
+    },
+    {
+      "epoch": 0.0354464894342195,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.96448087431694e-05,
+      "loss": 0.8703,
+      "step": 13
+    },
+    {
+      "epoch": 0.038173142467621,
+      "grad_norm": 3.109375,
+      "learning_rate": 4.9617486338797816e-05,
+      "loss": 0.8864,
+      "step": 14
+    },
+    {
+      "epoch": 0.0408997955010225,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.959016393442623e-05,
+      "loss": 0.963,
+      "step": 15
+    },
+    {
+      "epoch": 0.043626448534423996,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.956284153005465e-05,
+      "loss": 0.8232,
+      "step": 16
+    },
+    {
+      "epoch": 0.046353101567825496,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.953551912568306e-05,
+      "loss": 0.9968,
+      "step": 17
+    },
+    {
+      "epoch": 0.049079754601226995,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.9508196721311476e-05,
+      "loss": 0.8152,
+      "step": 18
+    },
+    {
+      "epoch": 0.051806407634628494,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.948087431693989e-05,
+      "loss": 0.8783,
+      "step": 19
+    },
+    {
+      "epoch": 0.054533060668029994,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.945355191256831e-05,
+      "loss": 0.7958,
+      "step": 20
+    },
+    {
+      "epoch": 0.05725971370143149,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.9426229508196726e-05,
+      "loss": 0.8389,
+      "step": 21
+    },
+    {
+      "epoch": 0.05998636673483299,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.9398907103825136e-05,
+      "loss": 0.9229,
+      "step": 22
+    },
+    {
+      "epoch": 0.0627130197682345,
+      "grad_norm": 2.375,
+      "learning_rate": 4.937158469945355e-05,
+      "loss": 0.8258,
+      "step": 23
+    },
+    {
+      "epoch": 0.065439672801636,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.934426229508197e-05,
+      "loss": 0.8485,
+      "step": 24
+    },
+    {
+      "epoch": 0.0681663258350375,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.9316939890710386e-05,
+      "loss": 0.7816,
+      "step": 25
+    },
+    {
+      "epoch": 0.070892978868439,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.92896174863388e-05,
+      "loss": 0.8281,
+      "step": 26
+    },
+    {
+      "epoch": 0.0736196319018405,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.926229508196721e-05,
+      "loss": 0.7671,
+      "step": 27
+    },
+    {
+      "epoch": 0.076346284935242,
+      "grad_norm": 3.078125,
+      "learning_rate": 4.923497267759563e-05,
+      "loss": 0.8681,
+      "step": 28
+    },
+    {
+      "epoch": 0.0790729379686435,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.9207650273224046e-05,
+      "loss": 0.827,
+      "step": 29
+    },
+    {
+      "epoch": 0.081799591002045,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.918032786885246e-05,
+      "loss": 0.8596,
+      "step": 30
+    },
+    {
+      "epoch": 0.0845262440354465,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.915300546448088e-05,
+      "loss": 0.7794,
+      "step": 31
+    },
+    {
+      "epoch": 0.08725289706884799,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.912568306010929e-05,
+      "loss": 0.811,
+      "step": 32
+    },
+    {
+      "epoch": 0.08997955010224949,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.9098360655737706e-05,
+      "loss": 0.8135,
+      "step": 33
+    },
+    {
+      "epoch": 0.09270620313565099,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.907103825136612e-05,
+      "loss": 0.7093,
+      "step": 34
+    },
+    {
+      "epoch": 0.09543285616905249,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.904371584699454e-05,
+      "loss": 0.7644,
+      "step": 35
+    },
+    {
+      "epoch": 0.09815950920245399,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.9016393442622957e-05,
+      "loss": 0.8183,
+      "step": 36
+    },
+    {
+      "epoch": 0.10088616223585549,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.8989071038251366e-05,
+      "loss": 0.7157,
+      "step": 37
+    },
+    {
+      "epoch": 0.10361281526925699,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.896174863387978e-05,
+      "loss": 0.7859,
+      "step": 38
+    },
+    {
+      "epoch": 0.10633946830265849,
+      "grad_norm": 3.421875,
+      "learning_rate": 4.89344262295082e-05,
+      "loss": 0.7499,
+      "step": 39
+    },
+    {
+      "epoch": 0.10906612133605999,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.890710382513661e-05,
+      "loss": 0.7009,
+      "step": 40
+    },
+    {
+      "epoch": 0.11179277436946149,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.887978142076503e-05,
+      "loss": 0.74,
+      "step": 41
+    },
+    {
+      "epoch": 0.11451942740286299,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.885245901639344e-05,
+      "loss": 0.8582,
+      "step": 42
+    },
+    {
+      "epoch": 0.11724608043626449,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.882513661202186e-05,
+      "loss": 0.8167,
+      "step": 43
+    },
+    {
+      "epoch": 0.11997273346966598,
+      "grad_norm": 2.125,
+      "learning_rate": 4.879781420765028e-05,
+      "loss": 0.7637,
+      "step": 44
+    },
+    {
+      "epoch": 0.12269938650306748,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.8770491803278687e-05,
+      "loss": 0.7251,
+      "step": 45
+    },
+    {
+      "epoch": 0.125426039536469,
+      "grad_norm": 2.25,
+      "learning_rate": 4.874316939890711e-05,
+      "loss": 0.7049,
+      "step": 46
+    },
+    {
+      "epoch": 0.1281526925698705,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.871584699453552e-05,
+      "loss": 0.7089,
+      "step": 47
+    },
+    {
+      "epoch": 0.130879345603272,
+      "grad_norm": 2.421875,
+      "learning_rate": 4.868852459016394e-05,
+      "loss": 0.7509,
+      "step": 48
+    },
+    {
+      "epoch": 0.1336059986366735,
+      "grad_norm": 2.5,
+      "learning_rate": 4.8661202185792353e-05,
+      "loss": 0.8138,
+      "step": 49
+    },
+    {
+      "epoch": 0.136332651670075,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.863387978142076e-05,
+      "loss": 0.7727,
+      "step": 50
+    },
+    {
+      "epoch": 0.136332651670075,
+      "eval_loss": 0.7231172919273376,
+      "eval_runtime": 10.7453,
+      "eval_samples_per_second": 273.048,
+      "eval_steps_per_second": 6.887,
+      "step": 50
+    },
+    {
+      "epoch": 0.1390593047034765,
+      "grad_norm": 2.25,
+      "learning_rate": 4.860655737704918e-05,
+      "loss": 0.8052,
+      "step": 51
+    },
+    {
+      "epoch": 0.141785957736878,
+      "grad_norm": 2.25,
+      "learning_rate": 4.85792349726776e-05,
+      "loss": 0.661,
+      "step": 52
+    },
+    {
+      "epoch": 0.1445126107702795,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.8551912568306013e-05,
+      "loss": 0.7028,
+      "step": 53
+    },
+    {
+      "epoch": 0.147239263803681,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.852459016393443e-05,
+      "loss": 0.7569,
+      "step": 54
+    },
+    {
+      "epoch": 0.1499659168370825,
+      "grad_norm": 1.9765625,
+      "learning_rate": 4.849726775956284e-05,
+      "loss": 0.6665,
+      "step": 55
+    },
+    {
+      "epoch": 0.152692569870484,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.846994535519126e-05,
+      "loss": 0.6757,
+      "step": 56
+    },
+    {
+      "epoch": 0.1554192229038855,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.844262295081968e-05,
+      "loss": 0.7102,
+      "step": 57
+    },
+    {
+      "epoch": 0.158145875937287,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.841530054644809e-05,
+      "loss": 0.612,
+      "step": 58
+    },
+    {
+      "epoch": 0.1608725289706885,
+      "grad_norm": 2.5,
+      "learning_rate": 4.838797814207651e-05,
+      "loss": 0.7282,
+      "step": 59
+    },
+    {
+      "epoch": 0.16359918200409,
+      "grad_norm": 2.109375,
+      "learning_rate": 4.836065573770492e-05,
+      "loss": 0.6647,
+      "step": 60
+    },
+    {
+      "epoch": 0.1663258350374915,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.8333333333333334e-05,
+      "loss": 0.8071,
+      "step": 61
+    },
+    {
+      "epoch": 0.169052488070893,
+      "grad_norm": 2.25,
+      "learning_rate": 4.830601092896175e-05,
+      "loss": 0.6777,
+      "step": 62
+    },
+    {
+      "epoch": 0.17177914110429449,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.827868852459017e-05,
+      "loss": 0.6732,
+      "step": 63
+    },
+    {
+      "epoch": 0.17450579413769599,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.8251366120218584e-05,
+      "loss": 0.6753,
+      "step": 64
+    },
+    {
+      "epoch": 0.17723244717109748,
+      "grad_norm": 3.421875,
+      "learning_rate": 4.8224043715846994e-05,
+      "loss": 0.6772,
+      "step": 65
+    },
+    {
+      "epoch": 0.17995910020449898,
+      "grad_norm": 2.453125,
+      "learning_rate": 4.819672131147541e-05,
+      "loss": 0.6985,
+      "step": 66
+    },
+    {
+      "epoch": 0.18268575323790048,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.816939890710383e-05,
+      "loss": 0.6628,
+      "step": 67
+    },
+    {
+      "epoch": 0.18541240627130198,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.8142076502732244e-05,
+      "loss": 0.6298,
+      "step": 68
+    },
+    {
+      "epoch": 0.18813905930470348,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.811475409836066e-05,
+      "loss": 0.6232,
+      "step": 69
+    },
+    {
+      "epoch": 0.19086571233810498,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.808743169398907e-05,
+      "loss": 0.6829,
+      "step": 70
+    },
+    {
+      "epoch": 0.19359236537150648,
+      "grad_norm": 1.984375,
+      "learning_rate": 4.806010928961749e-05,
+      "loss": 0.6489,
+      "step": 71
+    },
+    {
+      "epoch": 0.19631901840490798,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.8032786885245904e-05,
+      "loss": 0.6067,
+      "step": 72
+    },
+    {
+      "epoch": 0.19904567143830948,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.800546448087432e-05,
+      "loss": 0.5674,
+      "step": 73
+    },
+    {
+      "epoch": 0.20177232447171098,
+      "grad_norm": 1.96875,
+      "learning_rate": 4.797814207650274e-05,
+      "loss": 0.5357,
+      "step": 74
+    },
+    {
+      "epoch": 0.20449897750511248,
+      "grad_norm": 2.625,
+      "learning_rate": 4.795081967213115e-05,
+      "loss": 0.6547,
+      "step": 75
+    },
+    {
+      "epoch": 0.20722563053851398,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.7923497267759564e-05,
+      "loss": 0.595,
+      "step": 76
+    },
+    {
+      "epoch": 0.20995228357191548,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.789617486338798e-05,
+      "loss": 0.6878,
+      "step": 77
+    },
+    {
+      "epoch": 0.21267893660531698,
+      "grad_norm": 3.125,
+      "learning_rate": 4.78688524590164e-05,
+      "loss": 0.72,
+      "step": 78
+    },
+    {
+      "epoch": 0.21540558963871848,
+      "grad_norm": 2.65625,
+      "learning_rate": 4.7841530054644814e-05,
+      "loss": 0.6089,
+      "step": 79
+    },
+    {
+      "epoch": 0.21813224267211997,
+      "grad_norm": 2.96875,
+      "learning_rate": 4.7814207650273224e-05,
+      "loss": 0.7294,
+      "step": 80
+    },
+    {
+      "epoch": 0.22085889570552147,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.778688524590164e-05,
+      "loss": 0.5832,
+      "step": 81
+    },
+    {
+      "epoch": 0.22358554873892297,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.775956284153006e-05,
+      "loss": 0.6013,
+      "step": 82
+    },
+    {
+      "epoch": 0.22631220177232447,
+      "grad_norm": 2.421875,
+      "learning_rate": 4.7732240437158474e-05,
+      "loss": 0.5792,
+      "step": 83
+    },
+    {
+      "epoch": 0.22903885480572597,
+      "grad_norm": 2.890625,
+      "learning_rate": 4.770491803278689e-05,
+      "loss": 0.586,
+      "step": 84
+    },
+    {
+      "epoch": 0.23176550783912747,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.76775956284153e-05,
+      "loss": 0.5848,
+      "step": 85
+    },
+    {
+      "epoch": 0.23449216087252897,
+      "grad_norm": 2.921875,
+      "learning_rate": 4.765027322404372e-05,
+      "loss": 0.5748,
+      "step": 86
+    },
+    {
+      "epoch": 0.23721881390593047,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.7622950819672134e-05,
+      "loss": 0.6521,
+      "step": 87
+    },
+    {
+      "epoch": 0.23994546693933197,
+      "grad_norm": 2.25,
+      "learning_rate": 4.7595628415300544e-05,
+      "loss": 0.564,
+      "step": 88
+    },
+    {
+      "epoch": 0.24267211997273347,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.756830601092897e-05,
+      "loss": 0.6276,
+      "step": 89
+    },
+    {
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.754098360655738e-05,
+      "loss": 0.5992,
+      "step": 90
+    },
+    {
+      "epoch": 0.24812542603953647,
+      "grad_norm": 2.5,
+      "learning_rate": 4.7513661202185794e-05,
+      "loss": 0.5384,
+      "step": 91
+    },
+    {
+      "epoch": 0.250852079072938,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.748633879781421e-05,
+      "loss": 0.565,
+      "step": 92
+    },
+    {
+      "epoch": 0.25357873210633947,
+      "grad_norm": 2.625,
+      "learning_rate": 4.745901639344262e-05,
+      "loss": 0.6831,
+      "step": 93
+    },
+    {
+      "epoch": 0.256305385139741,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.7431693989071044e-05,
+      "loss": 0.6071,
+      "step": 94
+    },
+    {
+      "epoch": 0.25903203817314246,
+      "grad_norm": 3.453125,
+      "learning_rate": 4.740437158469946e-05,
+      "loss": 0.6907,
+      "step": 95
+    },
+    {
+      "epoch": 0.261758691206544,
+      "grad_norm": 4.5625,
+      "learning_rate": 4.737704918032787e-05,
+      "loss": 0.6481,
+      "step": 96
+    },
+    {
+      "epoch": 0.26448534423994546,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.734972677595629e-05,
+      "loss": 0.5598,
+      "step": 97
+    },
+    {
+      "epoch": 0.267211997273347,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.73224043715847e-05,
+      "loss": 0.5822,
+      "step": 98
+    },
+    {
+      "epoch": 0.26993865030674846,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.7295081967213114e-05,
+      "loss": 0.563,
+      "step": 99
+    },
+    {
+      "epoch": 0.27266530334015,
+      "grad_norm": 3.34375,
+      "learning_rate": 4.726775956284154e-05,
+      "loss": 0.5038,
+      "step": 100
+    },
+    {
+      "epoch": 0.27266530334015,
+      "eval_loss": 0.5729438066482544,
+      "eval_runtime": 10.4575,
+      "eval_samples_per_second": 280.565,
+      "eval_steps_per_second": 7.076,
+      "step": 100
+    },
+    {
+      "epoch": 0.27539195637355146,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.724043715846995e-05,
+      "loss": 0.5735,
+      "step": 101
+    },
+    {
+      "epoch": 0.278118609406953,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.7213114754098365e-05,
+      "loss": 0.5225,
+      "step": 102
+    },
+    {
+      "epoch": 0.28084526244035446,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.7185792349726774e-05,
+      "loss": 0.4861,
+      "step": 103
+    },
+    {
+      "epoch": 0.283571915473756,
+      "grad_norm": 2.5,
+      "learning_rate": 4.715846994535519e-05,
+      "loss": 0.5463,
+      "step": 104
+    },
+    {
+      "epoch": 0.28629856850715746,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.713114754098361e-05,
+      "loss": 0.5383,
+      "step": 105
+    },
+    {
+      "epoch": 0.289025221540559,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.7103825136612025e-05,
+      "loss": 0.5128,
+      "step": 106
+    },
+    {
+      "epoch": 0.29175187457396046,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.707650273224044e-05,
+      "loss": 0.5668,
+      "step": 107
+    },
+    {
+      "epoch": 0.294478527607362,
+      "grad_norm": 2.421875,
+      "learning_rate": 4.704918032786885e-05,
+      "loss": 0.5424,
+      "step": 108
+    },
+    {
+      "epoch": 0.29720518064076346,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.702185792349727e-05,
+      "loss": 0.4634,
+      "step": 109
+    },
+    {
+      "epoch": 0.299931833674165,
+      "grad_norm": 3.078125,
+      "learning_rate": 4.6994535519125685e-05,
+      "loss": 0.635,
+      "step": 110
+    },
+    {
+      "epoch": 0.30265848670756645,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.69672131147541e-05,
+      "loss": 0.5235,
+      "step": 111
+    },
+    {
+      "epoch": 0.305385139740968,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.693989071038252e-05,
+      "loss": 0.4973,
+      "step": 112
+    },
+    {
+      "epoch": 0.30811179277436945,
+      "grad_norm": 2.5,
+      "learning_rate": 4.691256830601093e-05,
+      "loss": 0.5911,
+      "step": 113
+    },
+    {
+      "epoch": 0.310838445807771,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.6885245901639345e-05,
+      "loss": 0.5065,
+      "step": 114
+    },
+    {
+      "epoch": 0.31356509884117245,
+      "grad_norm": 3.703125,
+      "learning_rate": 4.685792349726776e-05,
+      "loss": 0.5406,
+      "step": 115
+    },
+    {
+      "epoch": 0.316291751874574,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.683060109289618e-05,
+      "loss": 0.4668,
+      "step": 116
+    },
+    {
+      "epoch": 0.31901840490797545,
+      "grad_norm": 2.828125,
+      "learning_rate": 4.6803278688524595e-05,
+      "loss": 0.5773,
+      "step": 117
+    },
+    {
+      "epoch": 0.321745057941377,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.6775956284153005e-05,
+      "loss": 0.4382,
+      "step": 118
+    },
+    {
+      "epoch": 0.32447171097477845,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.674863387978142e-05,
+      "loss": 0.5338,
+      "step": 119
+    },
+    {
+      "epoch": 0.32719836400818,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.672131147540984e-05,
+      "loss": 0.4963,
+      "step": 120
+    },
+    {
+      "epoch": 0.32992501704158145,
+      "grad_norm": 2.875,
+      "learning_rate": 4.6693989071038255e-05,
+      "loss": 0.4586,
+      "step": 121
+    },
+    {
+      "epoch": 0.332651670074983,
+      "grad_norm": 3.0625,
+      "learning_rate": 4.666666666666667e-05,
+      "loss": 0.5493,
+      "step": 122
+    },
+    {
+      "epoch": 0.33537832310838445,
+      "grad_norm": 2.953125,
+      "learning_rate": 4.663934426229508e-05,
+      "loss": 0.5659,
+      "step": 123
+    },
+    {
+      "epoch": 0.338104976141786,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.66120218579235e-05,
+      "loss": 0.4902,
+      "step": 124
+    },
+    {
+      "epoch": 0.34083162917518744,
+      "grad_norm": 2.640625,
+      "learning_rate": 4.6584699453551915e-05,
+      "loss": 0.5991,
+      "step": 125
+    },
+    {
+      "epoch": 0.34355828220858897,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.655737704918033e-05,
+      "loss": 0.4989,
+      "step": 126
+    },
+    {
+      "epoch": 0.34628493524199044,
+      "grad_norm": 2.609375,
+      "learning_rate": 4.653005464480875e-05,
+      "loss": 0.517,
+      "step": 127
+    },
+    {
+      "epoch": 0.34901158827539197,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.650273224043716e-05,
+      "loss": 0.5073,
+      "step": 128
+    },
+    {
+      "epoch": 0.35173824130879344,
+      "grad_norm": 2.640625,
+      "learning_rate": 4.6475409836065575e-05,
+      "loss": 0.5392,
+      "step": 129
+    },
+    {
+      "epoch": 0.35446489434219497,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.644808743169399e-05,
+      "loss": 0.5008,
+      "step": 130
+    },
+    {
+      "epoch": 0.35719154737559644,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.64207650273224e-05,
+      "loss": 0.419,
+      "step": 131
+    },
+    {
+      "epoch": 0.35991820040899797,
+      "grad_norm": 2.375,
+      "learning_rate": 4.6393442622950825e-05,
+      "loss": 0.5474,
+      "step": 132
+    },
+    {
+      "epoch": 0.36264485344239944,
+      "grad_norm": 2.046875,
+      "learning_rate": 4.636612021857924e-05,
+      "loss": 0.3931,
+      "step": 133
+    },
+    {
+      "epoch": 0.36537150647580097,
+      "grad_norm": 2.65625,
+      "learning_rate": 4.633879781420765e-05,
+      "loss": 0.5517,
+      "step": 134
+    },
+    {
+      "epoch": 0.36809815950920244,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.631147540983607e-05,
+      "loss": 0.5651,
+      "step": 135
+    },
+    {
+      "epoch": 0.37082481254260397,
+      "grad_norm": 2.9375,
+      "learning_rate": 4.628415300546448e-05,
+      "loss": 0.5045,
+      "step": 136
+    },
+    {
+      "epoch": 0.37355146557600544,
+      "grad_norm": 3.03125,
+      "learning_rate": 4.62568306010929e-05,
+      "loss": 0.4672,
+      "step": 137
+    },
+    {
+      "epoch": 0.37627811860940696,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.622950819672132e-05,
+      "loss": 0.529,
+      "step": 138
+    },
+    {
+      "epoch": 0.37900477164280844,
+      "grad_norm": 3.0,
+      "learning_rate": 4.620218579234973e-05,
+      "loss": 0.4876,
+      "step": 139
+    },
+    {
+      "epoch": 0.38173142467620996,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.6174863387978145e-05,
+      "loss": 0.4881,
+      "step": 140
+    },
+    {
+      "epoch": 0.38445807770961143,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.6147540983606555e-05,
+      "loss": 0.5111,
+      "step": 141
+    },
+    {
+      "epoch": 0.38718473074301296,
+      "grad_norm": 2.25,
+      "learning_rate": 4.612021857923497e-05,
+      "loss": 0.4724,
+      "step": 142
+    },
+    {
+      "epoch": 0.38991138377641443,
+      "grad_norm": 2.625,
+      "learning_rate": 4.6092896174863395e-05,
+      "loss": 0.5158,
+      "step": 143
+    },
+    {
+      "epoch": 0.39263803680981596,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.6065573770491805e-05,
+      "loss": 0.4613,
+      "step": 144
+    },
+    {
+      "epoch": 0.39536468984321743,
+      "grad_norm": 2.5,
+      "learning_rate": 4.603825136612022e-05,
+      "loss": 0.5163,
+      "step": 145
+    },
+    {
+      "epoch": 0.39809134287661896,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.601092896174863e-05,
+      "loss": 0.5,
+      "step": 146
+    },
+    {
+      "epoch": 0.40081799591002043,
+      "grad_norm": 2.65625,
+      "learning_rate": 4.598360655737705e-05,
+      "loss": 0.5184,
+      "step": 147
+    },
+    {
+      "epoch": 0.40354464894342196,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.595628415300547e-05,
+      "loss": 0.3804,
+      "step": 148
+    },
+    {
+      "epoch": 0.40627130197682343,
+      "grad_norm": 2.125,
+      "learning_rate": 4.592896174863388e-05,
+      "loss": 0.4577,
+      "step": 149
+    },
+    {
+      "epoch": 0.40899795501022496,
+      "grad_norm": 2.96875,
+      "learning_rate": 4.59016393442623e-05,
+      "loss": 0.4672,
+      "step": 150
+    },
+    {
+      "epoch": 0.40899795501022496,
+      "eval_loss": 0.46329960227012634,
+      "eval_runtime": 10.4884,
+      "eval_samples_per_second": 279.737,
+      "eval_steps_per_second": 7.055,
+      "step": 150
+    },
+    {
+      "epoch": 0.4117246080436264,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.587431693989071e-05,
+      "loss": 0.455,
+      "step": 151
+    },
+    {
+      "epoch": 0.41445126107702795,
+      "grad_norm": 2.515625,
+      "learning_rate": 4.5846994535519125e-05,
+      "loss": 0.5093,
+      "step": 152
+    },
+    {
+      "epoch": 0.4171779141104294,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.581967213114754e-05,
+      "loss": 0.3648,
+      "step": 153
+    },
+    {
+      "epoch": 0.41990456714383095,
+      "grad_norm": 3.0,
+      "learning_rate": 4.579234972677596e-05,
+      "loss": 0.483,
+      "step": 154
+    },
+    {
+      "epoch": 0.4226312201772324,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.5765027322404376e-05,
+      "loss": 0.4301,
+      "step": 155
+    },
+    {
+      "epoch": 0.42535787321063395,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.5737704918032786e-05,
+      "loss": 0.4387,
+      "step": 156
+    },
+    {
+      "epoch": 0.4280845262440354,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.57103825136612e-05,
+      "loss": 0.4438,
+      "step": 157
+    },
+    {
+      "epoch": 0.43081117927743695,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.568306010928962e-05,
+      "loss": 0.4754,
+      "step": 158
+    },
+    {
+      "epoch": 0.4335378323108384,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.5655737704918036e-05,
+      "loss": 0.4232,
+      "step": 159
+    },
+    {
+      "epoch": 0.43626448534423995,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.562841530054645e-05,
+      "loss": 0.4849,
+      "step": 160
+    },
+    {
+      "epoch": 0.4389911383776414,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.560109289617486e-05,
+      "loss": 0.389,
+      "step": 161
+    },
+    {
+      "epoch": 0.44171779141104295,
+      "grad_norm": 2.484375,
+      "learning_rate": 4.557377049180328e-05,
+      "loss": 0.4942,
+      "step": 162
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.5546448087431696e-05,
+      "loss": 0.4641,
+      "step": 163
+    },
+    {
+      "epoch": 0.44717109747784595,
+      "grad_norm": 2.75,
+      "learning_rate": 4.551912568306011e-05,
+      "loss": 0.4902,
+      "step": 164
+    },
+    {
+      "epoch": 0.4498977505112474,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.549180327868853e-05,
+      "loss": 0.3432,
+      "step": 165
+    },
+    {
+      "epoch": 0.45262440354464895,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.5464480874316946e-05,
+      "loss": 0.5141,
+      "step": 166
+    },
+    {
+      "epoch": 0.4553510565780504,
+      "grad_norm": 2.25,
+      "learning_rate": 4.5437158469945356e-05,
+      "loss": 0.4729,
+      "step": 167
+    },
+    {
+      "epoch": 0.45807770961145194,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.540983606557377e-05,
+      "loss": 0.3816,
+      "step": 168
+    },
+    {
+      "epoch": 0.4608043626448534,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.538251366120219e-05,
+      "loss": 0.3922,
+      "step": 169
+    },
+    {
+      "epoch": 0.46353101567825494,
+      "grad_norm": 3.140625,
+      "learning_rate": 4.5355191256830606e-05,
+      "loss": 0.417,
+      "step": 170
+    },
+    {
+      "epoch": 0.4662576687116564,
+      "grad_norm": 2.609375,
+      "learning_rate": 4.532786885245902e-05,
+      "loss": 0.4133,
+      "step": 171
+    },
+    {
+      "epoch": 0.46898432174505794,
+      "grad_norm": 2.421875,
+      "learning_rate": 4.530054644808743e-05,
+      "loss": 0.4086,
+      "step": 172
+    },
+    {
+      "epoch": 0.47171097477845947,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.527322404371585e-05,
+      "loss": 0.4356,
+      "step": 173
+    },
+    {
+      "epoch": 0.47443762781186094,
+      "grad_norm": 2.890625,
+      "learning_rate": 4.524590163934426e-05,
+      "loss": 0.3996,
+      "step": 174
+    },
+    {
+      "epoch": 0.47716428084526247,
+      "grad_norm": 3.75,
+      "learning_rate": 4.521857923497268e-05,
+      "loss": 0.4559,
+      "step": 175
+    },
+    {
+      "epoch": 0.47989093387866394,
+      "grad_norm": 2.25,
+      "learning_rate": 4.51912568306011e-05,
+      "loss": 0.4188,
+      "step": 176
+    },
+    {
+      "epoch": 0.48261758691206547,
+      "grad_norm": 2.984375,
+      "learning_rate": 4.516393442622951e-05,
+      "loss": 0.4375,
+      "step": 177
+    },
+    {
+      "epoch": 0.48534423994546694,
+      "grad_norm": 2.484375,
+      "learning_rate": 4.5136612021857926e-05,
+      "loss": 0.3962,
+      "step": 178
+    },
+    {
+      "epoch": 0.48807089297886846,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.5109289617486336e-05,
+      "loss": 0.466,
+      "step": 179
+    },
+    {
+      "epoch": 0.49079754601226994,
+      "grad_norm": 2.609375,
+      "learning_rate": 4.508196721311476e-05,
+      "loss": 0.434,
+      "step": 180
+    },
+    {
+      "epoch": 0.49352419904567146,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.5054644808743176e-05,
+      "loss": 0.4035,
+      "step": 181
+    },
+    {
+      "epoch": 0.49625085207907293,
+      "grad_norm": 2.828125,
+      "learning_rate": 4.5027322404371586e-05,
+      "loss": 0.441,
+      "step": 182
+    },
+    {
+      "epoch": 0.49897750511247446,
+      "grad_norm": 2.375,
+      "learning_rate": 4.5e-05,
+      "loss": 0.422,
+      "step": 183
+    },
+    {
+      "epoch": 0.501704158145876,
+      "grad_norm": 2.5,
+      "learning_rate": 4.497267759562841e-05,
+      "loss": 0.465,
+      "step": 184
+    },
+    {
+      "epoch": 0.5044308111792775,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.494535519125683e-05,
+      "loss": 0.5081,
+      "step": 185
+    },
+    {
+      "epoch": 0.5071574642126789,
+      "grad_norm": 1.9375,
+      "learning_rate": 4.491803278688525e-05,
+      "loss": 0.3645,
+      "step": 186
+    },
+    {
+      "epoch": 0.5098841172460804,
+      "grad_norm": 2.625,
+      "learning_rate": 4.489071038251366e-05,
+      "loss": 0.4343,
+      "step": 187
+    },
+    {
+      "epoch": 0.512610770279482,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.486338797814208e-05,
+      "loss": 0.4445,
+      "step": 188
+    },
+    {
+      "epoch": 0.5153374233128835,
+      "grad_norm": 2.375,
+      "learning_rate": 4.483606557377049e-05,
+      "loss": 0.3488,
+      "step": 189
+    },
+    {
+      "epoch": 0.5180640763462849,
+      "grad_norm": 1.8515625,
+      "learning_rate": 4.4808743169398906e-05,
+      "loss": 0.3044,
+      "step": 190
+    },
+    {
+      "epoch": 0.5207907293796864,
+      "grad_norm": 2.25,
+      "learning_rate": 4.478142076502733e-05,
+      "loss": 0.3984,
+      "step": 191
+    },
+    {
+      "epoch": 0.523517382413088,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.475409836065574e-05,
+      "loss": 0.44,
+      "step": 192
+    },
+    {
+      "epoch": 0.5262440354464895,
+      "grad_norm": 3.0625,
+      "learning_rate": 4.4726775956284156e-05,
+      "loss": 0.489,
+      "step": 193
+    },
+    {
+      "epoch": 0.5289706884798909,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.4699453551912566e-05,
+      "loss": 0.4554,
+      "step": 194
+    },
+    {
+      "epoch": 0.5316973415132924,
+      "grad_norm": 2.375,
+      "learning_rate": 4.467213114754098e-05,
+      "loss": 0.4633,
+      "step": 195
+    },
+    {
+      "epoch": 0.534423994546694,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.46448087431694e-05,
+      "loss": 0.3736,
+      "step": 196
+    },
+    {
+      "epoch": 0.5371506475800955,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.4617486338797816e-05,
+      "loss": 0.3669,
+      "step": 197
+    },
+    {
+      "epoch": 0.5398773006134969,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.459016393442623e-05,
+      "loss": 0.3789,
+      "step": 198
+    },
+    {
+      "epoch": 0.5426039536468984,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.456284153005464e-05,
+      "loss": 0.3337,
+      "step": 199
+    },
+    {
+      "epoch": 0.5453306066803,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.453551912568306e-05,
+      "loss": 0.4465,
+      "step": 200
+    },
+    {
+      "epoch": 0.5453306066803,
+      "eval_loss": 0.39008229970932007,
+      "eval_runtime": 10.465,
+      "eval_samples_per_second": 280.362,
+      "eval_steps_per_second": 7.071,
+      "step": 200
+    },
+    {
+      "epoch": 0.5480572597137015,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.4508196721311477e-05,
+      "loss": 0.382,
+      "step": 201
+    },
+    {
+      "epoch": 0.5507839127471029,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.448087431693989e-05,
+      "loss": 0.3955,
+      "step": 202
+    },
+    {
+      "epoch": 0.5535105657805044,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.445355191256831e-05,
+      "loss": 0.3881,
+      "step": 203
+    },
+    {
+      "epoch": 0.556237218813906,
+      "grad_norm": 2.46875,
+      "learning_rate": 4.442622950819673e-05,
+      "loss": 0.4748,
+      "step": 204
+    },
+    {
+      "epoch": 0.5589638718473074,
+      "grad_norm": 2.515625,
+      "learning_rate": 4.4398907103825137e-05,
+      "loss": 0.3953,
+      "step": 205
+    },
+    {
+      "epoch": 0.5616905248807089,
+      "grad_norm": 3.578125,
+      "learning_rate": 4.437158469945355e-05,
+      "loss": 0.4119,
+      "step": 206
+    },
+    {
+      "epoch": 0.5644171779141104,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.434426229508197e-05,
+      "loss": 0.3849,
+      "step": 207
+    },
+    {
+      "epoch": 0.567143830947512,
+      "grad_norm": 2.25,
+      "learning_rate": 4.431693989071039e-05,
+      "loss": 0.3207,
+      "step": 208
+    },
+    {
+      "epoch": 0.5698704839809134,
+      "grad_norm": 2.46875,
+      "learning_rate": 4.4289617486338803e-05,
+      "loss": 0.3867,
+      "step": 209
+    },
+    {
+      "epoch": 0.5725971370143149,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.426229508196721e-05,
+      "loss": 0.3162,
+      "step": 210
+    },
+    {
+      "epoch": 0.5753237900477164,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.423497267759563e-05,
+      "loss": 0.3598,
+      "step": 211
+    },
+    {
+      "epoch": 0.578050443081118,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.420765027322405e-05,
+      "loss": 0.2998,
+      "step": 212
+    },
+    {
+      "epoch": 0.5807770961145194,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.4180327868852463e-05,
+      "loss": 0.3746,
+      "step": 213
+    },
+    {
+      "epoch": 0.5835037491479209,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.415300546448088e-05,
+      "loss": 0.4078,
+      "step": 214
+    },
+    {
+      "epoch": 0.5862304021813224,
+      "grad_norm": 3.15625,
+      "learning_rate": 4.412568306010929e-05,
+      "loss": 0.4022,
+      "step": 215
+    },
+    {
+      "epoch": 0.588957055214724,
+      "grad_norm": 2.75,
+      "learning_rate": 4.409836065573771e-05,
+      "loss": 0.3979,
+      "step": 216
+    },
+    {
+      "epoch": 0.5916837082481254,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.4071038251366124e-05,
+      "loss": 0.403,
+      "step": 217
+    },
+    {
+      "epoch": 0.5944103612815269,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.404371584699454e-05,
+      "loss": 0.4574,
+      "step": 218
+    },
+    {
+      "epoch": 0.5971370143149284,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.401639344262296e-05,
+      "loss": 0.3731,
+      "step": 219
+    },
+    {
+      "epoch": 0.59986366734833,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.398907103825137e-05,
+      "loss": 0.3485,
+      "step": 220
+    },
+    {
+      "epoch": 0.6025903203817314,
+      "grad_norm": 2.125,
+      "learning_rate": 4.3961748633879784e-05,
+      "loss": 0.3154,
+      "step": 221
+    },
+    {
+      "epoch": 0.6053169734151329,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.3934426229508194e-05,
+      "loss": 0.3349,
+      "step": 222
+    },
+    {
+      "epoch": 0.6080436264485344,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.390710382513662e-05,
+      "loss": 0.3175,
+      "step": 223
+    },
+    {
+      "epoch": 0.610770279481936,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.3879781420765034e-05,
+      "loss": 0.3651,
+      "step": 224
+    },
+    {
+      "epoch": 0.6134969325153374,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.3852459016393444e-05,
+      "loss": 0.4127,
+      "step": 225
+    },
+    {
+      "epoch": 0.6162235855487389,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.382513661202186e-05,
+      "loss": 0.3268,
+      "step": 226
+    },
+    {
+      "epoch": 0.6189502385821404,
+      "grad_norm": 2.75,
+      "learning_rate": 4.379781420765027e-05,
+      "loss": 0.4127,
+      "step": 227
+    },
+    {
+      "epoch": 0.621676891615542,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.377049180327869e-05,
+      "loss": 0.3801,
+      "step": 228
+    },
+    {
+      "epoch": 0.6244035446489434,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.374316939890711e-05,
+      "loss": 0.347,
+      "step": 229
+    },
+    {
+      "epoch": 0.6271301976823449,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.371584699453552e-05,
+      "loss": 0.3824,
+      "step": 230
+    },
+    {
+      "epoch": 0.6298568507157464,
+      "grad_norm": 2.65625,
+      "learning_rate": 4.368852459016394e-05,
+      "loss": 0.3975,
+      "step": 231
+    },
+    {
+      "epoch": 0.632583503749148,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.366120218579235e-05,
+      "loss": 0.279,
+      "step": 232
+    },
+    {
+      "epoch": 0.6353101567825494,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.3633879781420764e-05,
+      "loss": 0.3823,
+      "step": 233
+    },
+    {
+      "epoch": 0.6380368098159509,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.360655737704919e-05,
+      "loss": 0.4145,
+      "step": 234
+    },
+    {
+      "epoch": 0.6407634628493524,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.35792349726776e-05,
+      "loss": 0.3667,
+      "step": 235
+    },
+    {
+      "epoch": 0.643490115882754,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.3551912568306014e-05,
+      "loss": 0.3748,
+      "step": 236
+    },
+    {
+      "epoch": 0.6462167689161554,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.3524590163934424e-05,
+      "loss": 0.3707,
+      "step": 237
+    },
+    {
+      "epoch": 0.6489434219495569,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.349726775956284e-05,
+      "loss": 0.4004,
+      "step": 238
+    },
+    {
+      "epoch": 0.6516700749829584,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.346994535519126e-05,
+      "loss": 0.3586,
+      "step": 239
+    },
+    {
+      "epoch": 0.65439672801636,
+      "grad_norm": 3.0,
+      "learning_rate": 4.3442622950819674e-05,
+      "loss": 0.4031,
+      "step": 240
+    },
+    {
+      "epoch": 0.6571233810497614,
+      "grad_norm": 1.75,
+      "learning_rate": 4.341530054644809e-05,
+      "loss": 0.3161,
+      "step": 241
+    },
+    {
+      "epoch": 0.6598500340831629,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.338797814207651e-05,
+      "loss": 0.3081,
+      "step": 242
+    },
+    {
+      "epoch": 0.6625766871165644,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.336065573770492e-05,
+      "loss": 0.3311,
+      "step": 243
+    },
+    {
+      "epoch": 0.665303340149966,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.3333333333333334e-05,
+      "loss": 0.373,
+      "step": 244
+    },
+    {
+      "epoch": 0.6680299931833674,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.330601092896175e-05,
+      "loss": 0.2988,
+      "step": 245
+    },
+    {
+      "epoch": 0.6707566462167689,
+      "grad_norm": 2.453125,
+      "learning_rate": 4.327868852459017e-05,
+      "loss": 0.3783,
+      "step": 246
+    },
+    {
+      "epoch": 0.6734832992501704,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.3251366120218584e-05,
+      "loss": 0.3752,
+      "step": 247
+    },
+    {
+      "epoch": 0.676209952283572,
+      "grad_norm": 3.09375,
+      "learning_rate": 4.3224043715846994e-05,
+      "loss": 0.3207,
+      "step": 248
+    },
+    {
+      "epoch": 0.6789366053169734,
+      "grad_norm": 2.125,
+      "learning_rate": 4.319672131147541e-05,
+      "loss": 0.3463,
+      "step": 249
+    },
+    {
+      "epoch": 0.6816632583503749,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.316939890710383e-05,
+      "loss": 0.3168,
+      "step": 250
+    },
+    {
+      "epoch": 0.6816632583503749,
+      "eval_loss": 0.345150887966156,
+      "eval_runtime": 10.4736,
+      "eval_samples_per_second": 280.134,
+      "eval_steps_per_second": 7.065,
+      "step": 250
+    },
+    {
+      "epoch": 0.6843899113837764,
+      "grad_norm": 2.515625,
+      "learning_rate": 4.3142076502732244e-05,
+      "loss": 0.4059,
+      "step": 251
+    },
+    {
+      "epoch": 0.6871165644171779,
+      "grad_norm": 2.453125,
+      "learning_rate": 4.311475409836066e-05,
+      "loss": 0.3459,
+      "step": 252
+    },
+    {
+      "epoch": 0.6898432174505794,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.308743169398907e-05,
+      "loss": 0.3831,
+      "step": 253
+    },
+    {
+      "epoch": 0.6925698704839809,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.306010928961749e-05,
+      "loss": 0.3112,
+      "step": 254
+    },
+    {
+      "epoch": 0.6952965235173824,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.3032786885245904e-05,
+      "loss": 0.3744,
+      "step": 255
+    },
+    {
+      "epoch": 0.6980231765507839,
+      "grad_norm": 1.8359375,
+      "learning_rate": 4.300546448087432e-05,
+      "loss": 0.3333,
+      "step": 256
+    },
+    {
+      "epoch": 0.7007498295841854,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.297814207650274e-05,
+      "loss": 0.3659,
+      "step": 257
+    },
+    {
+      "epoch": 0.7034764826175869,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.295081967213115e-05,
+      "loss": 0.3733,
+      "step": 258
+    },
+    {
+      "epoch": 0.7062031356509885,
+      "grad_norm": 1.765625,
+      "learning_rate": 4.2923497267759564e-05,
+      "loss": 0.2948,
+      "step": 259
+    },
+    {
+      "epoch": 0.7089297886843899,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.289617486338798e-05,
+      "loss": 0.3185,
+      "step": 260
+    },
+    {
+      "epoch": 0.7116564417177914,
+      "grad_norm": 1.96875,
+      "learning_rate": 4.28688524590164e-05,
+      "loss": 0.3239,
+      "step": 261
+    },
+    {
+      "epoch": 0.7143830947511929,
+      "grad_norm": 2.0625,
+      "learning_rate": 4.2841530054644815e-05,
+      "loss": 0.3663,
+      "step": 262
+    },
+    {
+      "epoch": 0.7171097477845945,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.2814207650273224e-05,
+      "loss": 0.3465,
+      "step": 263
+    },
+    {
+      "epoch": 0.7198364008179959,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.278688524590164e-05,
+      "loss": 0.2601,
+      "step": 264
+    },
+    {
+      "epoch": 0.7225630538513974,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.275956284153005e-05,
+      "loss": 0.3734,
+      "step": 265
+    },
+    {
+      "epoch": 0.7252897068847989,
+      "grad_norm": 2.046875,
+      "learning_rate": 4.2732240437158475e-05,
+      "loss": 0.2817,
+      "step": 266
+    },
+    {
+      "epoch": 0.7280163599182005,
+      "grad_norm": 2.875,
+      "learning_rate": 4.270491803278689e-05,
+      "loss": 0.3996,
+      "step": 267
+    },
+    {
+      "epoch": 0.7307430129516019,
+      "grad_norm": 2.0,
+      "learning_rate": 4.26775956284153e-05,
+      "loss": 0.3014,
+      "step": 268
+    },
+    {
+      "epoch": 0.7334696659850034,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.265027322404372e-05,
+      "loss": 0.3414,
+      "step": 269
+    },
+    {
+      "epoch": 0.7361963190184049,
+      "grad_norm": 2.59375,
+      "learning_rate": 4.262295081967213e-05,
+      "loss": 0.3217,
+      "step": 270
+    },
+    {
+      "epoch": 0.7389229720518065,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.259562841530055e-05,
+      "loss": 0.3516,
+      "step": 271
+    },
+    {
+      "epoch": 0.7416496250852079,
+      "grad_norm": 2.25,
+      "learning_rate": 4.256830601092897e-05,
+      "loss": 0.35,
+      "step": 272
+    },
+    {
+      "epoch": 0.7443762781186094,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.254098360655738e-05,
+      "loss": 0.2774,
+      "step": 273
+    },
+    {
+      "epoch": 0.7471029311520109,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.2513661202185795e-05,
+      "loss": 0.3698,
+      "step": 274
+    },
+    {
+      "epoch": 0.7498295841854125,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.248633879781421e-05,
+      "loss": 0.4043,
+      "step": 275
+    },
+    {
+      "epoch": 0.7525562372188139,
+      "grad_norm": 2.5,
+      "learning_rate": 4.245901639344262e-05,
+      "loss": 0.3422,
+      "step": 276
+    },
+    {
+      "epoch": 0.7552828902522154,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.2431693989071045e-05,
+      "loss": 0.3581,
+      "step": 277
+    },
+    {
+      "epoch": 0.7580095432856169,
+      "grad_norm": 1.9296875,
+      "learning_rate": 4.2404371584699455e-05,
+      "loss": 0.3219,
+      "step": 278
+    },
+    {
+      "epoch": 0.7607361963190185,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.237704918032787e-05,
+      "loss": 0.3004,
+      "step": 279
+    },
+    {
+      "epoch": 0.7634628493524199,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.234972677595629e-05,
+      "loss": 0.3274,
+      "step": 280
+    },
+    {
+      "epoch": 0.7661895023858214,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.23224043715847e-05,
+      "loss": 0.3725,
+      "step": 281
+    },
+    {
+      "epoch": 0.7689161554192229,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.229508196721312e-05,
+      "loss": 0.3143,
+      "step": 282
+    },
+    {
+      "epoch": 0.7716428084526245,
+      "grad_norm": 1.875,
+      "learning_rate": 4.226775956284153e-05,
+      "loss": 0.3363,
+      "step": 283
+    },
+    {
+      "epoch": 0.7743694614860259,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.224043715846995e-05,
+      "loss": 0.3279,
+      "step": 284
+    },
+    {
+      "epoch": 0.7770961145194274,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.2213114754098365e-05,
+      "loss": 0.323,
+      "step": 285
+    },
+    {
+      "epoch": 0.7798227675528289,
+      "grad_norm": 2.0,
+      "learning_rate": 4.2185792349726775e-05,
+      "loss": 0.2939,
+      "step": 286
+    },
+    {
+      "epoch": 0.7825494205862304,
+      "grad_norm": 8.1875,
+      "learning_rate": 4.215846994535519e-05,
+      "loss": 0.374,
+      "step": 287
+    },
+    {
+      "epoch": 0.7852760736196319,
+      "grad_norm": 1.859375,
+      "learning_rate": 4.213114754098361e-05,
+      "loss": 0.2506,
+      "step": 288
+    },
+    {
+      "epoch": 0.7880027266530334,
+      "grad_norm": 2.09375,
+      "learning_rate": 4.2103825136612025e-05,
+      "loss": 0.3148,
+      "step": 289
+    },
+    {
+      "epoch": 0.7907293796864349,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.207650273224044e-05,
+      "loss": 0.3724,
+      "step": 290
+    },
+    {
+      "epoch": 0.7934560327198364,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.204918032786885e-05,
+      "loss": 0.3193,
+      "step": 291
+    },
+    {
+      "epoch": 0.7961826857532379,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.202185792349727e-05,
+      "loss": 0.3298,
+      "step": 292
+    },
+    {
+      "epoch": 0.7989093387866394,
+      "grad_norm": 1.9375,
+      "learning_rate": 4.1994535519125685e-05,
+      "loss": 0.3128,
+      "step": 293
+    },
+    {
+      "epoch": 0.8016359918200409,
+      "grad_norm": 1.9921875,
+      "learning_rate": 4.19672131147541e-05,
+      "loss": 0.3287,
+      "step": 294
+    },
+    {
+      "epoch": 0.8043626448534424,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.193989071038252e-05,
+      "loss": 0.3946,
+      "step": 295
+    },
+    {
+      "epoch": 0.8070892978868439,
+      "grad_norm": 1.96875,
+      "learning_rate": 4.191256830601093e-05,
+      "loss": 0.291,
+      "step": 296
+    },
+    {
+      "epoch": 0.8098159509202454,
+      "grad_norm": 1.8828125,
+      "learning_rate": 4.1885245901639345e-05,
+      "loss": 0.3178,
+      "step": 297
+    },
+    {
+      "epoch": 0.8125426039536469,
+      "grad_norm": 2.0,
+      "learning_rate": 4.185792349726776e-05,
+      "loss": 0.3077,
+      "step": 298
+    },
+    {
+      "epoch": 0.8152692569870484,
+      "grad_norm": 2.0,
+      "learning_rate": 4.183060109289618e-05,
+      "loss": 0.2937,
+      "step": 299
+    },
+    {
+      "epoch": 0.8179959100204499,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.1803278688524595e-05,
+      "loss": 0.2838,
+      "step": 300
+    },
+    {
+      "epoch": 0.8179959100204499,
+      "eval_loss": 0.3145870566368103,
+      "eval_runtime": 10.4998,
+      "eval_samples_per_second": 279.435,
+      "eval_steps_per_second": 7.048,
+      "step": 300
+    },
+    {
+      "epoch": 0.8207225630538514,
+      "grad_norm": 1.8671875,
+      "learning_rate": 4.1775956284153005e-05,
+      "loss": 0.3108,
+      "step": 301
+    },
+    {
+      "epoch": 0.8234492160872529,
+      "grad_norm": 1.90625,
+      "learning_rate": 4.174863387978142e-05,
+      "loss": 0.2609,
+      "step": 302
+    },
+    {
+      "epoch": 0.8261758691206544,
+      "grad_norm": 2.125,
+      "learning_rate": 4.172131147540984e-05,
+      "loss": 0.3314,
+      "step": 303
+    },
+    {
+      "epoch": 0.8289025221540559,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.1693989071038255e-05,
+      "loss": 0.3246,
+      "step": 304
+    },
+    {
+      "epoch": 0.8316291751874574,
+      "grad_norm": 1.8203125,
+      "learning_rate": 4.166666666666667e-05,
+      "loss": 0.3313,
+      "step": 305
+    },
+    {
+      "epoch": 0.8343558282208589,
+      "grad_norm": 1.921875,
+      "learning_rate": 4.163934426229508e-05,
+      "loss": 0.2834,
+      "step": 306
+    },
+    {
+      "epoch": 0.8370824812542604,
+      "grad_norm": 1.984375,
+      "learning_rate": 4.16120218579235e-05,
+      "loss": 0.3551,
+      "step": 307
+    },
+    {
+      "epoch": 0.8398091342876619,
+      "grad_norm": 1.90625,
+      "learning_rate": 4.158469945355191e-05,
+      "loss": 0.3313,
+      "step": 308
+    },
+    {
+      "epoch": 0.8425357873210634,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.155737704918033e-05,
+      "loss": 0.2649,
+      "step": 309
+    },
+    {
+      "epoch": 0.8452624403544649,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.153005464480875e-05,
+      "loss": 0.3665,
+      "step": 310
+    },
+    {
+      "epoch": 0.8479890933878664,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.150273224043716e-05,
+      "loss": 0.3643,
+      "step": 311
+    },
+    {
+      "epoch": 0.8507157464212679,
+      "grad_norm": 2.25,
+      "learning_rate": 4.1475409836065575e-05,
+      "loss": 0.3363,
+      "step": 312
+    },
+    {
+      "epoch": 0.8534423994546694,
+      "grad_norm": 1.9765625,
+      "learning_rate": 4.144808743169399e-05,
+      "loss": 0.3337,
+      "step": 313
+    },
+    {
+      "epoch": 0.8561690524880708,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.142076502732241e-05,
+      "loss": 0.3088,
+      "step": 314
+    },
+    {
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.78125,
+      "learning_rate": 4.1393442622950826e-05,
+      "loss": 0.3007,
+      "step": 315
+    },
+    {
+      "epoch": 0.8616223585548739,
+      "grad_norm": 2.84375,
+      "learning_rate": 4.1366120218579236e-05,
+      "loss": 0.3634,
+      "step": 316
+    },
+    {
+      "epoch": 0.8643490115882754,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.133879781420765e-05,
+      "loss": 0.3111,
+      "step": 317
+    },
+    {
+      "epoch": 0.8670756646216768,
+      "grad_norm": 2.125,
+      "learning_rate": 4.131147540983607e-05,
+      "loss": 0.3224,
+      "step": 318
+    },
+    {
+      "epoch": 0.8698023176550784,
+      "grad_norm": 1.9765625,
+      "learning_rate": 4.128415300546448e-05,
+      "loss": 0.3059,
+      "step": 319
+    },
+    {
+      "epoch": 0.8725289706884799,
+      "grad_norm": 2.375,
+      "learning_rate": 4.12568306010929e-05,
+      "loss": 0.3294,
+      "step": 320
+    },
+    {
+      "epoch": 0.8752556237218814,
+      "grad_norm": 2.34375,
+      "learning_rate": 4.122950819672131e-05,
+      "loss": 0.3469,
+      "step": 321
+    },
+    {
+      "epoch": 0.8779822767552828,
+      "grad_norm": 1.9765625,
+      "learning_rate": 4.120218579234973e-05,
+      "loss": 0.3265,
+      "step": 322
+    },
+    {
+      "epoch": 0.8807089297886844,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.1174863387978146e-05,
+      "loss": 0.3285,
+      "step": 323
+    },
+    {
+      "epoch": 0.8834355828220859,
+      "grad_norm": 1.9921875,
+      "learning_rate": 4.1147540983606556e-05,
+      "loss": 0.2867,
+      "step": 324
+    },
+    {
+      "epoch": 0.8861622358554874,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.112021857923498e-05,
+      "loss": 0.3807,
+      "step": 325
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 1.828125,
+      "learning_rate": 4.109289617486339e-05,
+      "loss": 0.2742,
+      "step": 326
+    },
+    {
+      "epoch": 0.8916155419222904,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.1065573770491806e-05,
+      "loss": 0.3621,
+      "step": 327
+    },
+    {
+      "epoch": 0.8943421949556919,
+      "grad_norm": 2.125,
+      "learning_rate": 4.103825136612022e-05,
+      "loss": 0.3563,
+      "step": 328
+    },
+    {
+      "epoch": 0.8970688479890934,
+      "grad_norm": 1.90625,
+      "learning_rate": 4.101092896174863e-05,
+      "loss": 0.3201,
+      "step": 329
+    },
+    {
+      "epoch": 0.8997955010224948,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.098360655737705e-05,
+      "loss": 0.3283,
+      "step": 330
+    },
+    {
+      "epoch": 0.9025221540558964,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.0956284153005466e-05,
+      "loss": 0.4133,
+      "step": 331
+    },
+    {
+      "epoch": 0.9052488070892979,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.092896174863388e-05,
+      "loss": 0.3423,
+      "step": 332
+    },
+    {
+      "epoch": 0.9079754601226994,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.09016393442623e-05,
+      "loss": 0.2898,
+      "step": 333
+    },
+    {
+      "epoch": 0.9107021131561008,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.087431693989071e-05,
+      "loss": 0.3519,
+      "step": 334
+    },
+    {
+      "epoch": 0.9134287661895024,
+      "grad_norm": 2.421875,
+      "learning_rate": 4.0846994535519126e-05,
+      "loss": 0.2869,
+      "step": 335
+    },
+    {
+      "epoch": 0.9161554192229039,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.081967213114754e-05,
+      "loss": 0.3479,
+      "step": 336
+    },
+    {
+      "epoch": 0.9188820722563054,
+      "grad_norm": 1.921875,
+      "learning_rate": 4.079234972677596e-05,
+      "loss": 0.2223,
+      "step": 337
+    },
+    {
+      "epoch": 0.9216087252897068,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.0765027322404376e-05,
+      "loss": 0.2925,
+      "step": 338
+    },
+    {
+      "epoch": 0.9243353783231084,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.0737704918032786e-05,
+      "loss": 0.2872,
+      "step": 339
+    },
+    {
+      "epoch": 0.9270620313565099,
+      "grad_norm": 1.7421875,
+      "learning_rate": 4.07103825136612e-05,
+      "loss": 0.2443,
+      "step": 340
+    },
+    {
+      "epoch": 0.9297886843899114,
+      "grad_norm": 1.875,
+      "learning_rate": 4.068306010928962e-05,
+      "loss": 0.282,
+      "step": 341
+    },
+    {
+      "epoch": 0.9325153374233128,
+      "grad_norm": 1.703125,
+      "learning_rate": 4.0655737704918036e-05,
+      "loss": 0.2714,
+      "step": 342
+    },
+    {
+      "epoch": 0.9352419904567144,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.062841530054645e-05,
+      "loss": 0.2933,
+      "step": 343
+    },
+    {
+      "epoch": 0.9379686434901159,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.060109289617486e-05,
+      "loss": 0.3431,
+      "step": 344
+    },
+    {
+      "epoch": 0.9406952965235174,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.057377049180328e-05,
+      "loss": 0.3025,
+      "step": 345
+    },
+    {
+      "epoch": 0.9434219495569189,
+      "grad_norm": 1.7890625,
+      "learning_rate": 4.0546448087431696e-05,
+      "loss": 0.2618,
+      "step": 346
+    },
+    {
+      "epoch": 0.9461486025903204,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.051912568306011e-05,
+      "loss": 0.3047,
+      "step": 347
+    },
+    {
+      "epoch": 0.9488752556237219,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.049180327868853e-05,
+      "loss": 0.374,
+      "step": 348
+    },
+    {
+      "epoch": 0.9516019086571234,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.046448087431694e-05,
+      "loss": 0.3283,
+      "step": 349
+    },
+    {
+      "epoch": 0.9543285616905249,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.0437158469945356e-05,
+      "loss": 0.3297,
+      "step": 350
+    },
+    {
+      "epoch": 0.9543285616905249,
+      "eval_loss": 0.2984359860420227,
+      "eval_runtime": 10.4694,
+      "eval_samples_per_second": 280.246,
+      "eval_steps_per_second": 7.068,
+      "step": 350
+    },
+    {
+      "epoch": 0.9570552147239264,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.040983606557377e-05,
+      "loss": 0.1863,
+      "step": 351
+    },
+    {
+      "epoch": 0.9597818677573279,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.038251366120219e-05,
+      "loss": 0.3341,
+      "step": 352
+    },
+    {
+      "epoch": 0.9625085207907293,
+      "grad_norm": 1.9296875,
+      "learning_rate": 4.0355191256830606e-05,
+      "loss": 0.2864,
+      "step": 353
+    },
+    {
+      "epoch": 0.9652351738241309,
+      "grad_norm": 1.7734375,
+      "learning_rate": 4.0327868852459016e-05,
+      "loss": 0.2823,
+      "step": 354
+    },
+    {
+      "epoch": 0.9679618268575324,
+      "grad_norm": 1.9140625,
+      "learning_rate": 4.030054644808743e-05,
+      "loss": 0.3156,
+      "step": 355
+    },
+    {
+      "epoch": 0.9706884798909339,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.027322404371585e-05,
+      "loss": 0.3024,
+      "step": 356
+    },
+    {
+      "epoch": 0.9734151329243353,
+      "grad_norm": 1.875,
+      "learning_rate": 4.0245901639344266e-05,
+      "loss": 0.2547,
+      "step": 357
+    },
+    {
+      "epoch": 0.9761417859577369,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.021857923497268e-05,
+      "loss": 0.3139,
+      "step": 358
+    },
+    {
+      "epoch": 0.9788684389911384,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.019125683060109e-05,
+      "loss": 0.3275,
+      "step": 359
+    },
+    {
+      "epoch": 0.9815950920245399,
+      "grad_norm": 1.78125,
+      "learning_rate": 4.016393442622951e-05,
+      "loss": 0.3016,
+      "step": 360
+    },
+    {
+      "epoch": 0.9843217450579413,
+      "grad_norm": 1.6484375,
+      "learning_rate": 4.0136612021857926e-05,
+      "loss": 0.2988,
+      "step": 361
+    },
+    {
+      "epoch": 0.9870483980913429,
+      "grad_norm": 2.1875,
+      "learning_rate": 4.0109289617486336e-05,
+      "loss": 0.3498,
+      "step": 362
+    },
+    {
+      "epoch": 0.9897750511247444,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.008196721311476e-05,
+      "loss": 0.3449,
+      "step": 363
+    },
+    {
+      "epoch": 0.9925017041581459,
+      "grad_norm": 1.8984375,
+      "learning_rate": 4.005464480874317e-05,
+      "loss": 0.2991,
+      "step": 364
+    },
+    {
+      "epoch": 0.9952283571915473,
+      "grad_norm": 1.78125,
+      "learning_rate": 4.0027322404371587e-05,
+      "loss": 0.2543,
+      "step": 365
+    },
+    {
+      "epoch": 0.9979550102249489,
+      "grad_norm": 2.0,
+      "learning_rate": 4e-05,
+      "loss": 0.2739,
+      "step": 366
+    },
+    {
+      "epoch": 1.0006816632583504,
+      "grad_norm": 2.203125,
+      "learning_rate": 3.997267759562841e-05,
+      "loss": 0.2851,
+      "step": 367
+    },
+    {
+      "epoch": 1.003408316291752,
+      "grad_norm": 1.8671875,
+      "learning_rate": 3.994535519125684e-05,
+      "loss": 0.2806,
+      "step": 368
+    },
+    {
+      "epoch": 1.0061349693251533,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.991803278688525e-05,
+      "loss": 0.198,
+      "step": 369
+    },
+    {
+      "epoch": 1.008861622358555,
+      "grad_norm": 1.6875,
+      "learning_rate": 3.989071038251366e-05,
+      "loss": 0.2471,
+      "step": 370
+    },
+    {
+      "epoch": 1.0115882753919563,
+      "grad_norm": 2.328125,
+      "learning_rate": 3.986338797814208e-05,
+      "loss": 0.1886,
+      "step": 371
+    },
+    {
+      "epoch": 1.0143149284253579,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.983606557377049e-05,
+      "loss": 0.1914,
+      "step": 372
+    },
+    {
+      "epoch": 1.0170415814587594,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.980874316939891e-05,
+      "loss": 0.1977,
+      "step": 373
+    },
+    {
+      "epoch": 1.0197682344921608,
+      "grad_norm": 2.65625,
+      "learning_rate": 3.9781420765027323e-05,
+      "loss": 0.2676,
+      "step": 374
+    },
+    {
+      "epoch": 1.0224948875255624,
+      "grad_norm": 1.765625,
+      "learning_rate": 3.975409836065574e-05,
+      "loss": 0.1977,
+      "step": 375
+    },
+    {
+      "epoch": 1.025221540558964,
+      "grad_norm": 2.09375,
+      "learning_rate": 3.972677595628416e-05,
+      "loss": 0.2324,
+      "step": 376
+    },
+    {
+      "epoch": 1.0279481935923653,
+      "grad_norm": 2.671875,
+      "learning_rate": 3.969945355191257e-05,
+      "loss": 0.2248,
+      "step": 377
+    },
+    {
+      "epoch": 1.030674846625767,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.9672131147540983e-05,
+      "loss": 0.2029,
+      "step": 378
+    },
+    {
+      "epoch": 1.0334014996591683,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.96448087431694e-05,
+      "loss": 0.2086,
+      "step": 379
+    },
+    {
+      "epoch": 1.0361281526925699,
+      "grad_norm": 2.546875,
+      "learning_rate": 3.961748633879782e-05,
+      "loss": 0.1856,
+      "step": 380
+    },
+    {
+      "epoch": 1.0388548057259714,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.9590163934426234e-05,
+      "loss": 0.2083,
+      "step": 381
+    },
+    {
+      "epoch": 1.0415814587593728,
+      "grad_norm": 1.7578125,
+      "learning_rate": 3.9562841530054644e-05,
+      "loss": 0.2241,
+      "step": 382
+    },
+    {
+      "epoch": 1.0443081117927744,
+      "grad_norm": 2.640625,
+      "learning_rate": 3.953551912568306e-05,
+      "loss": 0.2362,
+      "step": 383
+    },
+    {
+      "epoch": 1.047034764826176,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.950819672131148e-05,
+      "loss": 0.233,
+      "step": 384
+    },
+    {
+      "epoch": 1.0497614178595773,
+      "grad_norm": 1.890625,
+      "learning_rate": 3.9480874316939894e-05,
+      "loss": 0.2266,
+      "step": 385
+    },
+    {
+      "epoch": 1.052488070892979,
+      "grad_norm": 1.9140625,
+      "learning_rate": 3.945355191256831e-05,
+      "loss": 0.2344,
+      "step": 386
+    },
+    {
+      "epoch": 1.0552147239263803,
+      "grad_norm": 1.890625,
+      "learning_rate": 3.942622950819672e-05,
+      "loss": 0.2532,
+      "step": 387
+    },
+    {
+      "epoch": 1.0579413769597819,
+      "grad_norm": 1.984375,
+      "learning_rate": 3.939890710382514e-05,
+      "loss": 0.2098,
+      "step": 388
+    },
+    {
+      "epoch": 1.0606680299931834,
+      "grad_norm": 1.671875,
+      "learning_rate": 3.9371584699453554e-05,
+      "loss": 0.1985,
+      "step": 389
+    },
+    {
+      "epoch": 1.0633946830265848,
+      "grad_norm": 1.703125,
+      "learning_rate": 3.934426229508197e-05,
+      "loss": 0.2066,
+      "step": 390
+    },
+    {
+      "epoch": 1.0661213360599864,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.931693989071039e-05,
+      "loss": 0.1626,
+      "step": 391
+    },
+    {
+      "epoch": 1.068847989093388,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.92896174863388e-05,
+      "loss": 0.2073,
+      "step": 392
+    },
+    {
+      "epoch": 1.0715746421267893,
+      "grad_norm": 2.0,
+      "learning_rate": 3.9262295081967214e-05,
+      "loss": 0.1669,
+      "step": 393
+    },
+    {
+      "epoch": 1.074301295160191,
+      "grad_norm": 1.6640625,
+      "learning_rate": 3.923497267759563e-05,
+      "loss": 0.1853,
+      "step": 394
+    },
+    {
+      "epoch": 1.0770279481935923,
+      "grad_norm": 2.265625,
+      "learning_rate": 3.920765027322405e-05,
+      "loss": 0.286,
+      "step": 395
+    },
+    {
+      "epoch": 1.0797546012269938,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.9180327868852464e-05,
+      "loss": 0.2732,
+      "step": 396
+    },
+    {
+      "epoch": 1.0824812542603954,
+      "grad_norm": 2.359375,
+      "learning_rate": 3.9153005464480874e-05,
+      "loss": 0.2388,
+      "step": 397
+    },
+    {
+      "epoch": 1.0852079072937968,
+      "grad_norm": 1.921875,
+      "learning_rate": 3.912568306010929e-05,
+      "loss": 0.2329,
+      "step": 398
+    },
+    {
+      "epoch": 1.0879345603271984,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.909836065573771e-05,
+      "loss": 0.1929,
+      "step": 399
+    },
+    {
+      "epoch": 1.0906612133606,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.9071038251366124e-05,
+      "loss": 0.2274,
+      "step": 400
+    },
+    {
+      "epoch": 1.0906612133606,
+      "eval_loss": 0.2957896888256073,
+      "eval_runtime": 10.4777,
+      "eval_samples_per_second": 280.022,
+      "eval_steps_per_second": 7.063,
+      "step": 400
+    },
+    {
+      "epoch": 1.0933878663940013,
+      "grad_norm": 1.7265625,
+      "learning_rate": 3.904371584699454e-05,
+      "loss": 0.2292,
+      "step": 401
+    },
+    {
+      "epoch": 1.096114519427403,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.901639344262295e-05,
+      "loss": 0.1957,
+      "step": 402
+    },
+    {
+      "epoch": 1.0988411724608043,
+      "grad_norm": 1.7578125,
+      "learning_rate": 3.898907103825137e-05,
+      "loss": 0.2606,
+      "step": 403
+    },
+    {
+      "epoch": 1.1015678254942058,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.8961748633879784e-05,
+      "loss": 0.1819,
+      "step": 404
+    },
+    {
+      "epoch": 1.1042944785276074,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.89344262295082e-05,
+      "loss": 0.2064,
+      "step": 405
+    },
+    {
+      "epoch": 1.1070211315610088,
+      "grad_norm": 1.6328125,
+      "learning_rate": 3.890710382513662e-05,
+      "loss": 0.203,
+      "step": 406
+    },
+    {
+      "epoch": 1.1097477845944104,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.887978142076503e-05,
+      "loss": 0.1988,
+      "step": 407
+    },
+    {
+      "epoch": 1.112474437627812,
+      "grad_norm": 1.671875,
+      "learning_rate": 3.8852459016393444e-05,
+      "loss": 0.212,
+      "step": 408
+    },
+    {
+      "epoch": 1.1152010906612133,
+      "grad_norm": 3.203125,
+      "learning_rate": 3.882513661202186e-05,
+      "loss": 0.2086,
+      "step": 409
+    },
+    {
+      "epoch": 1.117927743694615,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.879781420765027e-05,
+      "loss": 0.1835,
+      "step": 410
+    },
+    {
+      "epoch": 1.1206543967280163,
+      "grad_norm": 1.6328125,
+      "learning_rate": 3.8770491803278694e-05,
+      "loss": 0.1587,
+      "step": 411
+    },
+    {
+      "epoch": 1.1233810497614178,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.8743169398907104e-05,
+      "loss": 0.217,
+      "step": 412
+    },
+    {
+      "epoch": 1.1261077027948194,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.871584699453552e-05,
+      "loss": 0.1817,
+      "step": 413
+    },
+    {
+      "epoch": 1.1288343558282208,
+      "grad_norm": 1.4921875,
+      "learning_rate": 3.868852459016394e-05,
+      "loss": 0.162,
+      "step": 414
+    },
+    {
+      "epoch": 1.1315610088616224,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.866120218579235e-05,
+      "loss": 0.2054,
+      "step": 415
+    },
+    {
+      "epoch": 1.134287661895024,
+      "grad_norm": 1.6796875,
+      "learning_rate": 3.8633879781420764e-05,
+      "loss": 0.1806,
+      "step": 416
+    },
+    {
+      "epoch": 1.1370143149284253,
+      "grad_norm": 1.75,
+      "learning_rate": 3.860655737704918e-05,
+      "loss": 0.2456,
+      "step": 417
+    },
+    {
+      "epoch": 1.139740967961827,
+      "grad_norm": 2.265625,
+      "learning_rate": 3.85792349726776e-05,
+      "loss": 0.2266,
+      "step": 418
+    },
+    {
+      "epoch": 1.1424676209952285,
+      "grad_norm": 2.609375,
+      "learning_rate": 3.8551912568306014e-05,
+      "loss": 0.2404,
+      "step": 419
+    },
+    {
+      "epoch": 1.1451942740286298,
+      "grad_norm": 2.453125,
+      "learning_rate": 3.8524590163934424e-05,
+      "loss": 0.254,
+      "step": 420
+    },
+    {
+      "epoch": 1.1479209270620314,
+      "grad_norm": 1.7265625,
+      "learning_rate": 3.849726775956284e-05,
+      "loss": 0.2092,
+      "step": 421
+    },
+    {
+      "epoch": 1.1506475800954328,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.8469945355191264e-05,
+      "loss": 0.2364,
+      "step": 422
+    },
+    {
+      "epoch": 1.1533742331288344,
+      "grad_norm": 1.90625,
+      "learning_rate": 3.8442622950819674e-05,
+      "loss": 0.2251,
+      "step": 423
+    },
+    {
+      "epoch": 1.156100886162236,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.841530054644809e-05,
+      "loss": 0.2403,
+      "step": 424
+    },
+    {
+      "epoch": 1.1588275391956373,
+      "grad_norm": 1.625,
+      "learning_rate": 3.83879781420765e-05,
+      "loss": 0.1998,
+      "step": 425
+    },
+    {
+      "epoch": 1.1615541922290389,
+      "grad_norm": 1.625,
+      "learning_rate": 3.836065573770492e-05,
+      "loss": 0.2005,
+      "step": 426
+    },
+    {
+      "epoch": 1.1642808452624402,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.8333333333333334e-05,
+      "loss": 0.2207,
+      "step": 427
+    },
+    {
+      "epoch": 1.1670074982958418,
+      "grad_norm": 1.4921875,
+      "learning_rate": 3.830601092896175e-05,
+      "loss": 0.2048,
+      "step": 428
+    },
+    {
+      "epoch": 1.1697341513292434,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.827868852459017e-05,
+      "loss": 0.2018,
+      "step": 429
+    },
+    {
+      "epoch": 1.1724608043626448,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.825136612021858e-05,
+      "loss": 0.2229,
+      "step": 430
+    },
+    {
+      "epoch": 1.1751874573960464,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.8224043715846995e-05,
+      "loss": 0.175,
+      "step": 431
+    },
+    {
+      "epoch": 1.177914110429448,
+      "grad_norm": 1.90625,
+      "learning_rate": 3.819672131147541e-05,
+      "loss": 0.2451,
+      "step": 432
+    },
+    {
+      "epoch": 1.1806407634628493,
+      "grad_norm": 1.7578125,
+      "learning_rate": 3.816939890710383e-05,
+      "loss": 0.2093,
+      "step": 433
+    },
+    {
+      "epoch": 1.1833674164962509,
+      "grad_norm": 1.96875,
+      "learning_rate": 3.8142076502732245e-05,
+      "loss": 0.255,
+      "step": 434
+    },
+    {
+      "epoch": 1.1860940695296525,
+      "grad_norm": 1.765625,
+      "learning_rate": 3.8114754098360655e-05,
+      "loss": 0.2104,
+      "step": 435
+    },
+    {
+      "epoch": 1.1888207225630538,
+      "grad_norm": 1.53125,
+      "learning_rate": 3.808743169398907e-05,
+      "loss": 0.1803,
+      "step": 436
+    },
+    {
+      "epoch": 1.1915473755964554,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.806010928961749e-05,
+      "loss": 0.2081,
+      "step": 437
+    },
+    {
+      "epoch": 1.1942740286298568,
+      "grad_norm": 1.90625,
+      "learning_rate": 3.8032786885245905e-05,
+      "loss": 0.2181,
+      "step": 438
+    },
+    {
+      "epoch": 1.1970006816632583,
+      "grad_norm": 1.671875,
+      "learning_rate": 3.800546448087432e-05,
+      "loss": 0.2288,
+      "step": 439
+    },
+    {
+      "epoch": 1.19972733469666,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.797814207650273e-05,
+      "loss": 0.1745,
+      "step": 440
+    },
+    {
+      "epoch": 1.2024539877300613,
+      "grad_norm": 1.578125,
+      "learning_rate": 3.795081967213115e-05,
+      "loss": 0.2039,
+      "step": 441
+    },
+    {
+      "epoch": 1.2051806407634629,
+      "grad_norm": 1.921875,
+      "learning_rate": 3.7923497267759565e-05,
+      "loss": 0.2602,
+      "step": 442
+    },
+    {
+      "epoch": 1.2079072937968642,
+      "grad_norm": 1.6796875,
+      "learning_rate": 3.789617486338798e-05,
+      "loss": 0.1903,
+      "step": 443
+    },
+    {
+      "epoch": 1.2106339468302658,
+      "grad_norm": 1.875,
+      "learning_rate": 3.78688524590164e-05,
+      "loss": 0.2494,
+      "step": 444
+    },
+    {
+      "epoch": 1.2133605998636674,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.784153005464481e-05,
+      "loss": 0.2091,
+      "step": 445
+    },
+    {
+      "epoch": 1.2160872528970688,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.7814207650273225e-05,
+      "loss": 0.2513,
+      "step": 446
+    },
+    {
+      "epoch": 1.2188139059304703,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.778688524590164e-05,
+      "loss": 0.2339,
+      "step": 447
+    },
+    {
+      "epoch": 1.221540558963872,
+      "grad_norm": 1.7265625,
+      "learning_rate": 3.775956284153006e-05,
+      "loss": 0.1883,
+      "step": 448
+    },
+    {
+      "epoch": 1.2242672119972733,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.7732240437158475e-05,
+      "loss": 0.1744,
+      "step": 449
+    },
+    {
+      "epoch": 1.2269938650306749,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.7704918032786885e-05,
+      "loss": 0.2394,
+      "step": 450
+    },
+    {
+      "epoch": 1.2269938650306749,
+      "eval_loss": 0.2854467034339905,
+      "eval_runtime": 14.3869,
+      "eval_samples_per_second": 203.935,
+      "eval_steps_per_second": 5.144,
+      "step": 450
+    },
+    {
+      "epoch": 1.2297205180640765,
+      "grad_norm": 1.6640625,
+      "learning_rate": 3.76775956284153e-05,
+      "loss": 0.1972,
+      "step": 451
+    },
+    {
+      "epoch": 1.2324471710974778,
+      "grad_norm": 1.796875,
+      "learning_rate": 3.765027322404372e-05,
+      "loss": 0.2047,
+      "step": 452
+    },
+    {
+      "epoch": 1.2351738241308794,
+      "grad_norm": 1.8125,
+      "learning_rate": 3.762295081967213e-05,
+      "loss": 0.2111,
+      "step": 453
+    },
+    {
+      "epoch": 1.2379004771642808,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.759562841530055e-05,
+      "loss": 0.2458,
+      "step": 454
+    },
+    {
+      "epoch": 1.2406271301976823,
+      "grad_norm": 1.5546875,
+      "learning_rate": 3.756830601092896e-05,
+      "loss": 0.2129,
+      "step": 455
+    },
+    {
+      "epoch": 1.243353783231084,
+      "grad_norm": 2.078125,
+      "learning_rate": 3.754098360655738e-05,
+      "loss": 0.2141,
+      "step": 456
+    },
+    {
+      "epoch": 1.2460804362644853,
+      "grad_norm": 1.875,
+      "learning_rate": 3.7513661202185795e-05,
+      "loss": 0.2098,
+      "step": 457
+    },
+    {
+      "epoch": 1.2488070892978869,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.7486338797814205e-05,
+      "loss": 0.2172,
+      "step": 458
+    },
+    {
+      "epoch": 1.2515337423312882,
+      "grad_norm": 1.6875,
+      "learning_rate": 3.745901639344263e-05,
+      "loss": 0.1959,
+      "step": 459
+    },
+    {
+      "epoch": 1.2542603953646898,
+      "grad_norm": 1.5390625,
+      "learning_rate": 3.7431693989071045e-05,
+      "loss": 0.1779,
+      "step": 460
+    },
+    {
+      "epoch": 1.2569870483980914,
+      "grad_norm": 2.40625,
+      "learning_rate": 3.7404371584699455e-05,
+      "loss": 0.1858,
+      "step": 461
+    },
+    {
+      "epoch": 1.259713701431493,
+      "grad_norm": 1.8359375,
+      "learning_rate": 3.737704918032787e-05,
+      "loss": 0.1951,
+      "step": 462
+    },
+    {
+      "epoch": 1.2624403544648943,
+      "grad_norm": 1.890625,
+      "learning_rate": 3.734972677595628e-05,
+      "loss": 0.2432,
+      "step": 463
+    },
+    {
+      "epoch": 1.265167007498296,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.73224043715847e-05,
+      "loss": 0.1922,
+      "step": 464
+    },
+    {
+      "epoch": 1.2678936605316973,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.729508196721312e-05,
+      "loss": 0.1822,
+      "step": 465
+    },
+    {
+      "epoch": 1.2706203135650989,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.726775956284153e-05,
+      "loss": 0.231,
+      "step": 466
+    },
+    {
+      "epoch": 1.2733469665985004,
+      "grad_norm": 2.3125,
+      "learning_rate": 3.724043715846995e-05,
+      "loss": 0.2652,
+      "step": 467
+    },
+    {
+      "epoch": 1.2760736196319018,
+      "grad_norm": 1.8984375,
+      "learning_rate": 3.721311475409836e-05,
+      "loss": 0.2187,
+      "step": 468
+    },
+    {
+      "epoch": 1.2788002726653034,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.7185792349726775e-05,
+      "loss": 0.1984,
+      "step": 469
+    },
+    {
+      "epoch": 1.2815269256987047,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.71584699453552e-05,
+      "loss": 0.2172,
+      "step": 470
+    },
+    {
+      "epoch": 1.2842535787321063,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.713114754098361e-05,
+      "loss": 0.2065,
+      "step": 471
+    },
+    {
+      "epoch": 1.286980231765508,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.7103825136612025e-05,
+      "loss": 0.197,
+      "step": 472
+    },
+    {
+      "epoch": 1.2897068847989093,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.7076502732240435e-05,
+      "loss": 0.1962,
+      "step": 473
+    },
+    {
+      "epoch": 1.2924335378323109,
+      "grad_norm": 1.7734375,
+      "learning_rate": 3.704918032786885e-05,
+      "loss": 0.222,
+      "step": 474
+    },
+    {
+      "epoch": 1.2951601908657122,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.702185792349727e-05,
+      "loss": 0.2102,
+      "step": 475
+    },
+    {
+      "epoch": 1.2978868438991138,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.6994535519125686e-05,
+      "loss": 0.2063,
+      "step": 476
+    },
+    {
+      "epoch": 1.3006134969325154,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.69672131147541e-05,
+      "loss": 0.1916,
+      "step": 477
+    },
+    {
+      "epoch": 1.303340149965917,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.693989071038251e-05,
+      "loss": 0.215,
+      "step": 478
+    },
+    {
+      "epoch": 1.3060668029993183,
+      "grad_norm": 1.5546875,
+      "learning_rate": 3.691256830601093e-05,
+      "loss": 0.1834,
+      "step": 479
+    },
+    {
+      "epoch": 1.30879345603272,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.6885245901639346e-05,
+      "loss": 0.22,
+      "step": 480
+    },
+    {
+      "epoch": 1.3115201090661213,
+      "grad_norm": 1.5078125,
+      "learning_rate": 3.685792349726776e-05,
+      "loss": 0.1889,
+      "step": 481
+    },
+    {
+      "epoch": 1.3142467620995228,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.683060109289618e-05,
+      "loss": 0.1976,
+      "step": 482
+    },
+    {
+      "epoch": 1.3169734151329244,
+      "grad_norm": 2.234375,
+      "learning_rate": 3.680327868852459e-05,
+      "loss": 0.1983,
+      "step": 483
+    },
+    {
+      "epoch": 1.3197000681663258,
+      "grad_norm": 1.65625,
+      "learning_rate": 3.6775956284153006e-05,
+      "loss": 0.21,
+      "step": 484
+    },
+    {
+      "epoch": 1.3224267211997274,
+      "grad_norm": 2.21875,
+      "learning_rate": 3.674863387978142e-05,
+      "loss": 0.2428,
+      "step": 485
+    },
+    {
+      "epoch": 1.3251533742331287,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.672131147540984e-05,
+      "loss": 0.2464,
+      "step": 486
+    },
+    {
+      "epoch": 1.3278800272665303,
+      "grad_norm": 1.6875,
+      "learning_rate": 3.6693989071038256e-05,
+      "loss": 0.1992,
+      "step": 487
+    },
+    {
+      "epoch": 1.330606680299932,
+      "grad_norm": 1.96875,
+      "learning_rate": 3.6666666666666666e-05,
+      "loss": 0.189,
+      "step": 488
+    },
+    {
+      "epoch": 1.3333333333333333,
+      "grad_norm": 1.8671875,
+      "learning_rate": 3.663934426229508e-05,
+      "loss": 0.2233,
+      "step": 489
+    },
+    {
+      "epoch": 1.3360599863667348,
+      "grad_norm": 1.78125,
+      "learning_rate": 3.66120218579235e-05,
+      "loss": 0.2253,
+      "step": 490
+    },
+    {
+      "epoch": 1.3387866394001362,
+      "grad_norm": 2.5,
+      "learning_rate": 3.6584699453551916e-05,
+      "loss": 0.203,
+      "step": 491
+    },
+    {
+      "epoch": 1.3415132924335378,
+      "grad_norm": 1.84375,
+      "learning_rate": 3.655737704918033e-05,
+      "loss": 0.2186,
+      "step": 492
+    },
+    {
+      "epoch": 1.3442399454669394,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.653005464480874e-05,
+      "loss": 0.261,
+      "step": 493
+    },
+    {
+      "epoch": 1.346966598500341,
+      "grad_norm": 1.7890625,
+      "learning_rate": 3.650273224043716e-05,
+      "loss": 0.258,
+      "step": 494
+    },
+    {
+      "epoch": 1.3496932515337423,
+      "grad_norm": 1.953125,
+      "learning_rate": 3.6475409836065576e-05,
+      "loss": 0.2413,
+      "step": 495
+    },
+    {
+      "epoch": 1.352419904567144,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.6448087431693986e-05,
+      "loss": 0.2122,
+      "step": 496
+    },
+    {
+      "epoch": 1.3551465576005453,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.642076502732241e-05,
+      "loss": 0.1779,
+      "step": 497
+    },
+    {
+      "epoch": 1.3578732106339468,
+      "grad_norm": 1.703125,
+      "learning_rate": 3.6393442622950826e-05,
+      "loss": 0.2392,
+      "step": 498
+    },
+    {
+      "epoch": 1.3605998636673484,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.6366120218579236e-05,
+      "loss": 0.21,
+      "step": 499
+    },
+    {
+      "epoch": 1.3633265167007498,
+      "grad_norm": 1.9140625,
+      "learning_rate": 3.633879781420765e-05,
+      "loss": 0.2348,
+      "step": 500
+    },
+    {
+      "epoch": 1.3633265167007498,
+      "eval_loss": 0.2783825099468231,
+      "eval_runtime": 10.5405,
+      "eval_samples_per_second": 278.355,
+      "eval_steps_per_second": 7.021,
+      "step": 500
+    },
+    {
+      "epoch": 1.3660531697341514,
+      "grad_norm": 1.9375,
+      "learning_rate": 3.631147540983606e-05,
+      "loss": 0.2224,
+      "step": 501
+    },
+    {
+      "epoch": 1.3687798227675527,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.6284153005464486e-05,
+      "loss": 0.2409,
+      "step": 502
+    },
+    {
+      "epoch": 1.3715064758009543,
+      "grad_norm": 1.890625,
+      "learning_rate": 3.62568306010929e-05,
+      "loss": 0.2172,
+      "step": 503
+    },
+    {
+      "epoch": 1.3742331288343559,
+      "grad_norm": 1.671875,
+      "learning_rate": 3.622950819672131e-05,
+      "loss": 0.1883,
+      "step": 504
+    },
+    {
+      "epoch": 1.3769597818677572,
+      "grad_norm": 1.6796875,
+      "learning_rate": 3.620218579234973e-05,
+      "loss": 0.198,
+      "step": 505
+    },
+    {
+      "epoch": 1.3796864349011588,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.617486338797814e-05,
+      "loss": 0.1821,
+      "step": 506
+    },
+    {
+      "epoch": 1.3824130879345602,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.6147540983606556e-05,
+      "loss": 0.1807,
+      "step": 507
+    },
+    {
+      "epoch": 1.3851397409679618,
+      "grad_norm": 2.0,
+      "learning_rate": 3.612021857923498e-05,
+      "loss": 0.2702,
+      "step": 508
+    },
+    {
+      "epoch": 1.3878663940013634,
+      "grad_norm": 1.671875,
+      "learning_rate": 3.609289617486339e-05,
+      "loss": 0.2172,
+      "step": 509
+    },
+    {
+      "epoch": 1.390593047034765,
+      "grad_norm": 1.5,
+      "learning_rate": 3.6065573770491806e-05,
+      "loss": 0.2034,
+      "step": 510
+    },
+    {
+      "epoch": 1.3933197000681663,
+      "grad_norm": 1.796875,
+      "learning_rate": 3.6038251366120216e-05,
+      "loss": 0.2197,
+      "step": 511
+    },
+    {
+      "epoch": 1.3960463531015679,
+      "grad_norm": 1.8984375,
+      "learning_rate": 3.601092896174863e-05,
+      "loss": 0.2258,
+      "step": 512
+    },
+    {
+      "epoch": 1.3987730061349692,
+      "grad_norm": 1.65625,
+      "learning_rate": 3.5983606557377056e-05,
+      "loss": 0.2059,
+      "step": 513
+    },
+    {
+      "epoch": 1.4014996591683708,
+      "grad_norm": 1.796875,
+      "learning_rate": 3.5956284153005466e-05,
+      "loss": 0.2198,
+      "step": 514
+    },
+    {
+      "epoch": 1.4042263122017724,
+      "grad_norm": 1.75,
+      "learning_rate": 3.592896174863388e-05,
+      "loss": 0.2053,
+      "step": 515
+    },
+    {
+      "epoch": 1.4069529652351738,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.590163934426229e-05,
+      "loss": 0.2173,
+      "step": 516
+    },
+    {
+      "epoch": 1.4096796182685754,
+      "grad_norm": 1.8515625,
+      "learning_rate": 3.587431693989071e-05,
+      "loss": 0.2264,
+      "step": 517
+    },
+    {
+      "epoch": 1.4124062713019767,
+      "grad_norm": 2.09375,
+      "learning_rate": 3.5846994535519126e-05,
+      "loss": 0.1772,
+      "step": 518
+    },
+    {
+      "epoch": 1.4151329243353783,
+      "grad_norm": 2.96875,
+      "learning_rate": 3.581967213114754e-05,
+      "loss": 0.1808,
+      "step": 519
+    },
+    {
+      "epoch": 1.4178595773687799,
+      "grad_norm": 1.546875,
+      "learning_rate": 3.579234972677596e-05,
+      "loss": 0.2128,
+      "step": 520
+    },
+    {
+      "epoch": 1.4205862304021815,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.576502732240437e-05,
+      "loss": 0.2321,
+      "step": 521
+    },
+    {
+      "epoch": 1.4233128834355828,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.5737704918032786e-05,
+      "loss": 0.2199,
+      "step": 522
+    },
+    {
+      "epoch": 1.4260395364689844,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.57103825136612e-05,
+      "loss": 0.2118,
+      "step": 523
+    },
+    {
+      "epoch": 1.4287661895023858,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.568306010928962e-05,
+      "loss": 0.1822,
+      "step": 524
+    },
+    {
+      "epoch": 1.4314928425357873,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.5655737704918037e-05,
+      "loss": 0.1833,
+      "step": 525
+    },
+    {
+      "epoch": 1.434219495569189,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.5628415300546446e-05,
+      "loss": 0.217,
+      "step": 526
+    },
+    {
+      "epoch": 1.4369461486025903,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.560109289617486e-05,
+      "loss": 0.1975,
+      "step": 527
+    },
+    {
+      "epoch": 1.4396728016359919,
+      "grad_norm": 1.765625,
+      "learning_rate": 3.557377049180328e-05,
+      "loss": 0.2436,
+      "step": 528
+    },
+    {
+      "epoch": 1.4423994546693932,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.55464480874317e-05,
+      "loss": 0.238,
+      "step": 529
+    },
+    {
+      "epoch": 1.4451261077027948,
+      "grad_norm": 1.65625,
+      "learning_rate": 3.551912568306011e-05,
+      "loss": 0.2092,
+      "step": 530
+    },
+    {
+      "epoch": 1.4478527607361964,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.549180327868853e-05,
+      "loss": 0.269,
+      "step": 531
+    },
+    {
+      "epoch": 1.4505794137695978,
+      "grad_norm": 1.609375,
+      "learning_rate": 3.546448087431694e-05,
+      "loss": 0.2096,
+      "step": 532
+    },
+    {
+      "epoch": 1.4533060668029993,
+      "grad_norm": 1.984375,
+      "learning_rate": 3.543715846994536e-05,
+      "loss": 0.2133,
+      "step": 533
+    },
+    {
+      "epoch": 1.4560327198364007,
+      "grad_norm": 1.546875,
+      "learning_rate": 3.5409836065573773e-05,
+      "loss": 0.2206,
+      "step": 534
+    },
+    {
+      "epoch": 1.4587593728698023,
+      "grad_norm": 1.609375,
+      "learning_rate": 3.538251366120219e-05,
+      "loss": 0.2184,
+      "step": 535
+    },
+    {
+      "epoch": 1.4614860259032039,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.535519125683061e-05,
+      "loss": 0.225,
+      "step": 536
+    },
+    {
+      "epoch": 1.4642126789366054,
+      "grad_norm": 1.5078125,
+      "learning_rate": 3.532786885245902e-05,
+      "loss": 0.159,
+      "step": 537
+    },
+    {
+      "epoch": 1.4669393319700068,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.5300546448087433e-05,
+      "loss": 0.2121,
+      "step": 538
+    },
+    {
+      "epoch": 1.4696659850034084,
+      "grad_norm": 1.8984375,
+      "learning_rate": 3.527322404371584e-05,
+      "loss": 0.2197,
+      "step": 539
+    },
+    {
+      "epoch": 1.4723926380368098,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.524590163934427e-05,
+      "loss": 0.2212,
+      "step": 540
+    },
+    {
+      "epoch": 1.4751192910702113,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.5218579234972684e-05,
+      "loss": 0.2078,
+      "step": 541
+    },
+    {
+      "epoch": 1.477845944103613,
+      "grad_norm": 1.7578125,
+      "learning_rate": 3.5191256830601094e-05,
+      "loss": 0.1861,
+      "step": 542
+    },
+    {
+      "epoch": 1.4805725971370143,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.516393442622951e-05,
+      "loss": 0.1659,
+      "step": 543
+    },
+    {
+      "epoch": 1.4832992501704159,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.513661202185792e-05,
+      "loss": 0.2232,
+      "step": 544
+    },
+    {
+      "epoch": 1.4860259032038172,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.5109289617486344e-05,
+      "loss": 0.2094,
+      "step": 545
+    },
+    {
+      "epoch": 1.4887525562372188,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.508196721311476e-05,
+      "loss": 0.2369,
+      "step": 546
+    },
+    {
+      "epoch": 1.4914792092706204,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.505464480874317e-05,
+      "loss": 0.234,
+      "step": 547
+    },
+    {
+      "epoch": 1.4942058623040217,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.502732240437159e-05,
+      "loss": 0.2192,
+      "step": 548
+    },
+    {
+      "epoch": 1.4969325153374233,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.5e-05,
+      "loss": 0.1848,
+      "step": 549
+    },
+    {
+      "epoch": 1.4996591683708247,
+      "grad_norm": 1.703125,
+      "learning_rate": 3.4972677595628414e-05,
+      "loss": 0.2117,
+      "step": 550
+    },
+    {
+      "epoch": 1.4996591683708247,
+      "eval_loss": 0.26996052265167236,
+      "eval_runtime": 13.0186,
+      "eval_samples_per_second": 225.369,
+      "eval_steps_per_second": 5.684,
+      "step": 550
+    },
+    {
+      "epoch": 1.5023858214042263,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.494535519125684e-05,
+      "loss": 0.2522,
+      "step": 551
+    },
+    {
+      "epoch": 1.5051124744376279,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.491803278688525e-05,
+      "loss": 0.1952,
+      "step": 552
+    },
+    {
+      "epoch": 1.5078391274710294,
+      "grad_norm": 1.6328125,
+      "learning_rate": 3.4890710382513664e-05,
+      "loss": 0.1976,
+      "step": 553
+    },
+    {
+      "epoch": 1.5105657805044308,
+      "grad_norm": 1.625,
+      "learning_rate": 3.4863387978142074e-05,
+      "loss": 0.2246,
+      "step": 554
+    },
+    {
+      "epoch": 1.5132924335378322,
+      "grad_norm": 1.8359375,
+      "learning_rate": 3.483606557377049e-05,
+      "loss": 0.2226,
+      "step": 555
+    },
+    {
+      "epoch": 1.5160190865712337,
+      "grad_norm": 1.640625,
+      "learning_rate": 3.4808743169398914e-05,
+      "loss": 0.1824,
+      "step": 556
+    },
+    {
+      "epoch": 1.5187457396046353,
+      "grad_norm": 1.90625,
+      "learning_rate": 3.4781420765027324e-05,
+      "loss": 0.2546,
+      "step": 557
+    },
+    {
+      "epoch": 1.521472392638037,
+      "grad_norm": 2.1875,
+      "learning_rate": 3.475409836065574e-05,
+      "loss": 0.1932,
+      "step": 558
+    },
+    {
+      "epoch": 1.5241990456714383,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.472677595628415e-05,
+      "loss": 0.1886,
+      "step": 559
+    },
+    {
+      "epoch": 1.5269256987048399,
+      "grad_norm": 1.875,
+      "learning_rate": 3.469945355191257e-05,
+      "loss": 0.2247,
+      "step": 560
+    },
+    {
+      "epoch": 1.5296523517382412,
+      "grad_norm": 1.578125,
+      "learning_rate": 3.4672131147540984e-05,
+      "loss": 0.1899,
+      "step": 561
+    },
+    {
+      "epoch": 1.5323790047716428,
+      "grad_norm": 2.1875,
+      "learning_rate": 3.46448087431694e-05,
+      "loss": 0.2542,
+      "step": 562
+    },
+    {
+      "epoch": 1.5351056578050444,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.461748633879782e-05,
+      "loss": 0.21,
+      "step": 563
+    },
+    {
+      "epoch": 1.537832310838446,
+      "grad_norm": 1.640625,
+      "learning_rate": 3.459016393442623e-05,
+      "loss": 0.211,
+      "step": 564
+    },
+    {
+      "epoch": 1.5405589638718473,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.4562841530054644e-05,
+      "loss": 0.2605,
+      "step": 565
+    },
+    {
+      "epoch": 1.5432856169052487,
+      "grad_norm": 1.8203125,
+      "learning_rate": 3.453551912568306e-05,
+      "loss": 0.2188,
+      "step": 566
+    },
+    {
+      "epoch": 1.5460122699386503,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.450819672131148e-05,
+      "loss": 0.1895,
+      "step": 567
+    },
+    {
+      "epoch": 1.5487389229720518,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.4480874316939894e-05,
+      "loss": 0.2003,
+      "step": 568
+    },
+    {
+      "epoch": 1.5514655760054534,
+      "grad_norm": 1.7578125,
+      "learning_rate": 3.445355191256831e-05,
+      "loss": 0.1952,
+      "step": 569
+    },
+    {
+      "epoch": 1.5541922290388548,
+      "grad_norm": 1.53125,
+      "learning_rate": 3.442622950819672e-05,
+      "loss": 0.1867,
+      "step": 570
+    },
+    {
+      "epoch": 1.5569188820722561,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.439890710382514e-05,
+      "loss": 0.1829,
+      "step": 571
+    },
+    {
+      "epoch": 1.5596455351056577,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.4371584699453554e-05,
+      "loss": 0.166,
+      "step": 572
+    },
+    {
+      "epoch": 1.5623721881390593,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.434426229508197e-05,
+      "loss": 0.2325,
+      "step": 573
+    },
+    {
+      "epoch": 1.565098841172461,
+      "grad_norm": 1.6328125,
+      "learning_rate": 3.431693989071039e-05,
+      "loss": 0.2518,
+      "step": 574
+    },
+    {
+      "epoch": 1.5678254942058623,
+      "grad_norm": 2.21875,
+      "learning_rate": 3.42896174863388e-05,
+      "loss": 0.1806,
+      "step": 575
+    },
+    {
+      "epoch": 1.5705521472392638,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.4262295081967214e-05,
+      "loss": 0.1933,
+      "step": 576
+    },
+    {
+      "epoch": 1.5732788002726652,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.423497267759563e-05,
+      "loss": 0.1794,
+      "step": 577
+    },
+    {
+      "epoch": 1.5760054533060668,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.420765027322405e-05,
+      "loss": 0.1966,
+      "step": 578
+    },
+    {
+      "epoch": 1.5787321063394684,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.4180327868852464e-05,
+      "loss": 0.221,
+      "step": 579
+    },
+    {
+      "epoch": 1.58145875937287,
+      "grad_norm": 1.7578125,
+      "learning_rate": 3.4153005464480874e-05,
+      "loss": 0.2284,
+      "step": 580
+    },
+    {
+      "epoch": 1.5841854124062713,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.412568306010929e-05,
+      "loss": 0.2029,
+      "step": 581
+    },
+    {
+      "epoch": 1.5869120654396727,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.409836065573771e-05,
+      "loss": 0.1657,
+      "step": 582
+    },
+    {
+      "epoch": 1.5896387184730743,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.4071038251366124e-05,
+      "loss": 0.1932,
+      "step": 583
+    },
+    {
+      "epoch": 1.5923653715064758,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.404371584699454e-05,
+      "loss": 0.191,
+      "step": 584
+    },
+    {
+      "epoch": 1.5950920245398774,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.401639344262295e-05,
+      "loss": 0.2095,
+      "step": 585
+    },
+    {
+      "epoch": 1.5978186775732788,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.398907103825137e-05,
+      "loss": 0.2283,
+      "step": 586
+    },
+    {
+      "epoch": 1.6005453306066801,
+      "grad_norm": 1.5078125,
+      "learning_rate": 3.396174863387978e-05,
+      "loss": 0.1891,
+      "step": 587
+    },
+    {
+      "epoch": 1.6032719836400817,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.39344262295082e-05,
+      "loss": 0.2488,
+      "step": 588
+    },
+    {
+      "epoch": 1.6059986366734833,
+      "grad_norm": 1.8515625,
+      "learning_rate": 3.390710382513662e-05,
+      "loss": 0.2381,
+      "step": 589
+    },
+    {
+      "epoch": 1.6087252897068849,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.387978142076503e-05,
+      "loss": 0.2196,
+      "step": 590
+    },
+    {
+      "epoch": 1.6114519427402862,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.3852459016393445e-05,
+      "loss": 0.1993,
+      "step": 591
+    },
+    {
+      "epoch": 1.6141785957736878,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.3825136612021854e-05,
+      "loss": 0.2325,
+      "step": 592
+    },
+    {
+      "epoch": 1.6169052488070892,
+      "grad_norm": 1.6328125,
+      "learning_rate": 3.379781420765028e-05,
+      "loss": 0.2193,
+      "step": 593
+    },
+    {
+      "epoch": 1.6196319018404908,
+      "grad_norm": 1.7734375,
+      "learning_rate": 3.3770491803278695e-05,
+      "loss": 0.1927,
+      "step": 594
+    },
+    {
+      "epoch": 1.6223585548738924,
+      "grad_norm": 1.8984375,
+      "learning_rate": 3.3743169398907105e-05,
+      "loss": 0.1883,
+      "step": 595
+    },
+    {
+      "epoch": 1.625085207907294,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.371584699453552e-05,
+      "loss": 0.2226,
+      "step": 596
+    },
+    {
+      "epoch": 1.6278118609406953,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.368852459016393e-05,
+      "loss": 0.19,
+      "step": 597
+    },
+    {
+      "epoch": 1.6305385139740967,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.366120218579235e-05,
+      "loss": 0.2088,
+      "step": 598
+    },
+    {
+      "epoch": 1.6332651670074982,
+      "grad_norm": 1.703125,
+      "learning_rate": 3.363387978142077e-05,
+      "loss": 0.228,
+      "step": 599
+    },
+    {
+      "epoch": 1.6359918200408998,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.360655737704918e-05,
+      "loss": 0.1875,
+      "step": 600
+    },
+    {
+      "epoch": 1.6359918200408998,
+      "eval_loss": 0.26377785205841064,
+      "eval_runtime": 10.4891,
+      "eval_samples_per_second": 279.718,
+      "eval_steps_per_second": 7.055,
+      "step": 600
+    },
+    {
+      "epoch": 1.6387184730743014,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.35792349726776e-05,
+      "loss": 0.2212,
+      "step": 601
+    },
+    {
+      "epoch": 1.6414451261077028,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.355191256830601e-05,
+      "loss": 0.2421,
+      "step": 602
+    },
+    {
+      "epoch": 1.6441717791411041,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.3524590163934425e-05,
+      "loss": 0.2132,
+      "step": 603
+    },
+    {
+      "epoch": 1.6468984321745057,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.349726775956284e-05,
+      "loss": 0.1702,
+      "step": 604
+    },
+    {
+      "epoch": 1.6496250852079073,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.346994535519126e-05,
+      "loss": 0.1796,
+      "step": 605
+    },
+    {
+      "epoch": 1.6523517382413089,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.3442622950819675e-05,
+      "loss": 0.1885,
+      "step": 606
+    },
+    {
+      "epoch": 1.6550783912747105,
+      "grad_norm": 1.75,
+      "learning_rate": 3.341530054644809e-05,
+      "loss": 0.168,
+      "step": 607
+    },
+    {
+      "epoch": 1.6578050443081118,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.33879781420765e-05,
+      "loss": 0.211,
+      "step": 608
+    },
+    {
+      "epoch": 1.6605316973415132,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.336065573770492e-05,
+      "loss": 0.1594,
+      "step": 609
+    },
+    {
+      "epoch": 1.6632583503749148,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.2155,
+      "step": 610
+    },
+    {
+      "epoch": 1.6659850034083163,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.330601092896175e-05,
+      "loss": 0.1873,
+      "step": 611
+    },
+    {
+      "epoch": 1.668711656441718,
+      "grad_norm": 1.6796875,
+      "learning_rate": 3.327868852459017e-05,
+      "loss": 0.2107,
+      "step": 612
+    },
+    {
+      "epoch": 1.6714383094751193,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.325136612021858e-05,
+      "loss": 0.1578,
+      "step": 613
+    },
+    {
+      "epoch": 1.6741649625085206,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.3224043715846995e-05,
+      "loss": 0.2218,
+      "step": 614
+    },
+    {
+      "epoch": 1.6768916155419222,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.319672131147541e-05,
+      "loss": 0.2206,
+      "step": 615
+    },
+    {
+      "epoch": 1.6796182685753238,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.316939890710383e-05,
+      "loss": 0.1859,
+      "step": 616
+    },
+    {
+      "epoch": 1.6823449216087254,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.3142076502732245e-05,
+      "loss": 0.2079,
+      "step": 617
+    },
+    {
+      "epoch": 1.6850715746421268,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.3114754098360655e-05,
+      "loss": 0.1674,
+      "step": 618
+    },
+    {
+      "epoch": 1.6877982276755283,
+      "grad_norm": 1.6640625,
+      "learning_rate": 3.308743169398907e-05,
+      "loss": 0.2176,
+      "step": 619
+    },
+    {
+      "epoch": 1.6905248807089297,
+      "grad_norm": 1.671875,
+      "learning_rate": 3.306010928961749e-05,
+      "loss": 0.2113,
+      "step": 620
+    },
+    {
+      "epoch": 1.6932515337423313,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.3032786885245905e-05,
+      "loss": 0.2109,
+      "step": 621
+    },
+    {
+      "epoch": 1.6959781867757329,
+      "grad_norm": 1.65625,
+      "learning_rate": 3.300546448087432e-05,
+      "loss": 0.1913,
+      "step": 622
+    },
+    {
+      "epoch": 1.6987048398091344,
+      "grad_norm": 1.5078125,
+      "learning_rate": 3.297814207650273e-05,
+      "loss": 0.207,
+      "step": 623
+    },
+    {
+      "epoch": 1.7014314928425358,
+      "grad_norm": 1.5,
+      "learning_rate": 3.295081967213115e-05,
+      "loss": 0.1971,
+      "step": 624
+    },
+    {
+      "epoch": 1.7041581458759372,
+      "grad_norm": 1.3203125,
+      "learning_rate": 3.2923497267759565e-05,
+      "loss": 0.1716,
+      "step": 625
+    },
+    {
+      "epoch": 1.7068847989093388,
+      "grad_norm": 1.375,
+      "learning_rate": 3.289617486338798e-05,
+      "loss": 0.2009,
+      "step": 626
+    },
+    {
+      "epoch": 1.7096114519427403,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.28688524590164e-05,
+      "loss": 0.2082,
+      "step": 627
+    },
+    {
+      "epoch": 1.712338104976142,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.284153005464481e-05,
+      "loss": 0.1856,
+      "step": 628
+    },
+    {
+      "epoch": 1.7150647580095433,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.2814207650273225e-05,
+      "loss": 0.2237,
+      "step": 629
+    },
+    {
+      "epoch": 1.7177914110429446,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.2786885245901635e-05,
+      "loss": 0.1963,
+      "step": 630
+    },
+    {
+      "epoch": 1.7205180640763462,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.275956284153006e-05,
+      "loss": 0.1947,
+      "step": 631
+    },
+    {
+      "epoch": 1.7232447171097478,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.2732240437158475e-05,
+      "loss": 0.208,
+      "step": 632
+    },
+    {
+      "epoch": 1.7259713701431494,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.2704918032786885e-05,
+      "loss": 0.1781,
+      "step": 633
+    },
+    {
+      "epoch": 1.7286980231765507,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.26775956284153e-05,
+      "loss": 0.1922,
+      "step": 634
+    },
+    {
+      "epoch": 1.7314246762099523,
+      "grad_norm": 1.4921875,
+      "learning_rate": 3.265027322404371e-05,
+      "loss": 0.1952,
+      "step": 635
+    },
+    {
+      "epoch": 1.7341513292433537,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.2622950819672136e-05,
+      "loss": 0.2208,
+      "step": 636
+    },
+    {
+      "epoch": 1.7368779822767553,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.259562841530055e-05,
+      "loss": 0.233,
+      "step": 637
+    },
+    {
+      "epoch": 1.7396046353101569,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.256830601092896e-05,
+      "loss": 0.1938,
+      "step": 638
+    },
+    {
+      "epoch": 1.7423312883435584,
+      "grad_norm": 1.8515625,
+      "learning_rate": 3.254098360655738e-05,
+      "loss": 0.2438,
+      "step": 639
+    },
+    {
+      "epoch": 1.7450579413769598,
+      "grad_norm": 1.640625,
+      "learning_rate": 3.251366120218579e-05,
+      "loss": 0.2006,
+      "step": 640
+    },
+    {
+      "epoch": 1.7477845944103612,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.2486338797814206e-05,
+      "loss": 0.1442,
+      "step": 641
+    },
+    {
+      "epoch": 1.7505112474437627,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.245901639344263e-05,
+      "loss": 0.2055,
+      "step": 642
+    },
+    {
+      "epoch": 1.7532379004771643,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.243169398907104e-05,
+      "loss": 0.1816,
+      "step": 643
+    },
+    {
+      "epoch": 1.755964553510566,
+      "grad_norm": 1.9296875,
+      "learning_rate": 3.2404371584699456e-05,
+      "loss": 0.2196,
+      "step": 644
+    },
+    {
+      "epoch": 1.7586912065439673,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.237704918032787e-05,
+      "loss": 0.1984,
+      "step": 645
+    },
+    {
+      "epoch": 1.7614178595773686,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.234972677595628e-05,
+      "loss": 0.1755,
+      "step": 646
+    },
+    {
+      "epoch": 1.7641445126107702,
+      "grad_norm": 1.6875,
+      "learning_rate": 3.2322404371584706e-05,
+      "loss": 0.2108,
+      "step": 647
+    },
+    {
+      "epoch": 1.7668711656441718,
+      "grad_norm": 1.4921875,
+      "learning_rate": 3.2295081967213116e-05,
+      "loss": 0.1759,
+      "step": 648
+    },
+    {
+      "epoch": 1.7695978186775734,
+      "grad_norm": 1.734375,
+      "learning_rate": 3.226775956284153e-05,
+      "loss": 0.2417,
+      "step": 649
+    },
+    {
+      "epoch": 1.7723244717109747,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.224043715846995e-05,
+      "loss": 0.2156,
+      "step": 650
+    },
+    {
+      "epoch": 1.7723244717109747,
+      "eval_loss": 0.25876903533935547,
+      "eval_runtime": 10.4509,
+      "eval_samples_per_second": 280.741,
+      "eval_steps_per_second": 7.081,
+      "step": 650
+    },
+    {
+      "epoch": 1.7750511247443763,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.221311475409836e-05,
+      "loss": 0.1863,
+      "step": 651
+    },
+    {
+      "epoch": 1.7777777777777777,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.2185792349726776e-05,
+      "loss": 0.2084,
+      "step": 652
+    },
+    {
+      "epoch": 1.7805044308111793,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.215846994535519e-05,
+      "loss": 0.2055,
+      "step": 653
+    },
+    {
+      "epoch": 1.7832310838445808,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.213114754098361e-05,
+      "loss": 0.1817,
+      "step": 654
+    },
+    {
+      "epoch": 1.7859577368779824,
+      "grad_norm": 1.578125,
+      "learning_rate": 3.2103825136612026e-05,
+      "loss": 0.1946,
+      "step": 655
+    },
+    {
+      "epoch": 1.7886843899113838,
+      "grad_norm": 1.8359375,
+      "learning_rate": 3.2076502732240436e-05,
+      "loss": 0.2335,
+      "step": 656
+    },
+    {
+      "epoch": 1.7914110429447851,
+      "grad_norm": 1.578125,
+      "learning_rate": 3.204918032786885e-05,
+      "loss": 0.2134,
+      "step": 657
+    },
+    {
+      "epoch": 1.7941376959781867,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.202185792349727e-05,
+      "loss": 0.1725,
+      "step": 658
+    },
+    {
+      "epoch": 1.7968643490115883,
+      "grad_norm": 1.375,
+      "learning_rate": 3.1994535519125686e-05,
+      "loss": 0.1658,
+      "step": 659
+    },
+    {
+      "epoch": 1.79959100204499,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.19672131147541e-05,
+      "loss": 0.2068,
+      "step": 660
+    },
+    {
+      "epoch": 1.8023176550783913,
+      "grad_norm": 2.125,
+      "learning_rate": 3.193989071038251e-05,
+      "loss": 0.203,
+      "step": 661
+    },
+    {
+      "epoch": 1.8050443081117926,
+      "grad_norm": 1.78125,
+      "learning_rate": 3.191256830601093e-05,
+      "loss": 0.1994,
+      "step": 662
+    },
+    {
+      "epoch": 1.8077709611451942,
+      "grad_norm": 1.953125,
+      "learning_rate": 3.1885245901639346e-05,
+      "loss": 0.1987,
+      "step": 663
+    },
+    {
+      "epoch": 1.8104976141785958,
+      "grad_norm": 1.8515625,
+      "learning_rate": 3.185792349726776e-05,
+      "loss": 0.2201,
+      "step": 664
+    },
+    {
+      "epoch": 1.8132242672119974,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.183060109289618e-05,
+      "loss": 0.207,
+      "step": 665
+    },
+    {
+      "epoch": 1.8159509202453987,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.180327868852459e-05,
+      "loss": 0.1725,
+      "step": 666
+    },
+    {
+      "epoch": 1.8186775732788003,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.1775956284153006e-05,
+      "loss": 0.2204,
+      "step": 667
+    },
+    {
+      "epoch": 1.8214042263122017,
+      "grad_norm": 1.765625,
+      "learning_rate": 3.174863387978142e-05,
+      "loss": 0.2289,
+      "step": 668
+    },
+    {
+      "epoch": 1.8241308793456033,
+      "grad_norm": 1.9609375,
+      "learning_rate": 3.172131147540984e-05,
+      "loss": 0.2366,
+      "step": 669
+    },
+    {
+      "epoch": 1.8268575323790048,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.1693989071038256e-05,
+      "loss": 0.1601,
+      "step": 670
+    },
+    {
+      "epoch": 1.8295841854124064,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.1666666666666666e-05,
+      "loss": 0.237,
+      "step": 671
+    },
+    {
+      "epoch": 1.8323108384458078,
+      "grad_norm": 1.90625,
+      "learning_rate": 3.163934426229508e-05,
+      "loss": 0.2129,
+      "step": 672
+    },
+    {
+      "epoch": 1.8350374914792091,
+      "grad_norm": 1.7890625,
+      "learning_rate": 3.161202185792349e-05,
+      "loss": 0.2035,
+      "step": 673
+    },
+    {
+      "epoch": 1.8377641445126107,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.1584699453551916e-05,
+      "loss": 0.204,
+      "step": 674
+    },
+    {
+      "epoch": 1.8404907975460123,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.155737704918033e-05,
+      "loss": 0.1954,
+      "step": 675
+    },
+    {
+      "epoch": 1.8432174505794139,
+      "grad_norm": 1.578125,
+      "learning_rate": 3.153005464480874e-05,
+      "loss": 0.2097,
+      "step": 676
+    },
+    {
+      "epoch": 1.8459441036128152,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.150273224043716e-05,
+      "loss": 0.1896,
+      "step": 677
+    },
+    {
+      "epoch": 1.8486707566462166,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.1475409836065576e-05,
+      "loss": 0.1965,
+      "step": 678
+    },
+    {
+      "epoch": 1.8513974096796182,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.144808743169399e-05,
+      "loss": 0.1523,
+      "step": 679
+    },
+    {
+      "epoch": 1.8541240627130198,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.142076502732241e-05,
+      "loss": 0.2106,
+      "step": 680
+    },
+    {
+      "epoch": 1.8568507157464214,
+      "grad_norm": 1.6875,
+      "learning_rate": 3.139344262295082e-05,
+      "loss": 0.2018,
+      "step": 681
+    },
+    {
+      "epoch": 1.8595773687798227,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.1366120218579236e-05,
+      "loss": 0.2062,
+      "step": 682
+    },
+    {
+      "epoch": 1.8623040218132243,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.133879781420765e-05,
+      "loss": 0.1732,
+      "step": 683
+    },
+    {
+      "epoch": 1.8650306748466257,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.131147540983606e-05,
+      "loss": 0.169,
+      "step": 684
+    },
+    {
+      "epoch": 1.8677573278800272,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.1284153005464487e-05,
+      "loss": 0.1973,
+      "step": 685
+    },
+    {
+      "epoch": 1.8704839809134288,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.1256830601092896e-05,
+      "loss": 0.1873,
+      "step": 686
+    },
+    {
+      "epoch": 1.8732106339468304,
+      "grad_norm": 1.8671875,
+      "learning_rate": 3.122950819672131e-05,
+      "loss": 0.2223,
+      "step": 687
+    },
+    {
+      "epoch": 1.8759372869802318,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3.120218579234973e-05,
+      "loss": 0.1571,
+      "step": 688
+    },
+    {
+      "epoch": 1.8786639400136331,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.117486338797814e-05,
+      "loss": 0.2049,
+      "step": 689
+    },
+    {
+      "epoch": 1.8813905930470347,
+      "grad_norm": 1.8046875,
+      "learning_rate": 3.114754098360656e-05,
+      "loss": 0.2364,
+      "step": 690
+    },
+    {
+      "epoch": 1.8841172460804363,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.112021857923497e-05,
+      "loss": 0.2153,
+      "step": 691
+    },
+    {
+      "epoch": 1.8868438991138379,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.109289617486339e-05,
+      "loss": 0.1776,
+      "step": 692
+    },
+    {
+      "epoch": 1.8895705521472392,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.106557377049181e-05,
+      "loss": 0.1955,
+      "step": 693
+    },
+    {
+      "epoch": 1.8922972051806408,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.1038251366120217e-05,
+      "loss": 0.2046,
+      "step": 694
+    },
+    {
+      "epoch": 1.8950238582140422,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.101092896174863e-05,
+      "loss": 0.1912,
+      "step": 695
+    },
+    {
+      "epoch": 1.8977505112474438,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.098360655737705e-05,
+      "loss": 0.1914,
+      "step": 696
+    },
+    {
+      "epoch": 1.9004771642808453,
+      "grad_norm": 1.375,
+      "learning_rate": 3.095628415300547e-05,
+      "loss": 0.1803,
+      "step": 697
+    },
+    {
+      "epoch": 1.903203817314247,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.0928961748633883e-05,
+      "loss": 0.2016,
+      "step": 698
+    },
+    {
+      "epoch": 1.9059304703476483,
+      "grad_norm": 1.3359375,
+      "learning_rate": 3.090163934426229e-05,
+      "loss": 0.1739,
+      "step": 699
+    },
+    {
+      "epoch": 1.9086571233810496,
+      "grad_norm": 1.859375,
+      "learning_rate": 3.087431693989071e-05,
+      "loss": 0.1869,
+      "step": 700
+    },
+    {
+      "epoch": 1.9086571233810496,
+      "eval_loss": 0.25387659668922424,
+      "eval_runtime": 13.7114,
+      "eval_samples_per_second": 213.983,
+      "eval_steps_per_second": 5.397,
+      "step": 700
+    },
+    {
+      "epoch": 1.9113837764144512,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.084699453551913e-05,
+      "loss": 0.1774,
+      "step": 701
+    },
+    {
+      "epoch": 1.9141104294478528,
+      "grad_norm": 1.953125,
+      "learning_rate": 3.0819672131147544e-05,
+      "loss": 0.2709,
+      "step": 702
+    },
+    {
+      "epoch": 1.9168370824812544,
+      "grad_norm": 1.78125,
+      "learning_rate": 3.079234972677596e-05,
+      "loss": 0.2049,
+      "step": 703
+    },
+    {
+      "epoch": 1.9195637355146558,
+      "grad_norm": 2.484375,
+      "learning_rate": 3.076502732240437e-05,
+      "loss": 0.2181,
+      "step": 704
+    },
+    {
+      "epoch": 1.9222903885480571,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.073770491803279e-05,
+      "loss": 0.188,
+      "step": 705
+    },
+    {
+      "epoch": 1.9250170415814587,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.0710382513661204e-05,
+      "loss": 0.1742,
+      "step": 706
+    },
+    {
+      "epoch": 1.9277436946148603,
+      "grad_norm": 1.703125,
+      "learning_rate": 3.068306010928962e-05,
+      "loss": 0.206,
+      "step": 707
+    },
+    {
+      "epoch": 1.9304703476482619,
+      "grad_norm": 1.765625,
+      "learning_rate": 3.065573770491804e-05,
+      "loss": 0.1779,
+      "step": 708
+    },
+    {
+      "epoch": 1.9331970006816632,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.062841530054645e-05,
+      "loss": 0.2112,
+      "step": 709
+    },
+    {
+      "epoch": 1.9359236537150648,
+      "grad_norm": 1.5390625,
+      "learning_rate": 3.0601092896174864e-05,
+      "loss": 0.1817,
+      "step": 710
+    },
+    {
+      "epoch": 1.9386503067484662,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.057377049180328e-05,
+      "loss": 0.2159,
+      "step": 711
+    },
+    {
+      "epoch": 1.9413769597818678,
+      "grad_norm": 1.640625,
+      "learning_rate": 3.05464480874317e-05,
+      "loss": 0.1992,
+      "step": 712
+    },
+    {
+      "epoch": 1.9441036128152693,
+      "grad_norm": 1.5859375,
+      "learning_rate": 3.0519125683060114e-05,
+      "loss": 0.1771,
+      "step": 713
+    },
+    {
+      "epoch": 1.946830265848671,
+      "grad_norm": 1.609375,
+      "learning_rate": 3.0491803278688524e-05,
+      "loss": 0.1784,
+      "step": 714
+    },
+    {
+      "epoch": 1.9495569188820723,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.046448087431694e-05,
+      "loss": 0.1602,
+      "step": 715
+    },
+    {
+      "epoch": 1.9522835719154736,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.043715846994536e-05,
+      "loss": 0.2056,
+      "step": 716
+    },
+    {
+      "epoch": 1.9550102249488752,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.040983606557377e-05,
+      "loss": 0.1936,
+      "step": 717
+    },
+    {
+      "epoch": 1.9577368779822768,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.0382513661202187e-05,
+      "loss": 0.1924,
+      "step": 718
+    },
+    {
+      "epoch": 1.9604635310156784,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.03551912568306e-05,
+      "loss": 0.1917,
+      "step": 719
+    },
+    {
+      "epoch": 1.9631901840490797,
+      "grad_norm": 1.6640625,
+      "learning_rate": 3.0327868852459017e-05,
+      "loss": 0.1844,
+      "step": 720
+    },
+    {
+      "epoch": 1.965916837082481,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.0300546448087437e-05,
+      "loss": 0.2059,
+      "step": 721
+    },
+    {
+      "epoch": 1.9686434901158827,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.0273224043715847e-05,
+      "loss": 0.1837,
+      "step": 722
+    },
+    {
+      "epoch": 1.9713701431492843,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.0245901639344264e-05,
+      "loss": 0.1796,
+      "step": 723
+    },
+    {
+      "epoch": 1.9740967961826859,
+      "grad_norm": 1.84375,
+      "learning_rate": 3.0218579234972677e-05,
+      "loss": 0.2047,
+      "step": 724
+    },
+    {
+      "epoch": 1.9768234492160872,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.0191256830601094e-05,
+      "loss": 0.212,
+      "step": 725
+    },
+    {
+      "epoch": 1.9795501022494888,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.016393442622951e-05,
+      "loss": 0.2343,
+      "step": 726
+    },
+    {
+      "epoch": 1.9822767552828902,
+      "grad_norm": 1.3515625,
+      "learning_rate": 3.0136612021857924e-05,
+      "loss": 0.1618,
+      "step": 727
+    },
+    {
+      "epoch": 1.9850034083162917,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.010928961748634e-05,
+      "loss": 0.2242,
+      "step": 728
+    },
+    {
+      "epoch": 1.9877300613496933,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.0081967213114754e-05,
+      "loss": 0.1894,
+      "step": 729
+    },
+    {
+      "epoch": 1.990456714383095,
+      "grad_norm": 1.7265625,
+      "learning_rate": 3.005464480874317e-05,
+      "loss": 0.2147,
+      "step": 730
+    },
+    {
+      "epoch": 1.9931833674164963,
+      "grad_norm": 1.828125,
+      "learning_rate": 3.0027322404371587e-05,
+      "loss": 0.2423,
+      "step": 731
+    },
+    {
+      "epoch": 1.9959100204498976,
+      "grad_norm": 1.546875,
+      "learning_rate": 3e-05,
+      "loss": 0.1763,
+      "step": 732
+    },
+    {
+      "epoch": 1.9986366734832992,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.9972677595628417e-05,
+      "loss": 0.1991,
+      "step": 733
+    },
+    {
+      "epoch": 2.001363326516701,
+      "grad_norm": 1.375,
+      "learning_rate": 2.994535519125683e-05,
+      "loss": 0.1528,
+      "step": 734
+    },
+    {
+      "epoch": 2.0040899795501024,
+      "grad_norm": 1.1328125,
+      "learning_rate": 2.9918032786885248e-05,
+      "loss": 0.1245,
+      "step": 735
+    },
+    {
+      "epoch": 2.006816632583504,
+      "grad_norm": 1.3125,
+      "learning_rate": 2.9890710382513664e-05,
+      "loss": 0.1446,
+      "step": 736
+    },
+    {
+      "epoch": 2.009543285616905,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.9863387978142078e-05,
+      "loss": 0.1325,
+      "step": 737
+    },
+    {
+      "epoch": 2.0122699386503067,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.9836065573770494e-05,
+      "loss": 0.137,
+      "step": 738
+    },
+    {
+      "epoch": 2.0149965916837083,
+      "grad_norm": 1.2265625,
+      "learning_rate": 2.9808743169398908e-05,
+      "loss": 0.127,
+      "step": 739
+    },
+    {
+      "epoch": 2.01772324471711,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2.9781420765027324e-05,
+      "loss": 0.1174,
+      "step": 740
+    },
+    {
+      "epoch": 2.0204498977505114,
+      "grad_norm": 2.21875,
+      "learning_rate": 2.975409836065574e-05,
+      "loss": 0.1886,
+      "step": 741
+    },
+    {
+      "epoch": 2.0231765507839126,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2.9726775956284154e-05,
+      "loss": 0.1623,
+      "step": 742
+    },
+    {
+      "epoch": 2.025903203817314,
+      "grad_norm": 1.71875,
+      "learning_rate": 2.969945355191257e-05,
+      "loss": 0.1096,
+      "step": 743
+    },
+    {
+      "epoch": 2.0286298568507157,
+      "grad_norm": 1.84375,
+      "learning_rate": 2.967213114754098e-05,
+      "loss": 0.1009,
+      "step": 744
+    },
+    {
+      "epoch": 2.0313565098841173,
+      "grad_norm": 2.421875,
+      "learning_rate": 2.96448087431694e-05,
+      "loss": 0.129,
+      "step": 745
+    },
+    {
+      "epoch": 2.034083162917519,
+      "grad_norm": 2.328125,
+      "learning_rate": 2.9617486338797818e-05,
+      "loss": 0.126,
+      "step": 746
+    },
+    {
+      "epoch": 2.03680981595092,
+      "grad_norm": 4.125,
+      "learning_rate": 2.9590163934426228e-05,
+      "loss": 0.1439,
+      "step": 747
+    },
+    {
+      "epoch": 2.0395364689843216,
+      "grad_norm": 2.0,
+      "learning_rate": 2.9562841530054648e-05,
+      "loss": 0.1212,
+      "step": 748
+    },
+    {
+      "epoch": 2.042263122017723,
+      "grad_norm": 2.140625,
+      "learning_rate": 2.9535519125683058e-05,
+      "loss": 0.1045,
+      "step": 749
+    },
+    {
+      "epoch": 2.044989775051125,
+      "grad_norm": 2.984375,
+      "learning_rate": 2.9508196721311478e-05,
+      "loss": 0.124,
+      "step": 750
+    },
+    {
+      "epoch": 2.044989775051125,
+      "eval_loss": 0.30449798703193665,
+      "eval_runtime": 10.471,
+      "eval_samples_per_second": 280.202,
+      "eval_steps_per_second": 7.067,
+      "step": 750
+    },
+    {
+      "epoch": 2.0477164280845264,
+      "grad_norm": 2.1875,
+      "learning_rate": 2.9480874316939895e-05,
+      "loss": 0.1237,
+      "step": 751
+    },
+    {
+      "epoch": 2.050443081117928,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.9453551912568304e-05,
+      "loss": 0.1463,
+      "step": 752
+    },
+    {
+      "epoch": 2.053169734151329,
+      "grad_norm": 1.984375,
+      "learning_rate": 2.9426229508196725e-05,
+      "loss": 0.1343,
+      "step": 753
+    },
+    {
+      "epoch": 2.0558963871847307,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.939890710382514e-05,
+      "loss": 0.1316,
+      "step": 754
+    },
+    {
+      "epoch": 2.0586230402181322,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.937158469945355e-05,
+      "loss": 0.1456,
+      "step": 755
+    },
+    {
+      "epoch": 2.061349693251534,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.934426229508197e-05,
+      "loss": 0.1312,
+      "step": 756
+    },
+    {
+      "epoch": 2.0640763462849354,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.931693989071038e-05,
+      "loss": 0.1216,
+      "step": 757
+    },
+    {
+      "epoch": 2.0668029993183366,
+      "grad_norm": 1.875,
+      "learning_rate": 2.9289617486338798e-05,
+      "loss": 0.1142,
+      "step": 758
+    },
+    {
+      "epoch": 2.069529652351738,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.9262295081967218e-05,
+      "loss": 0.1217,
+      "step": 759
+    },
+    {
+      "epoch": 2.0722563053851397,
+      "grad_norm": 1.34375,
+      "learning_rate": 2.9234972677595628e-05,
+      "loss": 0.1182,
+      "step": 760
+    },
+    {
+      "epoch": 2.0749829584185413,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.9207650273224048e-05,
+      "loss": 0.1347,
+      "step": 761
+    },
+    {
+      "epoch": 2.077709611451943,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.9180327868852458e-05,
+      "loss": 0.1171,
+      "step": 762
+    },
+    {
+      "epoch": 2.0804362644853445,
+      "grad_norm": 1.2265625,
+      "learning_rate": 2.9153005464480875e-05,
+      "loss": 0.0848,
+      "step": 763
+    },
+    {
+      "epoch": 2.0831629175187456,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.9125683060109295e-05,
+      "loss": 0.1303,
+      "step": 764
+    },
+    {
+      "epoch": 2.085889570552147,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.9098360655737705e-05,
+      "loss": 0.0987,
+      "step": 765
+    },
+    {
+      "epoch": 2.0886162235855488,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.907103825136612e-05,
+      "loss": 0.1319,
+      "step": 766
+    },
+    {
+      "epoch": 2.0913428766189504,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.9043715846994535e-05,
+      "loss": 0.1282,
+      "step": 767
+    },
+    {
+      "epoch": 2.094069529652352,
+      "grad_norm": 1.90625,
+      "learning_rate": 2.901639344262295e-05,
+      "loss": 0.1432,
+      "step": 768
+    },
+    {
+      "epoch": 2.096796182685753,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.8989071038251368e-05,
+      "loss": 0.1389,
+      "step": 769
+    },
+    {
+      "epoch": 2.0995228357191547,
+      "grad_norm": 1.734375,
+      "learning_rate": 2.896174863387978e-05,
+      "loss": 0.1268,
+      "step": 770
+    },
+    {
+      "epoch": 2.1022494887525562,
+      "grad_norm": 1.875,
+      "learning_rate": 2.8934426229508198e-05,
+      "loss": 0.1131,
+      "step": 771
+    },
+    {
+      "epoch": 2.104976141785958,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.890710382513661e-05,
+      "loss": 0.1094,
+      "step": 772
+    },
+    {
+      "epoch": 2.1077027948193594,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.8879781420765028e-05,
+      "loss": 0.1328,
+      "step": 773
+    },
+    {
+      "epoch": 2.1104294478527605,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.8852459016393445e-05,
+      "loss": 0.1357,
+      "step": 774
+    },
+    {
+      "epoch": 2.113156100886162,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.8825136612021858e-05,
+      "loss": 0.1558,
+      "step": 775
+    },
+    {
+      "epoch": 2.1158827539195637,
+      "grad_norm": 1.59375,
+      "learning_rate": 2.8797814207650275e-05,
+      "loss": 0.1215,
+      "step": 776
+    },
+    {
+      "epoch": 2.1186094069529653,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.877049180327869e-05,
+      "loss": 0.1253,
+      "step": 777
+    },
+    {
+      "epoch": 2.121336059986367,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2.8743169398907105e-05,
+      "loss": 0.1476,
+      "step": 778
+    },
+    {
+      "epoch": 2.124062713019768,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.8715846994535522e-05,
+      "loss": 0.113,
+      "step": 779
+    },
+    {
+      "epoch": 2.1267893660531696,
+      "grad_norm": 3.84375,
+      "learning_rate": 2.8688524590163935e-05,
+      "loss": 0.1733,
+      "step": 780
+    },
+    {
+      "epoch": 2.129516019086571,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.8661202185792352e-05,
+      "loss": 0.1346,
+      "step": 781
+    },
+    {
+      "epoch": 2.1322426721199728,
+      "grad_norm": 1.59375,
+      "learning_rate": 2.8633879781420765e-05,
+      "loss": 0.127,
+      "step": 782
+    },
+    {
+      "epoch": 2.1349693251533743,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.8606557377049182e-05,
+      "loss": 0.1237,
+      "step": 783
+    },
+    {
+      "epoch": 2.137695978186776,
+      "grad_norm": 2.34375,
+      "learning_rate": 2.85792349726776e-05,
+      "loss": 0.163,
+      "step": 784
+    },
+    {
+      "epoch": 2.140422631220177,
+      "grad_norm": 1.625,
+      "learning_rate": 2.8551912568306012e-05,
+      "loss": 0.1474,
+      "step": 785
+    },
+    {
+      "epoch": 2.1431492842535786,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.852459016393443e-05,
+      "loss": 0.1208,
+      "step": 786
+    },
+    {
+      "epoch": 2.1458759372869802,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.8497267759562845e-05,
+      "loss": 0.0973,
+      "step": 787
+    },
+    {
+      "epoch": 2.148602590320382,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.846994535519126e-05,
+      "loss": 0.1343,
+      "step": 788
+    },
+    {
+      "epoch": 2.1513292433537834,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.8442622950819675e-05,
+      "loss": 0.1351,
+      "step": 789
+    },
+    {
+      "epoch": 2.1540558963871845,
+      "grad_norm": 2.3125,
+      "learning_rate": 2.841530054644809e-05,
+      "loss": 0.1322,
+      "step": 790
+    },
+    {
+      "epoch": 2.156782549420586,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.8387978142076505e-05,
+      "loss": 0.1458,
+      "step": 791
+    },
+    {
+      "epoch": 2.1595092024539877,
+      "grad_norm": 1.5,
+      "learning_rate": 2.8360655737704922e-05,
+      "loss": 0.1271,
+      "step": 792
+    },
+    {
+      "epoch": 2.1622358554873893,
+      "grad_norm": 1.984375,
+      "learning_rate": 2.8333333333333335e-05,
+      "loss": 0.1484,
+      "step": 793
+    },
+    {
+      "epoch": 2.164962508520791,
+      "grad_norm": 1.890625,
+      "learning_rate": 2.8306010928961752e-05,
+      "loss": 0.1314,
+      "step": 794
+    },
+    {
+      "epoch": 2.1676891615541924,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.8278688524590162e-05,
+      "loss": 0.1271,
+      "step": 795
+    },
+    {
+      "epoch": 2.1704158145875936,
+      "grad_norm": 2.0625,
+      "learning_rate": 2.8251366120218582e-05,
+      "loss": 0.1255,
+      "step": 796
+    },
+    {
+      "epoch": 2.173142467620995,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2.8224043715847e-05,
+      "loss": 0.1578,
+      "step": 797
+    },
+    {
+      "epoch": 2.1758691206543967,
+      "grad_norm": 1.515625,
+      "learning_rate": 2.819672131147541e-05,
+      "loss": 0.1196,
+      "step": 798
+    },
+    {
+      "epoch": 2.1785957736877983,
+      "grad_norm": 2.546875,
+      "learning_rate": 2.816939890710383e-05,
+      "loss": 0.1323,
+      "step": 799
+    },
+    {
+      "epoch": 2.1813224267212,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.814207650273224e-05,
+      "loss": 0.1368,
+      "step": 800
+    },
+    {
+      "epoch": 2.1813224267212,
+      "eval_loss": 0.2813376486301422,
+      "eval_runtime": 10.4513,
+      "eval_samples_per_second": 280.732,
+      "eval_steps_per_second": 7.08,
+      "step": 800
+    },
+    {
+      "epoch": 2.184049079754601,
+      "grad_norm": 1.515625,
+      "learning_rate": 2.811475409836066e-05,
+      "loss": 0.1409,
+      "step": 801
+    },
+    {
+      "epoch": 2.1867757327880026,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.8087431693989076e-05,
+      "loss": 0.1158,
+      "step": 802
+    },
+    {
+      "epoch": 2.189502385821404,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2.8060109289617486e-05,
+      "loss": 0.1319,
+      "step": 803
+    },
+    {
+      "epoch": 2.192229038854806,
+      "grad_norm": 1.5,
+      "learning_rate": 2.8032786885245906e-05,
+      "loss": 0.1246,
+      "step": 804
+    },
+    {
+      "epoch": 2.1949556918882074,
+      "grad_norm": 1.71875,
+      "learning_rate": 2.8005464480874316e-05,
+      "loss": 0.1536,
+      "step": 805
+    },
+    {
+      "epoch": 2.1976823449216085,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2.7978142076502732e-05,
+      "loss": 0.1477,
+      "step": 806
+    },
+    {
+      "epoch": 2.20040899795501,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.7950819672131152e-05,
+      "loss": 0.1376,
+      "step": 807
+    },
+    {
+      "epoch": 2.2031356509884117,
+      "grad_norm": 2.328125,
+      "learning_rate": 2.7923497267759562e-05,
+      "loss": 0.1282,
+      "step": 808
+    },
+    {
+      "epoch": 2.2058623040218133,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.789617486338798e-05,
+      "loss": 0.126,
+      "step": 809
+    },
+    {
+      "epoch": 2.208588957055215,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.7868852459016392e-05,
+      "loss": 0.1363,
+      "step": 810
+    },
+    {
+      "epoch": 2.211315610088616,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.784153005464481e-05,
+      "loss": 0.1372,
+      "step": 811
+    },
+    {
+      "epoch": 2.2140422631220176,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2.7814207650273226e-05,
+      "loss": 0.1307,
+      "step": 812
+    },
+    {
+      "epoch": 2.216768916155419,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2.778688524590164e-05,
+      "loss": 0.1259,
+      "step": 813
+    },
+    {
+      "epoch": 2.2194955691888207,
+      "grad_norm": 2.0,
+      "learning_rate": 2.7759562841530056e-05,
+      "loss": 0.136,
+      "step": 814
+    },
+    {
+      "epoch": 2.2222222222222223,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.773224043715847e-05,
+      "loss": 0.1238,
+      "step": 815
+    },
+    {
+      "epoch": 2.224948875255624,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.7704918032786886e-05,
+      "loss": 0.1252,
+      "step": 816
+    },
+    {
+      "epoch": 2.227675528289025,
+      "grad_norm": 1.9375,
+      "learning_rate": 2.7677595628415303e-05,
+      "loss": 0.16,
+      "step": 817
+    },
+    {
+      "epoch": 2.2304021813224266,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.7650273224043716e-05,
+      "loss": 0.1643,
+      "step": 818
+    },
+    {
+      "epoch": 2.233128834355828,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.7622950819672133e-05,
+      "loss": 0.1445,
+      "step": 819
+    },
+    {
+      "epoch": 2.23585548738923,
+      "grad_norm": 1.8125,
+      "learning_rate": 2.7595628415300546e-05,
+      "loss": 0.1597,
+      "step": 820
+    },
+    {
+      "epoch": 2.2385821404226314,
+      "grad_norm": 2.546875,
+      "learning_rate": 2.7568306010928963e-05,
+      "loss": 0.1602,
+      "step": 821
+    },
+    {
+      "epoch": 2.2413087934560325,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2.754098360655738e-05,
+      "loss": 0.1423,
+      "step": 822
+    },
+    {
+      "epoch": 2.244035446489434,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.7513661202185793e-05,
+      "loss": 0.1178,
+      "step": 823
+    },
+    {
+      "epoch": 2.2467620995228357,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.748633879781421e-05,
+      "loss": 0.1407,
+      "step": 824
+    },
+    {
+      "epoch": 2.2494887525562373,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.7459016393442626e-05,
+      "loss": 0.1391,
+      "step": 825
+    },
+    {
+      "epoch": 2.252215405589639,
+      "grad_norm": 2.0625,
+      "learning_rate": 2.743169398907104e-05,
+      "loss": 0.1248,
+      "step": 826
+    },
+    {
+      "epoch": 2.2549420586230404,
+      "grad_norm": 1.765625,
+      "learning_rate": 2.7404371584699456e-05,
+      "loss": 0.132,
+      "step": 827
+    },
+    {
+      "epoch": 2.2576687116564416,
+      "grad_norm": 1.703125,
+      "learning_rate": 2.737704918032787e-05,
+      "loss": 0.1342,
+      "step": 828
+    },
+    {
+      "epoch": 2.260395364689843,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.7349726775956286e-05,
+      "loss": 0.1194,
+      "step": 829
+    },
+    {
+      "epoch": 2.2631220177232447,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2.7322404371584703e-05,
+      "loss": 0.138,
+      "step": 830
+    },
+    {
+      "epoch": 2.2658486707566463,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.7295081967213116e-05,
+      "loss": 0.1642,
+      "step": 831
+    },
+    {
+      "epoch": 2.268575323790048,
+      "grad_norm": 1.875,
+      "learning_rate": 2.7267759562841533e-05,
+      "loss": 0.1368,
+      "step": 832
+    },
+    {
+      "epoch": 2.271301976823449,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.7240437158469946e-05,
+      "loss": 0.1232,
+      "step": 833
+    },
+    {
+      "epoch": 2.2740286298568506,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2.7213114754098363e-05,
+      "loss": 0.1255,
+      "step": 834
+    },
+    {
+      "epoch": 2.276755282890252,
+      "grad_norm": 1.734375,
+      "learning_rate": 2.718579234972678e-05,
+      "loss": 0.1399,
+      "step": 835
+    },
+    {
+      "epoch": 2.279481935923654,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.7158469945355193e-05,
+      "loss": 0.1274,
+      "step": 836
+    },
+    {
+      "epoch": 2.2822085889570554,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.713114754098361e-05,
+      "loss": 0.1383,
+      "step": 837
+    },
+    {
+      "epoch": 2.284935241990457,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2.710382513661202e-05,
+      "loss": 0.1189,
+      "step": 838
+    },
+    {
+      "epoch": 2.287661895023858,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.707650273224044e-05,
+      "loss": 0.1553,
+      "step": 839
+    },
+    {
+      "epoch": 2.2903885480572597,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.7049180327868856e-05,
+      "loss": 0.133,
+      "step": 840
+    },
+    {
+      "epoch": 2.2931152010906612,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2.7021857923497266e-05,
+      "loss": 0.1355,
+      "step": 841
+    },
+    {
+      "epoch": 2.295841854124063,
+      "grad_norm": 2.1875,
+      "learning_rate": 2.6994535519125686e-05,
+      "loss": 0.1662,
+      "step": 842
+    },
+    {
+      "epoch": 2.298568507157464,
+      "grad_norm": 2.265625,
+      "learning_rate": 2.6967213114754096e-05,
+      "loss": 0.1237,
+      "step": 843
+    },
+    {
+      "epoch": 2.3012951601908656,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.6939890710382516e-05,
+      "loss": 0.1362,
+      "step": 844
+    },
+    {
+      "epoch": 2.304021813224267,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.6912568306010933e-05,
+      "loss": 0.1544,
+      "step": 845
+    },
+    {
+      "epoch": 2.3067484662576687,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.6885245901639343e-05,
+      "loss": 0.1158,
+      "step": 846
+    },
+    {
+      "epoch": 2.3094751192910703,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.6857923497267763e-05,
+      "loss": 0.1424,
+      "step": 847
+    },
+    {
+      "epoch": 2.312201772324472,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.6830601092896173e-05,
+      "loss": 0.1182,
+      "step": 848
+    },
+    {
+      "epoch": 2.314928425357873,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2.680327868852459e-05,
+      "loss": 0.1315,
+      "step": 849
+    },
+    {
+      "epoch": 2.3176550783912746,
+      "grad_norm": 1.6875,
+      "learning_rate": 2.677595628415301e-05,
+      "loss": 0.1293,
+      "step": 850
+    },
+    {
+      "epoch": 2.3176550783912746,
+      "eval_loss": 0.278757244348526,
+      "eval_runtime": 10.468,
+      "eval_samples_per_second": 280.283,
+      "eval_steps_per_second": 7.069,
+      "step": 850
+    },
+    {
+      "epoch": 2.320381731424676,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2.674863387978142e-05,
+      "loss": 0.1268,
+      "step": 851
+    },
+    {
+      "epoch": 2.3231083844580778,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.6721311475409837e-05,
+      "loss": 0.1016,
+      "step": 852
+    },
+    {
+      "epoch": 2.3258350374914794,
+      "grad_norm": 1.5625,
+      "learning_rate": 2.669398907103825e-05,
+      "loss": 0.1364,
+      "step": 853
+    },
+    {
+      "epoch": 2.3285616905248805,
+      "grad_norm": 2.625,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 0.1397,
+      "step": 854
+    },
+    {
+      "epoch": 2.331288343558282,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.6639344262295087e-05,
+      "loss": 0.1403,
+      "step": 855
+    },
+    {
+      "epoch": 2.3340149965916837,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.6612021857923497e-05,
+      "loss": 0.133,
+      "step": 856
+    },
+    {
+      "epoch": 2.3367416496250852,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.6584699453551913e-05,
+      "loss": 0.1119,
+      "step": 857
+    },
+    {
+      "epoch": 2.339468302658487,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.6557377049180327e-05,
+      "loss": 0.1207,
+      "step": 858
+    },
+    {
+      "epoch": 2.3421949556918884,
+      "grad_norm": 1.59375,
+      "learning_rate": 2.6530054644808743e-05,
+      "loss": 0.1156,
+      "step": 859
+    },
+    {
+      "epoch": 2.3449216087252895,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.650273224043716e-05,
+      "loss": 0.1422,
+      "step": 860
+    },
+    {
+      "epoch": 2.347648261758691,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.6475409836065573e-05,
+      "loss": 0.1288,
+      "step": 861
+    },
+    {
+      "epoch": 2.3503749147920927,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.644808743169399e-05,
+      "loss": 0.1229,
+      "step": 862
+    },
+    {
+      "epoch": 2.3531015678254943,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.6420765027322407e-05,
+      "loss": 0.1069,
+      "step": 863
+    },
+    {
+      "epoch": 2.355828220858896,
+      "grad_norm": 1.8125,
+      "learning_rate": 2.639344262295082e-05,
+      "loss": 0.1357,
+      "step": 864
+    },
+    {
+      "epoch": 2.358554873892297,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.6366120218579237e-05,
+      "loss": 0.1462,
+      "step": 865
+    },
+    {
+      "epoch": 2.3612815269256986,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2.633879781420765e-05,
+      "loss": 0.1183,
+      "step": 866
+    },
+    {
+      "epoch": 2.3640081799591,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2.6311475409836067e-05,
+      "loss": 0.1333,
+      "step": 867
+    },
+    {
+      "epoch": 2.3667348329925018,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.6284153005464484e-05,
+      "loss": 0.1525,
+      "step": 868
+    },
+    {
+      "epoch": 2.3694614860259033,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.6256830601092897e-05,
+      "loss": 0.1264,
+      "step": 869
+    },
+    {
+      "epoch": 2.372188139059305,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2.6229508196721314e-05,
+      "loss": 0.1349,
+      "step": 870
+    },
+    {
+      "epoch": 2.374914792092706,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2.6202185792349727e-05,
+      "loss": 0.1222,
+      "step": 871
+    },
+    {
+      "epoch": 2.3776414451261076,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.6174863387978144e-05,
+      "loss": 0.1652,
+      "step": 872
+    },
+    {
+      "epoch": 2.3803680981595092,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.614754098360656e-05,
+      "loss": 0.1412,
+      "step": 873
+    },
+    {
+      "epoch": 2.383094751192911,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.6120218579234974e-05,
+      "loss": 0.1277,
+      "step": 874
+    },
+    {
+      "epoch": 2.3858214042263124,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.609289617486339e-05,
+      "loss": 0.1418,
+      "step": 875
+    },
+    {
+      "epoch": 2.3885480572597135,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2.6065573770491804e-05,
+      "loss": 0.1155,
+      "step": 876
+    },
+    {
+      "epoch": 2.391274710293115,
+      "grad_norm": 1.6875,
+      "learning_rate": 2.603825136612022e-05,
+      "loss": 0.1495,
+      "step": 877
+    },
+    {
+      "epoch": 2.3940013633265167,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2.6010928961748637e-05,
+      "loss": 0.1408,
+      "step": 878
+    },
+    {
+      "epoch": 2.3967280163599183,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.598360655737705e-05,
+      "loss": 0.1381,
+      "step": 879
+    },
+    {
+      "epoch": 2.39945466939332,
+      "grad_norm": 1.5625,
+      "learning_rate": 2.5956284153005467e-05,
+      "loss": 0.1336,
+      "step": 880
+    },
+    {
+      "epoch": 2.4021813224267214,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2.5928961748633877e-05,
+      "loss": 0.1232,
+      "step": 881
+    },
+    {
+      "epoch": 2.4049079754601226,
+      "grad_norm": 1.640625,
+      "learning_rate": 2.5901639344262297e-05,
+      "loss": 0.1325,
+      "step": 882
+    },
+    {
+      "epoch": 2.407634628493524,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.5874316939890714e-05,
+      "loss": 0.1256,
+      "step": 883
+    },
+    {
+      "epoch": 2.4103612815269257,
+      "grad_norm": 2.625,
+      "learning_rate": 2.5846994535519127e-05,
+      "loss": 0.1461,
+      "step": 884
+    },
+    {
+      "epoch": 2.4130879345603273,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.5819672131147544e-05,
+      "loss": 0.1104,
+      "step": 885
+    },
+    {
+      "epoch": 2.4158145875937285,
+      "grad_norm": 2.5,
+      "learning_rate": 2.5792349726775954e-05,
+      "loss": 0.157,
+      "step": 886
+    },
+    {
+      "epoch": 2.41854124062713,
+      "grad_norm": 1.921875,
+      "learning_rate": 2.5765027322404374e-05,
+      "loss": 0.136,
+      "step": 887
+    },
+    {
+      "epoch": 2.4212678936605316,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2.573770491803279e-05,
+      "loss": 0.1057,
+      "step": 888
+    },
+    {
+      "epoch": 2.423994546693933,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.57103825136612e-05,
+      "loss": 0.1135,
+      "step": 889
+    },
+    {
+      "epoch": 2.426721199727335,
+      "grad_norm": 1.609375,
+      "learning_rate": 2.568306010928962e-05,
+      "loss": 0.1197,
+      "step": 890
+    },
+    {
+      "epoch": 2.4294478527607364,
+      "grad_norm": 1.71875,
+      "learning_rate": 2.565573770491803e-05,
+      "loss": 0.1542,
+      "step": 891
+    },
+    {
+      "epoch": 2.4321745057941375,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2.5628415300546447e-05,
+      "loss": 0.1422,
+      "step": 892
+    },
+    {
+      "epoch": 2.434901158827539,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.5601092896174867e-05,
+      "loss": 0.128,
+      "step": 893
+    },
+    {
+      "epoch": 2.4376278118609407,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.5573770491803277e-05,
+      "loss": 0.1129,
+      "step": 894
+    },
+    {
+      "epoch": 2.4403544648943423,
+      "grad_norm": 1.921875,
+      "learning_rate": 2.5546448087431697e-05,
+      "loss": 0.1378,
+      "step": 895
+    },
+    {
+      "epoch": 2.443081117927744,
+      "grad_norm": 1.8125,
+      "learning_rate": 2.5519125683060107e-05,
+      "loss": 0.1482,
+      "step": 896
+    },
+    {
+      "epoch": 2.445807770961145,
+      "grad_norm": 1.75,
+      "learning_rate": 2.5491803278688524e-05,
+      "loss": 0.1122,
+      "step": 897
+    },
+    {
+      "epoch": 2.4485344239945466,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.5464480874316944e-05,
+      "loss": 0.1425,
+      "step": 898
+    },
+    {
+      "epoch": 2.451261077027948,
+      "grad_norm": 1.640625,
+      "learning_rate": 2.5437158469945354e-05,
+      "loss": 0.1223,
+      "step": 899
+    },
+    {
+      "epoch": 2.4539877300613497,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.540983606557377e-05,
+      "loss": 0.1613,
+      "step": 900
+    },
+    {
+      "epoch": 2.4539877300613497,
+      "eval_loss": 0.27992647886276245,
+      "eval_runtime": 10.4815,
+      "eval_samples_per_second": 279.922,
+      "eval_steps_per_second": 7.06,
+      "step": 900
+    },
+    {
+      "epoch": 2.4567143830947513,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.538251366120219e-05,
+      "loss": 0.1244,
+      "step": 901
+    },
+    {
+      "epoch": 2.459441036128153,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.53551912568306e-05,
+      "loss": 0.161,
+      "step": 902
+    },
+    {
+      "epoch": 2.462167689161554,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.5327868852459018e-05,
+      "loss": 0.1346,
+      "step": 903
+    },
+    {
+      "epoch": 2.4648943421949556,
+      "grad_norm": 1.984375,
+      "learning_rate": 2.530054644808743e-05,
+      "loss": 0.139,
+      "step": 904
+    },
+    {
+      "epoch": 2.467620995228357,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2.5273224043715848e-05,
+      "loss": 0.1149,
+      "step": 905
+    },
+    {
+      "epoch": 2.470347648261759,
+      "grad_norm": 1.984375,
+      "learning_rate": 2.5245901639344264e-05,
+      "loss": 0.1382,
+      "step": 906
+    },
+    {
+      "epoch": 2.4730743012951604,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.5218579234972678e-05,
+      "loss": 0.0984,
+      "step": 907
+    },
+    {
+      "epoch": 2.4758009543285615,
+      "grad_norm": 1.84375,
+      "learning_rate": 2.5191256830601094e-05,
+      "loss": 0.1421,
+      "step": 908
+    },
+    {
+      "epoch": 2.478527607361963,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2.5163934426229508e-05,
+      "loss": 0.1348,
+      "step": 909
+    },
+    {
+      "epoch": 2.4812542603953647,
+      "grad_norm": 1.609375,
+      "learning_rate": 2.5136612021857924e-05,
+      "loss": 0.133,
+      "step": 910
+    },
+    {
+      "epoch": 2.4839809134287663,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.510928961748634e-05,
+      "loss": 0.1305,
+      "step": 911
+    },
+    {
+      "epoch": 2.486707566462168,
+      "grad_norm": 2.296875,
+      "learning_rate": 2.5081967213114754e-05,
+      "loss": 0.1048,
+      "step": 912
+    },
+    {
+      "epoch": 2.4894342194955694,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.505464480874317e-05,
+      "loss": 0.1408,
+      "step": 913
+    },
+    {
+      "epoch": 2.4921608725289706,
+      "grad_norm": 1.5,
+      "learning_rate": 2.5027322404371584e-05,
+      "loss": 0.0913,
+      "step": 914
+    },
+    {
+      "epoch": 2.494887525562372,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.5e-05,
+      "loss": 0.1243,
+      "step": 915
+    },
+    {
+      "epoch": 2.4976141785957737,
+      "grad_norm": 1.6875,
+      "learning_rate": 2.4972677595628415e-05,
+      "loss": 0.1243,
+      "step": 916
+    },
+    {
+      "epoch": 2.5003408316291753,
+      "grad_norm": 1.890625,
+      "learning_rate": 2.494535519125683e-05,
+      "loss": 0.1554,
+      "step": 917
+    },
+    {
+      "epoch": 2.5030674846625764,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.4918032786885248e-05,
+      "loss": 0.1523,
+      "step": 918
+    },
+    {
+      "epoch": 2.505794137695978,
+      "grad_norm": 1.734375,
+      "learning_rate": 2.489071038251366e-05,
+      "loss": 0.1266,
+      "step": 919
+    },
+    {
+      "epoch": 2.5085207907293796,
+      "grad_norm": 1.765625,
+      "learning_rate": 2.4863387978142078e-05,
+      "loss": 0.1228,
+      "step": 920
+    },
+    {
+      "epoch": 2.511247443762781,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2.483606557377049e-05,
+      "loss": 0.1557,
+      "step": 921
+    },
+    {
+      "epoch": 2.5139740967961828,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.4808743169398908e-05,
+      "loss": 0.1475,
+      "step": 922
+    },
+    {
+      "epoch": 2.5167007498295844,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.4781420765027325e-05,
+      "loss": 0.1213,
+      "step": 923
+    },
+    {
+      "epoch": 2.519427402862986,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2.4754098360655738e-05,
+      "loss": 0.1167,
+      "step": 924
+    },
+    {
+      "epoch": 2.522154055896387,
+      "grad_norm": 1.8125,
+      "learning_rate": 2.4726775956284155e-05,
+      "loss": 0.0924,
+      "step": 925
+    },
+    {
+      "epoch": 2.5248807089297887,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.4699453551912568e-05,
+      "loss": 0.1235,
+      "step": 926
+    },
+    {
+      "epoch": 2.5276073619631902,
+      "grad_norm": 1.75,
+      "learning_rate": 2.4672131147540985e-05,
+      "loss": 0.1525,
+      "step": 927
+    },
+    {
+      "epoch": 2.530334014996592,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.46448087431694e-05,
+      "loss": 0.116,
+      "step": 928
+    },
+    {
+      "epoch": 2.533060668029993,
+      "grad_norm": 1.6875,
+      "learning_rate": 2.4617486338797815e-05,
+      "loss": 0.1234,
+      "step": 929
+    },
+    {
+      "epoch": 2.5357873210633946,
+      "grad_norm": 1.65625,
+      "learning_rate": 2.459016393442623e-05,
+      "loss": 0.1362,
+      "step": 930
+    },
+    {
+      "epoch": 2.538513974096796,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.4562841530054645e-05,
+      "loss": 0.1447,
+      "step": 931
+    },
+    {
+      "epoch": 2.5412406271301977,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2.453551912568306e-05,
+      "loss": 0.1345,
+      "step": 932
+    },
+    {
+      "epoch": 2.5439672801635993,
+      "grad_norm": 1.71875,
+      "learning_rate": 2.4508196721311478e-05,
+      "loss": 0.1164,
+      "step": 933
+    },
+    {
+      "epoch": 2.546693933197001,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.448087431693989e-05,
+      "loss": 0.1182,
+      "step": 934
+    },
+    {
+      "epoch": 2.549420586230402,
+      "grad_norm": 1.59375,
+      "learning_rate": 2.4453551912568305e-05,
+      "loss": 0.1269,
+      "step": 935
+    },
+    {
+      "epoch": 2.5521472392638036,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.442622950819672e-05,
+      "loss": 0.124,
+      "step": 936
+    },
+    {
+      "epoch": 2.554873892297205,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2.439890710382514e-05,
+      "loss": 0.1458,
+      "step": 937
+    },
+    {
+      "epoch": 2.5576005453306068,
+      "grad_norm": 1.8125,
+      "learning_rate": 2.4371584699453555e-05,
+      "loss": 0.1377,
+      "step": 938
+    },
+    {
+      "epoch": 2.560327198364008,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2.434426229508197e-05,
+      "loss": 0.1341,
+      "step": 939
+    },
+    {
+      "epoch": 2.5630538513974095,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.431693989071038e-05,
+      "loss": 0.1399,
+      "step": 940
+    },
+    {
+      "epoch": 2.565780504430811,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.42896174863388e-05,
+      "loss": 0.138,
+      "step": 941
+    },
+    {
+      "epoch": 2.5685071574642127,
+      "grad_norm": 1.5,
+      "learning_rate": 2.4262295081967215e-05,
+      "loss": 0.1159,
+      "step": 942
+    },
+    {
+      "epoch": 2.5712338104976142,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.423497267759563e-05,
+      "loss": 0.1405,
+      "step": 943
+    },
+    {
+      "epoch": 2.573960463531016,
+      "grad_norm": 1.5,
+      "learning_rate": 2.4207650273224045e-05,
+      "loss": 0.1383,
+      "step": 944
+    },
+    {
+      "epoch": 2.5766871165644174,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.418032786885246e-05,
+      "loss": 0.1232,
+      "step": 945
+    },
+    {
+      "epoch": 2.5794137695978185,
+      "grad_norm": 1.3125,
+      "learning_rate": 2.4153005464480875e-05,
+      "loss": 0.0966,
+      "step": 946
+    },
+    {
+      "epoch": 2.58214042263122,
+      "grad_norm": 1.78125,
+      "learning_rate": 2.4125683060109292e-05,
+      "loss": 0.1257,
+      "step": 947
+    },
+    {
+      "epoch": 2.5848670756646217,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.4098360655737705e-05,
+      "loss": 0.105,
+      "step": 948
+    },
+    {
+      "epoch": 2.5875937286980233,
+      "grad_norm": 1.75,
+      "learning_rate": 2.4071038251366122e-05,
+      "loss": 0.1144,
+      "step": 949
+    },
+    {
+      "epoch": 2.5903203817314244,
+      "grad_norm": 2.5625,
+      "learning_rate": 2.4043715846994535e-05,
+      "loss": 0.1526,
+      "step": 950
+    },
+    {
+      "epoch": 2.5903203817314244,
+      "eval_loss": 0.2760230600833893,
+      "eval_runtime": 10.5585,
+      "eval_samples_per_second": 277.882,
+      "eval_steps_per_second": 7.009,
+      "step": 950
+    },
+    {
+      "epoch": 2.593047034764826,
+      "grad_norm": 1.5625,
+      "learning_rate": 2.4016393442622952e-05,
+      "loss": 0.1295,
+      "step": 951
+    },
+    {
+      "epoch": 2.5957736877982276,
+      "grad_norm": 1.96875,
+      "learning_rate": 2.398907103825137e-05,
+      "loss": 0.1368,
+      "step": 952
+    },
+    {
+      "epoch": 2.598500340831629,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.3961748633879782e-05,
+      "loss": 0.1278,
+      "step": 953
+    },
+    {
+      "epoch": 2.6012269938650308,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2.39344262295082e-05,
+      "loss": 0.1272,
+      "step": 954
+    },
+    {
+      "epoch": 2.6039536468984323,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.3907103825136612e-05,
+      "loss": 0.1381,
+      "step": 955
+    },
+    {
+      "epoch": 2.606680299931834,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.387978142076503e-05,
+      "loss": 0.117,
+      "step": 956
+    },
+    {
+      "epoch": 2.609406952965235,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.3852459016393445e-05,
+      "loss": 0.1476,
+      "step": 957
+    },
+    {
+      "epoch": 2.6121336059986366,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2.382513661202186e-05,
+      "loss": 0.1061,
+      "step": 958
+    },
+    {
+      "epoch": 2.6148602590320382,
+      "grad_norm": 1.5625,
+      "learning_rate": 2.3797814207650272e-05,
+      "loss": 0.1242,
+      "step": 959
+    },
+    {
+      "epoch": 2.61758691206544,
+      "grad_norm": 1.65625,
+      "learning_rate": 2.377049180327869e-05,
+      "loss": 0.1243,
+      "step": 960
+    },
+    {
+      "epoch": 2.620313565098841,
+      "grad_norm": 1.5625,
+      "learning_rate": 2.3743169398907105e-05,
+      "loss": 0.1282,
+      "step": 961
+    },
+    {
+      "epoch": 2.6230402181322425,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.3715846994535522e-05,
+      "loss": 0.1264,
+      "step": 962
+    },
+    {
+      "epoch": 2.625766871165644,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.3688524590163936e-05,
+      "loss": 0.1267,
+      "step": 963
+    },
+    {
+      "epoch": 2.6284935241990457,
+      "grad_norm": 1.6875,
+      "learning_rate": 2.366120218579235e-05,
+      "loss": 0.1357,
+      "step": 964
+    },
+    {
+      "epoch": 2.6312201772324473,
+      "grad_norm": 1.609375,
+      "learning_rate": 2.363387978142077e-05,
+      "loss": 0.1113,
+      "step": 965
+    },
+    {
+      "epoch": 2.633946830265849,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.3606557377049182e-05,
+      "loss": 0.1188,
+      "step": 966
+    },
+    {
+      "epoch": 2.6366734832992504,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.3579234972677596e-05,
+      "loss": 0.1318,
+      "step": 967
+    },
+    {
+      "epoch": 2.6394001363326516,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.3551912568306012e-05,
+      "loss": 0.1416,
+      "step": 968
+    },
+    {
+      "epoch": 2.642126789366053,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.3524590163934426e-05,
+      "loss": 0.1109,
+      "step": 969
+    },
+    {
+      "epoch": 2.6448534423994547,
+      "grad_norm": 1.90625,
+      "learning_rate": 2.3497267759562842e-05,
+      "loss": 0.134,
+      "step": 970
+    },
+    {
+      "epoch": 2.6475800954328563,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.346994535519126e-05,
+      "loss": 0.1093,
+      "step": 971
+    },
+    {
+      "epoch": 2.6503067484662575,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.3442622950819672e-05,
+      "loss": 0.1249,
+      "step": 972
+    },
+    {
+      "epoch": 2.653033401499659,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2.341530054644809e-05,
+      "loss": 0.1377,
+      "step": 973
+    },
+    {
+      "epoch": 2.6557600545330606,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.3387978142076502e-05,
+      "loss": 0.1255,
+      "step": 974
+    },
+    {
+      "epoch": 2.658486707566462,
+      "grad_norm": 1.734375,
+      "learning_rate": 2.336065573770492e-05,
+      "loss": 0.1029,
+      "step": 975
+    },
+    {
+      "epoch": 2.661213360599864,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.3333333333333336e-05,
+      "loss": 0.1248,
+      "step": 976
+    },
+    {
+      "epoch": 2.6639400136332654,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.330601092896175e-05,
+      "loss": 0.1308,
+      "step": 977
+    },
+    {
+      "epoch": 2.6666666666666665,
+      "grad_norm": 1.75,
+      "learning_rate": 2.3278688524590166e-05,
+      "loss": 0.1206,
+      "step": 978
+    },
+    {
+      "epoch": 2.669393319700068,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.325136612021858e-05,
+      "loss": 0.1124,
+      "step": 979
+    },
+    {
+      "epoch": 2.6721199727334697,
+      "grad_norm": 1.765625,
+      "learning_rate": 2.3224043715846996e-05,
+      "loss": 0.1345,
+      "step": 980
+    },
+    {
+      "epoch": 2.6748466257668713,
+      "grad_norm": 1.765625,
+      "learning_rate": 2.3196721311475413e-05,
+      "loss": 0.1408,
+      "step": 981
+    },
+    {
+      "epoch": 2.6775732788002724,
+      "grad_norm": 1.78125,
+      "learning_rate": 2.3169398907103826e-05,
+      "loss": 0.1263,
+      "step": 982
+    },
+    {
+      "epoch": 2.680299931833674,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2.314207650273224e-05,
+      "loss": 0.1449,
+      "step": 983
+    },
+    {
+      "epoch": 2.6830265848670756,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.311475409836066e-05,
+      "loss": 0.1279,
+      "step": 984
+    },
+    {
+      "epoch": 2.685753237900477,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.3087431693989073e-05,
+      "loss": 0.1307,
+      "step": 985
+    },
+    {
+      "epoch": 2.6884798909338787,
+      "grad_norm": 1.515625,
+      "learning_rate": 2.3060109289617486e-05,
+      "loss": 0.1096,
+      "step": 986
+    },
+    {
+      "epoch": 2.6912065439672803,
+      "grad_norm": 2.640625,
+      "learning_rate": 2.3032786885245903e-05,
+      "loss": 0.129,
+      "step": 987
+    },
+    {
+      "epoch": 2.693933197000682,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.3005464480874316e-05,
+      "loss": 0.1329,
+      "step": 988
+    },
+    {
+      "epoch": 2.696659850034083,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2.2978142076502736e-05,
+      "loss": 0.1546,
+      "step": 989
+    },
+    {
+      "epoch": 2.6993865030674846,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.295081967213115e-05,
+      "loss": 0.1368,
+      "step": 990
+    },
+    {
+      "epoch": 2.702113156100886,
+      "grad_norm": 1.640625,
+      "learning_rate": 2.2923497267759563e-05,
+      "loss": 0.1262,
+      "step": 991
+    },
+    {
+      "epoch": 2.704839809134288,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.289617486338798e-05,
+      "loss": 0.125,
+      "step": 992
+    },
+    {
+      "epoch": 2.707566462167689,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.2868852459016393e-05,
+      "loss": 0.1651,
+      "step": 993
+    },
+    {
+      "epoch": 2.7102931152010905,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2.284153005464481e-05,
+      "loss": 0.121,
+      "step": 994
+    },
+    {
+      "epoch": 2.713019768234492,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2.2814207650273226e-05,
+      "loss": 0.1409,
+      "step": 995
+    },
+    {
+      "epoch": 2.7157464212678937,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.278688524590164e-05,
+      "loss": 0.1118,
+      "step": 996
+    },
+    {
+      "epoch": 2.7184730743012953,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.2759562841530056e-05,
+      "loss": 0.1111,
+      "step": 997
+    },
+    {
+      "epoch": 2.721199727334697,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.2732240437158473e-05,
+      "loss": 0.1084,
+      "step": 998
+    },
+    {
+      "epoch": 2.7239263803680984,
+      "grad_norm": 1.890625,
+      "learning_rate": 2.2704918032786886e-05,
+      "loss": 0.1338,
+      "step": 999
+    },
+    {
+      "epoch": 2.7266530334014996,
+      "grad_norm": 1.78125,
+      "learning_rate": 2.2677595628415303e-05,
+      "loss": 0.1445,
+      "step": 1000
+    },
+    {
+      "epoch": 2.7266530334014996,
+      "eval_loss": 0.27945783734321594,
+      "eval_runtime": 10.5629,
+      "eval_samples_per_second": 277.764,
+      "eval_steps_per_second": 7.006,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1830,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 9.160867720721203e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}