diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,6218 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 500,
+  "global_step": 884,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.014728639868492,
+      "learning_rate": 0.0001,
+      "loss": 3.3438,
+      "step": 1
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.07497563513899,
+      "learning_rate": 0.0001,
+      "loss": 3.4844,
+      "step": 2
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.0683045636143302,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 3
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.341314586764068,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 4
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.1805565704564955,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 5
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.244329296572686,
+      "learning_rate": 0.0001,
+      "loss": 3.6562,
+      "step": 6
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.595711743609493,
+      "learning_rate": 0.0001,
+      "loss": 3.7344,
+      "step": 7
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.4778447381697863,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 8
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.9898264241527004,
+      "learning_rate": 0.0001,
+      "loss": 3.4219,
+      "step": 9
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.542104616744462,
+      "learning_rate": 0.0001,
+      "loss": 3.2656,
+      "step": 10
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.4277356413883804,
+      "learning_rate": 0.0001,
+      "loss": 3.2188,
+      "step": 11
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.2437025610823427,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 12
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.305702485290415,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 13
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.7399497184119506,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 14
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.203877548043482,
+      "learning_rate": 0.0001,
+      "loss": 3.3281,
+      "step": 15
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.7598651340996145,
+      "learning_rate": 0.0001,
+      "loss": 3.4531,
+      "step": 16
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.8794034742734076,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 17
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.716863791265328,
+      "learning_rate": 0.0001,
+      "loss": 3.2812,
+      "step": 18
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.6209919053501953,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 19
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.648879316727216,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 20
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.110008991960499,
+      "learning_rate": 0.0001,
+      "loss": 3.2188,
+      "step": 21
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.9713952473944705,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 22
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.7018343790938353,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 23
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.4131899574811118,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 24
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.298790786105439,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 25
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.5966196850680148,
+      "learning_rate": 0.0001,
+      "loss": 3.2031,
+      "step": 26
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.875902211805855,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 27
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.765858168009748,
+      "learning_rate": 0.0001,
+      "loss": 3.2031,
+      "step": 28
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.070259667441601,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 29
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.3005916640548,
+      "learning_rate": 0.0001,
+      "loss": 3.25,
+      "step": 30
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.7606571016769332,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 31
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.1981380713510323,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 32
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.5672449144444,
+      "learning_rate": 0.0001,
+      "loss": 3.3125,
+      "step": 33
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.646603965279102,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 34
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.2248743875212385,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 35
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.8724583239645565,
+      "learning_rate": 0.0001,
+      "loss": 3.3438,
+      "step": 36
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.5297221873351754,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 37
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.5275460394914675,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 38
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.358392438696745,
+      "learning_rate": 0.0001,
+      "loss": 3.2344,
+      "step": 39
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.55912275345912,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 40
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.363728235132881,
+      "learning_rate": 0.0001,
+      "loss": 3.2344,
+      "step": 41
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.175630702744953,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 42
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.805357918124218,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 43
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.993779953713576,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 44
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.039174383300467,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 45
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.4583649396718203,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 46
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.400384244467908,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 47
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.9016353799852213,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 48
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.12493864757068,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 49
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.8938775159043613,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 50
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.890651173471067,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 51
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.614938570857581,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 52
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.969969594116141,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 53
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.7352993614346293,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 54
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.1359556182594974,
+      "learning_rate": 0.0001,
+      "loss": 3.2188,
+      "step": 55
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.843122039034664,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 56
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.3288764649012377,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 57
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.8165528057214244,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 58
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.5191401715518427,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 59
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.1379077630223975,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.124369818711383,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 61
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.132972808670283,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 62
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.9614040202740735,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 63
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.8972234185293777,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 64
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.6678504925494466,
+      "learning_rate": 0.0001,
+      "loss": 3.0781,
+      "step": 65
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.911760340298777,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 66
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.522073404974725,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 67
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.873975472150959,
+      "learning_rate": 0.0001,
+      "loss": 3.0781,
+      "step": 68
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.248241805309187,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 69
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.734654438703655,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 70
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.914708142756069,
+      "learning_rate": 0.0001,
+      "loss": 3.4688,
+      "step": 71
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.0704529793274205,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 72
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.7714788446837417,
+      "learning_rate": 0.0001,
+      "loss": 3.2344,
+      "step": 73
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.1777578325193923,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 74
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.7344906874924058,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 75
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.8946934280105747,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 76
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.1542104061738323,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 77
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.762285352941279,
+      "learning_rate": 0.0001,
+      "loss": 3.3125,
+      "step": 78
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.5687661240845703,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 79
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.354199830221798,
+      "learning_rate": 0.0001,
+      "loss": 3.2344,
+      "step": 80
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.982271568285038,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 81
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.6629483529030193,
+      "learning_rate": 0.0001,
+      "loss": 3.2656,
+      "step": 82
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.097410495508935,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 83
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.4293559919795342,
+      "learning_rate": 0.0001,
+      "loss": 3.1875,
+      "step": 84
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.016797304151073,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 85
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.6266954910547033,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 86
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.691102686844325,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 87
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.0573753479866896,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 88
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.248776325364024,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 89
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.684435268310462,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 90
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.9459506953365486,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 91
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.6777239852969212,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 92
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.687978088563062,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 93
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.625427110996173,
+      "learning_rate": 0.0001,
+      "loss": 3.1875,
+      "step": 94
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.364572491761355,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 95
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.003287810433365,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 96
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.7602829324063642,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 97
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.8680654062417377,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 98
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.3377680005099393,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 99
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.080756825063626,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 100
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.5824032205107357,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 101
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.572874096937035,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 102
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.173554273811993,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 103
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.374295066463168,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 104
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.546675683646743,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 105
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.8610685415267687,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 106
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.739045406962109,
+      "learning_rate": 0.0001,
+      "loss": 3.25,
+      "step": 107
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.963531109735375,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 108
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.269922121525073,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 109
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.1227369643458402,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 110
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.0619044281415593,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 111
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 3.502108277442133,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 112
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 3.4217887133458893,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 113
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 4.071458514694356,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 114
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 3.8820655057810574,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 115
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.425526328303125,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 116
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.5725735012774393,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 117
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.570192135060652,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 118
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 4.518786389858919,
+      "learning_rate": 0.0001,
+      "loss": 3.3906,
+      "step": 119
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.781462280343676,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 120
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 4.0169315946259525,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 121
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 3.3904299413281547,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 122
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.767985407126416,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 123
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.350209228567863,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 124
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 3.7716820325330063,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 125
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.6172676918246065,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 126
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.8615732602702617,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 127
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 3.1977593475641566,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 128
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 3.7044980154353344,
+      "learning_rate": 0.0001,
+      "loss": 3.0781,
+      "step": 129
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.4708896342997333,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 130
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.973855966147942,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 131
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.7119122676528393,
+      "learning_rate": 0.0001,
+      "loss": 2.3594,
+      "step": 132
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.6702654024783024,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 133
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 3.7431706355808334,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 134
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 5.077035030726395,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 135
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 3.690111157612484,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 136
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.9422207378285203,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 137
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 4.791349492274715,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 138
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 2.5077421362522108,
+      "learning_rate": 0.0001,
+      "loss": 3.0781,
+      "step": 139
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 3.1415791107859667,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 140
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 3.655269108842213,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 141
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 3.1238567068777745,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 142
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.6177206881858,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 143
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.843884596485035,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 144
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 4.63778601469586,
+      "learning_rate": 0.0001,
+      "loss": 3.4062,
+      "step": 145
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 2.2470659627608582,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 146
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.597522080096179,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 147
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.3910575990800633,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 148
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.0899424465475698,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 149
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 2.8859452606243186,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 150
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 4.272675820285552,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 151
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 3.258577402368087,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 152
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 4.136534066435656,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 153
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 3.503047467842312,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 154
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 3.2423987815492694,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 155
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.4377222090009774,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 156
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.1680307697838144,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 157
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.7903929164222423,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 158
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 3.9152200948164633,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 159
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 3.148242879456836,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 160
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 3.2016489522036387,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 161
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 2.6290757877359283,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 162
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 3.3331963669130316,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 163
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 3.7860541087442985,
+      "learning_rate": 0.0001,
+      "loss": 3.3906,
+      "step": 164
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 2.6724816282540425,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 165
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 2.7967393894511416,
+      "learning_rate": 0.0001,
+      "loss": 3.375,
+      "step": 166
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.3937925854061146,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 167
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 4.384654503841222,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 168
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.9320332448378066,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 169
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 3.8025261019973815,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 170
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.7469268151387083,
+      "learning_rate": 0.0001,
+      "loss": 3.3438,
+      "step": 171
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 2.728521201067668,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 172
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 2.945552120230828,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 173
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 3.5474098558831555,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 174
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 2.682553203483424,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 175
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 3.454442321478634,
+      "learning_rate": 0.0001,
+      "loss": 3.2031,
+      "step": 176
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 3.4711419275650197,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 177
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.96972797034164,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 178
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.241804222438612,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 179
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.591008613141231,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 180
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 3.1836736658810905,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 181
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 2.6696567939120377,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 182
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 4.153665896714593,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 183
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 3.632729399655711,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 184
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 2.3690150169249673,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 185
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 3.606188859835758,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 186
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.8610059159346135,
+      "learning_rate": 0.0001,
+      "loss": 3.0781,
+      "step": 187
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.162526657575551,
+      "learning_rate": 0.0001,
+      "loss": 3.25,
+      "step": 188
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 3.3799373047880463,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 189
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 3.636158335840892,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 190
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.628330891784058,
+      "learning_rate": 0.0001,
+      "loss": 2.1719,
+      "step": 191
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 3.1369675986549543,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 192
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 2.9451234006615112,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 193
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 3.1049660335818894,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 194
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.746393657093893,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 195
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.313442015397048,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 196
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.5237073828607777,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 197
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 3.993671122520398,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 198
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 3.7104137607661096,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 199
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.63732458277305,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 200
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.637164098858904,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 201
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.7067823222285443,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 202
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 3.20499440039486,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 203
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 3.390167457388061,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 204
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.9755718114652785,
+      "learning_rate": 0.0001,
+      "loss": 3.25,
+      "step": 205
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.9858024159992387,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 206
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 3.49719960170014,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 207
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.760481519803184,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 208
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 3.5361985653549035,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 209
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.7189516787553702,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 210
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.7444161167117778,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 211
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 4.178779298351167,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 212
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 5.1382962163861245,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 213
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 4.777877381510249,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 214
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 4.483777499230272,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 215
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 2.6505694947027414,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 216
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 2.8266381222956296,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 217
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.7454666967878922,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 218
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 3.2359889427504243,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 219
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 3.2617408981943914,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 220
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 3.761832594055263,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 221
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 3.429839617150036,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 222
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 3.472012057501413,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 223
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 4.347859361063026,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 224
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.7603120152326337,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 225
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.141198806856796,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 226
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 2.27575220665026,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 227
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 3.292652786163134,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 228
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 2.8521097618658215,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 229
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 3.5683909346483116,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 230
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 2.048666439224908,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 231
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 3.226473060669876,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 232
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 2.560468630545825,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 233
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 2.8610577465558875,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 234
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 3.276597524431846,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 235
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 3.11183393883472,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 236
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 2.731944444220084,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 237
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 3.439580204033313,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 238
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 3.850548595357673,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 239
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 3.032256388909145,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 240
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 2.816746313771955,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 241
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.555308458291825,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 242
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.1232590262180104,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 243
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 2.841004178819089,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 244
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 3.169879937448116,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 245
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 3.756621242224477,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 246
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.291591115732283,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 247
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.9491635337617876,
+      "learning_rate": 0.0001,
+      "loss": 3.2031,
+      "step": 248
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 4.5454750211170705,
+      "learning_rate": 0.0001,
+      "loss": 2.4219,
+      "step": 249
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 3.1194188990231173,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 250
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.1708709674001154,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 251
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.769197611076157,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 252
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.1034571175550734,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 253
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.3624038577718443,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 254
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.695977386674938,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 255
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 3.520629660314111,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 256
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 3.6643635174993214,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 257
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 3.6400728396728788,
+      "learning_rate": 0.0001,
+      "loss": 2.3281,
+      "step": 258
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 3.1938129371502533,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 259
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 4.070065400459323,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 260
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 3.8599798135960857,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 261
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 4.0800675518707425,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 262
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 2.88070935560625,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 263
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 3.6143748192147958,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 264
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 4.033368015293004,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 265
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 3.1096713230086754,
+      "learning_rate": 0.0001,
+      "loss": 3.1562,
+      "step": 266
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 3.53880392811516,
+      "learning_rate": 0.0001,
+      "loss": 2.2188,
+      "step": 267
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 3.9253800416766644,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 268
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 3.4594381842770394,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 269
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 3.868070923795578,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 270
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.234829607238394,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 271
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.823186245574893,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 272
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 3.0994313335043002,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 273
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 3.6618700341670536,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 274
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.245984922959924,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 275
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 3.413612437224495,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 276
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 3.6451564739213547,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 277
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 4.166476678137405,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 278
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 3.724493167963291,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 279
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 3.395942991582062,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 280
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 3.5527811358540693,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 281
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 5.687097991739583,
+      "learning_rate": 0.0001,
+      "loss": 3.2188,
+      "step": 282
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 4.6245839813569045,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 283
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.001745122577989,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 284
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.2065745171783675,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 285
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 3.763842873535832,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 286
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 3.4297749721988753,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 287
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 5.112060409623577,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 288
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.765001988028911,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 289
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.025477764290351,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 290
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 3.533207269511524,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 291
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 3.894629067717557,
+      "learning_rate": 0.0001,
+      "loss": 2.3438,
+      "step": 292
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 3.8265063420175554,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 293
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 5.387518896603435,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 294
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 3.462770499090204,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 295
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 3.999081454001175,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 296
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 3.26791217269049,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 297
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 4.01576665487724,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 298
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 3.4357297956359223,
+      "learning_rate": 0.0001,
+      "loss": 3.2188,
+      "step": 299
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.991044885131129,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 300
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.2208290782529088,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 301
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 3.5358550835047375,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 302
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 3.25374956312475,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 303
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 3.260985818723821,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 304
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 3.9257270591651534,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 305
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 3.167203120292308,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 306
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 4.694459033055643,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 307
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 3.6064245290727452,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 308
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.4765920276238136,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 309
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 4.168438398393131,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 310
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 4.641284910447617,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 311
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 3.165019356899751,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 312
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 6.293667183946401,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 313
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 3.8581829580543574,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 314
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 4.0595835955067505,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 315
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 3.804012147939095,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 316
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.8757598500054873,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 317
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 3.473056721459923,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 318
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 4.030824580859273,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 319
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 3.7575238271559557,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 320
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.319465423783974,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 321
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 3.4270491578278315,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 322
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 4.597605413378048,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 323
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 5.242593092794404,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 324
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 3.5690314241373007,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 325
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.170604037406789,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 326
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 4.6838516697328,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 327
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.2080896344291854,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 328
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 5.326504169409197,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 329
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 3.2789255992893596,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 330
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 3.085516640743533,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 331
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 4.194716526521928,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 332
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 3.2525926179968034,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 333
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 3.0977142194617313,
+      "learning_rate": 0.0001,
+      "loss": 2.5469,
+      "step": 334
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.760665526282257,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 335
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.7818001961714445,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 336
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.6609563888659973,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 337
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 3.982276940149541,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 338
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 2.8042721280801484,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 339
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 3.2615173671408555,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 340
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 3.6914782762730924,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 341
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 4.899453093729484,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 342
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 4.291638036327811,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 343
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 2.6028433980069865,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 344
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 2.7549032124651736,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 345
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 3.352486062521882,
+      "learning_rate": 0.0001,
+      "loss": 3.1562,
+      "step": 346
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.8708894259256845,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 347
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 5.037035733284411,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 348
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 3.594206741562048,
+      "learning_rate": 0.0001,
+      "loss": 2.3438,
+      "step": 349
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 5.159307734311026,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 350
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 5.9233278621953644,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 351
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 4.0715030156474254,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 352
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 3.044655552777047,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 353
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 3.7734627420835953,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 354
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 5.180693081625301,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 355
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.3219223924038745,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 356
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 4.142269921917345,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 357
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 3.3735403369190893,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 358
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 3.5239894283719653,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 359
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 3.8828475802810245,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 360
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 4.693796163234429,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 361
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.4075539488803925,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 362
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.664547612895459,
+      "learning_rate": 0.0001,
+      "loss": 3.2188,
+      "step": 363
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 5.938694208857688,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 364
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 3.409013748608653,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 365
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 4.4658543768768935,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 366
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.371693408259016,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 367
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 4.559788858289379,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 368
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.628331869322542,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 369
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 4.39455926251329,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 370
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 3.853396943835766,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 371
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 3.9765219333825494,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 372
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.061152055123867,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 373
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 4.618483012917297,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 374
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 3.3260259398876393,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 375
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 3.3538962131850387,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 376
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 5.8575854324075864,
+      "learning_rate": 0.0001,
+      "loss": 2.2344,
+      "step": 377
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 4.168971058411045,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 378
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 3.6920069784108116,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 379
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.170712811496067,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 380
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.793007744358583,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 381
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 4.025355549587713,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 382
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 3.4870718987013296,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 383
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 3.3667477898409186,
+      "learning_rate": 0.0001,
+      "loss": 2.2969,
+      "step": 384
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 3.7242254076840884,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 385
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 3.1840545349531495,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 386
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 4.823197200047613,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 387
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 3.988386839545984,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 388
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 5.057343299049697,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 389
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 5.161242480912821,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 390
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 4.480668734671967,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 391
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 3.9365190921112907,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 392
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 4.662495774692652,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 393
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 3.5240918262495318,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 394
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.526636196082364,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 395
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.469697798668837,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 396
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 5.4459922507695495,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 397
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 3.236562942073936,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 398
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 3.770016835883372,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 399
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 3.9591980759103653,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 400
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 7.860080427566192,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 401
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 3.5735897118338418,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 402
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 3.253545354981559,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 403
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 3.079257326273105,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 404
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 4.47108730912994,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 405
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 4.921590616999781,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 406
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 3.270607047698539,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 407
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 4.207173315577373,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 408
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 3.7871838368285267,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 409
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.936529474935983,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 410
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 3.8957693898330037,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 411
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.960143029053001,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 412
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 4.03955454269378,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 413
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 8.311598719730528,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 414
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 4.078913705129825,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 415
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 3.2725183823603006,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 416
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 5.107210559438947,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 417
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 4.293656342570625,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 418
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 3.5007632189651408,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 419
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 2.870288914226923,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 420
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 4.013292523806265,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 421
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 2.944258225094142,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 422
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 5.0008572219675,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 423
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 3.4668064948996435,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 424
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 3.282414339798479,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 425
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 4.3660030510317736,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 426
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 4.277759844521119,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 427
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 6.495737721951466,
+      "learning_rate": 0.0001,
+      "loss": 2.5469,
+      "step": 428
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 3.6749746067665052,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 429
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 3.8285972640178647,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 430
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 4.329624774415669,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 431
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 3.5664735497666173,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 432
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 3.814296312027705,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 433
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 3.255543154666781,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 434
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 6.098448774362813,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 435
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 4.594773365704486,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 436
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 4.256945975023405,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 437
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 3.45198248082701,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 438
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.645343343544797,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 439
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 2.8494264914506013,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 440
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 3.2371106907573353,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 441
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.351841237094819,
+      "learning_rate": 0.0001,
+      "loss": 2.5469,
+      "step": 442
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.885380513377714,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 443
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 4.141226131557114,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 444
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 3.571384859469851,
+      "learning_rate": 0.0001,
+      "loss": 3.2031,
+      "step": 445
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 4.143653406347033,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 446
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 3.376258580527233,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 447
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 5.041611047731054,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 448
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.7510995579910897,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 449
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 4.794901656709599,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 450
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 3.2071039166350794,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 451
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.6160535680470733,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 452
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.4951696700796053,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 453
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 4.052957529710831,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 454
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 4.5548596168461035,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 455
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 4.692304645656616,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 456
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 4.251726697426816,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 457
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 5.055218343110248,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 458
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 4.0514218848787475,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 459
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 5.471343199254358,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 460
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 4.364540022104307,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 461
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 4.347289897028965,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 462
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 3.320864844261629,
+      "learning_rate": 0.0001,
+      "loss": 2.1719,
+      "step": 463
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 4.277320044723722,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 464
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 5.2453592445245745,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 465
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 5.520158779920925,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 466
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 5.09919550518767,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 467
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 5.7882890121116075,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 468
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 4.166784108590436,
+      "learning_rate": 0.0001,
+      "loss": 2.3906,
+      "step": 469
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 5.532799561595931,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 470
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 4.199106278412939,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 471
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 5.148294228201523,
+      "learning_rate": 0.0001,
+      "loss": 2.4844,
+      "step": 472
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 6.06473338860914,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 473
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 4.4020944879221435,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 474
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 6.37168824745839,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 475
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 6.2089065293516335,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 476
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 5.9677659863997174,
+      "learning_rate": 0.0001,
+      "loss": 2.3281,
+      "step": 477
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 4.535209473651245,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 478
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 4.758021060904884,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 479
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 4.6744206823178605,
+      "learning_rate": 0.0001,
+      "loss": 2.2969,
+      "step": 480
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 4.748502122462707,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 481
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 3.8572636341399447,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 482
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 4.399471923346186,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 483
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 3.646671632648375,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 484
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 4.00567569136169,
+      "learning_rate": 0.0001,
+      "loss": 3.1562,
+      "step": 485
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 3.65464670971521,
+      "learning_rate": 0.0001,
+      "loss": 2.3906,
+      "step": 486
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 3.91732048961442,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 487
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 5.317268932471553,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 488
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 5.341302649630214,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 489
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 4.383122101856098,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 490
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 3.8860693666995063,
+      "learning_rate": 0.0001,
+      "loss": 2.4844,
+      "step": 491
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 4.364787342162316,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 492
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 4.470798240244141,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 493
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 6.104282384032258,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 494
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 5.006875450612742,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 495
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 5.331678950721802,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 496
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 4.04281058614812,
+      "learning_rate": 0.0001,
+      "loss": 2.3594,
+      "step": 497
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 5.345612164193795,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 498
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 5.341596960250008,
+      "learning_rate": 0.0001,
+      "loss": 2.3438,
+      "step": 499
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 5.739876231264038,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 500
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 5.142026268720496,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 501
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 4.15681316206977,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 502
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 4.817970447202398,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 503
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 8.308610641023103,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 504
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 5.00252311689909,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 505
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 4.380395189338173,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 506
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 4.670956184899664,
+      "learning_rate": 0.0001,
+      "loss": 3.1719,
+      "step": 507
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 4.621499756596695,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 508
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 6.29671870323029,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 509
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 4.932336285101389,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 510
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 4.3245121078513655,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 511
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 7.030111858054767,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 512
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 4.89566917127695,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 513
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 4.195486069953785,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 514
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 5.136743915157165,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 515
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 3.948875651747143,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 516
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 3.231795023359959,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 517
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 4.9551613025707475,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 518
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 5.865415887327459,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 519
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 4.7737465818996885,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 520
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 6.733184400927877,
+      "learning_rate": 0.0001,
+      "loss": 2.125,
+      "step": 521
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 5.290951724234347,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 522
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 4.585822054902166,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 523
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 5.845655683014917,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 524
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 4.058548342806645,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 525
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 4.730863583042205,
+      "learning_rate": 0.0001,
+      "loss": 2.3125,
+      "step": 526
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 3.9460443083159125,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 527
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 5.985763082632476,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 528
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 4.034177525824554,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 529
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 6.0743244504041,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 530
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 6.144275879007958,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 531
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 5.208959046160335,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 532
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 4.505033267544332,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 533
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 7.682902684774122,
+      "learning_rate": 0.0001,
+      "loss": 2.3594,
+      "step": 534
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 5.4360470024869905,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 535
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 5.211817151957024,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 536
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 4.959284674849101,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 537
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 5.4968430538522215,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 538
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 4.1351746678748755,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 539
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 5.320761577416053,
+      "learning_rate": 0.0001,
+      "loss": 3.1406,
+      "step": 540
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 6.655636255576953,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 541
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 4.458839149948617,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 542
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 4.972007553913938,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 543
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 3.5505879907694755,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 544
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 5.0186604630551255,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 545
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 4.717499667281885,
+      "learning_rate": 0.0001,
+      "loss": 3.0938,
+      "step": 546
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 4.38634574908589,
+      "learning_rate": 0.0001,
+      "loss": 2.3125,
+      "step": 547
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 5.633061705506686,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 548
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 5.028007311949011,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 549
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 6.111968054796198,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 550
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 4.563137449239842,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 551
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 4.540680149103331,
+      "learning_rate": 0.0001,
+      "loss": 2.9844,
+      "step": 552
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 7.279079756719769,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 553
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 3.9211600361134695,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 554
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 4.376188709334436,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 555
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.7536996299053405,
+      "learning_rate": 0.0001,
+      "loss": 1.9062,
+      "step": 556
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.730154548981628,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 557
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.483638601741794,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 558
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 5.017903669312139,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 559
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 4.497681177312601,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 560
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 3.6938673030692635,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 561
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 4.627269663966374,
+      "learning_rate": 0.0001,
+      "loss": 3.0156,
+      "step": 562
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 6.028249115854368,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 563
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 5.895270257080288,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 564
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 5.782086264164467,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 565
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 4.832572929404261,
+      "learning_rate": 0.0001,
+      "loss": 2.25,
+      "step": 566
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 4.777919636172623,
+      "learning_rate": 0.0001,
+      "loss": 2.5469,
+      "step": 567
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 5.844834596317805,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 568
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 4.3943966123305165,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 569
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 4.3493264028288285,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 570
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 4.579129290042329,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 571
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 5.106305124633349,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 572
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 3.8121132157014146,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 573
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 7.674035029680783,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 574
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 4.0155468016806255,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 575
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 4.535174372431354,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 576
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 7.713765819686499,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 577
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 6.992348998956431,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 578
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 5.366316115730502,
+      "learning_rate": 0.0001,
+      "loss": 3.2969,
+      "step": 579
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 4.628717491988879,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 580
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 3.736490314420957,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 581
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 3.270159000979569,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 582
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 4.209971095000819,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 583
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 5.904260931835911,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 584
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 4.56368296212557,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 585
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 4.356143809551981,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 586
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 5.192460484725346,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 587
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 4.970184066855408,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 588
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 3.668363909507462,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 589
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 4.975508877974084,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 590
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 4.426312606397656,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 591
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 4.871684486582093,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 592
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 3.8454122192166054,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 593
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 6.046171861939742,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 594
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 4.236762915206439,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 595
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 4.754038548279519,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 596
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 5.3018074744010235,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 597
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 5.138289855563257,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 598
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 5.478633422952733,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 599
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 6.855786567331759,
+      "learning_rate": 0.0001,
+      "loss": 2.4844,
+      "step": 600
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 4.265407093713752,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 601
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 5.036701430071399,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 602
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 4.849399765693369,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 603
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 6.244463203398971,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 604
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 4.911627851437594,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 605
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 5.648660311489794,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 606
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 5.510436851938899,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 607
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 5.73305511033946,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 608
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 5.035663640250028,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 609
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 5.706713259761806,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 610
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 3.9388716131805177,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 611
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 5.208282554862349,
+      "learning_rate": 0.0001,
+      "loss": 2.2656,
+      "step": 612
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 4.030323203128877,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 613
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 4.982573254410871,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 614
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 4.282555681716531,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 615
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 5.254295196214892,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 616
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 6.428623089456714,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 617
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 4.767141810934083,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 618
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 5.419475658643534,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 619
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 5.472135282653432,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 620
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 6.675110370195235,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 621
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 4.366266038957683,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 622
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 3.896695556817304,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 623
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 6.3050055621586605,
+      "learning_rate": 0.0001,
+      "loss": 2.4844,
+      "step": 624
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 4.491540640178512,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 625
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 6.046498345815951,
+      "learning_rate": 0.0001,
+      "loss": 2.4844,
+      "step": 626
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 6.403430130057227,
+      "learning_rate": 0.0001,
+      "loss": 3.0781,
+      "step": 627
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 5.363921266764386,
+      "learning_rate": 0.0001,
+      "loss": 2.1875,
+      "step": 628
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 3.827451350516171,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 629
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 5.50903534823105,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 630
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 4.175065236833921,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 631
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 5.037111092373576,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 632
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 4.6903714504924725,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 633
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 5.433323796199888,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 634
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 6.2352792612834,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 635
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 5.2711186382571515,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 636
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 7.882069923983431,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 637
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 4.487399648438269,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 638
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 4.582740219582226,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 639
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 5.253149476772823,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 640
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 4.717261080461104,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 641
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 4.170314259960067,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 642
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 4.9179394418307,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 643
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 5.211586048763502,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 644
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 4.138246002031018,
+      "learning_rate": 0.0001,
+      "loss": 2.2656,
+      "step": 645
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 4.040019263183381,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 646
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 5.016054162159075,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 647
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 4.160869140788706,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 648
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 4.319009327023701,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 649
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 4.788179201751919,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 650
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 4.081130439995906,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 651
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 5.50642508525935,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 652
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 5.585953512031478,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 653
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 5.31807917503442,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 654
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 3.59491039054475,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 655
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 4.4689533270014135,
+      "learning_rate": 0.0001,
+      "loss": 3.2656,
+      "step": 656
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 4.968753065799484,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 657
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 5.528728807749279,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 658
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 5.035149173943463,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 659
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 5.1939238070569935,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 660
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 7.036562828613889,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 661
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 6.297262433586043,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 662
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 5.58699604459073,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 663
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 4.092479681857709,
+      "learning_rate": 0.0001,
+      "loss": 3.1094,
+      "step": 664
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 5.358390139052102,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 665
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 5.280630772569093,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 666
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 4.491504971073838,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 667
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 4.969900611469726,
+      "learning_rate": 0.0001,
+      "loss": 2.0312,
+      "step": 668
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 5.26280232160558,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 669
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 4.34447668238551,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 670
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 4.233433870902845,
+      "learning_rate": 0.0001,
+      "loss": 2.1719,
+      "step": 671
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 4.471039084021881,
+      "learning_rate": 0.0001,
+      "loss": 2.2969,
+      "step": 672
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 5.759691256699786,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 673
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 5.222531825557149,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 674
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 4.719114936711103,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 675
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 5.322770709604783,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 676
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 4.411446154538555,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 677
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 4.64030284036272,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 678
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 4.138101002033864,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 679
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 5.086493549779005,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 680
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 3.930014385181362,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 681
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 4.729385169407767,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 682
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 3.9189890006972665,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 683
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 5.903640555457615,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 684
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 6.904659388771962,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 685
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 8.989355711263677,
+      "learning_rate": 0.0001,
+      "loss": 2.0,
+      "step": 686
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 4.7695501889288145,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 687
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 5.69647405964174,
+      "learning_rate": 0.0001,
+      "loss": 2.5469,
+      "step": 688
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 6.100773782892059,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 689
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 5.686663254617055,
+      "learning_rate": 0.0001,
+      "loss": 2.0469,
+      "step": 690
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 6.571330530953064,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 691
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 5.718821223772928,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 692
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 4.373723630863873,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 693
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 5.085229478433662,
+      "learning_rate": 0.0001,
+      "loss": 2.8594,
+      "step": 694
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 7.021754317383601,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 695
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 6.04687595749246,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 696
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 6.788165163318837,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 697
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 5.561311682085884,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 698
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 4.608936689327367,
+      "learning_rate": 0.0001,
+      "loss": 2.4219,
+      "step": 699
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 5.968780759931178,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 700
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 5.103044298583068,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 701
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 6.45236411621874,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 702
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 4.8281182531637095,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 703
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 4.389882657051867,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 704
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 4.962384851428117,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 705
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 4.982045946774825,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 706
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 6.41107777745504,
+      "learning_rate": 0.0001,
+      "loss": 2.1406,
+      "step": 707
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 5.090681442306407,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 708
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 5.165328647890187,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 709
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 4.705626239369721,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 710
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 7.677966149358577,
+      "learning_rate": 0.0001,
+      "loss": 2.3281,
+      "step": 711
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 5.659659609102272,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 712
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 4.766715309460445,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 713
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 5.570413898672718,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 714
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 6.482364017731351,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 715
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 7.676974890118262,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 716
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 7.269455563092696,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 717
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 7.4750489121354695,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 718
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 5.096852230759679,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 719
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 9.551421687804485,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 720
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 8.308410042464198,
+      "learning_rate": 0.0001,
+      "loss": 2.3438,
+      "step": 721
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 7.249812504415762,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 722
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 5.625497642517863,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 723
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 5.528162218971885,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 724
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 5.97119760259338,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 725
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 5.0051682876370736,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 726
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 6.153680263773301,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 727
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 12.408064278903167,
+      "learning_rate": 0.0001,
+      "loss": 2.2188,
+      "step": 728
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 6.967676512014587,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 729
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 5.596141752113706,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 730
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 6.15457256616309,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 731
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 5.7616559342316584,
+      "learning_rate": 0.0001,
+      "loss": 2.4844,
+      "step": 732
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 5.9891271652349936,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 733
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 6.652398069643558,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 734
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 6.356574056404722,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 735
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 4.593278190996901,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 736
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 6.0605354589139635,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 737
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 5.70706586433816,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 738
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 4.893608346712496,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 739
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 6.088522365219184,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 740
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 5.431869278014897,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 741
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 5.596003857292665,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 742
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 5.97698909669504,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 743
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 5.805355147115389,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 744
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 5.6494434657268515,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 745
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 6.6389660268244945,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 746
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 5.25191874182185,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 747
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 4.828238204533096,
+      "learning_rate": 0.0001,
+      "loss": 2.3281,
+      "step": 748
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 5.71616111332261,
+      "learning_rate": 0.0001,
+      "loss": 2.2656,
+      "step": 749
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 5.33198279309472,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 750
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 6.676847054708395,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 751
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 6.987831720140158,
+      "learning_rate": 0.0001,
+      "loss": 2.4375,
+      "step": 752
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 5.246383776912542,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 753
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 6.2816140273248235,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 754
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 7.924891850428938,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 755
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 6.172178443029792,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 756
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 4.599788973324431,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 757
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 4.9529853584549075,
+      "learning_rate": 0.0001,
+      "loss": 2.2656,
+      "step": 758
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 6.319742548596082,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 759
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 7.205622577272047,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 760
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 9.18574662601195,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 761
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 6.956078746670551,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 762
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 5.0104025328964115,
+      "learning_rate": 0.0001,
+      "loss": 2.2656,
+      "step": 763
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 5.418890629156982,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 764
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 5.404410173821834,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 765
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 6.568137897267971,
+      "learning_rate": 0.0001,
+      "loss": 2.1719,
+      "step": 766
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 5.578470702623748,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 767
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 6.235890736509007,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 768
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 5.114739271673889,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 769
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 5.631412619020986,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 770
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 6.0619101459204305,
+      "learning_rate": 0.0001,
+      "loss": 2.2344,
+      "step": 771
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 6.715082841558436,
+      "learning_rate": 0.0001,
+      "loss": 2.3125,
+      "step": 772
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 7.162414120906969,
+      "learning_rate": 0.0001,
+      "loss": 2.1406,
+      "step": 773
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 6.361814355074576,
+      "learning_rate": 0.0001,
+      "loss": 2.375,
+      "step": 774
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 5.233411385933616,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 775
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 5.773944915421188,
+      "learning_rate": 0.0001,
+      "loss": 2.6094,
+      "step": 776
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 6.980243552018795,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 777
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 5.40460163775666,
+      "learning_rate": 0.0001,
+      "loss": 2.9688,
+      "step": 778
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 8.48311753833824,
+      "learning_rate": 0.0001,
+      "loss": 2.2031,
+      "step": 779
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 5.876697470240219,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 780
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 6.706613026904167,
+      "learning_rate": 0.0001,
+      "loss": 3.1875,
+      "step": 781
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 7.2764400516796846,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 782
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 6.094306397049338,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 783
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 6.278005609224948,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 784
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 5.3086920461585185,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 785
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 5.684767409346208,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 786
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 5.9289557000568704,
+      "learning_rate": 0.0001,
+      "loss": 2.6875,
+      "step": 787
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 5.5746989859790945,
+      "learning_rate": 0.0001,
+      "loss": 2.1719,
+      "step": 788
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 9.301668372300803,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 789
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 6.513123479044599,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 790
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 4.777033577512223,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 791
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 4.457114397241286,
+      "learning_rate": 0.0001,
+      "loss": 3.0,
+      "step": 792
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 6.689527406548995,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 793
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 6.983770238015634,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 794
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 6.802187681181655,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 795
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 7.033033169580362,
+      "learning_rate": 0.0001,
+      "loss": 2.3594,
+      "step": 796
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 6.26808782701995,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 797
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 6.005513689302465,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 798
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 7.767577684581086,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 799
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 8.120626893479068,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 800
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 7.580438309650423,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 801
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 6.601396559848888,
+      "learning_rate": 0.0001,
+      "loss": 3.2031,
+      "step": 802
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 5.589941827621371,
+      "learning_rate": 0.0001,
+      "loss": 2.7656,
+      "step": 803
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 7.904399035942282,
+      "learning_rate": 0.0001,
+      "loss": 2.4531,
+      "step": 804
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 4.281379429844708,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 805
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 6.356739367521081,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 806
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 6.96355742617709,
+      "learning_rate": 0.0001,
+      "loss": 2.5469,
+      "step": 807
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 7.726157531308813,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 808
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 4.480569716014928,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 809
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 5.8390738745875055,
+      "learning_rate": 0.0001,
+      "loss": 3.0312,
+      "step": 810
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 6.615258635722885,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 811
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 7.391480221820631,
+      "learning_rate": 0.0001,
+      "loss": 2.2812,
+      "step": 812
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 5.191253582921654,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 813
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 5.4670864088124045,
+      "learning_rate": 0.0001,
+      "loss": 3.125,
+      "step": 814
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 5.624287979287397,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 815
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 6.313425088933223,
+      "learning_rate": 0.0001,
+      "loss": 2.5938,
+      "step": 816
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 6.818221819002743,
+      "learning_rate": 0.0001,
+      "loss": 1.7891,
+      "step": 817
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 6.464530097139393,
+      "learning_rate": 0.0001,
+      "loss": 2.75,
+      "step": 818
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 5.69636118536784,
+      "learning_rate": 0.0001,
+      "loss": 2.7969,
+      "step": 819
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 6.3130472400300635,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 820
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 6.164694991173136,
+      "learning_rate": 0.0001,
+      "loss": 2.25,
+      "step": 821
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 7.070295165751384,
+      "learning_rate": 0.0001,
+      "loss": 1.9844,
+      "step": 822
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 5.154099348952669,
+      "learning_rate": 0.0001,
+      "loss": 2.9062,
+      "step": 823
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 4.925492637519167,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 824
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 6.390984018933043,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 825
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 6.9138596085917285,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 826
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 5.369883533158641,
+      "learning_rate": 0.0001,
+      "loss": 2.8125,
+      "step": 827
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 7.456914473954453,
+      "learning_rate": 0.0001,
+      "loss": 2.7812,
+      "step": 828
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 5.791370397982031,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 829
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 4.473564456526789,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 830
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 6.377472917090085,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 831
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 7.061870345826034,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 832
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 7.475056565446695,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 833
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 5.943474193334696,
+      "learning_rate": 0.0001,
+      "loss": 2.0781,
+      "step": 834
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 5.786165179983016,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 835
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 6.540859565742491,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 836
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 6.852178210980113,
+      "learning_rate": 0.0001,
+      "loss": 2.2969,
+      "step": 837
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 6.427660937563824,
+      "learning_rate": 0.0001,
+      "loss": 3.0469,
+      "step": 838
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 7.404832954994649,
+      "learning_rate": 0.0001,
+      "loss": 2.875,
+      "step": 839
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 5.15547466947869,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 840
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 9.284304247251512,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 841
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 6.44211346051645,
+      "learning_rate": 0.0001,
+      "loss": 2.4062,
+      "step": 842
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 7.407204225068616,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 843
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 6.536030004163721,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 844
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 5.664816059826595,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 845
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 7.074285974063112,
+      "learning_rate": 0.0001,
+      "loss": 2.3281,
+      "step": 846
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 4.930926666384628,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 847
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 5.663530450660167,
+      "learning_rate": 0.0001,
+      "loss": 3.0625,
+      "step": 848
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 5.635849522363945,
+      "learning_rate": 0.0001,
+      "loss": 2.9375,
+      "step": 849
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 8.186385957751863,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 850
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 5.9945008515729254,
+      "learning_rate": 0.0001,
+      "loss": 2.5312,
+      "step": 851
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 5.472155827201553,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 852
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 9.14877945827067,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 853
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 6.100888417967896,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 854
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 7.028201495974476,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 855
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 7.158110848813972,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 856
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 5.9746126891715985,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 857
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 6.194895879605958,
+      "learning_rate": 0.0001,
+      "loss": 2.2812,
+      "step": 858
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 5.7717568499039755,
+      "learning_rate": 0.0001,
+      "loss": 2.7031,
+      "step": 859
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 7.816780702683526,
+      "learning_rate": 0.0001,
+      "loss": 2.8906,
+      "step": 860
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 6.419624396463192,
+      "learning_rate": 0.0001,
+      "loss": 2.5625,
+      "step": 861
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 7.185649890506834,
+      "learning_rate": 0.0001,
+      "loss": 2.8438,
+      "step": 862
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 5.6908416343601935,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 863
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 4.859486347190332,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 864
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 6.285213197221704,
+      "learning_rate": 0.0001,
+      "loss": 2.625,
+      "step": 865
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 6.738698450535726,
+      "learning_rate": 0.0001,
+      "loss": 2.9219,
+      "step": 866
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 5.9454993899618875,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 867
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 8.998641819830656,
+      "learning_rate": 0.0001,
+      "loss": 2.0625,
+      "step": 868
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 7.536207240118511,
+      "learning_rate": 0.0001,
+      "loss": 2.5,
+      "step": 869
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 5.971690600825191,
+      "learning_rate": 0.0001,
+      "loss": 2.6406,
+      "step": 870
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 4.808198125624626,
+      "learning_rate": 0.0001,
+      "loss": 2.6719,
+      "step": 871
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 5.456368344689776,
+      "learning_rate": 0.0001,
+      "loss": 2.3438,
+      "step": 872
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 5.88624229180227,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 873
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 6.792604392226326,
+      "learning_rate": 0.0001,
+      "loss": 2.9531,
+      "step": 874
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 8.842989968926629,
+      "learning_rate": 0.0001,
+      "loss": 2.4219,
+      "step": 875
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 6.812604406395962,
+      "learning_rate": 0.0001,
+      "loss": 2.3281,
+      "step": 876
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 8.040123847392087,
+      "learning_rate": 0.0001,
+      "loss": 2.7188,
+      "step": 877
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 5.413687121021301,
+      "learning_rate": 0.0001,
+      "loss": 2.8281,
+      "step": 878
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 5.85591948084603,
+      "learning_rate": 0.0001,
+      "loss": 2.6562,
+      "step": 879
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 7.731647984076021,
+      "learning_rate": 0.0001,
+      "loss": 2.2656,
+      "step": 880
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 5.3911022010810346,
+      "learning_rate": 0.0001,
+      "loss": 2.5781,
+      "step": 881
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 7.088453541216916,
+      "learning_rate": 0.0001,
+      "loss": 2.4688,
+      "step": 882
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 6.873463198730782,
+      "learning_rate": 0.0001,
+      "loss": 2.7344,
+      "step": 883
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 8.452510173369372,
+      "learning_rate": 0.0001,
+      "loss": 2.5156,
+      "step": 884
+    },
+    {
+      "epoch": 4.0,
+      "step": 884,
+      "total_flos": 17145224232960.0,
+      "train_loss": 2.7685502686651584,
+      "train_runtime": 2858.5803,
+      "train_samples_per_second": 2.47,
+      "train_steps_per_second": 0.309
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 884,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 500,
+  "total_flos": 17145224232960.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}