diff --git "a/checkpoint-1000/trainer_state.json" "b/checkpoint-1000/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-1000/trainer_state.json"
@@ -0,0 +1,7033 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.01,
+  "eval_steps": 500,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 1e-05,
+      "grad_norm": 1.631775788945859,
+      "learning_rate": 3e-06,
+      "loss": 10.8658,
+      "step": 1
+    },
+    {
+      "epoch": 2e-05,
+      "grad_norm": 1.6173222705173065,
+      "learning_rate": 6e-06,
+      "loss": 10.8645,
+      "step": 2
+    },
+    {
+      "epoch": 3e-05,
+      "grad_norm": 1.6387507243802044,
+      "learning_rate": 9e-06,
+      "loss": 10.8646,
+      "step": 3
+    },
+    {
+      "epoch": 4e-05,
+      "grad_norm": 1.5975767011448438,
+      "learning_rate": 1.2e-05,
+      "loss": 10.8638,
+      "step": 4
+    },
+    {
+      "epoch": 5e-05,
+      "grad_norm": 1.6454022013221787,
+      "learning_rate": 1.5e-05,
+      "loss": 10.8605,
+      "step": 5
+    },
+    {
+      "epoch": 6e-05,
+      "grad_norm": 1.6407987097684302,
+      "learning_rate": 1.8e-05,
+      "loss": 10.8581,
+      "step": 6
+    },
+    {
+      "epoch": 7e-05,
+      "grad_norm": 1.609586764602888,
+      "learning_rate": 2.1000000000000002e-05,
+      "loss": 10.8444,
+      "step": 7
+    },
+    {
+      "epoch": 8e-05,
+      "grad_norm": 1.4683048428970586,
+      "learning_rate": 2.4e-05,
+      "loss": 10.8173,
+      "step": 8
+    },
+    {
+      "epoch": 9e-05,
+      "grad_norm": 1.3933728304700357,
+      "learning_rate": 2.7e-05,
+      "loss": 10.8102,
+      "step": 9
+    },
+    {
+      "epoch": 0.0001,
+      "grad_norm": 1.3326098319804733,
+      "learning_rate": 3e-05,
+      "loss": 10.7958,
+      "step": 10
+    },
+    {
+      "epoch": 0.00011,
+      "grad_norm": 1.217237966643813,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 10.779,
+      "step": 11
+    },
+    {
+      "epoch": 0.00012,
+      "grad_norm": 1.1764691634458027,
+      "learning_rate": 3.6e-05,
+      "loss": 10.7677,
+      "step": 12
+    },
+    {
+      "epoch": 0.00013,
+      "grad_norm": 1.1304717895604097,
+      "learning_rate": 3.9e-05,
+      "loss": 10.7486,
+      "step": 13
+    },
+    {
+      "epoch": 0.00014,
+      "grad_norm": 1.115888694899127,
+      "learning_rate": 4.2000000000000004e-05,
+      "loss": 10.7378,
+      "step": 14
+    },
+    {
+      "epoch": 0.00015,
+      "grad_norm": 1.1065220670447153,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 10.7288,
+      "step": 15
+    },
+    {
+      "epoch": 0.00016,
+      "grad_norm": 1.075226769828573,
+      "learning_rate": 4.8e-05,
+      "loss": 10.7115,
+      "step": 16
+    },
+    {
+      "epoch": 0.00017,
+      "grad_norm": 1.0461893681391048,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 10.6956,
+      "step": 17
+    },
+    {
+      "epoch": 0.00018,
+      "grad_norm": 1.0230892582928048,
+      "learning_rate": 5.4e-05,
+      "loss": 10.6795,
+      "step": 18
+    },
+    {
+      "epoch": 0.00019,
+      "grad_norm": 0.9869910790408369,
+      "learning_rate": 5.7e-05,
+      "loss": 10.6637,
+      "step": 19
+    },
+    {
+      "epoch": 0.0002,
+      "grad_norm": 0.9729421058544264,
+      "learning_rate": 6e-05,
+      "loss": 10.6515,
+      "step": 20
+    },
+    {
+      "epoch": 0.00021,
+      "grad_norm": 0.943988636941647,
+      "learning_rate": 6.3e-05,
+      "loss": 10.6378,
+      "step": 21
+    },
+    {
+      "epoch": 0.00022,
+      "grad_norm": 0.9278112602221015,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 10.6233,
+      "step": 22
+    },
+    {
+      "epoch": 0.00023,
+      "grad_norm": 0.920877936489298,
+      "learning_rate": 6.9e-05,
+      "loss": 10.6091,
+      "step": 23
+    },
+    {
+      "epoch": 0.00024,
+      "grad_norm": 0.9178015901962371,
+      "learning_rate": 7.2e-05,
+      "loss": 10.5981,
+      "step": 24
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 0.9193853191136445,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 10.5835,
+      "step": 25
+    },
+    {
+      "epoch": 0.00026,
+      "grad_norm": 0.9191743831022944,
+      "learning_rate": 7.8e-05,
+      "loss": 10.5705,
+      "step": 26
+    },
+    {
+      "epoch": 0.00027,
+      "grad_norm": 0.9136913401152261,
+      "learning_rate": 8.1e-05,
+      "loss": 10.5585,
+      "step": 27
+    },
+    {
+      "epoch": 0.00028,
+      "grad_norm": 0.9119853724531574,
+      "learning_rate": 8.400000000000001e-05,
+      "loss": 10.5456,
+      "step": 28
+    },
+    {
+      "epoch": 0.00029,
+      "grad_norm": 0.9130829908837624,
+      "learning_rate": 8.7e-05,
+      "loss": 10.5312,
+      "step": 29
+    },
+    {
+      "epoch": 0.0003,
+      "grad_norm": 0.9186726006674357,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 10.5159,
+      "step": 30
+    },
+    {
+      "epoch": 0.00031,
+      "grad_norm": 0.9155120967133267,
+      "learning_rate": 9.3e-05,
+      "loss": 10.5012,
+      "step": 31
+    },
+    {
+      "epoch": 0.00032,
+      "grad_norm": 0.9096055134642034,
+      "learning_rate": 9.6e-05,
+      "loss": 10.4871,
+      "step": 32
+    },
+    {
+      "epoch": 0.00033,
+      "grad_norm": 0.91013596598753,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 10.4706,
+      "step": 33
+    },
+    {
+      "epoch": 0.00034,
+      "grad_norm": 0.9103576711685224,
+      "learning_rate": 0.00010200000000000001,
+      "loss": 10.4543,
+      "step": 34
+    },
+    {
+      "epoch": 0.00035,
+      "grad_norm": 0.9164612613814794,
+      "learning_rate": 0.00010500000000000002,
+      "loss": 10.4377,
+      "step": 35
+    },
+    {
+      "epoch": 0.00036,
+      "grad_norm": 0.9129652960092816,
+      "learning_rate": 0.000108,
+      "loss": 10.4202,
+      "step": 36
+    },
+    {
+      "epoch": 0.00037,
+      "grad_norm": 0.9029365420083331,
+      "learning_rate": 0.000111,
+      "loss": 10.4036,
+      "step": 37
+    },
+    {
+      "epoch": 0.00038,
+      "grad_norm": 0.9075151789728013,
+      "learning_rate": 0.000114,
+      "loss": 10.3847,
+      "step": 38
+    },
+    {
+      "epoch": 0.00039,
+      "grad_norm": 0.9102185085970206,
+      "learning_rate": 0.000117,
+      "loss": 10.3654,
+      "step": 39
+    },
+    {
+      "epoch": 0.0004,
+      "grad_norm": 0.9144897951886523,
+      "learning_rate": 0.00012,
+      "loss": 10.3432,
+      "step": 40
+    },
+    {
+      "epoch": 0.00041,
+      "grad_norm": 0.9044769084629607,
+      "learning_rate": 0.000123,
+      "loss": 10.3253,
+      "step": 41
+    },
+    {
+      "epoch": 0.00042,
+      "grad_norm": 0.9101249031053068,
+      "learning_rate": 0.000126,
+      "loss": 10.3047,
+      "step": 42
+    },
+    {
+      "epoch": 0.00043,
+      "grad_norm": 0.9147428510616606,
+      "learning_rate": 0.000129,
+      "loss": 10.282,
+      "step": 43
+    },
+    {
+      "epoch": 0.00044,
+      "grad_norm": 0.9138907921510238,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 10.2606,
+      "step": 44
+    },
+    {
+      "epoch": 0.00045,
+      "grad_norm": 0.9165726299867081,
+      "learning_rate": 0.000135,
+      "loss": 10.2377,
+      "step": 45
+    },
+    {
+      "epoch": 0.00046,
+      "grad_norm": 0.906013196308877,
+      "learning_rate": 0.000138,
+      "loss": 10.216,
+      "step": 46
+    },
+    {
+      "epoch": 0.00047,
+      "grad_norm": 0.9116233570839986,
+      "learning_rate": 0.000141,
+      "loss": 10.1904,
+      "step": 47
+    },
+    {
+      "epoch": 0.00048,
+      "grad_norm": 0.9086131145608887,
+      "learning_rate": 0.000144,
+      "loss": 10.1674,
+      "step": 48
+    },
+    {
+      "epoch": 0.00049,
+      "grad_norm": 0.9094213201037699,
+      "learning_rate": 0.000147,
+      "loss": 10.1435,
+      "step": 49
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 0.9128293203892458,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 10.1173,
+      "step": 50
+    },
+    {
+      "epoch": 0.00051,
+      "grad_norm": 0.9140261923858894,
+      "learning_rate": 0.000153,
+      "loss": 10.0916,
+      "step": 51
+    },
+    {
+      "epoch": 0.00052,
+      "grad_norm": 0.9121253786029146,
+      "learning_rate": 0.000156,
+      "loss": 10.0657,
+      "step": 52
+    },
+    {
+      "epoch": 0.00053,
+      "grad_norm": 0.9034950978450355,
+      "learning_rate": 0.000159,
+      "loss": 10.0413,
+      "step": 53
+    },
+    {
+      "epoch": 0.00054,
+      "grad_norm": 0.9241206201798855,
+      "learning_rate": 0.000162,
+      "loss": 10.0092,
+      "step": 54
+    },
+    {
+      "epoch": 0.00055,
+      "grad_norm": 0.9149154841379399,
+      "learning_rate": 0.000165,
+      "loss": 9.9852,
+      "step": 55
+    },
+    {
+      "epoch": 0.00056,
+      "grad_norm": 0.9070429364957124,
+      "learning_rate": 0.00016800000000000002,
+      "loss": 9.9561,
+      "step": 56
+    },
+    {
+      "epoch": 0.00057,
+      "grad_norm": 0.9122859536360214,
+      "learning_rate": 0.000171,
+      "loss": 9.9285,
+      "step": 57
+    },
+    {
+      "epoch": 0.00058,
+      "grad_norm": 0.9101615769667563,
+      "learning_rate": 0.000174,
+      "loss": 9.9038,
+      "step": 58
+    },
+    {
+      "epoch": 0.00059,
+      "grad_norm": 0.91462118557662,
+      "learning_rate": 0.000177,
+      "loss": 9.872,
+      "step": 59
+    },
+    {
+      "epoch": 0.0006,
+      "grad_norm": 0.902238261354272,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 9.846,
+      "step": 60
+    },
+    {
+      "epoch": 0.00061,
+      "grad_norm": 0.9010263217287604,
+      "learning_rate": 0.000183,
+      "loss": 9.8177,
+      "step": 61
+    },
+    {
+      "epoch": 0.00062,
+      "grad_norm": 0.9073977759483284,
+      "learning_rate": 0.000186,
+      "loss": 9.7857,
+      "step": 62
+    },
+    {
+      "epoch": 0.00063,
+      "grad_norm": 0.9002035349554564,
+      "learning_rate": 0.000189,
+      "loss": 9.7585,
+      "step": 63
+    },
+    {
+      "epoch": 0.00064,
+      "grad_norm": 0.8964472690991813,
+      "learning_rate": 0.000192,
+      "loss": 9.7283,
+      "step": 64
+    },
+    {
+      "epoch": 0.00065,
+      "grad_norm": 0.9038015098943822,
+      "learning_rate": 0.00019500000000000002,
+      "loss": 9.696,
+      "step": 65
+    },
+    {
+      "epoch": 0.00066,
+      "grad_norm": 0.8969444374465311,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 9.6719,
+      "step": 66
+    },
+    {
+      "epoch": 0.00067,
+      "grad_norm": 0.898249772566312,
+      "learning_rate": 0.000201,
+      "loss": 9.642,
+      "step": 67
+    },
+    {
+      "epoch": 0.00068,
+      "grad_norm": 0.9048900549589218,
+      "learning_rate": 0.00020400000000000003,
+      "loss": 9.6091,
+      "step": 68
+    },
+    {
+      "epoch": 0.00069,
+      "grad_norm": 0.8901553719298861,
+      "learning_rate": 0.00020700000000000002,
+      "loss": 9.5816,
+      "step": 69
+    },
+    {
+      "epoch": 0.0007,
+      "grad_norm": 0.903386863411625,
+      "learning_rate": 0.00021000000000000004,
+      "loss": 9.5447,
+      "step": 70
+    },
+    {
+      "epoch": 0.00071,
+      "grad_norm": 0.8933303377898588,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 9.518,
+      "step": 71
+    },
+    {
+      "epoch": 0.00072,
+      "grad_norm": 0.8969213907091691,
+      "learning_rate": 0.000216,
+      "loss": 9.4906,
+      "step": 72
+    },
+    {
+      "epoch": 0.00073,
+      "grad_norm": 0.8960518018191052,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 9.4566,
+      "step": 73
+    },
+    {
+      "epoch": 0.00074,
+      "grad_norm": 0.8964935065687102,
+      "learning_rate": 0.000222,
+      "loss": 9.4297,
+      "step": 74
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 0.8917043311520961,
+      "learning_rate": 0.000225,
+      "loss": 9.3947,
+      "step": 75
+    },
+    {
+      "epoch": 0.00076,
+      "grad_norm": 0.8997723193205578,
+      "learning_rate": 0.000228,
+      "loss": 9.3676,
+      "step": 76
+    },
+    {
+      "epoch": 0.00077,
+      "grad_norm": 0.8937482133829812,
+      "learning_rate": 0.000231,
+      "loss": 9.3393,
+      "step": 77
+    },
+    {
+      "epoch": 0.00078,
+      "grad_norm": 0.886737139439046,
+      "learning_rate": 0.000234,
+      "loss": 9.3091,
+      "step": 78
+    },
+    {
+      "epoch": 0.00079,
+      "grad_norm": 0.8895258541637481,
+      "learning_rate": 0.00023700000000000001,
+      "loss": 9.2733,
+      "step": 79
+    },
+    {
+      "epoch": 0.0008,
+      "grad_norm": 0.8909958989870267,
+      "learning_rate": 0.00024,
+      "loss": 9.2384,
+      "step": 80
+    },
+    {
+      "epoch": 0.00081,
+      "grad_norm": 0.8966003836256963,
+      "learning_rate": 0.00024300000000000002,
+      "loss": 9.2045,
+      "step": 81
+    },
+    {
+      "epoch": 0.00082,
+      "grad_norm": 0.9055894629552318,
+      "learning_rate": 0.000246,
+      "loss": 9.1791,
+      "step": 82
+    },
+    {
+      "epoch": 0.00083,
+      "grad_norm": 0.8961362604582432,
+      "learning_rate": 0.00024900000000000004,
+      "loss": 9.1431,
+      "step": 83
+    },
+    {
+      "epoch": 0.00084,
+      "grad_norm": 0.8980147614185676,
+      "learning_rate": 0.000252,
+      "loss": 9.1116,
+      "step": 84
+    },
+    {
+      "epoch": 0.00085,
+      "grad_norm": 0.8928314575907911,
+      "learning_rate": 0.000255,
+      "loss": 9.0884,
+      "step": 85
+    },
+    {
+      "epoch": 0.00086,
+      "grad_norm": 0.8950046881177521,
+      "learning_rate": 0.000258,
+      "loss": 9.0518,
+      "step": 86
+    },
+    {
+      "epoch": 0.00087,
+      "grad_norm": 0.8877891352831231,
+      "learning_rate": 0.000261,
+      "loss": 9.0297,
+      "step": 87
+    },
+    {
+      "epoch": 0.00088,
+      "grad_norm": 0.8929677780146621,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 8.992,
+      "step": 88
+    },
+    {
+      "epoch": 0.00089,
+      "grad_norm": 0.8871031960638883,
+      "learning_rate": 0.000267,
+      "loss": 8.9663,
+      "step": 89
+    },
+    {
+      "epoch": 0.0009,
+      "grad_norm": 0.8863614322091727,
+      "learning_rate": 0.00027,
+      "loss": 8.9387,
+      "step": 90
+    },
+    {
+      "epoch": 0.00091,
+      "grad_norm": 0.8806256133072948,
+      "learning_rate": 0.000273,
+      "loss": 8.9083,
+      "step": 91
+    },
+    {
+      "epoch": 0.00092,
+      "grad_norm": 0.8826316050497074,
+      "learning_rate": 0.000276,
+      "loss": 8.8757,
+      "step": 92
+    },
+    {
+      "epoch": 0.00093,
+      "grad_norm": 0.8833603577684496,
+      "learning_rate": 0.000279,
+      "loss": 8.8461,
+      "step": 93
+    },
+    {
+      "epoch": 0.00094,
+      "grad_norm": 0.8819538724809766,
+      "learning_rate": 0.000282,
+      "loss": 8.82,
+      "step": 94
+    },
+    {
+      "epoch": 0.00095,
+      "grad_norm": 0.8776473753829749,
+      "learning_rate": 0.000285,
+      "loss": 8.7909,
+      "step": 95
+    },
+    {
+      "epoch": 0.00096,
+      "grad_norm": 0.8854898687433331,
+      "learning_rate": 0.000288,
+      "loss": 8.7608,
+      "step": 96
+    },
+    {
+      "epoch": 0.00097,
+      "grad_norm": 0.8763526561707659,
+      "learning_rate": 0.000291,
+      "loss": 8.7376,
+      "step": 97
+    },
+    {
+      "epoch": 0.00098,
+      "grad_norm": 0.8773720509513535,
+      "learning_rate": 0.000294,
+      "loss": 8.7001,
+      "step": 98
+    },
+    {
+      "epoch": 0.00099,
+      "grad_norm": 0.8782537783818637,
+      "learning_rate": 0.000297,
+      "loss": 8.6785,
+      "step": 99
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 0.876084031734807,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 8.6509,
+      "step": 100
+    },
+    {
+      "epoch": 0.00101,
+      "grad_norm": 0.8775511931302766,
+      "learning_rate": 0.00030300000000000005,
+      "loss": 8.6162,
+      "step": 101
+    },
+    {
+      "epoch": 0.00102,
+      "grad_norm": 0.8666072056658197,
+      "learning_rate": 0.000306,
+      "loss": 8.5962,
+      "step": 102
+    },
+    {
+      "epoch": 0.00103,
+      "grad_norm": 0.8733315070806934,
+      "learning_rate": 0.000309,
+      "loss": 8.5716,
+      "step": 103
+    },
+    {
+      "epoch": 0.00104,
+      "grad_norm": 0.8664419648151436,
+      "learning_rate": 0.000312,
+      "loss": 8.5503,
+      "step": 104
+    },
+    {
+      "epoch": 0.00105,
+      "grad_norm": 0.8699404252946213,
+      "learning_rate": 0.000315,
+      "loss": 8.5232,
+      "step": 105
+    },
+    {
+      "epoch": 0.00106,
+      "grad_norm": 0.8630101619311507,
+      "learning_rate": 0.000318,
+      "loss": 8.4946,
+      "step": 106
+    },
+    {
+      "epoch": 0.00107,
+      "grad_norm": 0.8533006978361278,
+      "learning_rate": 0.000321,
+      "loss": 8.4694,
+      "step": 107
+    },
+    {
+      "epoch": 0.00108,
+      "grad_norm": 0.8571194919918376,
+      "learning_rate": 0.000324,
+      "loss": 8.4408,
+      "step": 108
+    },
+    {
+      "epoch": 0.00109,
+      "grad_norm": 0.8496626885878062,
+      "learning_rate": 0.000327,
+      "loss": 8.4213,
+      "step": 109
+    },
+    {
+      "epoch": 0.0011,
+      "grad_norm": 0.8617458268479945,
+      "learning_rate": 0.00033,
+      "loss": 8.3989,
+      "step": 110
+    },
+    {
+      "epoch": 0.00111,
+      "grad_norm": 0.874730580725934,
+      "learning_rate": 0.000333,
+      "loss": 8.3705,
+      "step": 111
+    },
+    {
+      "epoch": 0.00112,
+      "grad_norm": 0.9211403811176949,
+      "learning_rate": 0.00033600000000000004,
+      "loss": 8.351,
+      "step": 112
+    },
+    {
+      "epoch": 0.00113,
+      "grad_norm": 0.9451163730301329,
+      "learning_rate": 0.000339,
+      "loss": 8.3126,
+      "step": 113
+    },
+    {
+      "epoch": 0.00114,
+      "grad_norm": 0.8518853453666535,
+      "learning_rate": 0.000342,
+      "loss": 8.3025,
+      "step": 114
+    },
+    {
+      "epoch": 0.00115,
+      "grad_norm": 0.8499246309464553,
+      "learning_rate": 0.00034500000000000004,
+      "loss": 8.2758,
+      "step": 115
+    },
+    {
+      "epoch": 0.00116,
+      "grad_norm": 0.8769128820472754,
+      "learning_rate": 0.000348,
+      "loss": 8.2536,
+      "step": 116
+    },
+    {
+      "epoch": 0.00117,
+      "grad_norm": 0.829578266212784,
+      "learning_rate": 0.000351,
+      "loss": 8.2211,
+      "step": 117
+    },
+    {
+      "epoch": 0.00118,
+      "grad_norm": 0.8587574762862499,
+      "learning_rate": 0.000354,
+      "loss": 8.2068,
+      "step": 118
+    },
+    {
+      "epoch": 0.00119,
+      "grad_norm": 0.8383808879241313,
+      "learning_rate": 0.000357,
+      "loss": 8.1942,
+      "step": 119
+    },
+    {
+      "epoch": 0.0012,
+      "grad_norm": 0.8155263912622424,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 8.1675,
+      "step": 120
+    },
+    {
+      "epoch": 0.00121,
+      "grad_norm": 0.8344307084821188,
+      "learning_rate": 0.000363,
+      "loss": 8.1409,
+      "step": 121
+    },
+    {
+      "epoch": 0.00122,
+      "grad_norm": 0.8097993043330719,
+      "learning_rate": 0.000366,
+      "loss": 8.1244,
+      "step": 122
+    },
+    {
+      "epoch": 0.00123,
+      "grad_norm": 0.8029969793277704,
+      "learning_rate": 0.000369,
+      "loss": 8.102,
+      "step": 123
+    },
+    {
+      "epoch": 0.00124,
+      "grad_norm": 0.7829455528112805,
+      "learning_rate": 0.000372,
+      "loss": 8.0811,
+      "step": 124
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 0.8110394816024603,
+      "learning_rate": 0.000375,
+      "loss": 8.0581,
+      "step": 125
+    },
+    {
+      "epoch": 0.00126,
+      "grad_norm": 0.8039928825408066,
+      "learning_rate": 0.000378,
+      "loss": 8.0463,
+      "step": 126
+    },
+    {
+      "epoch": 0.00127,
+      "grad_norm": 0.8369020039958236,
+      "learning_rate": 0.000381,
+      "loss": 8.0243,
+      "step": 127
+    },
+    {
+      "epoch": 0.00128,
+      "grad_norm": 0.9124681746054819,
+      "learning_rate": 0.000384,
+      "loss": 8.0062,
+      "step": 128
+    },
+    {
+      "epoch": 0.00129,
+      "grad_norm": 0.9497790585671452,
+      "learning_rate": 0.00038700000000000003,
+      "loss": 7.997,
+      "step": 129
+    },
+    {
+      "epoch": 0.0013,
+      "grad_norm": 0.895034767303024,
+      "learning_rate": 0.00039000000000000005,
+      "loss": 7.9709,
+      "step": 130
+    },
+    {
+      "epoch": 0.00131,
+      "grad_norm": 0.7344919418773682,
+      "learning_rate": 0.000393,
+      "loss": 7.9348,
+      "step": 131
+    },
+    {
+      "epoch": 0.00132,
+      "grad_norm": 0.8078203789615136,
+      "learning_rate": 0.00039600000000000003,
+      "loss": 7.9285,
+      "step": 132
+    },
+    {
+      "epoch": 0.00133,
+      "grad_norm": 0.8012273324997825,
+      "learning_rate": 0.00039900000000000005,
+      "loss": 7.907,
+      "step": 133
+    },
+    {
+      "epoch": 0.00134,
+      "grad_norm": 0.7227755159346104,
+      "learning_rate": 0.000402,
+      "loss": 7.8939,
+      "step": 134
+    },
+    {
+      "epoch": 0.00135,
+      "grad_norm": 0.7235157354092677,
+      "learning_rate": 0.00040500000000000003,
+      "loss": 7.8661,
+      "step": 135
+    },
+    {
+      "epoch": 0.00136,
+      "grad_norm": 0.7822895950244824,
+      "learning_rate": 0.00040800000000000005,
+      "loss": 7.852,
+      "step": 136
+    },
+    {
+      "epoch": 0.00137,
+      "grad_norm": 0.7608062709985561,
+      "learning_rate": 0.000411,
+      "loss": 7.8315,
+      "step": 137
+    },
+    {
+      "epoch": 0.00138,
+      "grad_norm": 0.7380360973204948,
+      "learning_rate": 0.00041400000000000003,
+      "loss": 7.8088,
+      "step": 138
+    },
+    {
+      "epoch": 0.00139,
+      "grad_norm": 0.684851228306475,
+      "learning_rate": 0.00041700000000000005,
+      "loss": 7.7952,
+      "step": 139
+    },
+    {
+      "epoch": 0.0014,
+      "grad_norm": 0.693462889960616,
+      "learning_rate": 0.00042000000000000007,
+      "loss": 7.7794,
+      "step": 140
+    },
+    {
+      "epoch": 0.00141,
+      "grad_norm": 0.7360428489985282,
+      "learning_rate": 0.000423,
+      "loss": 7.7672,
+      "step": 141
+    },
+    {
+      "epoch": 0.00142,
+      "grad_norm": 0.8088792714181905,
+      "learning_rate": 0.00042599999999999995,
+      "loss": 7.7572,
+      "step": 142
+    },
+    {
+      "epoch": 0.00143,
+      "grad_norm": 0.8184868537412088,
+      "learning_rate": 0.00042899999999999997,
+      "loss": 7.7297,
+      "step": 143
+    },
+    {
+      "epoch": 0.00144,
+      "grad_norm": 0.7328054873489724,
+      "learning_rate": 0.000432,
+      "loss": 7.7108,
+      "step": 144
+    },
+    {
+      "epoch": 0.00145,
+      "grad_norm": 0.742383079085953,
+      "learning_rate": 0.000435,
+      "loss": 7.7067,
+      "step": 145
+    },
+    {
+      "epoch": 0.00146,
+      "grad_norm": 0.8017743593965694,
+      "learning_rate": 0.00043799999999999997,
+      "loss": 7.6737,
+      "step": 146
+    },
+    {
+      "epoch": 0.00147,
+      "grad_norm": 0.7202665590443219,
+      "learning_rate": 0.000441,
+      "loss": 7.6628,
+      "step": 147
+    },
+    {
+      "epoch": 0.00148,
+      "grad_norm": 0.5746342281678257,
+      "learning_rate": 0.000444,
+      "loss": 7.6477,
+      "step": 148
+    },
+    {
+      "epoch": 0.00149,
+      "grad_norm": 0.5929410187247641,
+      "learning_rate": 0.00044699999999999997,
+      "loss": 7.6424,
+      "step": 149
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 0.7151318764270516,
+      "learning_rate": 0.00045,
+      "loss": 7.6133,
+      "step": 150
+    },
+    {
+      "epoch": 0.00151,
+      "grad_norm": 0.6796514647913489,
+      "learning_rate": 0.000453,
+      "loss": 7.5947,
+      "step": 151
+    },
+    {
+      "epoch": 0.00152,
+      "grad_norm": 0.543758313944587,
+      "learning_rate": 0.000456,
+      "loss": 7.5852,
+      "step": 152
+    },
+    {
+      "epoch": 0.00153,
+      "grad_norm": 0.7299692440583161,
+      "learning_rate": 0.000459,
+      "loss": 7.5768,
+      "step": 153
+    },
+    {
+      "epoch": 0.00154,
+      "grad_norm": 0.8631376359970574,
+      "learning_rate": 0.000462,
+      "loss": 7.5644,
+      "step": 154
+    },
+    {
+      "epoch": 0.00155,
+      "grad_norm": 0.9628027146132815,
+      "learning_rate": 0.000465,
+      "loss": 7.546,
+      "step": 155
+    },
+    {
+      "epoch": 0.00156,
+      "grad_norm": 1.6485250632015214,
+      "learning_rate": 0.000468,
+      "loss": 7.5336,
+      "step": 156
+    },
+    {
+      "epoch": 0.00157,
+      "grad_norm": 0.9248857473926935,
+      "learning_rate": 0.000471,
+      "loss": 7.5216,
+      "step": 157
+    },
+    {
+      "epoch": 0.00158,
+      "grad_norm": 0.7465186792591986,
+      "learning_rate": 0.00047400000000000003,
+      "loss": 7.4981,
+      "step": 158
+    },
+    {
+      "epoch": 0.00159,
+      "grad_norm": 0.5902384572619932,
+      "learning_rate": 0.000477,
+      "loss": 7.4827,
+      "step": 159
+    },
+    {
+      "epoch": 0.0016,
+      "grad_norm": 0.799266704152554,
+      "learning_rate": 0.00048,
+      "loss": 7.4675,
+      "step": 160
+    },
+    {
+      "epoch": 0.00161,
+      "grad_norm": 0.7827143911710401,
+      "learning_rate": 0.00048300000000000003,
+      "loss": 7.4466,
+      "step": 161
+    },
+    {
+      "epoch": 0.00162,
+      "grad_norm": 0.9218921339316959,
+      "learning_rate": 0.00048600000000000005,
+      "loss": 7.4513,
+      "step": 162
+    },
+    {
+      "epoch": 0.00163,
+      "grad_norm": 0.7287660540574216,
+      "learning_rate": 0.0004890000000000001,
+      "loss": 7.4357,
+      "step": 163
+    },
+    {
+      "epoch": 0.00164,
+      "grad_norm": 0.5250441215321361,
+      "learning_rate": 0.000492,
+      "loss": 7.4302,
+      "step": 164
+    },
+    {
+      "epoch": 0.00165,
+      "grad_norm": 0.741808200483857,
+      "learning_rate": 0.000495,
+      "loss": 7.4077,
+      "step": 165
+    },
+    {
+      "epoch": 0.00166,
+      "grad_norm": 0.6759515654254841,
+      "learning_rate": 0.0004980000000000001,
+      "loss": 7.3847,
+      "step": 166
+    },
+    {
+      "epoch": 0.00167,
+      "grad_norm": 0.5877266102377413,
+      "learning_rate": 0.000501,
+      "loss": 7.373,
+      "step": 167
+    },
+    {
+      "epoch": 0.00168,
+      "grad_norm": 0.571053515725499,
+      "learning_rate": 0.000504,
+      "loss": 7.3498,
+      "step": 168
+    },
+    {
+      "epoch": 0.00169,
+      "grad_norm": 0.6044099115482432,
+      "learning_rate": 0.0005070000000000001,
+      "loss": 7.3473,
+      "step": 169
+    },
+    {
+      "epoch": 0.0017,
+      "grad_norm": 0.46081682030640647,
+      "learning_rate": 0.00051,
+      "loss": 7.3262,
+      "step": 170
+    },
+    {
+      "epoch": 0.00171,
+      "grad_norm": 0.6775750897944629,
+      "learning_rate": 0.000513,
+      "loss": 7.3116,
+      "step": 171
+    },
+    {
+      "epoch": 0.00172,
+      "grad_norm": 0.57210862548929,
+      "learning_rate": 0.000516,
+      "loss": 7.3117,
+      "step": 172
+    },
+    {
+      "epoch": 0.00173,
+      "grad_norm": 0.5300582190464731,
+      "learning_rate": 0.0005189999999999999,
+      "loss": 7.2934,
+      "step": 173
+    },
+    {
+      "epoch": 0.00174,
+      "grad_norm": 0.7575900839335431,
+      "learning_rate": 0.000522,
+      "loss": 7.3114,
+      "step": 174
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 0.7613961222663432,
+      "learning_rate": 0.000525,
+      "loss": 7.2747,
+      "step": 175
+    },
+    {
+      "epoch": 0.00176,
+      "grad_norm": 0.6765759081090318,
+      "learning_rate": 0.0005279999999999999,
+      "loss": 7.2505,
+      "step": 176
+    },
+    {
+      "epoch": 0.00177,
+      "grad_norm": 0.8933831251648804,
+      "learning_rate": 0.000531,
+      "loss": 7.2576,
+      "step": 177
+    },
+    {
+      "epoch": 0.00178,
+      "grad_norm": 0.7076429855739662,
+      "learning_rate": 0.000534,
+      "loss": 7.2471,
+      "step": 178
+    },
+    {
+      "epoch": 0.00179,
+      "grad_norm": 0.49163795680938555,
+      "learning_rate": 0.000537,
+      "loss": 7.2319,
+      "step": 179
+    },
+    {
+      "epoch": 0.0018,
+      "grad_norm": 0.6284573091622804,
+      "learning_rate": 0.00054,
+      "loss": 7.2064,
+      "step": 180
+    },
+    {
+      "epoch": 0.00181,
+      "grad_norm": 0.5715580305485367,
+      "learning_rate": 0.000543,
+      "loss": 7.1941,
+      "step": 181
+    },
+    {
+      "epoch": 0.00182,
+      "grad_norm": 0.4282299340658738,
+      "learning_rate": 0.000546,
+      "loss": 7.1864,
+      "step": 182
+    },
+    {
+      "epoch": 0.00183,
+      "grad_norm": 0.5948913741099119,
+      "learning_rate": 0.000549,
+      "loss": 7.1781,
+      "step": 183
+    },
+    {
+      "epoch": 0.00184,
+      "grad_norm": 0.4755161712268706,
+      "learning_rate": 0.000552,
+      "loss": 7.1637,
+      "step": 184
+    },
+    {
+      "epoch": 0.00185,
+      "grad_norm": 0.46412280386502286,
+      "learning_rate": 0.000555,
+      "loss": 7.14,
+      "step": 185
+    },
+    {
+      "epoch": 0.00186,
+      "grad_norm": 0.526893652843914,
+      "learning_rate": 0.000558,
+      "loss": 7.1495,
+      "step": 186
+    },
+    {
+      "epoch": 0.00187,
+      "grad_norm": 0.46987960377579885,
+      "learning_rate": 0.000561,
+      "loss": 7.1236,
+      "step": 187
+    },
+    {
+      "epoch": 0.00188,
+      "grad_norm": 0.41993376006980737,
+      "learning_rate": 0.000564,
+      "loss": 7.1004,
+      "step": 188
+    },
+    {
+      "epoch": 0.00189,
+      "grad_norm": 0.5433001024887105,
+      "learning_rate": 0.000567,
+      "loss": 7.0986,
+      "step": 189
+    },
+    {
+      "epoch": 0.0019,
+      "grad_norm": 0.42284758794841465,
+      "learning_rate": 0.00057,
+      "loss": 7.0796,
+      "step": 190
+    },
+    {
+      "epoch": 0.00191,
+      "grad_norm": 0.47158596225286625,
+      "learning_rate": 0.000573,
+      "loss": 7.0701,
+      "step": 191
+    },
+    {
+      "epoch": 0.00192,
+      "grad_norm": 0.4590137866807191,
+      "learning_rate": 0.000576,
+      "loss": 7.0611,
+      "step": 192
+    },
+    {
+      "epoch": 0.00193,
+      "grad_norm": 0.43849327433792495,
+      "learning_rate": 0.000579,
+      "loss": 7.0612,
+      "step": 193
+    },
+    {
+      "epoch": 0.00194,
+      "grad_norm": 0.4326532382468588,
+      "learning_rate": 0.000582,
+      "loss": 7.0527,
+      "step": 194
+    },
+    {
+      "epoch": 0.00195,
+      "grad_norm": 0.508356146068285,
+      "learning_rate": 0.000585,
+      "loss": 7.0348,
+      "step": 195
+    },
+    {
+      "epoch": 0.00196,
+      "grad_norm": 0.534972126993015,
+      "learning_rate": 0.000588,
+      "loss": 7.0227,
+      "step": 196
+    },
+    {
+      "epoch": 0.00197,
+      "grad_norm": 0.693386583628206,
+      "learning_rate": 0.000591,
+      "loss": 7.0198,
+      "step": 197
+    },
+    {
+      "epoch": 0.00198,
+      "grad_norm": 1.4796610947894584,
+      "learning_rate": 0.000594,
+      "loss": 7.0311,
+      "step": 198
+    },
+    {
+      "epoch": 0.00199,
+      "grad_norm": 0.8596271717690976,
+      "learning_rate": 0.0005970000000000001,
+      "loss": 7.0019,
+      "step": 199
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 0.5642900417004818,
+      "learning_rate": 0.0006000000000000001,
+      "loss": 6.991,
+      "step": 200
+    },
+    {
+      "epoch": 0.00201,
+      "grad_norm": 0.8692961874823039,
+      "learning_rate": 0.000603,
+      "loss": 6.9753,
+      "step": 201
+    },
+    {
+      "epoch": 0.00202,
+      "grad_norm": 0.6679575424248658,
+      "learning_rate": 0.0006060000000000001,
+      "loss": 6.9579,
+      "step": 202
+    },
+    {
+      "epoch": 0.00203,
+      "grad_norm": 0.8457647455673973,
+      "learning_rate": 0.0006090000000000001,
+      "loss": 6.9683,
+      "step": 203
+    },
+    {
+      "epoch": 0.00204,
+      "grad_norm": 0.9450482633193271,
+      "learning_rate": 0.000612,
+      "loss": 6.9556,
+      "step": 204
+    },
+    {
+      "epoch": 0.00205,
+      "grad_norm": 1.3090292735766231,
+      "learning_rate": 0.000615,
+      "loss": 6.9523,
+      "step": 205
+    },
+    {
+      "epoch": 0.00206,
+      "grad_norm": 0.6638095483572409,
+      "learning_rate": 0.000618,
+      "loss": 6.9303,
+      "step": 206
+    },
+    {
+      "epoch": 0.00207,
+      "grad_norm": 0.4840297186697793,
+      "learning_rate": 0.000621,
+      "loss": 6.9174,
+      "step": 207
+    },
+    {
+      "epoch": 0.00208,
+      "grad_norm": 0.7834386214905589,
+      "learning_rate": 0.000624,
+      "loss": 6.9192,
+      "step": 208
+    },
+    {
+      "epoch": 0.00209,
+      "grad_norm": 0.69675327719343,
+      "learning_rate": 0.000627,
+      "loss": 6.9018,
+      "step": 209
+    },
+    {
+      "epoch": 0.0021,
+      "grad_norm": 0.5517351762495105,
+      "learning_rate": 0.00063,
+      "loss": 6.8834,
+      "step": 210
+    },
+    {
+      "epoch": 0.00211,
+      "grad_norm": 0.6866030941755482,
+      "learning_rate": 0.000633,
+      "loss": 6.8831,
+      "step": 211
+    },
+    {
+      "epoch": 0.00212,
+      "grad_norm": 0.4815850149383259,
+      "learning_rate": 0.000636,
+      "loss": 6.8711,
+      "step": 212
+    },
+    {
+      "epoch": 0.00213,
+      "grad_norm": 0.5445114180165818,
+      "learning_rate": 0.000639,
+      "loss": 6.8513,
+      "step": 213
+    },
+    {
+      "epoch": 0.00214,
+      "grad_norm": 0.5893761744890885,
+      "learning_rate": 0.000642,
+      "loss": 6.8511,
+      "step": 214
+    },
+    {
+      "epoch": 0.00215,
+      "grad_norm": 0.3989068076407255,
+      "learning_rate": 0.000645,
+      "loss": 6.8428,
+      "step": 215
+    },
+    {
+      "epoch": 0.00216,
+      "grad_norm": 0.4674719209271809,
+      "learning_rate": 0.000648,
+      "loss": 6.8374,
+      "step": 216
+    },
+    {
+      "epoch": 0.00217,
+      "grad_norm": 0.556887374004828,
+      "learning_rate": 0.000651,
+      "loss": 6.8123,
+      "step": 217
+    },
+    {
+      "epoch": 0.00218,
+      "grad_norm": 0.5348764881519483,
+      "learning_rate": 0.000654,
+      "loss": 6.815,
+      "step": 218
+    },
+    {
+      "epoch": 0.00219,
+      "grad_norm": 0.6261358280484484,
+      "learning_rate": 0.000657,
+      "loss": 6.7906,
+      "step": 219
+    },
+    {
+      "epoch": 0.0022,
+      "grad_norm": 0.5490386646627615,
+      "learning_rate": 0.00066,
+      "loss": 6.8058,
+      "step": 220
+    },
+    {
+      "epoch": 0.00221,
+      "grad_norm": 0.5297655781082383,
+      "learning_rate": 0.0006630000000000001,
+      "loss": 6.7838,
+      "step": 221
+    },
+    {
+      "epoch": 0.00222,
+      "grad_norm": 0.53116953133404,
+      "learning_rate": 0.000666,
+      "loss": 6.7711,
+      "step": 222
+    },
+    {
+      "epoch": 0.00223,
+      "grad_norm": 0.46985855580572156,
+      "learning_rate": 0.000669,
+      "loss": 6.7662,
+      "step": 223
+    },
+    {
+      "epoch": 0.00224,
+      "grad_norm": 0.48892819667849163,
+      "learning_rate": 0.0006720000000000001,
+      "loss": 6.7615,
+      "step": 224
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 0.5426443115029689,
+      "learning_rate": 0.000675,
+      "loss": 6.7505,
+      "step": 225
+    },
+    {
+      "epoch": 0.00226,
+      "grad_norm": 0.47341745143430014,
+      "learning_rate": 0.000678,
+      "loss": 6.7457,
+      "step": 226
+    },
+    {
+      "epoch": 0.00227,
+      "grad_norm": 0.47753897999990824,
+      "learning_rate": 0.0006810000000000001,
+      "loss": 6.7186,
+      "step": 227
+    },
+    {
+      "epoch": 0.00228,
+      "grad_norm": 0.43835516232945165,
+      "learning_rate": 0.000684,
+      "loss": 6.721,
+      "step": 228
+    },
+    {
+      "epoch": 0.00229,
+      "grad_norm": 0.3666587821660354,
+      "learning_rate": 0.000687,
+      "loss": 6.7162,
+      "step": 229
+    },
+    {
+      "epoch": 0.0023,
+      "grad_norm": 0.5954344273025705,
+      "learning_rate": 0.0006900000000000001,
+      "loss": 6.6971,
+      "step": 230
+    },
+    {
+      "epoch": 0.00231,
+      "grad_norm": 0.8324250780860898,
+      "learning_rate": 0.000693,
+      "loss": 6.6912,
+      "step": 231
+    },
+    {
+      "epoch": 0.00232,
+      "grad_norm": 1.1082992895496584,
+      "learning_rate": 0.000696,
+      "loss": 6.7117,
+      "step": 232
+    },
+    {
+      "epoch": 0.00233,
+      "grad_norm": 0.8989391942429391,
+      "learning_rate": 0.0006990000000000001,
+      "loss": 6.6931,
+      "step": 233
+    },
+    {
+      "epoch": 0.00234,
+      "grad_norm": 0.8501087453831264,
+      "learning_rate": 0.000702,
+      "loss": 6.6816,
+      "step": 234
+    },
+    {
+      "epoch": 0.00235,
+      "grad_norm": 0.9709457331919181,
+      "learning_rate": 0.000705,
+      "loss": 6.6715,
+      "step": 235
+    },
+    {
+      "epoch": 0.00236,
+      "grad_norm": 0.8996312948341649,
+      "learning_rate": 0.000708,
+      "loss": 6.6542,
+      "step": 236
+    },
+    {
+      "epoch": 0.00237,
+      "grad_norm": 0.7941572817187773,
+      "learning_rate": 0.0007109999999999999,
+      "loss": 6.6634,
+      "step": 237
+    },
+    {
+      "epoch": 0.00238,
+      "grad_norm": 0.649710293154646,
+      "learning_rate": 0.000714,
+      "loss": 6.6483,
+      "step": 238
+    },
+    {
+      "epoch": 0.00239,
+      "grad_norm": 0.7175873388046764,
+      "learning_rate": 0.000717,
+      "loss": 6.6317,
+      "step": 239
+    },
+    {
+      "epoch": 0.0024,
+      "grad_norm": 1.0726439429102004,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 6.6535,
+      "step": 240
+    },
+    {
+      "epoch": 0.00241,
+      "grad_norm": 1.1551390926973517,
+      "learning_rate": 0.000723,
+      "loss": 6.6508,
+      "step": 241
+    },
+    {
+      "epoch": 0.00242,
+      "grad_norm": 0.8245355038796127,
+      "learning_rate": 0.000726,
+      "loss": 6.615,
+      "step": 242
+    },
+    {
+      "epoch": 0.00243,
+      "grad_norm": 0.7119399485811939,
+      "learning_rate": 0.000729,
+      "loss": 6.6026,
+      "step": 243
+    },
+    {
+      "epoch": 0.00244,
+      "grad_norm": 0.6396700306701443,
+      "learning_rate": 0.000732,
+      "loss": 6.6042,
+      "step": 244
+    },
+    {
+      "epoch": 0.00245,
+      "grad_norm": 0.668492143187707,
+      "learning_rate": 0.000735,
+      "loss": 6.5953,
+      "step": 245
+    },
+    {
+      "epoch": 0.00246,
+      "grad_norm": 0.6209038847600604,
+      "learning_rate": 0.000738,
+      "loss": 6.5759,
+      "step": 246
+    },
+    {
+      "epoch": 0.00247,
+      "grad_norm": 0.49470830317055475,
+      "learning_rate": 0.000741,
+      "loss": 6.5677,
+      "step": 247
+    },
+    {
+      "epoch": 0.00248,
+      "grad_norm": 0.5745764116827149,
+      "learning_rate": 0.000744,
+      "loss": 6.5775,
+      "step": 248
+    },
+    {
+      "epoch": 0.00249,
+      "grad_norm": 0.5319509172858093,
+      "learning_rate": 0.000747,
+      "loss": 6.5558,
+      "step": 249
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 0.482084360804442,
+      "learning_rate": 0.00075,
+      "loss": 6.556,
+      "step": 250
+    },
+    {
+      "epoch": 0.00251,
+      "grad_norm": 0.46516739275647623,
+      "learning_rate": 0.000753,
+      "loss": 6.521,
+      "step": 251
+    },
+    {
+      "epoch": 0.00252,
+      "grad_norm": 0.4629119115625355,
+      "learning_rate": 0.000756,
+      "loss": 6.531,
+      "step": 252
+    },
+    {
+      "epoch": 0.00253,
+      "grad_norm": 0.37719629506333596,
+      "learning_rate": 0.000759,
+      "loss": 6.519,
+      "step": 253
+    },
+    {
+      "epoch": 0.00254,
+      "grad_norm": 0.44323602664762185,
+      "learning_rate": 0.000762,
+      "loss": 6.5149,
+      "step": 254
+    },
+    {
+      "epoch": 0.00255,
+      "grad_norm": 0.38153047495099895,
+      "learning_rate": 0.0007650000000000001,
+      "loss": 6.5129,
+      "step": 255
+    },
+    {
+      "epoch": 0.00256,
+      "grad_norm": 0.5270908471121704,
+      "learning_rate": 0.000768,
+      "loss": 6.4934,
+      "step": 256
+    },
+    {
+      "epoch": 0.00257,
+      "grad_norm": 0.6201344591076082,
+      "learning_rate": 0.000771,
+      "loss": 6.4997,
+      "step": 257
+    },
+    {
+      "epoch": 0.00258,
+      "grad_norm": 0.6391276132887356,
+      "learning_rate": 0.0007740000000000001,
+      "loss": 6.477,
+      "step": 258
+    },
+    {
+      "epoch": 0.00259,
+      "grad_norm": 0.6374758421191778,
+      "learning_rate": 0.000777,
+      "loss": 6.4787,
+      "step": 259
+    },
+    {
+      "epoch": 0.0026,
+      "grad_norm": 0.5597091224464362,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 6.4607,
+      "step": 260
+    },
+    {
+      "epoch": 0.00261,
+      "grad_norm": 0.587169694241395,
+      "learning_rate": 0.0007830000000000001,
+      "loss": 6.4722,
+      "step": 261
+    },
+    {
+      "epoch": 0.00262,
+      "grad_norm": 0.6112267949829847,
+      "learning_rate": 0.000786,
+      "loss": 6.4511,
+      "step": 262
+    },
+    {
+      "epoch": 0.00263,
+      "grad_norm": 0.5933922824160996,
+      "learning_rate": 0.0007890000000000001,
+      "loss": 6.4574,
+      "step": 263
+    },
+    {
+      "epoch": 0.00264,
+      "grad_norm": 0.6560299493456899,
+      "learning_rate": 0.0007920000000000001,
+      "loss": 6.4408,
+      "step": 264
+    },
+    {
+      "epoch": 0.00265,
+      "grad_norm": 0.9913628812090025,
+      "learning_rate": 0.000795,
+      "loss": 6.4422,
+      "step": 265
+    },
+    {
+      "epoch": 0.00266,
+      "grad_norm": 1.340981155098937,
+      "learning_rate": 0.0007980000000000001,
+      "loss": 6.4533,
+      "step": 266
+    },
+    {
+      "epoch": 0.00267,
+      "grad_norm": 0.8266116512325479,
+      "learning_rate": 0.0008010000000000001,
+      "loss": 6.4319,
+      "step": 267
+    },
+    {
+      "epoch": 0.00268,
+      "grad_norm": 0.9896228951890642,
+      "learning_rate": 0.000804,
+      "loss": 6.4378,
+      "step": 268
+    },
+    {
+      "epoch": 0.00269,
+      "grad_norm": 1.2352739008881923,
+      "learning_rate": 0.0008070000000000001,
+      "loss": 6.4279,
+      "step": 269
+    },
+    {
+      "epoch": 0.0027,
+      "grad_norm": 1.1652427209458782,
+      "learning_rate": 0.0008100000000000001,
+      "loss": 6.4326,
+      "step": 270
+    },
+    {
+      "epoch": 0.00271,
+      "grad_norm": 1.0407181933539849,
+      "learning_rate": 0.000813,
+      "loss": 6.4319,
+      "step": 271
+    },
+    {
+      "epoch": 0.00272,
+      "grad_norm": 0.8880696455452757,
+      "learning_rate": 0.0008160000000000001,
+      "loss": 6.4138,
+      "step": 272
+    },
+    {
+      "epoch": 0.00273,
+      "grad_norm": 0.8477724135782442,
+      "learning_rate": 0.0008190000000000001,
+      "loss": 6.404,
+      "step": 273
+    },
+    {
+      "epoch": 0.00274,
+      "grad_norm": 0.7818547901656048,
+      "learning_rate": 0.000822,
+      "loss": 6.383,
+      "step": 274
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 0.6915610404761925,
+      "learning_rate": 0.0008250000000000001,
+      "loss": 6.3888,
+      "step": 275
+    },
+    {
+      "epoch": 0.00276,
+      "grad_norm": 0.6369714396426732,
+      "learning_rate": 0.0008280000000000001,
+      "loss": 6.3775,
+      "step": 276
+    },
+    {
+      "epoch": 0.00277,
+      "grad_norm": 0.6792843462530734,
+      "learning_rate": 0.0008310000000000001,
+      "loss": 6.3726,
+      "step": 277
+    },
+    {
+      "epoch": 0.00278,
+      "grad_norm": 0.6716653191335978,
+      "learning_rate": 0.0008340000000000001,
+      "loss": 6.3561,
+      "step": 278
+    },
+    {
+      "epoch": 0.00279,
+      "grad_norm": 0.6104270551210891,
+      "learning_rate": 0.0008370000000000001,
+      "loss": 6.3562,
+      "step": 279
+    },
+    {
+      "epoch": 0.0028,
+      "grad_norm": 0.5327216367370322,
+      "learning_rate": 0.0008400000000000001,
+      "loss": 6.3379,
+      "step": 280
+    },
+    {
+      "epoch": 0.00281,
+      "grad_norm": 0.4495801132850456,
+      "learning_rate": 0.0008430000000000001,
+      "loss": 6.3253,
+      "step": 281
+    },
+    {
+      "epoch": 0.00282,
+      "grad_norm": 0.4185635012011635,
+      "learning_rate": 0.000846,
+      "loss": 6.3254,
+      "step": 282
+    },
+    {
+      "epoch": 0.00283,
+      "grad_norm": 0.41306707794715253,
+      "learning_rate": 0.0008489999999999999,
+      "loss": 6.3154,
+      "step": 283
+    },
+    {
+      "epoch": 0.00284,
+      "grad_norm": 0.447351018324713,
+      "learning_rate": 0.0008519999999999999,
+      "loss": 6.3075,
+      "step": 284
+    },
+    {
+      "epoch": 0.00285,
+      "grad_norm": 0.4656314656211844,
+      "learning_rate": 0.000855,
+      "loss": 6.3102,
+      "step": 285
+    },
+    {
+      "epoch": 0.00286,
+      "grad_norm": 0.5287748664566101,
+      "learning_rate": 0.0008579999999999999,
+      "loss": 6.2881,
+      "step": 286
+    },
+    {
+      "epoch": 0.00287,
+      "grad_norm": 0.60454227039484,
+      "learning_rate": 0.000861,
+      "loss": 6.2937,
+      "step": 287
+    },
+    {
+      "epoch": 0.00288,
+      "grad_norm": 0.6409086244349441,
+      "learning_rate": 0.000864,
+      "loss": 6.2743,
+      "step": 288
+    },
+    {
+      "epoch": 0.00289,
+      "grad_norm": 0.7540915605033448,
+      "learning_rate": 0.0008669999999999999,
+      "loss": 6.2909,
+      "step": 289
+    },
+    {
+      "epoch": 0.0029,
+      "grad_norm": 0.9532532853232767,
+      "learning_rate": 0.00087,
+      "loss": 6.2777,
+      "step": 290
+    },
+    {
+      "epoch": 0.00291,
+      "grad_norm": 0.9297633606905631,
+      "learning_rate": 0.000873,
+      "loss": 6.278,
+      "step": 291
+    },
+    {
+      "epoch": 0.00292,
+      "grad_norm": 0.6544361038243887,
+      "learning_rate": 0.0008759999999999999,
+      "loss": 6.2635,
+      "step": 292
+    },
+    {
+      "epoch": 0.00293,
+      "grad_norm": 0.9586546136156446,
+      "learning_rate": 0.000879,
+      "loss": 6.2582,
+      "step": 293
+    },
+    {
+      "epoch": 0.00294,
+      "grad_norm": 0.8674924960686783,
+      "learning_rate": 0.000882,
+      "loss": 6.2684,
+      "step": 294
+    },
+    {
+      "epoch": 0.00295,
+      "grad_norm": 0.8596325280201164,
+      "learning_rate": 0.0008849999999999999,
+      "loss": 6.2363,
+      "step": 295
+    },
+    {
+      "epoch": 0.00296,
+      "grad_norm": 0.9927641151458286,
+      "learning_rate": 0.000888,
+      "loss": 6.2541,
+      "step": 296
+    },
+    {
+      "epoch": 0.00297,
+      "grad_norm": 1.342485766358639,
+      "learning_rate": 0.000891,
+      "loss": 6.2408,
+      "step": 297
+    },
+    {
+      "epoch": 0.00298,
+      "grad_norm": 1.1878316085061287,
+      "learning_rate": 0.0008939999999999999,
+      "loss": 6.2537,
+      "step": 298
+    },
+    {
+      "epoch": 0.00299,
+      "grad_norm": 0.9496422749623654,
+      "learning_rate": 0.000897,
+      "loss": 6.2242,
+      "step": 299
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 1.122193003605518,
+      "learning_rate": 0.0009,
+      "loss": 6.2361,
+      "step": 300
+    },
+    {
+      "epoch": 0.00301,
+      "grad_norm": 1.129970594986655,
+      "learning_rate": 0.0009029999999999999,
+      "loss": 6.2273,
+      "step": 301
+    },
+    {
+      "epoch": 0.00302,
+      "grad_norm": 1.0740447263196922,
+      "learning_rate": 0.000906,
+      "loss": 6.2071,
+      "step": 302
+    },
+    {
+      "epoch": 0.00303,
+      "grad_norm": 1.1900410452977912,
+      "learning_rate": 0.000909,
+      "loss": 6.2313,
+      "step": 303
+    },
+    {
+      "epoch": 0.00304,
+      "grad_norm": 0.804691464481299,
+      "learning_rate": 0.000912,
+      "loss": 6.2111,
+      "step": 304
+    },
+    {
+      "epoch": 0.00305,
+      "grad_norm": 0.7167209084416579,
+      "learning_rate": 0.000915,
+      "loss": 6.2106,
+      "step": 305
+    },
+    {
+      "epoch": 0.00306,
+      "grad_norm": 0.5686498260282739,
+      "learning_rate": 0.000918,
+      "loss": 6.1897,
+      "step": 306
+    },
+    {
+      "epoch": 0.00307,
+      "grad_norm": 0.5740516870647188,
+      "learning_rate": 0.000921,
+      "loss": 6.1847,
+      "step": 307
+    },
+    {
+      "epoch": 0.00308,
+      "grad_norm": 0.5214022662741855,
+      "learning_rate": 0.000924,
+      "loss": 6.1668,
+      "step": 308
+    },
+    {
+      "epoch": 0.00309,
+      "grad_norm": 0.489157506496739,
+      "learning_rate": 0.000927,
+      "loss": 6.1798,
+      "step": 309
+    },
+    {
+      "epoch": 0.0031,
+      "grad_norm": 0.4872945232166538,
+      "learning_rate": 0.00093,
+      "loss": 6.1622,
+      "step": 310
+    },
+    {
+      "epoch": 0.00311,
+      "grad_norm": 0.4909949625440354,
+      "learning_rate": 0.000933,
+      "loss": 6.1533,
+      "step": 311
+    },
+    {
+      "epoch": 0.00312,
+      "grad_norm": 0.4186129744309998,
+      "learning_rate": 0.000936,
+      "loss": 6.1314,
+      "step": 312
+    },
+    {
+      "epoch": 0.00313,
+      "grad_norm": 0.36050967020968366,
+      "learning_rate": 0.0009390000000000001,
+      "loss": 6.1442,
+      "step": 313
+    },
+    {
+      "epoch": 0.00314,
+      "grad_norm": 0.3818285660239077,
+      "learning_rate": 0.000942,
+      "loss": 6.1495,
+      "step": 314
+    },
+    {
+      "epoch": 0.00315,
+      "grad_norm": 0.42967169925093956,
+      "learning_rate": 0.000945,
+      "loss": 6.126,
+      "step": 315
+    },
+    {
+      "epoch": 0.00316,
+      "grad_norm": 0.46511434454587514,
+      "learning_rate": 0.0009480000000000001,
+      "loss": 6.1004,
+      "step": 316
+    },
+    {
+      "epoch": 0.00317,
+      "grad_norm": 0.5237888199450732,
+      "learning_rate": 0.000951,
+      "loss": 6.1201,
+      "step": 317
+    },
+    {
+      "epoch": 0.00318,
+      "grad_norm": 0.6610672935792641,
+      "learning_rate": 0.000954,
+      "loss": 6.1161,
+      "step": 318
+    },
+    {
+      "epoch": 0.00319,
+      "grad_norm": 0.9099491192879063,
+      "learning_rate": 0.0009570000000000001,
+      "loss": 6.1122,
+      "step": 319
+    },
+    {
+      "epoch": 0.0032,
+      "grad_norm": 1.0329714723925014,
+      "learning_rate": 0.00096,
+      "loss": 6.1198,
+      "step": 320
+    },
+    {
+      "epoch": 0.00321,
+      "grad_norm": 0.9944509511152075,
+      "learning_rate": 0.000963,
+      "loss": 6.1077,
+      "step": 321
+    },
+    {
+      "epoch": 0.00322,
+      "grad_norm": 1.41191394849347,
+      "learning_rate": 0.0009660000000000001,
+      "loss": 6.1101,
+      "step": 322
+    },
+    {
+      "epoch": 0.00323,
+      "grad_norm": 0.9172707652477707,
+      "learning_rate": 0.000969,
+      "loss": 6.097,
+      "step": 323
+    },
+    {
+      "epoch": 0.00324,
+      "grad_norm": 1.1261423941310122,
+      "learning_rate": 0.0009720000000000001,
+      "loss": 6.1132,
+      "step": 324
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 0.6623590740718236,
+      "learning_rate": 0.0009750000000000001,
+      "loss": 6.0626,
+      "step": 325
+    },
+    {
+      "epoch": 0.00326,
+      "grad_norm": 0.7364273563271467,
+      "learning_rate": 0.0009780000000000001,
+      "loss": 6.0809,
+      "step": 326
+    },
+    {
+      "epoch": 0.00327,
+      "grad_norm": 0.8106867198528734,
+      "learning_rate": 0.000981,
+      "loss": 6.0853,
+      "step": 327
+    },
+    {
+      "epoch": 0.00328,
+      "grad_norm": 0.9008187294951384,
+      "learning_rate": 0.000984,
+      "loss": 6.0637,
+      "step": 328
+    },
+    {
+      "epoch": 0.00329,
+      "grad_norm": 1.0311811602663732,
+      "learning_rate": 0.000987,
+      "loss": 6.0736,
+      "step": 329
+    },
+    {
+      "epoch": 0.0033,
+      "grad_norm": 0.7413155368855245,
+      "learning_rate": 0.00099,
+      "loss": 6.0572,
+      "step": 330
+    },
+    {
+      "epoch": 0.00331,
+      "grad_norm": 0.6745658849207387,
+      "learning_rate": 0.0009930000000000002,
+      "loss": 6.0599,
+      "step": 331
+    },
+    {
+      "epoch": 0.00332,
+      "grad_norm": 0.5913240343902441,
+      "learning_rate": 0.0009960000000000001,
+      "loss": 6.041,
+      "step": 332
+    },
+    {
+      "epoch": 0.00333,
+      "grad_norm": 0.5668749800176679,
+      "learning_rate": 0.000999,
+      "loss": 6.025,
+      "step": 333
+    },
+    {
+      "epoch": 0.00334,
+      "grad_norm": 0.5007608052342689,
+      "learning_rate": 0.001002,
+      "loss": 6.0336,
+      "step": 334
+    },
+    {
+      "epoch": 0.00335,
+      "grad_norm": 0.3983861566645405,
+      "learning_rate": 0.001005,
+      "loss": 6.0284,
+      "step": 335
+    },
+    {
+      "epoch": 0.00336,
+      "grad_norm": 0.4274260388302738,
+      "learning_rate": 0.001008,
+      "loss": 6.0181,
+      "step": 336
+    },
+    {
+      "epoch": 0.00337,
+      "grad_norm": 0.5335498119421307,
+      "learning_rate": 0.0010110000000000002,
+      "loss": 6.0004,
+      "step": 337
+    },
+    {
+      "epoch": 0.00338,
+      "grad_norm": 0.4831054617031532,
+      "learning_rate": 0.0010140000000000001,
+      "loss": 6.0126,
+      "step": 338
+    },
+    {
+      "epoch": 0.00339,
+      "grad_norm": 0.5437142240239534,
+      "learning_rate": 0.0010170000000000001,
+      "loss": 6.0197,
+      "step": 339
+    },
+    {
+      "epoch": 0.0034,
+      "grad_norm": 0.5265472037464025,
+      "learning_rate": 0.00102,
+      "loss": 5.9884,
+      "step": 340
+    },
+    {
+      "epoch": 0.00341,
+      "grad_norm": 0.4912689865797111,
+      "learning_rate": 0.001023,
+      "loss": 5.9692,
+      "step": 341
+    },
+    {
+      "epoch": 0.00342,
+      "grad_norm": 0.42997171211054086,
+      "learning_rate": 0.001026,
+      "loss": 5.974,
+      "step": 342
+    },
+    {
+      "epoch": 0.00343,
+      "grad_norm": 0.5197303002983154,
+      "learning_rate": 0.0010290000000000002,
+      "loss": 5.9763,
+      "step": 343
+    },
+    {
+      "epoch": 0.00344,
+      "grad_norm": 0.8117900330313431,
+      "learning_rate": 0.001032,
+      "loss": 5.9747,
+      "step": 344
+    },
+    {
+      "epoch": 0.00345,
+      "grad_norm": 1.1753813945983669,
+      "learning_rate": 0.001035,
+      "loss": 5.9788,
+      "step": 345
+    },
+    {
+      "epoch": 0.00346,
+      "grad_norm": 0.8135676081857764,
+      "learning_rate": 0.0010379999999999999,
+      "loss": 5.9748,
+      "step": 346
+    },
+    {
+      "epoch": 0.00347,
+      "grad_norm": 1.0182912247404574,
+      "learning_rate": 0.001041,
+      "loss": 5.9557,
+      "step": 347
+    },
+    {
+      "epoch": 0.00348,
+      "grad_norm": 1.1407576555241683,
+      "learning_rate": 0.001044,
+      "loss": 5.978,
+      "step": 348
+    },
+    {
+      "epoch": 0.00349,
+      "grad_norm": 0.7853068136249622,
+      "learning_rate": 0.001047,
+      "loss": 5.9412,
+      "step": 349
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 1.0427704318540805,
+      "learning_rate": 0.00105,
+      "loss": 5.9779,
+      "step": 350
+    },
+    {
+      "epoch": 0.00351,
+      "grad_norm": 0.8821399606009466,
+      "learning_rate": 0.001053,
+      "loss": 5.9701,
+      "step": 351
+    },
+    {
+      "epoch": 0.00352,
+      "grad_norm": 0.9582157894617032,
+      "learning_rate": 0.0010559999999999999,
+      "loss": 5.955,
+      "step": 352
+    },
+    {
+      "epoch": 0.00353,
+      "grad_norm": 1.0526665256553966,
+      "learning_rate": 0.001059,
+      "loss": 5.958,
+      "step": 353
+    },
+    {
+      "epoch": 0.00354,
+      "grad_norm": 1.045275747166985,
+      "learning_rate": 0.001062,
+      "loss": 5.9353,
+      "step": 354
+    },
+    {
+      "epoch": 0.00355,
+      "grad_norm": 1.1505195376317356,
+      "learning_rate": 0.001065,
+      "loss": 5.9542,
+      "step": 355
+    },
+    {
+      "epoch": 0.00356,
+      "grad_norm": 1.0355197503433216,
+      "learning_rate": 0.001068,
+      "loss": 5.9425,
+      "step": 356
+    },
+    {
+      "epoch": 0.00357,
+      "grad_norm": 1.127934795973434,
+      "learning_rate": 0.001071,
+      "loss": 5.9396,
+      "step": 357
+    },
+    {
+      "epoch": 0.00358,
+      "grad_norm": 0.7430425972675007,
+      "learning_rate": 0.001074,
+      "loss": 5.9201,
+      "step": 358
+    },
+    {
+      "epoch": 0.00359,
+      "grad_norm": 0.6597065121039739,
+      "learning_rate": 0.001077,
+      "loss": 5.9099,
+      "step": 359
+    },
+    {
+      "epoch": 0.0036,
+      "grad_norm": 0.6034653307534226,
+      "learning_rate": 0.00108,
+      "loss": 5.9081,
+      "step": 360
+    },
+    {
+      "epoch": 0.00361,
+      "grad_norm": 0.5960018023982208,
+      "learning_rate": 0.001083,
+      "loss": 5.9047,
+      "step": 361
+    },
+    {
+      "epoch": 0.00362,
+      "grad_norm": 0.4563634882449727,
+      "learning_rate": 0.001086,
+      "loss": 5.884,
+      "step": 362
+    },
+    {
+      "epoch": 0.00363,
+      "grad_norm": 0.49274399902142996,
+      "learning_rate": 0.001089,
+      "loss": 5.8759,
+      "step": 363
+    },
+    {
+      "epoch": 0.00364,
+      "grad_norm": 0.4937234603270663,
+      "learning_rate": 0.001092,
+      "loss": 5.8901,
+      "step": 364
+    },
+    {
+      "epoch": 0.00365,
+      "grad_norm": 0.5102012627619638,
+      "learning_rate": 0.001095,
+      "loss": 5.888,
+      "step": 365
+    },
+    {
+      "epoch": 0.00366,
+      "grad_norm": 0.4676595798467989,
+      "learning_rate": 0.001098,
+      "loss": 5.862,
+      "step": 366
+    },
+    {
+      "epoch": 0.00367,
+      "grad_norm": 0.49526135096535867,
+      "learning_rate": 0.001101,
+      "loss": 5.8667,
+      "step": 367
+    },
+    {
+      "epoch": 0.00368,
+      "grad_norm": 0.47887378181150303,
+      "learning_rate": 0.001104,
+      "loss": 5.8643,
+      "step": 368
+    },
+    {
+      "epoch": 0.00369,
+      "grad_norm": 0.48887117156741833,
+      "learning_rate": 0.001107,
+      "loss": 5.8686,
+      "step": 369
+    },
+    {
+      "epoch": 0.0037,
+      "grad_norm": 0.4473709149836047,
+      "learning_rate": 0.00111,
+      "loss": 5.8472,
+      "step": 370
+    },
+    {
+      "epoch": 0.00371,
+      "grad_norm": 0.38589559577094035,
+      "learning_rate": 0.001113,
+      "loss": 5.8158,
+      "step": 371
+    },
+    {
+      "epoch": 0.00372,
+      "grad_norm": 0.3912315505838062,
+      "learning_rate": 0.001116,
+      "loss": 5.8379,
+      "step": 372
+    },
+    {
+      "epoch": 0.00373,
+      "grad_norm": 0.38616823047071297,
+      "learning_rate": 0.001119,
+      "loss": 5.8267,
+      "step": 373
+    },
+    {
+      "epoch": 0.00374,
+      "grad_norm": 0.45854090440574513,
+      "learning_rate": 0.001122,
+      "loss": 5.8316,
+      "step": 374
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 0.5169440196993219,
+      "learning_rate": 0.0011250000000000001,
+      "loss": 5.8332,
+      "step": 375
+    },
+    {
+      "epoch": 0.00376,
+      "grad_norm": 0.5067806568705457,
+      "learning_rate": 0.001128,
+      "loss": 5.8287,
+      "step": 376
+    },
+    {
+      "epoch": 0.00377,
+      "grad_norm": 0.48558945502532774,
+      "learning_rate": 0.001131,
+      "loss": 5.8236,
+      "step": 377
+    },
+    {
+      "epoch": 0.00378,
+      "grad_norm": 0.47384141098896654,
+      "learning_rate": 0.001134,
+      "loss": 5.8187,
+      "step": 378
+    },
+    {
+      "epoch": 0.00379,
+      "grad_norm": 0.5705731390544022,
+      "learning_rate": 0.001137,
+      "loss": 5.8065,
+      "step": 379
+    },
+    {
+      "epoch": 0.0038,
+      "grad_norm": 0.8415616570321116,
+      "learning_rate": 0.00114,
+      "loss": 5.8323,
+      "step": 380
+    },
+    {
+      "epoch": 0.00381,
+      "grad_norm": 1.152388235651458,
+      "learning_rate": 0.0011430000000000001,
+      "loss": 5.8155,
+      "step": 381
+    },
+    {
+      "epoch": 0.00382,
+      "grad_norm": 0.7784536663385624,
+      "learning_rate": 0.001146,
+      "loss": 5.7896,
+      "step": 382
+    },
+    {
+      "epoch": 0.00383,
+      "grad_norm": 1.2096458575940454,
+      "learning_rate": 0.001149,
+      "loss": 5.8132,
+      "step": 383
+    },
+    {
+      "epoch": 0.00384,
+      "grad_norm": 1.2032626959449177,
+      "learning_rate": 0.001152,
+      "loss": 5.8295,
+      "step": 384
+    },
+    {
+      "epoch": 0.00385,
+      "grad_norm": 1.2258405640081835,
+      "learning_rate": 0.001155,
+      "loss": 5.8193,
+      "step": 385
+    },
+    {
+      "epoch": 0.00386,
+      "grad_norm": 1.060557976067675,
+      "learning_rate": 0.001158,
+      "loss": 5.8182,
+      "step": 386
+    },
+    {
+      "epoch": 0.00387,
+      "grad_norm": 1.6852101829047932,
+      "learning_rate": 0.0011610000000000001,
+      "loss": 5.8306,
+      "step": 387
+    },
+    {
+      "epoch": 0.00388,
+      "grad_norm": 0.7125426173667109,
+      "learning_rate": 0.001164,
+      "loss": 5.7875,
+      "step": 388
+    },
+    {
+      "epoch": 0.00389,
+      "grad_norm": 0.9333298966305301,
+      "learning_rate": 0.001167,
+      "loss": 5.8092,
+      "step": 389
+    },
+    {
+      "epoch": 0.0039,
+      "grad_norm": 0.7871116791575423,
+      "learning_rate": 0.00117,
+      "loss": 5.7842,
+      "step": 390
+    },
+    {
+      "epoch": 0.00391,
+      "grad_norm": 0.9033950769229127,
+      "learning_rate": 0.001173,
+      "loss": 5.7945,
+      "step": 391
+    },
+    {
+      "epoch": 0.00392,
+      "grad_norm": 1.0985861295177402,
+      "learning_rate": 0.001176,
+      "loss": 5.8091,
+      "step": 392
+    },
+    {
+      "epoch": 0.00393,
+      "grad_norm": 0.9893983760666882,
+      "learning_rate": 0.0011790000000000001,
+      "loss": 5.787,
+      "step": 393
+    },
+    {
+      "epoch": 0.00394,
+      "grad_norm": 1.0087630537900902,
+      "learning_rate": 0.001182,
+      "loss": 5.7718,
+      "step": 394
+    },
+    {
+      "epoch": 0.00395,
+      "grad_norm": 0.9357634093540522,
+      "learning_rate": 0.001185,
+      "loss": 5.7577,
+      "step": 395
+    },
+    {
+      "epoch": 0.00396,
+      "grad_norm": 0.8613606742928634,
+      "learning_rate": 0.001188,
+      "loss": 5.7674,
+      "step": 396
+    },
+    {
+      "epoch": 0.00397,
+      "grad_norm": 0.9393680367248612,
+      "learning_rate": 0.001191,
+      "loss": 5.7666,
+      "step": 397
+    },
+    {
+      "epoch": 0.00398,
+      "grad_norm": 0.8380984764873387,
+      "learning_rate": 0.0011940000000000002,
+      "loss": 5.7669,
+      "step": 398
+    },
+    {
+      "epoch": 0.00399,
+      "grad_norm": 0.7495495962771003,
+      "learning_rate": 0.0011970000000000001,
+      "loss": 5.7689,
+      "step": 399
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 0.6237821646680863,
+      "learning_rate": 0.0012000000000000001,
+      "loss": 5.751,
+      "step": 400
+    },
+    {
+      "epoch": 0.00401,
+      "grad_norm": 0.6042562364668606,
+      "learning_rate": 0.001203,
+      "loss": 5.7286,
+      "step": 401
+    },
+    {
+      "epoch": 0.00402,
+      "grad_norm": 0.6800421237430357,
+      "learning_rate": 0.001206,
+      "loss": 5.7387,
+      "step": 402
+    },
+    {
+      "epoch": 0.00403,
+      "grad_norm": 0.5349967773183291,
+      "learning_rate": 0.001209,
+      "loss": 5.7296,
+      "step": 403
+    },
+    {
+      "epoch": 0.00404,
+      "grad_norm": 0.4491885962138907,
+      "learning_rate": 0.0012120000000000002,
+      "loss": 5.7204,
+      "step": 404
+    },
+    {
+      "epoch": 0.00405,
+      "grad_norm": 0.4231295613571548,
+      "learning_rate": 0.0012150000000000002,
+      "loss": 5.7133,
+      "step": 405
+    },
+    {
+      "epoch": 0.00406,
+      "grad_norm": 0.4344671315280792,
+      "learning_rate": 0.0012180000000000001,
+      "loss": 5.7007,
+      "step": 406
+    },
+    {
+      "epoch": 0.00407,
+      "grad_norm": 0.3917681066216858,
+      "learning_rate": 0.0012209999999999999,
+      "loss": 5.6947,
+      "step": 407
+    },
+    {
+      "epoch": 0.00408,
+      "grad_norm": 0.3526446903520604,
+      "learning_rate": 0.001224,
+      "loss": 5.7113,
+      "step": 408
+    },
+    {
+      "epoch": 0.00409,
+      "grad_norm": 0.3601774169994176,
+      "learning_rate": 0.001227,
+      "loss": 5.689,
+      "step": 409
+    },
+    {
+      "epoch": 0.0041,
+      "grad_norm": 0.3999316895065895,
+      "learning_rate": 0.00123,
+      "loss": 5.6821,
+      "step": 410
+    },
+    {
+      "epoch": 0.00411,
+      "grad_norm": 0.5215468120681382,
+      "learning_rate": 0.001233,
+      "loss": 5.6975,
+      "step": 411
+    },
+    {
+      "epoch": 0.00412,
+      "grad_norm": 0.8731356141140694,
+      "learning_rate": 0.001236,
+      "loss": 5.697,
+      "step": 412
+    },
+    {
+      "epoch": 0.00413,
+      "grad_norm": 1.2920248463477522,
+      "learning_rate": 0.0012389999999999999,
+      "loss": 5.7158,
+      "step": 413
+    },
+    {
+      "epoch": 0.00414,
+      "grad_norm": 0.7474803494460109,
+      "learning_rate": 0.001242,
+      "loss": 5.6771,
+      "step": 414
+    },
+    {
+      "epoch": 0.00415,
+      "grad_norm": 0.9736431117993121,
+      "learning_rate": 0.001245,
+      "loss": 5.6888,
+      "step": 415
+    },
+    {
+      "epoch": 0.00416,
+      "grad_norm": 0.8653333577780613,
+      "learning_rate": 0.001248,
+      "loss": 5.6728,
+      "step": 416
+    },
+    {
+      "epoch": 0.00417,
+      "grad_norm": 0.6891363999339204,
+      "learning_rate": 0.001251,
+      "loss": 5.695,
+      "step": 417
+    },
+    {
+      "epoch": 0.00418,
+      "grad_norm": 0.7955125411502495,
+      "learning_rate": 0.001254,
+      "loss": 5.6734,
+      "step": 418
+    },
+    {
+      "epoch": 0.00419,
+      "grad_norm": 0.8034523576562718,
+      "learning_rate": 0.0012569999999999999,
+      "loss": 5.6601,
+      "step": 419
+    },
+    {
+      "epoch": 0.0042,
+      "grad_norm": 0.7731586474207807,
+      "learning_rate": 0.00126,
+      "loss": 5.6748,
+      "step": 420
+    },
+    {
+      "epoch": 0.00421,
+      "grad_norm": 0.747486262420627,
+      "learning_rate": 0.001263,
+      "loss": 5.6666,
+      "step": 421
+    },
+    {
+      "epoch": 0.00422,
+      "grad_norm": 0.7917981329409665,
+      "learning_rate": 0.001266,
+      "loss": 5.6544,
+      "step": 422
+    },
+    {
+      "epoch": 0.00423,
+      "grad_norm": 1.0889555078416353,
+      "learning_rate": 0.001269,
+      "loss": 5.6655,
+      "step": 423
+    },
+    {
+      "epoch": 0.00424,
+      "grad_norm": 0.9654337501414605,
+      "learning_rate": 0.001272,
+      "loss": 5.6614,
+      "step": 424
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 0.9055610792467201,
+      "learning_rate": 0.001275,
+      "loss": 5.6624,
+      "step": 425
+    },
+    {
+      "epoch": 0.00426,
+      "grad_norm": 0.8212981627676188,
+      "learning_rate": 0.001278,
+      "loss": 5.6637,
+      "step": 426
+    },
+    {
+      "epoch": 0.00427,
+      "grad_norm": 0.7602647353125763,
+      "learning_rate": 0.001281,
+      "loss": 5.6467,
+      "step": 427
+    },
+    {
+      "epoch": 0.00428,
+      "grad_norm": 0.677777266675102,
+      "learning_rate": 0.001284,
+      "loss": 5.6204,
+      "step": 428
+    },
+    {
+      "epoch": 0.00429,
+      "grad_norm": 0.5947091658499406,
+      "learning_rate": 0.001287,
+      "loss": 5.6311,
+      "step": 429
+    },
+    {
+      "epoch": 0.0043,
+      "grad_norm": 0.6377204770277832,
+      "learning_rate": 0.00129,
+      "loss": 5.6309,
+      "step": 430
+    },
+    {
+      "epoch": 0.00431,
+      "grad_norm": 0.6897719280155576,
+      "learning_rate": 0.001293,
+      "loss": 5.6193,
+      "step": 431
+    },
+    {
+      "epoch": 0.00432,
+      "grad_norm": 0.6884919593361081,
+      "learning_rate": 0.001296,
+      "loss": 5.6258,
+      "step": 432
+    },
+    {
+      "epoch": 0.00433,
+      "grad_norm": 0.6913913571918432,
+      "learning_rate": 0.001299,
+      "loss": 5.6177,
+      "step": 433
+    },
+    {
+      "epoch": 0.00434,
+      "grad_norm": 0.7261280979587743,
+      "learning_rate": 0.001302,
+      "loss": 5.6176,
+      "step": 434
+    },
+    {
+      "epoch": 0.00435,
+      "grad_norm": 0.8547702731757605,
+      "learning_rate": 0.001305,
+      "loss": 5.6162,
+      "step": 435
+    },
+    {
+      "epoch": 0.00436,
+      "grad_norm": 0.9457491419795808,
+      "learning_rate": 0.001308,
+      "loss": 5.5986,
+      "step": 436
+    },
+    {
+      "epoch": 0.00437,
+      "grad_norm": 0.9092672289397813,
+      "learning_rate": 0.001311,
+      "loss": 5.6144,
+      "step": 437
+    },
+    {
+      "epoch": 0.00438,
+      "grad_norm": 0.9049337850080227,
+      "learning_rate": 0.001314,
+      "loss": 5.6026,
+      "step": 438
+    },
+    {
+      "epoch": 0.00439,
+      "grad_norm": 0.7237349559204094,
+      "learning_rate": 0.001317,
+      "loss": 5.622,
+      "step": 439
+    },
+    {
+      "epoch": 0.0044,
+      "grad_norm": 0.8693791239531735,
+      "learning_rate": 0.00132,
+      "loss": 5.6008,
+      "step": 440
+    },
+    {
+      "epoch": 0.00441,
+      "grad_norm": 0.8508838859779835,
+      "learning_rate": 0.001323,
+      "loss": 5.5985,
+      "step": 441
+    },
+    {
+      "epoch": 0.00442,
+      "grad_norm": 0.6987140204651114,
+      "learning_rate": 0.0013260000000000001,
+      "loss": 5.587,
+      "step": 442
+    },
+    {
+      "epoch": 0.00443,
+      "grad_norm": 0.5719525863328404,
+      "learning_rate": 0.001329,
+      "loss": 5.5843,
+      "step": 443
+    },
+    {
+      "epoch": 0.00444,
+      "grad_norm": 0.5407139364493208,
+      "learning_rate": 0.001332,
+      "loss": 5.5841,
+      "step": 444
+    },
+    {
+      "epoch": 0.00445,
+      "grad_norm": 0.5170533332401992,
+      "learning_rate": 0.001335,
+      "loss": 5.5667,
+      "step": 445
+    },
+    {
+      "epoch": 0.00446,
+      "grad_norm": 0.43806698904849195,
+      "learning_rate": 0.001338,
+      "loss": 5.5666,
+      "step": 446
+    },
+    {
+      "epoch": 0.00447,
+      "grad_norm": 0.49048920433285326,
+      "learning_rate": 0.001341,
+      "loss": 5.5671,
+      "step": 447
+    },
+    {
+      "epoch": 0.00448,
+      "grad_norm": 0.46215050883864656,
+      "learning_rate": 0.0013440000000000001,
+      "loss": 5.5475,
+      "step": 448
+    },
+    {
+      "epoch": 0.00449,
+      "grad_norm": 0.5259389714982564,
+      "learning_rate": 0.001347,
+      "loss": 5.5523,
+      "step": 449
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 0.6261840891481112,
+      "learning_rate": 0.00135,
+      "loss": 5.5542,
+      "step": 450
+    },
+    {
+      "epoch": 0.00451,
+      "grad_norm": 0.659112753346069,
+      "learning_rate": 0.001353,
+      "loss": 5.5431,
+      "step": 451
+    },
+    {
+      "epoch": 0.00452,
+      "grad_norm": 0.6800042712218282,
+      "learning_rate": 0.001356,
+      "loss": 5.5311,
+      "step": 452
+    },
+    {
+      "epoch": 0.00453,
+      "grad_norm": 0.5745322110996829,
+      "learning_rate": 0.001359,
+      "loss": 5.5188,
+      "step": 453
+    },
+    {
+      "epoch": 0.00454,
+      "grad_norm": 0.5230006416449293,
+      "learning_rate": 0.0013620000000000001,
+      "loss": 5.5319,
+      "step": 454
+    },
+    {
+      "epoch": 0.00455,
+      "grad_norm": 0.5858648520183006,
+      "learning_rate": 0.0013650000000000001,
+      "loss": 5.5314,
+      "step": 455
+    },
+    {
+      "epoch": 0.00456,
+      "grad_norm": 0.5800568870191161,
+      "learning_rate": 0.001368,
+      "loss": 5.5269,
+      "step": 456
+    },
+    {
+      "epoch": 0.00457,
+      "grad_norm": 0.5535894536098482,
+      "learning_rate": 0.001371,
+      "loss": 5.5162,
+      "step": 457
+    },
+    {
+      "epoch": 0.00458,
+      "grad_norm": 0.56095957103827,
+      "learning_rate": 0.001374,
+      "loss": 5.5224,
+      "step": 458
+    },
+    {
+      "epoch": 0.00459,
+      "grad_norm": 0.7492551931077938,
+      "learning_rate": 0.0013770000000000002,
+      "loss": 5.517,
+      "step": 459
+    },
+    {
+      "epoch": 0.0046,
+      "grad_norm": 0.8251083608050601,
+      "learning_rate": 0.0013800000000000002,
+      "loss": 5.5084,
+      "step": 460
+    },
+    {
+      "epoch": 0.00461,
+      "grad_norm": 0.7810512714683711,
+      "learning_rate": 0.0013830000000000001,
+      "loss": 5.5038,
+      "step": 461
+    },
+    {
+      "epoch": 0.00462,
+      "grad_norm": 0.8065032793416945,
+      "learning_rate": 0.001386,
+      "loss": 5.5174,
+      "step": 462
+    },
+    {
+      "epoch": 0.00463,
+      "grad_norm": 1.0894770209329594,
+      "learning_rate": 0.001389,
+      "loss": 5.5013,
+      "step": 463
+    },
+    {
+      "epoch": 0.00464,
+      "grad_norm": 1.3225439160647088,
+      "learning_rate": 0.001392,
+      "loss": 5.5348,
+      "step": 464
+    },
+    {
+      "epoch": 0.00465,
+      "grad_norm": 1.0604196603169807,
+      "learning_rate": 0.0013950000000000002,
+      "loss": 5.4973,
+      "step": 465
+    },
+    {
+      "epoch": 0.00466,
+      "grad_norm": 1.279638390325454,
+      "learning_rate": 0.0013980000000000002,
+      "loss": 5.5365,
+      "step": 466
+    },
+    {
+      "epoch": 0.00467,
+      "grad_norm": 0.8376473861337255,
+      "learning_rate": 0.0014010000000000001,
+      "loss": 5.5034,
+      "step": 467
+    },
+    {
+      "epoch": 0.00468,
+      "grad_norm": 0.8787509291075707,
+      "learning_rate": 0.001404,
+      "loss": 5.4981,
+      "step": 468
+    },
+    {
+      "epoch": 0.00469,
+      "grad_norm": 0.8315105482947757,
+      "learning_rate": 0.001407,
+      "loss": 5.4995,
+      "step": 469
+    },
+    {
+      "epoch": 0.0047,
+      "grad_norm": 0.900233910053011,
+      "learning_rate": 0.00141,
+      "loss": 5.5098,
+      "step": 470
+    },
+    {
+      "epoch": 0.00471,
+      "grad_norm": 1.1782268624389831,
+      "learning_rate": 0.001413,
+      "loss": 5.5031,
+      "step": 471
+    },
+    {
+      "epoch": 0.00472,
+      "grad_norm": 0.8433457613569132,
+      "learning_rate": 0.001416,
+      "loss": 5.4989,
+      "step": 472
+    },
+    {
+      "epoch": 0.00473,
+      "grad_norm": 0.8984284318795871,
+      "learning_rate": 0.001419,
+      "loss": 5.5107,
+      "step": 473
+    },
+    {
+      "epoch": 0.00474,
+      "grad_norm": 0.8057807296189134,
+      "learning_rate": 0.0014219999999999999,
+      "loss": 5.4892,
+      "step": 474
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 0.8485388443307728,
+      "learning_rate": 0.001425,
+      "loss": 5.4826,
+      "step": 475
+    },
+    {
+      "epoch": 0.00476,
+      "grad_norm": 0.9809665505076786,
+      "learning_rate": 0.001428,
+      "loss": 5.5192,
+      "step": 476
+    },
+    {
+      "epoch": 0.00477,
+      "grad_norm": 0.9686040040277449,
+      "learning_rate": 0.001431,
+      "loss": 5.4785,
+      "step": 477
+    },
+    {
+      "epoch": 0.00478,
+      "grad_norm": 0.8580634771679295,
+      "learning_rate": 0.001434,
+      "loss": 5.4949,
+      "step": 478
+    },
+    {
+      "epoch": 0.00479,
+      "grad_norm": 0.9699411511566143,
+      "learning_rate": 0.001437,
+      "loss": 5.4782,
+      "step": 479
+    },
+    {
+      "epoch": 0.0048,
+      "grad_norm": 0.8190893004419723,
+      "learning_rate": 0.0014399999999999999,
+      "loss": 5.4711,
+      "step": 480
+    },
+    {
+      "epoch": 0.00481,
+      "grad_norm": 0.7019568417012634,
+      "learning_rate": 0.001443,
+      "loss": 5.4711,
+      "step": 481
+    },
+    {
+      "epoch": 0.00482,
+      "grad_norm": 0.6677085458766991,
+      "learning_rate": 0.001446,
+      "loss": 5.4413,
+      "step": 482
+    },
+    {
+      "epoch": 0.00483,
+      "grad_norm": 0.6622223138809283,
+      "learning_rate": 0.001449,
+      "loss": 5.4499,
+      "step": 483
+    },
+    {
+      "epoch": 0.00484,
+      "grad_norm": 0.6831178312322733,
+      "learning_rate": 0.001452,
+      "loss": 5.4529,
+      "step": 484
+    },
+    {
+      "epoch": 0.00485,
+      "grad_norm": 0.6786720110326826,
+      "learning_rate": 0.001455,
+      "loss": 5.4548,
+      "step": 485
+    },
+    {
+      "epoch": 0.00486,
+      "grad_norm": 0.639008514866701,
+      "learning_rate": 0.001458,
+      "loss": 5.4237,
+      "step": 486
+    },
+    {
+      "epoch": 0.00487,
+      "grad_norm": 0.6663550617928226,
+      "learning_rate": 0.001461,
+      "loss": 5.4384,
+      "step": 487
+    },
+    {
+      "epoch": 0.00488,
+      "grad_norm": 0.5492133897414355,
+      "learning_rate": 0.001464,
+      "loss": 5.4132,
+      "step": 488
+    },
+    {
+      "epoch": 0.00489,
+      "grad_norm": 0.5801507624750007,
+      "learning_rate": 0.001467,
+      "loss": 5.4021,
+      "step": 489
+    },
+    {
+      "epoch": 0.0049,
+      "grad_norm": 0.7369258331072537,
+      "learning_rate": 0.00147,
+      "loss": 5.4206,
+      "step": 490
+    },
+    {
+      "epoch": 0.00491,
+      "grad_norm": 0.8149385883601376,
+      "learning_rate": 0.001473,
+      "loss": 5.4211,
+      "step": 491
+    },
+    {
+      "epoch": 0.00492,
+      "grad_norm": 0.7605903135404127,
+      "learning_rate": 0.001476,
+      "loss": 5.4167,
+      "step": 492
+    },
+    {
+      "epoch": 0.00493,
+      "grad_norm": 0.5930344173404182,
+      "learning_rate": 0.001479,
+      "loss": 5.4047,
+      "step": 493
+    },
+    {
+      "epoch": 0.00494,
+      "grad_norm": 0.7636413305061132,
+      "learning_rate": 0.001482,
+      "loss": 5.4167,
+      "step": 494
+    },
+    {
+      "epoch": 0.00495,
+      "grad_norm": 0.7369014833667976,
+      "learning_rate": 0.001485,
+      "loss": 5.4191,
+      "step": 495
+    },
+    {
+      "epoch": 0.00496,
+      "grad_norm": 0.8504550792082386,
+      "learning_rate": 0.001488,
+      "loss": 5.4031,
+      "step": 496
+    },
+    {
+      "epoch": 0.00497,
+      "grad_norm": 0.8843425860263048,
+      "learning_rate": 0.001491,
+      "loss": 5.3921,
+      "step": 497
+    },
+    {
+      "epoch": 0.00498,
+      "grad_norm": 0.7438750793797253,
+      "learning_rate": 0.001494,
+      "loss": 5.4145,
+      "step": 498
+    },
+    {
+      "epoch": 0.00499,
+      "grad_norm": 0.7036650868069556,
+      "learning_rate": 0.001497,
+      "loss": 5.3822,
+      "step": 499
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 0.7877256477029045,
+      "learning_rate": 0.0015,
+      "loss": 5.3896,
+      "step": 500
+    },
+    {
+      "epoch": 0.00501,
+      "grad_norm": 0.7990985096145019,
+      "learning_rate": 0.001503,
+      "loss": 5.3912,
+      "step": 501
+    },
+    {
+      "epoch": 0.00502,
+      "grad_norm": 0.5932348165440957,
+      "learning_rate": 0.001506,
+      "loss": 5.3876,
+      "step": 502
+    },
+    {
+      "epoch": 0.00503,
+      "grad_norm": 0.6328380348360387,
+      "learning_rate": 0.0015090000000000001,
+      "loss": 5.391,
+      "step": 503
+    },
+    {
+      "epoch": 0.00504,
+      "grad_norm": 0.5819727032922326,
+      "learning_rate": 0.001512,
+      "loss": 5.3693,
+      "step": 504
+    },
+    {
+      "epoch": 0.00505,
+      "grad_norm": 0.5953710061568833,
+      "learning_rate": 0.001515,
+      "loss": 5.3594,
+      "step": 505
+    },
+    {
+      "epoch": 0.00506,
+      "grad_norm": 0.5986845177656173,
+      "learning_rate": 0.001518,
+      "loss": 5.3624,
+      "step": 506
+    },
+    {
+      "epoch": 0.00507,
+      "grad_norm": 0.623690249743195,
+      "learning_rate": 0.001521,
+      "loss": 5.3571,
+      "step": 507
+    },
+    {
+      "epoch": 0.00508,
+      "grad_norm": 0.653996676321799,
+      "learning_rate": 0.001524,
+      "loss": 5.3588,
+      "step": 508
+    },
+    {
+      "epoch": 0.00509,
+      "grad_norm": 0.7417086851753733,
+      "learning_rate": 0.0015270000000000001,
+      "loss": 5.3422,
+      "step": 509
+    },
+    {
+      "epoch": 0.0051,
+      "grad_norm": 0.7033408638361137,
+      "learning_rate": 0.0015300000000000001,
+      "loss": 5.3598,
+      "step": 510
+    },
+    {
+      "epoch": 0.00511,
+      "grad_norm": 0.7013752626190988,
+      "learning_rate": 0.001533,
+      "loss": 5.3361,
+      "step": 511
+    },
+    {
+      "epoch": 0.00512,
+      "grad_norm": 0.7403626060663853,
+      "learning_rate": 0.001536,
+      "loss": 5.3344,
+      "step": 512
+    },
+    {
+      "epoch": 0.00513,
+      "grad_norm": 0.7668257914395668,
+      "learning_rate": 0.001539,
+      "loss": 5.3488,
+      "step": 513
+    },
+    {
+      "epoch": 0.00514,
+      "grad_norm": 0.8677889222141009,
+      "learning_rate": 0.001542,
+      "loss": 5.3327,
+      "step": 514
+    },
+    {
+      "epoch": 0.00515,
+      "grad_norm": 0.896065553430359,
+      "learning_rate": 0.0015450000000000001,
+      "loss": 5.3422,
+      "step": 515
+    },
+    {
+      "epoch": 0.00516,
+      "grad_norm": 1.0837566571017694,
+      "learning_rate": 0.0015480000000000001,
+      "loss": 5.3497,
+      "step": 516
+    },
+    {
+      "epoch": 0.00517,
+      "grad_norm": 0.8071431981996826,
+      "learning_rate": 0.001551,
+      "loss": 5.3323,
+      "step": 517
+    },
+    {
+      "epoch": 0.00518,
+      "grad_norm": 0.7918860105262308,
+      "learning_rate": 0.001554,
+      "loss": 5.3156,
+      "step": 518
+    },
+    {
+      "epoch": 0.00519,
+      "grad_norm": 0.7777992304037674,
+      "learning_rate": 0.001557,
+      "loss": 5.3213,
+      "step": 519
+    },
+    {
+      "epoch": 0.0052,
+      "grad_norm": 0.8275508154311308,
+      "learning_rate": 0.0015600000000000002,
+      "loss": 5.3297,
+      "step": 520
+    },
+    {
+      "epoch": 0.00521,
+      "grad_norm": 1.081326682572488,
+      "learning_rate": 0.0015630000000000002,
+      "loss": 5.3161,
+      "step": 521
+    },
+    {
+      "epoch": 0.00522,
+      "grad_norm": 1.0769033841173197,
+      "learning_rate": 0.0015660000000000001,
+      "loss": 5.3235,
+      "step": 522
+    },
+    {
+      "epoch": 0.00523,
+      "grad_norm": 1.1142920349652348,
+      "learning_rate": 0.001569,
+      "loss": 5.3418,
+      "step": 523
+    },
+    {
+      "epoch": 0.00524,
+      "grad_norm": 0.9680819380144772,
+      "learning_rate": 0.001572,
+      "loss": 5.3387,
+      "step": 524
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 1.042843464512002,
+      "learning_rate": 0.001575,
+      "loss": 5.3364,
+      "step": 525
+    },
+    {
+      "epoch": 0.00526,
+      "grad_norm": 0.8760110028730904,
+      "learning_rate": 0.0015780000000000002,
+      "loss": 5.3079,
+      "step": 526
+    },
+    {
+      "epoch": 0.00527,
+      "grad_norm": 0.7131338611439731,
+      "learning_rate": 0.0015810000000000002,
+      "loss": 5.3127,
+      "step": 527
+    },
+    {
+      "epoch": 0.00528,
+      "grad_norm": 0.6786352126644868,
+      "learning_rate": 0.0015840000000000001,
+      "loss": 5.2868,
+      "step": 528
+    },
+    {
+      "epoch": 0.00529,
+      "grad_norm": 0.6952357562094686,
+      "learning_rate": 0.001587,
+      "loss": 5.2922,
+      "step": 529
+    },
+    {
+      "epoch": 0.0053,
+      "grad_norm": 0.8086799159810172,
+      "learning_rate": 0.00159,
+      "loss": 5.3039,
+      "step": 530
+    },
+    {
+      "epoch": 0.00531,
+      "grad_norm": 0.828973806141186,
+      "learning_rate": 0.001593,
+      "loss": 5.2873,
+      "step": 531
+    },
+    {
+      "epoch": 0.00532,
+      "grad_norm": 0.7467931538676229,
+      "learning_rate": 0.0015960000000000002,
+      "loss": 5.2943,
+      "step": 532
+    },
+    {
+      "epoch": 0.00533,
+      "grad_norm": 0.7141354989500697,
+      "learning_rate": 0.0015990000000000002,
+      "loss": 5.2786,
+      "step": 533
+    },
+    {
+      "epoch": 0.00534,
+      "grad_norm": 0.9764493557723114,
+      "learning_rate": 0.0016020000000000001,
+      "loss": 5.2728,
+      "step": 534
+    },
+    {
+      "epoch": 0.00535,
+      "grad_norm": 1.3880068471110967,
+      "learning_rate": 0.001605,
+      "loss": 5.2984,
+      "step": 535
+    },
+    {
+      "epoch": 0.00536,
+      "grad_norm": 0.9041341459356813,
+      "learning_rate": 0.001608,
+      "loss": 5.2855,
+      "step": 536
+    },
+    {
+      "epoch": 0.00537,
+      "grad_norm": 0.762740194970871,
+      "learning_rate": 0.0016110000000000002,
+      "loss": 5.2723,
+      "step": 537
+    },
+    {
+      "epoch": 0.00538,
+      "grad_norm": 0.787312661332683,
+      "learning_rate": 0.0016140000000000002,
+      "loss": 5.2506,
+      "step": 538
+    },
+    {
+      "epoch": 0.00539,
+      "grad_norm": 0.6102453826005042,
+      "learning_rate": 0.0016170000000000002,
+      "loss": 5.2365,
+      "step": 539
+    },
+    {
+      "epoch": 0.0054,
+      "grad_norm": 0.6664103859218952,
+      "learning_rate": 0.0016200000000000001,
+      "loss": 5.2513,
+      "step": 540
+    },
+    {
+      "epoch": 0.00541,
+      "grad_norm": 0.7228434818484509,
+      "learning_rate": 0.001623,
+      "loss": 5.2273,
+      "step": 541
+    },
+    {
+      "epoch": 0.00542,
+      "grad_norm": 0.9646545444558308,
+      "learning_rate": 0.001626,
+      "loss": 5.2641,
+      "step": 542
+    },
+    {
+      "epoch": 0.00543,
+      "grad_norm": 1.1121220265997553,
+      "learning_rate": 0.0016290000000000002,
+      "loss": 5.2329,
+      "step": 543
+    },
+    {
+      "epoch": 0.00544,
+      "grad_norm": 0.7994777164441184,
+      "learning_rate": 0.0016320000000000002,
+      "loss": 5.2404,
+      "step": 544
+    },
+    {
+      "epoch": 0.00545,
+      "grad_norm": 0.7226008260314222,
+      "learning_rate": 0.0016350000000000002,
+      "loss": 5.2461,
+      "step": 545
+    },
+    {
+      "epoch": 0.00546,
+      "grad_norm": 0.7699535423166085,
+      "learning_rate": 0.0016380000000000001,
+      "loss": 5.2193,
+      "step": 546
+    },
+    {
+      "epoch": 0.00547,
+      "grad_norm": 0.6548240326600666,
+      "learning_rate": 0.001641,
+      "loss": 5.2108,
+      "step": 547
+    },
+    {
+      "epoch": 0.00548,
+      "grad_norm": 0.6332922946851393,
+      "learning_rate": 0.001644,
+      "loss": 5.2061,
+      "step": 548
+    },
+    {
+      "epoch": 0.00549,
+      "grad_norm": 0.6231528959927674,
+      "learning_rate": 0.0016470000000000002,
+      "loss": 5.2,
+      "step": 549
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 0.7419840881368932,
+      "learning_rate": 0.0016500000000000002,
+      "loss": 5.2179,
+      "step": 550
+    },
+    {
+      "epoch": 0.00551,
+      "grad_norm": 0.7180205816820676,
+      "learning_rate": 0.0016530000000000002,
+      "loss": 5.2057,
+      "step": 551
+    },
+    {
+      "epoch": 0.00552,
+      "grad_norm": 0.5920069574731561,
+      "learning_rate": 0.0016560000000000001,
+      "loss": 5.1823,
+      "step": 552
+    },
+    {
+      "epoch": 0.00553,
+      "grad_norm": 0.7996998429214144,
+      "learning_rate": 0.001659,
+      "loss": 5.1806,
+      "step": 553
+    },
+    {
+      "epoch": 0.00554,
+      "grad_norm": 1.0229110500291838,
+      "learning_rate": 0.0016620000000000003,
+      "loss": 5.1965,
+      "step": 554
+    },
+    {
+      "epoch": 0.00555,
+      "grad_norm": 1.1118473608885646,
+      "learning_rate": 0.0016650000000000002,
+      "loss": 5.1994,
+      "step": 555
+    },
+    {
+      "epoch": 0.00556,
+      "grad_norm": 0.9366759039894813,
+      "learning_rate": 0.0016680000000000002,
+      "loss": 5.1806,
+      "step": 556
+    },
+    {
+      "epoch": 0.00557,
+      "grad_norm": 0.9046668934887724,
+      "learning_rate": 0.0016710000000000002,
+      "loss": 5.1671,
+      "step": 557
+    },
+    {
+      "epoch": 0.00558,
+      "grad_norm": 1.142251826676036,
+      "learning_rate": 0.0016740000000000001,
+      "loss": 5.2009,
+      "step": 558
+    },
+    {
+      "epoch": 0.00559,
+      "grad_norm": 1.0520781475504497,
+      "learning_rate": 0.001677,
+      "loss": 5.1865,
+      "step": 559
+    },
+    {
+      "epoch": 0.0056,
+      "grad_norm": 1.0780070897638405,
+      "learning_rate": 0.0016800000000000003,
+      "loss": 5.1609,
+      "step": 560
+    },
+    {
+      "epoch": 0.00561,
+      "grad_norm": 0.8904071170090557,
+      "learning_rate": 0.0016830000000000003,
+      "loss": 5.1755,
+      "step": 561
+    },
+    {
+      "epoch": 0.00562,
+      "grad_norm": 0.8189640026396579,
+      "learning_rate": 0.0016860000000000002,
+      "loss": 5.168,
+      "step": 562
+    },
+    {
+      "epoch": 0.00563,
+      "grad_norm": 0.746495696524217,
+      "learning_rate": 0.001689,
+      "loss": 5.1552,
+      "step": 563
+    },
+    {
+      "epoch": 0.00564,
+      "grad_norm": 0.7249953066463264,
+      "learning_rate": 0.001692,
+      "loss": 5.1416,
+      "step": 564
+    },
+    {
+      "epoch": 0.00565,
+      "grad_norm": 0.6193711615047397,
+      "learning_rate": 0.001695,
+      "loss": 5.1336,
+      "step": 565
+    },
+    {
+      "epoch": 0.00566,
+      "grad_norm": 0.8661212922050541,
+      "learning_rate": 0.0016979999999999999,
+      "loss": 5.1381,
+      "step": 566
+    },
+    {
+      "epoch": 0.00567,
+      "grad_norm": 0.9452019797636565,
+      "learning_rate": 0.0017009999999999998,
+      "loss": 5.1333,
+      "step": 567
+    },
+    {
+      "epoch": 0.00568,
+      "grad_norm": 0.8863756714851743,
+      "learning_rate": 0.0017039999999999998,
+      "loss": 5.1455,
+      "step": 568
+    },
+    {
+      "epoch": 0.00569,
+      "grad_norm": 0.8164512297006329,
+      "learning_rate": 0.001707,
+      "loss": 5.1087,
+      "step": 569
+    },
+    {
+      "epoch": 0.0057,
+      "grad_norm": 0.8055756655780417,
+      "learning_rate": 0.00171,
+      "loss": 5.1416,
+      "step": 570
+    },
+    {
+      "epoch": 0.00571,
+      "grad_norm": 0.9556127682537684,
+      "learning_rate": 0.001713,
+      "loss": 5.1421,
+      "step": 571
+    },
+    {
+      "epoch": 0.00572,
+      "grad_norm": 1.1121438340859977,
+      "learning_rate": 0.0017159999999999999,
+      "loss": 5.1242,
+      "step": 572
+    },
+    {
+      "epoch": 0.00573,
+      "grad_norm": 0.8538691427356556,
+      "learning_rate": 0.0017189999999999998,
+      "loss": 5.1261,
+      "step": 573
+    },
+    {
+      "epoch": 0.00574,
+      "grad_norm": 0.754134808897758,
+      "learning_rate": 0.001722,
+      "loss": 5.1186,
+      "step": 574
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 0.6045959777005846,
+      "learning_rate": 0.001725,
+      "loss": 5.0826,
+      "step": 575
+    },
+    {
+      "epoch": 0.00576,
+      "grad_norm": 0.5849168439848929,
+      "learning_rate": 0.001728,
+      "loss": 5.0868,
+      "step": 576
+    },
+    {
+      "epoch": 0.00577,
+      "grad_norm": 0.5881868482585118,
+      "learning_rate": 0.001731,
+      "loss": 5.0984,
+      "step": 577
+    },
+    {
+      "epoch": 0.00578,
+      "grad_norm": 0.6496481817365951,
+      "learning_rate": 0.0017339999999999999,
+      "loss": 5.0795,
+      "step": 578
+    },
+    {
+      "epoch": 0.00579,
+      "grad_norm": 0.7126042661301508,
+      "learning_rate": 0.0017369999999999998,
+      "loss": 5.0702,
+      "step": 579
+    },
+    {
+      "epoch": 0.0058,
+      "grad_norm": 0.691634070596695,
+      "learning_rate": 0.00174,
+      "loss": 5.0826,
+      "step": 580
+    },
+    {
+      "epoch": 0.00581,
+      "grad_norm": 0.6405819953602082,
+      "learning_rate": 0.001743,
+      "loss": 5.0809,
+      "step": 581
+    },
+    {
+      "epoch": 0.00582,
+      "grad_norm": 0.6144348123489994,
+      "learning_rate": 0.001746,
+      "loss": 5.0509,
+      "step": 582
+    },
+    {
+      "epoch": 0.00583,
+      "grad_norm": 0.5400038424579979,
+      "learning_rate": 0.001749,
+      "loss": 5.0752,
+      "step": 583
+    },
+    {
+      "epoch": 0.00584,
+      "grad_norm": 0.4936797939946059,
+      "learning_rate": 0.0017519999999999999,
+      "loss": 5.0634,
+      "step": 584
+    },
+    {
+      "epoch": 0.00585,
+      "grad_norm": 0.5420757595953297,
+      "learning_rate": 0.0017549999999999998,
+      "loss": 5.0509,
+      "step": 585
+    },
+    {
+      "epoch": 0.00586,
+      "grad_norm": 0.6461298473240921,
+      "learning_rate": 0.001758,
+      "loss": 5.0463,
+      "step": 586
+    },
+    {
+      "epoch": 0.00587,
+      "grad_norm": 1.0127747465457377,
+      "learning_rate": 0.001761,
+      "loss": 5.0477,
+      "step": 587
+    },
+    {
+      "epoch": 0.00588,
+      "grad_norm": 1.312605638466154,
+      "learning_rate": 0.001764,
+      "loss": 5.0646,
+      "step": 588
+    },
+    {
+      "epoch": 0.00589,
+      "grad_norm": 0.7336414507180539,
+      "learning_rate": 0.001767,
+      "loss": 5.0246,
+      "step": 589
+    },
+    {
+      "epoch": 0.0059,
+      "grad_norm": 0.9403709566188089,
+      "learning_rate": 0.0017699999999999999,
+      "loss": 5.0344,
+      "step": 590
+    },
+    {
+      "epoch": 0.00591,
+      "grad_norm": 1.1082528668309917,
+      "learning_rate": 0.001773,
+      "loss": 5.0455,
+      "step": 591
+    },
+    {
+      "epoch": 0.00592,
+      "grad_norm": 1.2840924567627583,
+      "learning_rate": 0.001776,
+      "loss": 5.0904,
+      "step": 592
+    },
+    {
+      "epoch": 0.00593,
+      "grad_norm": 0.7010705426983365,
+      "learning_rate": 0.001779,
+      "loss": 5.0507,
+      "step": 593
+    },
+    {
+      "epoch": 0.00594,
+      "grad_norm": 0.7515579064184676,
+      "learning_rate": 0.001782,
+      "loss": 5.0452,
+      "step": 594
+    },
+    {
+      "epoch": 0.00595,
+      "grad_norm": 0.8237589608472985,
+      "learning_rate": 0.001785,
+      "loss": 5.0574,
+      "step": 595
+    },
+    {
+      "epoch": 0.00596,
+      "grad_norm": 0.7511193245039597,
+      "learning_rate": 0.0017879999999999999,
+      "loss": 5.0458,
+      "step": 596
+    },
+    {
+      "epoch": 0.00597,
+      "grad_norm": 0.6951714106885373,
+      "learning_rate": 0.001791,
+      "loss": 5.0059,
+      "step": 597
+    },
+    {
+      "epoch": 0.00598,
+      "grad_norm": 0.6637745885790589,
+      "learning_rate": 0.001794,
+      "loss": 5.0231,
+      "step": 598
+    },
+    {
+      "epoch": 0.00599,
+      "grad_norm": 0.7127858481763457,
+      "learning_rate": 0.001797,
+      "loss": 5.0133,
+      "step": 599
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 0.6761974733345899,
+      "learning_rate": 0.0018,
+      "loss": 5.0135,
+      "step": 600
+    },
+    {
+      "epoch": 0.00601,
+      "grad_norm": 0.6625605364634614,
+      "learning_rate": 0.001803,
+      "loss": 5.0015,
+      "step": 601
+    },
+    {
+      "epoch": 0.00602,
+      "grad_norm": 0.6742478727145375,
+      "learning_rate": 0.0018059999999999999,
+      "loss": 4.9862,
+      "step": 602
+    },
+    {
+      "epoch": 0.00603,
+      "grad_norm": 1.00015732542698,
+      "learning_rate": 0.001809,
+      "loss": 5.0193,
+      "step": 603
+    },
+    {
+      "epoch": 0.00604,
+      "grad_norm": 1.387382192884798,
+      "learning_rate": 0.001812,
+      "loss": 5.0251,
+      "step": 604
+    },
+    {
+      "epoch": 0.00605,
+      "grad_norm": 0.6727404947716551,
+      "learning_rate": 0.001815,
+      "loss": 5.0023,
+      "step": 605
+    },
+    {
+      "epoch": 0.00606,
+      "grad_norm": 0.9044609854709968,
+      "learning_rate": 0.001818,
+      "loss": 5.0189,
+      "step": 606
+    },
+    {
+      "epoch": 0.00607,
+      "grad_norm": 1.08596708759871,
+      "learning_rate": 0.001821,
+      "loss": 5.0221,
+      "step": 607
+    },
+    {
+      "epoch": 0.00608,
+      "grad_norm": 1.0369936566425986,
+      "learning_rate": 0.001824,
+      "loss": 5.008,
+      "step": 608
+    },
+    {
+      "epoch": 0.00609,
+      "grad_norm": 1.0935517991120203,
+      "learning_rate": 0.001827,
+      "loss": 5.0109,
+      "step": 609
+    },
+    {
+      "epoch": 0.0061,
+      "grad_norm": 0.9727711844599547,
+      "learning_rate": 0.00183,
+      "loss": 4.9666,
+      "step": 610
+    },
+    {
+      "epoch": 0.00611,
+      "grad_norm": 0.9492725313696737,
+      "learning_rate": 0.001833,
+      "loss": 4.9894,
+      "step": 611
+    },
+    {
+      "epoch": 0.00612,
+      "grad_norm": 0.845936691035656,
+      "learning_rate": 0.001836,
+      "loss": 4.9768,
+      "step": 612
+    },
+    {
+      "epoch": 0.00613,
+      "grad_norm": 0.917579763764549,
+      "learning_rate": 0.001839,
+      "loss": 4.9836,
+      "step": 613
+    },
+    {
+      "epoch": 0.00614,
+      "grad_norm": 0.8975809320202123,
+      "learning_rate": 0.001842,
+      "loss": 5.0024,
+      "step": 614
+    },
+    {
+      "epoch": 0.00615,
+      "grad_norm": 1.1935315831043936,
+      "learning_rate": 0.001845,
+      "loss": 5.0018,
+      "step": 615
+    },
+    {
+      "epoch": 0.00616,
+      "grad_norm": 0.9948318214992812,
+      "learning_rate": 0.001848,
+      "loss": 4.9924,
+      "step": 616
+    },
+    {
+      "epoch": 0.00617,
+      "grad_norm": 0.8063669388844663,
+      "learning_rate": 0.001851,
+      "loss": 4.9919,
+      "step": 617
+    },
+    {
+      "epoch": 0.00618,
+      "grad_norm": 0.8184910219660716,
+      "learning_rate": 0.001854,
+      "loss": 4.9666,
+      "step": 618
+    },
+    {
+      "epoch": 0.00619,
+      "grad_norm": 0.7780464806882716,
+      "learning_rate": 0.001857,
+      "loss": 4.9753,
+      "step": 619
+    },
+    {
+      "epoch": 0.0062,
+      "grad_norm": 0.7430630101852395,
+      "learning_rate": 0.00186,
+      "loss": 4.9566,
+      "step": 620
+    },
+    {
+      "epoch": 0.00621,
+      "grad_norm": 0.8040699289060931,
+      "learning_rate": 0.001863,
+      "loss": 4.9542,
+      "step": 621
+    },
+    {
+      "epoch": 0.00622,
+      "grad_norm": 0.8423285566803137,
+      "learning_rate": 0.001866,
+      "loss": 4.9653,
+      "step": 622
+    },
+    {
+      "epoch": 0.00623,
+      "grad_norm": 0.6802855865245334,
+      "learning_rate": 0.001869,
+      "loss": 4.9365,
+      "step": 623
+    },
+    {
+      "epoch": 0.00624,
+      "grad_norm": 0.7045868643772514,
+      "learning_rate": 0.001872,
+      "loss": 4.9425,
+      "step": 624
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 0.69605003901388,
+      "learning_rate": 0.001875,
+      "loss": 4.9397,
+      "step": 625
+    },
+    {
+      "epoch": 0.00626,
+      "grad_norm": 0.8788947819856907,
+      "learning_rate": 0.0018780000000000001,
+      "loss": 4.9403,
+      "step": 626
+    },
+    {
+      "epoch": 0.00627,
+      "grad_norm": 0.8580113274469313,
+      "learning_rate": 0.001881,
+      "loss": 4.9238,
+      "step": 627
+    },
+    {
+      "epoch": 0.00628,
+      "grad_norm": 0.7437087045232712,
+      "learning_rate": 0.001884,
+      "loss": 4.9553,
+      "step": 628
+    },
+    {
+      "epoch": 0.00629,
+      "grad_norm": 0.673794469112573,
+      "learning_rate": 0.001887,
+      "loss": 4.9059,
+      "step": 629
+    },
+    {
+      "epoch": 0.0063,
+      "grad_norm": 0.7529443514224647,
+      "learning_rate": 0.00189,
+      "loss": 4.9225,
+      "step": 630
+    },
+    {
+      "epoch": 0.00631,
+      "grad_norm": 0.7882316002133182,
+      "learning_rate": 0.0018930000000000002,
+      "loss": 4.9159,
+      "step": 631
+    },
+    {
+      "epoch": 0.00632,
+      "grad_norm": 0.7345089369079263,
+      "learning_rate": 0.0018960000000000001,
+      "loss": 4.9318,
+      "step": 632
+    },
+    {
+      "epoch": 0.00633,
+      "grad_norm": 0.807557335679046,
+      "learning_rate": 0.001899,
+      "loss": 4.9156,
+      "step": 633
+    },
+    {
+      "epoch": 0.00634,
+      "grad_norm": 0.856273971211143,
+      "learning_rate": 0.001902,
+      "loss": 4.9086,
+      "step": 634
+    },
+    {
+      "epoch": 0.00635,
+      "grad_norm": 0.8041095750030954,
+      "learning_rate": 0.001905,
+      "loss": 4.8993,
+      "step": 635
+    },
+    {
+      "epoch": 0.00636,
+      "grad_norm": 0.8334087326563642,
+      "learning_rate": 0.001908,
+      "loss": 4.9117,
+      "step": 636
+    },
+    {
+      "epoch": 0.00637,
+      "grad_norm": 0.8711627404236827,
+      "learning_rate": 0.0019110000000000002,
+      "loss": 4.9242,
+      "step": 637
+    },
+    {
+      "epoch": 0.00638,
+      "grad_norm": 0.950273886749592,
+      "learning_rate": 0.0019140000000000001,
+      "loss": 4.918,
+      "step": 638
+    },
+    {
+      "epoch": 0.00639,
+      "grad_norm": 0.9763758019156279,
+      "learning_rate": 0.001917,
+      "loss": 4.8946,
+      "step": 639
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 0.9069546349974866,
+      "learning_rate": 0.00192,
+      "loss": 4.9049,
+      "step": 640
+    },
+    {
+      "epoch": 0.00641,
+      "grad_norm": 0.7602914411110755,
+      "learning_rate": 0.001923,
+      "loss": 4.8967,
+      "step": 641
+    },
+    {
+      "epoch": 0.00642,
+      "grad_norm": 0.6358369958975738,
+      "learning_rate": 0.001926,
+      "loss": 4.8823,
+      "step": 642
+    },
+    {
+      "epoch": 0.00643,
+      "grad_norm": 0.6298192488222032,
+      "learning_rate": 0.0019290000000000002,
+      "loss": 4.8984,
+      "step": 643
+    },
+    {
+      "epoch": 0.00644,
+      "grad_norm": 0.5835488483304159,
+      "learning_rate": 0.0019320000000000001,
+      "loss": 4.8719,
+      "step": 644
+    },
+    {
+      "epoch": 0.00645,
+      "grad_norm": 0.509893487039198,
+      "learning_rate": 0.001935,
+      "loss": 4.8697,
+      "step": 645
+    },
+    {
+      "epoch": 0.00646,
+      "grad_norm": 0.5016743009567477,
+      "learning_rate": 0.001938,
+      "loss": 4.8822,
+      "step": 646
+    },
+    {
+      "epoch": 0.00647,
+      "grad_norm": 0.4834671196673339,
+      "learning_rate": 0.001941,
+      "loss": 4.8737,
+      "step": 647
+    },
+    {
+      "epoch": 0.00648,
+      "grad_norm": 0.4900666309904975,
+      "learning_rate": 0.0019440000000000002,
+      "loss": 4.8665,
+      "step": 648
+    },
+    {
+      "epoch": 0.00649,
+      "grad_norm": 0.6242032394190251,
+      "learning_rate": 0.0019470000000000002,
+      "loss": 4.8569,
+      "step": 649
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 0.7946730011730083,
+      "learning_rate": 0.0019500000000000001,
+      "loss": 4.8944,
+      "step": 650
+    },
+    {
+      "epoch": 0.00651,
+      "grad_norm": 0.7813096551019217,
+      "learning_rate": 0.001953,
+      "loss": 4.8797,
+      "step": 651
+    },
+    {
+      "epoch": 0.00652,
+      "grad_norm": 0.5708054833927125,
+      "learning_rate": 0.0019560000000000003,
+      "loss": 4.8538,
+      "step": 652
+    },
+    {
+      "epoch": 0.00653,
+      "grad_norm": 0.6416767177196502,
+      "learning_rate": 0.0019590000000000002,
+      "loss": 4.8612,
+      "step": 653
+    },
+    {
+      "epoch": 0.00654,
+      "grad_norm": 0.6414247107018044,
+      "learning_rate": 0.001962,
+      "loss": 4.8324,
+      "step": 654
+    },
+    {
+      "epoch": 0.00655,
+      "grad_norm": 0.5608777579230684,
+      "learning_rate": 0.001965,
+      "loss": 4.8581,
+      "step": 655
+    },
+    {
+      "epoch": 0.00656,
+      "grad_norm": 0.4812696659686437,
+      "learning_rate": 0.001968,
+      "loss": 4.8497,
+      "step": 656
+    },
+    {
+      "epoch": 0.00657,
+      "grad_norm": 0.5196607021803705,
+      "learning_rate": 0.001971,
+      "loss": 4.8212,
+      "step": 657
+    },
+    {
+      "epoch": 0.00658,
+      "grad_norm": 0.5384007134004025,
+      "learning_rate": 0.001974,
+      "loss": 4.8422,
+      "step": 658
+    },
+    {
+      "epoch": 0.00659,
+      "grad_norm": 0.6084877834513672,
+      "learning_rate": 0.001977,
+      "loss": 4.8215,
+      "step": 659
+    },
+    {
+      "epoch": 0.0066,
+      "grad_norm": 0.7589081819730935,
+      "learning_rate": 0.00198,
+      "loss": 4.8434,
+      "step": 660
+    },
+    {
+      "epoch": 0.00661,
+      "grad_norm": 0.7941713837035096,
+      "learning_rate": 0.001983,
+      "loss": 4.8155,
+      "step": 661
+    },
+    {
+      "epoch": 0.00662,
+      "grad_norm": 0.7978547164974868,
+      "learning_rate": 0.0019860000000000004,
+      "loss": 4.8127,
+      "step": 662
+    },
+    {
+      "epoch": 0.00663,
+      "grad_norm": 0.893378015840618,
+      "learning_rate": 0.0019890000000000003,
+      "loss": 4.8265,
+      "step": 663
+    },
+    {
+      "epoch": 0.00664,
+      "grad_norm": 0.899949440268495,
+      "learning_rate": 0.0019920000000000003,
+      "loss": 4.8283,
+      "step": 664
+    },
+    {
+      "epoch": 0.00665,
+      "grad_norm": 0.8127988991300923,
+      "learning_rate": 0.0019950000000000002,
+      "loss": 4.7986,
+      "step": 665
+    },
+    {
+      "epoch": 0.00666,
+      "grad_norm": 0.9200900241429067,
+      "learning_rate": 0.001998,
+      "loss": 4.8226,
+      "step": 666
+    },
+    {
+      "epoch": 0.00667,
+      "grad_norm": 1.037264674390151,
+      "learning_rate": 0.002001,
+      "loss": 4.8324,
+      "step": 667
+    },
+    {
+      "epoch": 0.00668,
+      "grad_norm": 0.8082146942337904,
+      "learning_rate": 0.002004,
+      "loss": 4.8427,
+      "step": 668
+    },
+    {
+      "epoch": 0.00669,
+      "grad_norm": 0.7033624756486074,
+      "learning_rate": 0.002007,
+      "loss": 4.8562,
+      "step": 669
+    },
+    {
+      "epoch": 0.0067,
+      "grad_norm": 0.751969455636164,
+      "learning_rate": 0.00201,
+      "loss": 4.8525,
+      "step": 670
+    },
+    {
+      "epoch": 0.00671,
+      "grad_norm": 0.736520529365372,
+      "learning_rate": 0.002013,
+      "loss": 4.8206,
+      "step": 671
+    },
+    {
+      "epoch": 0.00672,
+      "grad_norm": 0.7466982774868701,
+      "learning_rate": 0.002016,
+      "loss": 4.8129,
+      "step": 672
+    },
+    {
+      "epoch": 0.00673,
+      "grad_norm": 0.7025220166479262,
+      "learning_rate": 0.002019,
+      "loss": 4.8146,
+      "step": 673
+    },
+    {
+      "epoch": 0.00674,
+      "grad_norm": 0.8461453039283889,
+      "learning_rate": 0.0020220000000000004,
+      "loss": 4.8144,
+      "step": 674
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 0.9399881649158435,
+      "learning_rate": 0.0020250000000000003,
+      "loss": 4.8482,
+      "step": 675
+    },
+    {
+      "epoch": 0.00676,
+      "grad_norm": 0.9357632097468723,
+      "learning_rate": 0.0020280000000000003,
+      "loss": 4.8268,
+      "step": 676
+    },
+    {
+      "epoch": 0.00677,
+      "grad_norm": 0.7758960619033557,
+      "learning_rate": 0.0020310000000000003,
+      "loss": 4.8204,
+      "step": 677
+    },
+    {
+      "epoch": 0.00678,
+      "grad_norm": 0.699292513140664,
+      "learning_rate": 0.0020340000000000002,
+      "loss": 4.8248,
+      "step": 678
+    },
+    {
+      "epoch": 0.00679,
+      "grad_norm": 0.7370787957429817,
+      "learning_rate": 0.002037,
+      "loss": 4.816,
+      "step": 679
+    },
+    {
+      "epoch": 0.0068,
+      "grad_norm": 0.8377547362902558,
+      "learning_rate": 0.00204,
+      "loss": 4.8174,
+      "step": 680
+    },
+    {
+      "epoch": 0.00681,
+      "grad_norm": 0.8259782799224379,
+      "learning_rate": 0.002043,
+      "loss": 4.8155,
+      "step": 681
+    },
+    {
+      "epoch": 0.00682,
+      "grad_norm": 0.7684261091318535,
+      "learning_rate": 0.002046,
+      "loss": 4.8082,
+      "step": 682
+    },
+    {
+      "epoch": 0.00683,
+      "grad_norm": 0.8487367019402318,
+      "learning_rate": 0.002049,
+      "loss": 4.7989,
+      "step": 683
+    },
+    {
+      "epoch": 0.00684,
+      "grad_norm": 0.8838018894616847,
+      "learning_rate": 0.002052,
+      "loss": 4.8194,
+      "step": 684
+    },
+    {
+      "epoch": 0.00685,
+      "grad_norm": 0.8860697203395584,
+      "learning_rate": 0.0020550000000000004,
+      "loss": 4.8252,
+      "step": 685
+    },
+    {
+      "epoch": 0.00686,
+      "grad_norm": 0.7336183086529302,
+      "learning_rate": 0.0020580000000000004,
+      "loss": 4.7837,
+      "step": 686
+    },
+    {
+      "epoch": 0.00687,
+      "grad_norm": 0.8176630379413288,
+      "learning_rate": 0.0020610000000000003,
+      "loss": 4.7944,
+      "step": 687
+    },
+    {
+      "epoch": 0.00688,
+      "grad_norm": 0.7703386551342313,
+      "learning_rate": 0.002064,
+      "loss": 4.7926,
+      "step": 688
+    },
+    {
+      "epoch": 0.00689,
+      "grad_norm": 0.6919162061146223,
+      "learning_rate": 0.002067,
+      "loss": 4.7965,
+      "step": 689
+    },
+    {
+      "epoch": 0.0069,
+      "grad_norm": 0.7424392154268248,
+      "learning_rate": 0.00207,
+      "loss": 4.7893,
+      "step": 690
+    },
+    {
+      "epoch": 0.00691,
+      "grad_norm": 0.6515618524352145,
+      "learning_rate": 0.0020729999999999998,
+      "loss": 4.7559,
+      "step": 691
+    },
+    {
+      "epoch": 0.00692,
+      "grad_norm": 0.6440846578393002,
+      "learning_rate": 0.0020759999999999997,
+      "loss": 4.776,
+      "step": 692
+    },
+    {
+      "epoch": 0.00693,
+      "grad_norm": 0.6847536481828279,
+      "learning_rate": 0.0020789999999999997,
+      "loss": 4.7889,
+      "step": 693
+    },
+    {
+      "epoch": 0.00694,
+      "grad_norm": 0.6321576161870056,
+      "learning_rate": 0.002082,
+      "loss": 4.7627,
+      "step": 694
+    },
+    {
+      "epoch": 0.00695,
+      "grad_norm": 0.5791129920715202,
+      "learning_rate": 0.002085,
+      "loss": 4.7609,
+      "step": 695
+    },
+    {
+      "epoch": 0.00696,
+      "grad_norm": 0.5895865438272808,
+      "learning_rate": 0.002088,
+      "loss": 4.7723,
+      "step": 696
+    },
+    {
+      "epoch": 0.00697,
+      "grad_norm": 0.5008187604770619,
+      "learning_rate": 0.002091,
+      "loss": 4.7695,
+      "step": 697
+    },
+    {
+      "epoch": 0.00698,
+      "grad_norm": 0.6970439697756265,
+      "learning_rate": 0.002094,
+      "loss": 4.7619,
+      "step": 698
+    },
+    {
+      "epoch": 0.00699,
+      "grad_norm": 0.8941704543265332,
+      "learning_rate": 0.002097,
+      "loss": 4.7572,
+      "step": 699
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 0.9068627730041655,
+      "learning_rate": 0.0021,
+      "loss": 4.787,
+      "step": 700
+    },
+    {
+      "epoch": 0.00701,
+      "grad_norm": 0.7146483381512303,
+      "learning_rate": 0.002103,
+      "loss": 4.7547,
+      "step": 701
+    },
+    {
+      "epoch": 0.00702,
+      "grad_norm": 0.9172255209446268,
+      "learning_rate": 0.002106,
+      "loss": 4.77,
+      "step": 702
+    },
+    {
+      "epoch": 0.00703,
+      "grad_norm": 0.9047172643914575,
+      "learning_rate": 0.0021089999999999998,
+      "loss": 4.7553,
+      "step": 703
+    },
+    {
+      "epoch": 0.00704,
+      "grad_norm": 0.7853692419556185,
+      "learning_rate": 0.0021119999999999997,
+      "loss": 4.7583,
+      "step": 704
+    },
+    {
+      "epoch": 0.00705,
+      "grad_norm": 0.7199878385614988,
+      "learning_rate": 0.002115,
+      "loss": 4.7725,
+      "step": 705
+    },
+    {
+      "epoch": 0.00706,
+      "grad_norm": 0.7213393080579115,
+      "learning_rate": 0.002118,
+      "loss": 4.7581,
+      "step": 706
+    },
+    {
+      "epoch": 0.00707,
+      "grad_norm": 0.7597119331851468,
+      "learning_rate": 0.002121,
+      "loss": 4.7413,
+      "step": 707
+    },
+    {
+      "epoch": 0.00708,
+      "grad_norm": 0.6864102182118973,
+      "learning_rate": 0.002124,
+      "loss": 4.7187,
+      "step": 708
+    },
+    {
+      "epoch": 0.00709,
+      "grad_norm": 0.7815902187763394,
+      "learning_rate": 0.002127,
+      "loss": 4.7572,
+      "step": 709
+    },
+    {
+      "epoch": 0.0071,
+      "grad_norm": 0.8451784595752648,
+      "learning_rate": 0.00213,
+      "loss": 4.7552,
+      "step": 710
+    },
+    {
+      "epoch": 0.00711,
+      "grad_norm": 1.0054655399528605,
+      "learning_rate": 0.002133,
+      "loss": 4.7414,
+      "step": 711
+    },
+    {
+      "epoch": 0.00712,
+      "grad_norm": 0.9031323884556907,
+      "learning_rate": 0.002136,
+      "loss": 4.7728,
+      "step": 712
+    },
+    {
+      "epoch": 0.00713,
+      "grad_norm": 0.961250906275713,
+      "learning_rate": 0.002139,
+      "loss": 4.7862,
+      "step": 713
+    },
+    {
+      "epoch": 0.00714,
+      "grad_norm": 0.9556615314074448,
+      "learning_rate": 0.002142,
+      "loss": 4.7819,
+      "step": 714
+    },
+    {
+      "epoch": 0.00715,
+      "grad_norm": 0.837203607680531,
+      "learning_rate": 0.0021449999999999998,
+      "loss": 4.7417,
+      "step": 715
+    },
+    {
+      "epoch": 0.00716,
+      "grad_norm": 0.7607986282551458,
+      "learning_rate": 0.002148,
+      "loss": 4.7359,
+      "step": 716
+    },
+    {
+      "epoch": 0.00717,
+      "grad_norm": 0.8703365352693242,
+      "learning_rate": 0.002151,
+      "loss": 4.7519,
+      "step": 717
+    },
+    {
+      "epoch": 0.00718,
+      "grad_norm": 0.8830641357048177,
+      "learning_rate": 0.002154,
+      "loss": 4.7536,
+      "step": 718
+    },
+    {
+      "epoch": 0.00719,
+      "grad_norm": 0.8090298717986324,
+      "learning_rate": 0.002157,
+      "loss": 4.7586,
+      "step": 719
+    },
+    {
+      "epoch": 0.0072,
+      "grad_norm": 0.7002439324520396,
+      "learning_rate": 0.00216,
+      "loss": 4.7466,
+      "step": 720
+    },
+    {
+      "epoch": 0.00721,
+      "grad_norm": 0.7540412799334538,
+      "learning_rate": 0.002163,
+      "loss": 4.7512,
+      "step": 721
+    },
+    {
+      "epoch": 0.00722,
+      "grad_norm": 0.7234067697970273,
+      "learning_rate": 0.002166,
+      "loss": 4.7241,
+      "step": 722
+    },
+    {
+      "epoch": 0.00723,
+      "grad_norm": 0.5796869415275953,
+      "learning_rate": 0.002169,
+      "loss": 4.73,
+      "step": 723
+    },
+    {
+      "epoch": 0.00724,
+      "grad_norm": 0.6360613090935692,
+      "learning_rate": 0.002172,
+      "loss": 4.7294,
+      "step": 724
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 0.6592111108932344,
+      "learning_rate": 0.002175,
+      "loss": 4.7232,
+      "step": 725
+    },
+    {
+      "epoch": 0.00726,
+      "grad_norm": 0.7000176967246123,
+      "learning_rate": 0.002178,
+      "loss": 4.7406,
+      "step": 726
+    },
+    {
+      "epoch": 0.00727,
+      "grad_norm": 0.6658154130327723,
+      "learning_rate": 0.0021809999999999998,
+      "loss": 4.7131,
+      "step": 727
+    },
+    {
+      "epoch": 0.00728,
+      "grad_norm": 0.500886702178687,
+      "learning_rate": 0.002184,
+      "loss": 4.7222,
+      "step": 728
+    },
+    {
+      "epoch": 0.00729,
+      "grad_norm": 0.553445989931654,
+      "learning_rate": 0.002187,
+      "loss": 4.7196,
+      "step": 729
+    },
+    {
+      "epoch": 0.0073,
+      "grad_norm": 0.5928953773304845,
+      "learning_rate": 0.00219,
+      "loss": 4.7153,
+      "step": 730
+    },
+    {
+      "epoch": 0.00731,
+      "grad_norm": 0.5280339609019513,
+      "learning_rate": 0.002193,
+      "loss": 4.7069,
+      "step": 731
+    },
+    {
+      "epoch": 0.00732,
+      "grad_norm": 0.4601497488067425,
+      "learning_rate": 0.002196,
+      "loss": 4.7146,
+      "step": 732
+    },
+    {
+      "epoch": 0.00733,
+      "grad_norm": 0.4831437067076967,
+      "learning_rate": 0.002199,
+      "loss": 4.6865,
+      "step": 733
+    },
+    {
+      "epoch": 0.00734,
+      "grad_norm": 0.48957731222976764,
+      "learning_rate": 0.002202,
+      "loss": 4.7176,
+      "step": 734
+    },
+    {
+      "epoch": 0.00735,
+      "grad_norm": 0.5029506248084066,
+      "learning_rate": 0.002205,
+      "loss": 4.7231,
+      "step": 735
+    },
+    {
+      "epoch": 0.00736,
+      "grad_norm": 0.5300436466729722,
+      "learning_rate": 0.002208,
+      "loss": 4.7045,
+      "step": 736
+    },
+    {
+      "epoch": 0.00737,
+      "grad_norm": 0.5354857814520255,
+      "learning_rate": 0.002211,
+      "loss": 4.6701,
+      "step": 737
+    },
+    {
+      "epoch": 0.00738,
+      "grad_norm": 0.6855959285026678,
+      "learning_rate": 0.002214,
+      "loss": 4.6857,
+      "step": 738
+    },
+    {
+      "epoch": 0.00739,
+      "grad_norm": 0.7193696222416395,
+      "learning_rate": 0.0022170000000000002,
+      "loss": 4.6773,
+      "step": 739
+    },
+    {
+      "epoch": 0.0074,
+      "grad_norm": 0.7000843676029133,
+      "learning_rate": 0.00222,
+      "loss": 4.686,
+      "step": 740
+    },
+    {
+      "epoch": 0.00741,
+      "grad_norm": 0.8262482718120322,
+      "learning_rate": 0.002223,
+      "loss": 4.6648,
+      "step": 741
+    },
+    {
+      "epoch": 0.00742,
+      "grad_norm": 0.8068053565529363,
+      "learning_rate": 0.002226,
+      "loss": 4.71,
+      "step": 742
+    },
+    {
+      "epoch": 0.00743,
+      "grad_norm": 0.7713935209386231,
+      "learning_rate": 0.002229,
+      "loss": 4.6667,
+      "step": 743
+    },
+    {
+      "epoch": 0.00744,
+      "grad_norm": 0.6806090978340125,
+      "learning_rate": 0.002232,
+      "loss": 4.668,
+      "step": 744
+    },
+    {
+      "epoch": 0.00745,
+      "grad_norm": 0.8150134425373752,
+      "learning_rate": 0.002235,
+      "loss": 4.6906,
+      "step": 745
+    },
+    {
+      "epoch": 0.00746,
+      "grad_norm": 0.9083405480081935,
+      "learning_rate": 0.002238,
+      "loss": 4.6909,
+      "step": 746
+    },
+    {
+      "epoch": 0.00747,
+      "grad_norm": 1.1312224699232258,
+      "learning_rate": 0.002241,
+      "loss": 4.6956,
+      "step": 747
+    },
+    {
+      "epoch": 0.00748,
+      "grad_norm": 0.8174601291810354,
+      "learning_rate": 0.002244,
+      "loss": 4.6973,
+      "step": 748
+    },
+    {
+      "epoch": 0.00749,
+      "grad_norm": 0.8423282795209855,
+      "learning_rate": 0.002247,
+      "loss": 4.6802,
+      "step": 749
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 0.8679250685858194,
+      "learning_rate": 0.0022500000000000003,
+      "loss": 4.7268,
+      "step": 750
+    },
+    {
+      "epoch": 0.00751,
+      "grad_norm": 1.0939871221004271,
+      "learning_rate": 0.0022530000000000002,
+      "loss": 4.7337,
+      "step": 751
+    },
+    {
+      "epoch": 0.00752,
+      "grad_norm": 0.9886618564303525,
+      "learning_rate": 0.002256,
+      "loss": 4.6822,
+      "step": 752
+    },
+    {
+      "epoch": 0.00753,
+      "grad_norm": 0.9258452428585717,
+      "learning_rate": 0.002259,
+      "loss": 4.7192,
+      "step": 753
+    },
+    {
+      "epoch": 0.00754,
+      "grad_norm": 0.8790583060068752,
+      "learning_rate": 0.002262,
+      "loss": 4.7345,
+      "step": 754
+    },
+    {
+      "epoch": 0.00755,
+      "grad_norm": 0.7727162685258321,
+      "learning_rate": 0.002265,
+      "loss": 4.6919,
+      "step": 755
+    },
+    {
+      "epoch": 0.00756,
+      "grad_norm": 0.8048369196552551,
+      "learning_rate": 0.002268,
+      "loss": 4.6801,
+      "step": 756
+    },
+    {
+      "epoch": 0.00757,
+      "grad_norm": 0.7307749692176225,
+      "learning_rate": 0.002271,
+      "loss": 4.6902,
+      "step": 757
+    },
+    {
+      "epoch": 0.00758,
+      "grad_norm": 0.7628918458541498,
+      "learning_rate": 0.002274,
+      "loss": 4.6882,
+      "step": 758
+    },
+    {
+      "epoch": 0.00759,
+      "grad_norm": 0.6811469027490352,
+      "learning_rate": 0.002277,
+      "loss": 4.693,
+      "step": 759
+    },
+    {
+      "epoch": 0.0076,
+      "grad_norm": 0.5530412513377371,
+      "learning_rate": 0.00228,
+      "loss": 4.6735,
+      "step": 760
+    },
+    {
+      "epoch": 0.00761,
+      "grad_norm": 0.5221449888101848,
+      "learning_rate": 0.002283,
+      "loss": 4.6405,
+      "step": 761
+    },
+    {
+      "epoch": 0.00762,
+      "grad_norm": 0.5687089394846262,
+      "learning_rate": 0.0022860000000000003,
+      "loss": 4.6208,
+      "step": 762
+    },
+    {
+      "epoch": 0.00763,
+      "grad_norm": 0.5814285508645848,
+      "learning_rate": 0.0022890000000000002,
+      "loss": 4.6718,
+      "step": 763
+    },
+    {
+      "epoch": 0.00764,
+      "grad_norm": 0.6388540284979518,
+      "learning_rate": 0.002292,
+      "loss": 4.683,
+      "step": 764
+    },
+    {
+      "epoch": 0.00765,
+      "grad_norm": 0.7278589788698283,
+      "learning_rate": 0.002295,
+      "loss": 4.6752,
+      "step": 765
+    },
+    {
+      "epoch": 0.00766,
+      "grad_norm": 0.7050667087459527,
+      "learning_rate": 0.002298,
+      "loss": 4.6715,
+      "step": 766
+    },
+    {
+      "epoch": 0.00767,
+      "grad_norm": 0.6023307971425665,
+      "learning_rate": 0.002301,
+      "loss": 4.6623,
+      "step": 767
+    },
+    {
+      "epoch": 0.00768,
+      "grad_norm": 0.7162182495110988,
+      "learning_rate": 0.002304,
+      "loss": 4.6687,
+      "step": 768
+    },
+    {
+      "epoch": 0.00769,
+      "grad_norm": 0.8754398583131885,
+      "learning_rate": 0.002307,
+      "loss": 4.6855,
+      "step": 769
+    },
+    {
+      "epoch": 0.0077,
+      "grad_norm": 0.9282950433205286,
+      "learning_rate": 0.00231,
+      "loss": 4.656,
+      "step": 770
+    },
+    {
+      "epoch": 0.00771,
+      "grad_norm": 0.8826983762517153,
+      "learning_rate": 0.002313,
+      "loss": 4.6927,
+      "step": 771
+    },
+    {
+      "epoch": 0.00772,
+      "grad_norm": 0.7955428299875815,
+      "learning_rate": 0.002316,
+      "loss": 4.6752,
+      "step": 772
+    },
+    {
+      "epoch": 0.00773,
+      "grad_norm": 0.6879017191297421,
+      "learning_rate": 0.0023190000000000003,
+      "loss": 4.6732,
+      "step": 773
+    },
+    {
+      "epoch": 0.00774,
+      "grad_norm": 0.5805904836674535,
+      "learning_rate": 0.0023220000000000003,
+      "loss": 4.6842,
+      "step": 774
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 0.5872309146101224,
+      "learning_rate": 0.0023250000000000002,
+      "loss": 4.6741,
+      "step": 775
+    },
+    {
+      "epoch": 0.00776,
+      "grad_norm": 0.6663800275805344,
+      "learning_rate": 0.002328,
+      "loss": 4.6762,
+      "step": 776
+    },
+    {
+      "epoch": 0.00777,
+      "grad_norm": 0.6475349218207965,
+      "learning_rate": 0.002331,
+      "loss": 4.6499,
+      "step": 777
+    },
+    {
+      "epoch": 0.00778,
+      "grad_norm": 0.6498373909650491,
+      "learning_rate": 0.002334,
+      "loss": 4.6643,
+      "step": 778
+    },
+    {
+      "epoch": 0.00779,
+      "grad_norm": 0.6357690952406082,
+      "learning_rate": 0.002337,
+      "loss": 4.6181,
+      "step": 779
+    },
+    {
+      "epoch": 0.0078,
+      "grad_norm": 0.6241849680287349,
+      "learning_rate": 0.00234,
+      "loss": 4.6419,
+      "step": 780
+    },
+    {
+      "epoch": 0.00781,
+      "grad_norm": 0.6808062655697796,
+      "learning_rate": 0.002343,
+      "loss": 4.6451,
+      "step": 781
+    },
+    {
+      "epoch": 0.00782,
+      "grad_norm": 0.7065267585011001,
+      "learning_rate": 0.002346,
+      "loss": 4.6436,
+      "step": 782
+    },
+    {
+      "epoch": 0.00783,
+      "grad_norm": 0.6381701638777938,
+      "learning_rate": 0.002349,
+      "loss": 4.6242,
+      "step": 783
+    },
+    {
+      "epoch": 0.00784,
+      "grad_norm": 0.647841330234918,
+      "learning_rate": 0.002352,
+      "loss": 4.6355,
+      "step": 784
+    },
+    {
+      "epoch": 0.00785,
+      "grad_norm": 0.60562299847623,
+      "learning_rate": 0.0023550000000000003,
+      "loss": 4.6345,
+      "step": 785
+    },
+    {
+      "epoch": 0.00786,
+      "grad_norm": 0.5148036449557751,
+      "learning_rate": 0.0023580000000000003,
+      "loss": 4.628,
+      "step": 786
+    },
+    {
+      "epoch": 0.00787,
+      "grad_norm": 0.7217903580979332,
+      "learning_rate": 0.0023610000000000003,
+      "loss": 4.641,
+      "step": 787
+    },
+    {
+      "epoch": 0.00788,
+      "grad_norm": 0.8515178253715305,
+      "learning_rate": 0.002364,
+      "loss": 4.6466,
+      "step": 788
+    },
+    {
+      "epoch": 0.00789,
+      "grad_norm": 0.8831947515557061,
+      "learning_rate": 0.002367,
+      "loss": 4.6753,
+      "step": 789
+    },
+    {
+      "epoch": 0.0079,
+      "grad_norm": 0.9816312249435458,
+      "learning_rate": 0.00237,
+      "loss": 4.6574,
+      "step": 790
+    },
+    {
+      "epoch": 0.00791,
+      "grad_norm": 0.8257428278422617,
+      "learning_rate": 0.002373,
+      "loss": 4.6081,
+      "step": 791
+    },
+    {
+      "epoch": 0.00792,
+      "grad_norm": 0.6889392893349975,
+      "learning_rate": 0.002376,
+      "loss": 4.6372,
+      "step": 792
+    },
+    {
+      "epoch": 0.00793,
+      "grad_norm": 0.5470340397913868,
+      "learning_rate": 0.002379,
+      "loss": 4.6272,
+      "step": 793
+    },
+    {
+      "epoch": 0.00794,
+      "grad_norm": 0.5333909794818302,
+      "learning_rate": 0.002382,
+      "loss": 4.6359,
+      "step": 794
+    },
+    {
+      "epoch": 0.00795,
+      "grad_norm": 0.47330116805918854,
+      "learning_rate": 0.002385,
+      "loss": 4.6105,
+      "step": 795
+    },
+    {
+      "epoch": 0.00796,
+      "grad_norm": 0.46996625387544017,
+      "learning_rate": 0.0023880000000000004,
+      "loss": 4.6049,
+      "step": 796
+    },
+    {
+      "epoch": 0.00797,
+      "grad_norm": 0.4648603328586337,
+      "learning_rate": 0.0023910000000000003,
+      "loss": 4.6461,
+      "step": 797
+    },
+    {
+      "epoch": 0.00798,
+      "grad_norm": 0.48147045136320854,
+      "learning_rate": 0.0023940000000000003,
+      "loss": 4.6125,
+      "step": 798
+    },
+    {
+      "epoch": 0.00799,
+      "grad_norm": 0.5520002928611956,
+      "learning_rate": 0.0023970000000000003,
+      "loss": 4.6461,
+      "step": 799
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 0.49520453644625784,
+      "learning_rate": 0.0024000000000000002,
+      "loss": 4.5958,
+      "step": 800
+    },
+    {
+      "epoch": 0.00801,
+      "grad_norm": 0.4617883447215667,
+      "learning_rate": 0.002403,
+      "loss": 4.6244,
+      "step": 801
+    },
+    {
+      "epoch": 0.00802,
+      "grad_norm": 0.6171622349156032,
+      "learning_rate": 0.002406,
+      "loss": 4.6206,
+      "step": 802
+    },
+    {
+      "epoch": 0.00803,
+      "grad_norm": 0.7819651777797723,
+      "learning_rate": 0.002409,
+      "loss": 4.605,
+      "step": 803
+    },
+    {
+      "epoch": 0.00804,
+      "grad_norm": 0.8072878053268496,
+      "learning_rate": 0.002412,
+      "loss": 4.6195,
+      "step": 804
+    },
+    {
+      "epoch": 0.00805,
+      "grad_norm": 0.6878812715646375,
+      "learning_rate": 0.002415,
+      "loss": 4.6081,
+      "step": 805
+    },
+    {
+      "epoch": 0.00806,
+      "grad_norm": 0.7628005366591507,
+      "learning_rate": 0.002418,
+      "loss": 4.6308,
+      "step": 806
+    },
+    {
+      "epoch": 0.00807,
+      "grad_norm": 0.7958527408861041,
+      "learning_rate": 0.0024210000000000004,
+      "loss": 4.6228,
+      "step": 807
+    },
+    {
+      "epoch": 0.00808,
+      "grad_norm": 0.7899389450719584,
+      "learning_rate": 0.0024240000000000004,
+      "loss": 4.6083,
+      "step": 808
+    },
+    {
+      "epoch": 0.00809,
+      "grad_norm": 1.193539574978161,
+      "learning_rate": 0.0024270000000000003,
+      "loss": 4.6201,
+      "step": 809
+    },
+    {
+      "epoch": 0.0081,
+      "grad_norm": 1.0067442084818319,
+      "learning_rate": 0.0024300000000000003,
+      "loss": 4.6554,
+      "step": 810
+    },
+    {
+      "epoch": 0.00811,
+      "grad_norm": 1.0302946243785736,
+      "learning_rate": 0.0024330000000000003,
+      "loss": 4.6338,
+      "step": 811
+    },
+    {
+      "epoch": 0.00812,
+      "grad_norm": 0.7966936317025509,
+      "learning_rate": 0.0024360000000000002,
+      "loss": 4.6044,
+      "step": 812
+    },
+    {
+      "epoch": 0.00813,
+      "grad_norm": 0.7843512654561826,
+      "learning_rate": 0.0024389999999999998,
+      "loss": 4.6449,
+      "step": 813
+    },
+    {
+      "epoch": 0.00814,
+      "grad_norm": 0.7981724618057067,
+      "learning_rate": 0.0024419999999999997,
+      "loss": 4.6482,
+      "step": 814
+    },
+    {
+      "epoch": 0.00815,
+      "grad_norm": 0.8382113380987876,
+      "learning_rate": 0.0024449999999999997,
+      "loss": 4.6588,
+      "step": 815
+    },
+    {
+      "epoch": 0.00816,
+      "grad_norm": 0.8204581966398267,
+      "learning_rate": 0.002448,
+      "loss": 4.6311,
+      "step": 816
+    },
+    {
+      "epoch": 0.00817,
+      "grad_norm": 1.02336938073518,
+      "learning_rate": 0.002451,
+      "loss": 4.6583,
+      "step": 817
+    },
+    {
+      "epoch": 0.00818,
+      "grad_norm": 0.815478540285485,
+      "learning_rate": 0.002454,
+      "loss": 4.6309,
+      "step": 818
+    },
+    {
+      "epoch": 0.00819,
+      "grad_norm": 0.9108711148393207,
+      "learning_rate": 0.002457,
+      "loss": 4.6483,
+      "step": 819
+    },
+    {
+      "epoch": 0.0082,
+      "grad_norm": 0.8773364113378127,
+      "learning_rate": 0.00246,
+      "loss": 4.6373,
+      "step": 820
+    },
+    {
+      "epoch": 0.00821,
+      "grad_norm": 0.8093045902813614,
+      "learning_rate": 0.002463,
+      "loss": 4.6126,
+      "step": 821
+    },
+    {
+      "epoch": 0.00822,
+      "grad_norm": 0.8775561270633004,
+      "learning_rate": 0.002466,
+      "loss": 4.6164,
+      "step": 822
+    },
+    {
+      "epoch": 0.00823,
+      "grad_norm": 0.8916321669966187,
+      "learning_rate": 0.002469,
+      "loss": 4.6298,
+      "step": 823
+    },
+    {
+      "epoch": 0.00824,
+      "grad_norm": 0.7939051533904264,
+      "learning_rate": 0.002472,
+      "loss": 4.6511,
+      "step": 824
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 0.9509206049767348,
+      "learning_rate": 0.0024749999999999998,
+      "loss": 4.6625,
+      "step": 825
+    },
+    {
+      "epoch": 0.00826,
+      "grad_norm": 1.0914612729506281,
+      "learning_rate": 0.0024779999999999997,
+      "loss": 4.6356,
+      "step": 826
+    },
+    {
+      "epoch": 0.00827,
+      "grad_norm": 1.0440371247088225,
+      "learning_rate": 0.002481,
+      "loss": 4.6709,
+      "step": 827
+    },
+    {
+      "epoch": 0.00828,
+      "grad_norm": 1.0292672719671891,
+      "learning_rate": 0.002484,
+      "loss": 4.6644,
+      "step": 828
+    },
+    {
+      "epoch": 0.00829,
+      "grad_norm": 1.002549149572049,
+      "learning_rate": 0.002487,
+      "loss": 4.6608,
+      "step": 829
+    },
+    {
+      "epoch": 0.0083,
+      "grad_norm": 0.8229445260626227,
+      "learning_rate": 0.00249,
+      "loss": 4.6642,
+      "step": 830
+    },
+    {
+      "epoch": 0.00831,
+      "grad_norm": 0.7223475739297199,
+      "learning_rate": 0.002493,
+      "loss": 4.6379,
+      "step": 831
+    },
+    {
+      "epoch": 0.00832,
+      "grad_norm": 0.6422365780456449,
+      "learning_rate": 0.002496,
+      "loss": 4.6349,
+      "step": 832
+    },
+    {
+      "epoch": 0.00833,
+      "grad_norm": 0.6433276699815419,
+      "learning_rate": 0.002499,
+      "loss": 4.626,
+      "step": 833
+    },
+    {
+      "epoch": 0.00834,
+      "grad_norm": 0.6136587911860008,
+      "learning_rate": 0.002502,
+      "loss": 4.6284,
+      "step": 834
+    },
+    {
+      "epoch": 0.00835,
+      "grad_norm": 0.5892258230535582,
+      "learning_rate": 0.002505,
+      "loss": 4.617,
+      "step": 835
+    },
+    {
+      "epoch": 0.00836,
+      "grad_norm": 0.5407454758774727,
+      "learning_rate": 0.002508,
+      "loss": 4.6416,
+      "step": 836
+    },
+    {
+      "epoch": 0.00837,
+      "grad_norm": 0.5840603338652609,
+      "learning_rate": 0.0025109999999999998,
+      "loss": 4.582,
+      "step": 837
+    },
+    {
+      "epoch": 0.00838,
+      "grad_norm": 0.5192725472759927,
+      "learning_rate": 0.0025139999999999997,
+      "loss": 4.6102,
+      "step": 838
+    },
+    {
+      "epoch": 0.00839,
+      "grad_norm": 0.5064380785759203,
+      "learning_rate": 0.002517,
+      "loss": 4.6034,
+      "step": 839
+    },
+    {
+      "epoch": 0.0084,
+      "grad_norm": 0.48476724687493267,
+      "learning_rate": 0.00252,
+      "loss": 4.5854,
+      "step": 840
+    },
+    {
+      "epoch": 0.00841,
+      "grad_norm": 0.44496684540968734,
+      "learning_rate": 0.002523,
+      "loss": 4.5786,
+      "step": 841
+    },
+    {
+      "epoch": 0.00842,
+      "grad_norm": 0.42609007840223895,
+      "learning_rate": 0.002526,
+      "loss": 4.5771,
+      "step": 842
+    },
+    {
+      "epoch": 0.00843,
+      "grad_norm": 0.4703393687667864,
+      "learning_rate": 0.002529,
+      "loss": 4.5751,
+      "step": 843
+    },
+    {
+      "epoch": 0.00844,
+      "grad_norm": 0.4976619641323943,
+      "learning_rate": 0.002532,
+      "loss": 4.5956,
+      "step": 844
+    },
+    {
+      "epoch": 0.00845,
+      "grad_norm": 0.530881004052944,
+      "learning_rate": 0.002535,
+      "loss": 4.5625,
+      "step": 845
+    },
+    {
+      "epoch": 0.00846,
+      "grad_norm": 0.5069253535552343,
+      "learning_rate": 0.002538,
+      "loss": 4.584,
+      "step": 846
+    },
+    {
+      "epoch": 0.00847,
+      "grad_norm": 0.4900054637856495,
+      "learning_rate": 0.002541,
+      "loss": 4.5637,
+      "step": 847
+    },
+    {
+      "epoch": 0.00848,
+      "grad_norm": 0.5361511355183629,
+      "learning_rate": 0.002544,
+      "loss": 4.5693,
+      "step": 848
+    },
+    {
+      "epoch": 0.00849,
+      "grad_norm": 0.6067359238432654,
+      "learning_rate": 0.002547,
+      "loss": 4.5644,
+      "step": 849
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 0.5519192846207763,
+      "learning_rate": 0.00255,
+      "loss": 4.573,
+      "step": 850
+    },
+    {
+      "epoch": 0.00851,
+      "grad_norm": 0.46694877598438245,
+      "learning_rate": 0.002553,
+      "loss": 4.5875,
+      "step": 851
+    },
+    {
+      "epoch": 0.00852,
+      "grad_norm": 0.477565098915178,
+      "learning_rate": 0.002556,
+      "loss": 4.5765,
+      "step": 852
+    },
+    {
+      "epoch": 0.00853,
+      "grad_norm": 0.5020213435824815,
+      "learning_rate": 0.002559,
+      "loss": 4.5556,
+      "step": 853
+    },
+    {
+      "epoch": 0.00854,
+      "grad_norm": 0.5171409161048013,
+      "learning_rate": 0.002562,
+      "loss": 4.5495,
+      "step": 854
+    },
+    {
+      "epoch": 0.00855,
+      "grad_norm": 0.46627459343076927,
+      "learning_rate": 0.002565,
+      "loss": 4.5252,
+      "step": 855
+    },
+    {
+      "epoch": 0.00856,
+      "grad_norm": 0.5139521756940325,
+      "learning_rate": 0.002568,
+      "loss": 4.5623,
+      "step": 856
+    },
+    {
+      "epoch": 0.00857,
+      "grad_norm": 0.6011403998041547,
+      "learning_rate": 0.002571,
+      "loss": 4.5577,
+      "step": 857
+    },
+    {
+      "epoch": 0.00858,
+      "grad_norm": 0.554768384377006,
+      "learning_rate": 0.002574,
+      "loss": 4.5487,
+      "step": 858
+    },
+    {
+      "epoch": 0.00859,
+      "grad_norm": 0.539858542755145,
+      "learning_rate": 0.002577,
+      "loss": 4.5383,
+      "step": 859
+    },
+    {
+      "epoch": 0.0086,
+      "grad_norm": 0.586599307397693,
+      "learning_rate": 0.00258,
+      "loss": 4.571,
+      "step": 860
+    },
+    {
+      "epoch": 0.00861,
+      "grad_norm": 0.7154666400015554,
+      "learning_rate": 0.0025830000000000002,
+      "loss": 4.5928,
+      "step": 861
+    },
+    {
+      "epoch": 0.00862,
+      "grad_norm": 0.7345971792604707,
+      "learning_rate": 0.002586,
+      "loss": 4.5402,
+      "step": 862
+    },
+    {
+      "epoch": 0.00863,
+      "grad_norm": 0.7491760821516434,
+      "learning_rate": 0.002589,
+      "loss": 4.5765,
+      "step": 863
+    },
+    {
+      "epoch": 0.00864,
+      "grad_norm": 0.9255705361922033,
+      "learning_rate": 0.002592,
+      "loss": 4.5322,
+      "step": 864
+    },
+    {
+      "epoch": 0.00865,
+      "grad_norm": 0.9964986146275199,
+      "learning_rate": 0.002595,
+      "loss": 4.5664,
+      "step": 865
+    },
+    {
+      "epoch": 0.00866,
+      "grad_norm": 0.7618488122087141,
+      "learning_rate": 0.002598,
+      "loss": 4.5836,
+      "step": 866
+    },
+    {
+      "epoch": 0.00867,
+      "grad_norm": 0.8524780083566116,
+      "learning_rate": 0.002601,
+      "loss": 4.5742,
+      "step": 867
+    },
+    {
+      "epoch": 0.00868,
+      "grad_norm": 0.9692981312410378,
+      "learning_rate": 0.002604,
+      "loss": 4.5808,
+      "step": 868
+    },
+    {
+      "epoch": 0.00869,
+      "grad_norm": 1.1822252043975705,
+      "learning_rate": 0.002607,
+      "loss": 4.6073,
+      "step": 869
+    },
+    {
+      "epoch": 0.0087,
+      "grad_norm": 0.9057663759386707,
+      "learning_rate": 0.00261,
+      "loss": 4.5844,
+      "step": 870
+    },
+    {
+      "epoch": 0.00871,
+      "grad_norm": 0.9457338978675252,
+      "learning_rate": 0.002613,
+      "loss": 4.6115,
+      "step": 871
+    },
+    {
+      "epoch": 0.00872,
+      "grad_norm": 0.9845348105394848,
+      "learning_rate": 0.002616,
+      "loss": 4.5975,
+      "step": 872
+    },
+    {
+      "epoch": 0.00873,
+      "grad_norm": 0.8202179076205192,
+      "learning_rate": 0.0026190000000000002,
+      "loss": 4.5967,
+      "step": 873
+    },
+    {
+      "epoch": 0.00874,
+      "grad_norm": 0.6587988147688274,
+      "learning_rate": 0.002622,
+      "loss": 4.6142,
+      "step": 874
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 0.6312495265838277,
+      "learning_rate": 0.002625,
+      "loss": 4.5549,
+      "step": 875
+    },
+    {
+      "epoch": 0.00876,
+      "grad_norm": 0.6646817876274769,
+      "learning_rate": 0.002628,
+      "loss": 4.5838,
+      "step": 876
+    },
+    {
+      "epoch": 0.00877,
+      "grad_norm": 0.632354886157607,
+      "learning_rate": 0.002631,
+      "loss": 4.6011,
+      "step": 877
+    },
+    {
+      "epoch": 0.00878,
+      "grad_norm": 0.5630676163174572,
+      "learning_rate": 0.002634,
+      "loss": 4.5288,
+      "step": 878
+    },
+    {
+      "epoch": 0.00879,
+      "grad_norm": 0.4918531988998375,
+      "learning_rate": 0.002637,
+      "loss": 4.5559,
+      "step": 879
+    },
+    {
+      "epoch": 0.0088,
+      "grad_norm": 0.42476181684324305,
+      "learning_rate": 0.00264,
+      "loss": 4.5634,
+      "step": 880
+    },
+    {
+      "epoch": 0.00881,
+      "grad_norm": 0.4573573466722849,
+      "learning_rate": 0.002643,
+      "loss": 4.5644,
+      "step": 881
+    },
+    {
+      "epoch": 0.00882,
+      "grad_norm": 0.5881448674370812,
+      "learning_rate": 0.002646,
+      "loss": 4.5659,
+      "step": 882
+    },
+    {
+      "epoch": 0.00883,
+      "grad_norm": 0.7764456560266775,
+      "learning_rate": 0.002649,
+      "loss": 4.5645,
+      "step": 883
+    },
+    {
+      "epoch": 0.00884,
+      "grad_norm": 0.9651176541039754,
+      "learning_rate": 0.0026520000000000003,
+      "loss": 4.5629,
+      "step": 884
+    },
+    {
+      "epoch": 0.00885,
+      "grad_norm": 1.0453630458113787,
+      "learning_rate": 0.0026550000000000002,
+      "loss": 4.5675,
+      "step": 885
+    },
+    {
+      "epoch": 0.00886,
+      "grad_norm": 0.8249295551343052,
+      "learning_rate": 0.002658,
+      "loss": 4.5811,
+      "step": 886
+    },
+    {
+      "epoch": 0.00887,
+      "grad_norm": 0.6632170677276661,
+      "learning_rate": 0.002661,
+      "loss": 4.5643,
+      "step": 887
+    },
+    {
+      "epoch": 0.00888,
+      "grad_norm": 0.7818922566742896,
+      "learning_rate": 0.002664,
+      "loss": 4.5709,
+      "step": 888
+    },
+    {
+      "epoch": 0.00889,
+      "grad_norm": 0.8131057291041344,
+      "learning_rate": 0.002667,
+      "loss": 4.5276,
+      "step": 889
+    },
+    {
+      "epoch": 0.0089,
+      "grad_norm": 0.7364786352062309,
+      "learning_rate": 0.00267,
+      "loss": 4.5735,
+      "step": 890
+    },
+    {
+      "epoch": 0.00891,
+      "grad_norm": 0.6174568923006037,
+      "learning_rate": 0.002673,
+      "loss": 4.5538,
+      "step": 891
+    },
+    {
+      "epoch": 0.00892,
+      "grad_norm": 0.6060396654742667,
+      "learning_rate": 0.002676,
+      "loss": 4.5678,
+      "step": 892
+    },
+    {
+      "epoch": 0.00893,
+      "grad_norm": 0.6503337239639668,
+      "learning_rate": 0.002679,
+      "loss": 4.5453,
+      "step": 893
+    },
+    {
+      "epoch": 0.00894,
+      "grad_norm": 0.6599395002026207,
+      "learning_rate": 0.002682,
+      "loss": 4.5291,
+      "step": 894
+    },
+    {
+      "epoch": 0.00895,
+      "grad_norm": 0.5989877186645693,
+      "learning_rate": 0.0026850000000000003,
+      "loss": 4.5412,
+      "step": 895
+    },
+    {
+      "epoch": 0.00896,
+      "grad_norm": 0.5286031214975206,
+      "learning_rate": 0.0026880000000000003,
+      "loss": 4.5273,
+      "step": 896
+    },
+    {
+      "epoch": 0.00897,
+      "grad_norm": 0.6246596167729576,
+      "learning_rate": 0.0026910000000000002,
+      "loss": 4.5504,
+      "step": 897
+    },
+    {
+      "epoch": 0.00898,
+      "grad_norm": 0.6886920087577523,
+      "learning_rate": 0.002694,
+      "loss": 4.5437,
+      "step": 898
+    },
+    {
+      "epoch": 0.00899,
+      "grad_norm": 0.7603324493631337,
+      "learning_rate": 0.002697,
+      "loss": 4.5543,
+      "step": 899
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 0.7773743953648492,
+      "learning_rate": 0.0027,
+      "loss": 4.5794,
+      "step": 900
+    },
+    {
+      "epoch": 0.00901,
+      "grad_norm": 0.683256197441996,
+      "learning_rate": 0.002703,
+      "loss": 4.5307,
+      "step": 901
+    },
+    {
+      "epoch": 0.00902,
+      "grad_norm": 0.5681357763332335,
+      "learning_rate": 0.002706,
+      "loss": 4.5356,
+      "step": 902
+    },
+    {
+      "epoch": 0.00903,
+      "grad_norm": 0.5420591540444755,
+      "learning_rate": 0.002709,
+      "loss": 4.5338,
+      "step": 903
+    },
+    {
+      "epoch": 0.00904,
+      "grad_norm": 0.5224631659490503,
+      "learning_rate": 0.002712,
+      "loss": 4.5093,
+      "step": 904
+    },
+    {
+      "epoch": 0.00905,
+      "grad_norm": 0.5026034590467293,
+      "learning_rate": 0.002715,
+      "loss": 4.5252,
+      "step": 905
+    },
+    {
+      "epoch": 0.00906,
+      "grad_norm": 0.5177890071237494,
+      "learning_rate": 0.002718,
+      "loss": 4.5378,
+      "step": 906
+    },
+    {
+      "epoch": 0.00907,
+      "grad_norm": 0.5764689015080159,
+      "learning_rate": 0.0027210000000000003,
+      "loss": 4.5536,
+      "step": 907
+    },
+    {
+      "epoch": 0.00908,
+      "grad_norm": 0.6259624722487185,
+      "learning_rate": 0.0027240000000000003,
+      "loss": 4.5265,
+      "step": 908
+    },
+    {
+      "epoch": 0.00909,
+      "grad_norm": 0.670091172363038,
+      "learning_rate": 0.0027270000000000003,
+      "loss": 4.5481,
+      "step": 909
+    },
+    {
+      "epoch": 0.0091,
+      "grad_norm": 0.7211417475777565,
+      "learning_rate": 0.0027300000000000002,
+      "loss": 4.555,
+      "step": 910
+    },
+    {
+      "epoch": 0.00911,
+      "grad_norm": 0.6734826041799787,
+      "learning_rate": 0.002733,
+      "loss": 4.5118,
+      "step": 911
+    },
+    {
+      "epoch": 0.00912,
+      "grad_norm": 0.5721394003951694,
+      "learning_rate": 0.002736,
+      "loss": 4.5134,
+      "step": 912
+    },
+    {
+      "epoch": 0.00913,
+      "grad_norm": 0.5576363751689392,
+      "learning_rate": 0.002739,
+      "loss": 4.5062,
+      "step": 913
+    },
+    {
+      "epoch": 0.00914,
+      "grad_norm": 0.659556953854551,
+      "learning_rate": 0.002742,
+      "loss": 4.5426,
+      "step": 914
+    },
+    {
+      "epoch": 0.00915,
+      "grad_norm": 0.7056386423863461,
+      "learning_rate": 0.002745,
+      "loss": 4.5041,
+      "step": 915
+    },
+    {
+      "epoch": 0.00916,
+      "grad_norm": 0.6615396800381155,
+      "learning_rate": 0.002748,
+      "loss": 4.4829,
+      "step": 916
+    },
+    {
+      "epoch": 0.00917,
+      "grad_norm": 0.6203895946828626,
+      "learning_rate": 0.002751,
+      "loss": 4.4952,
+      "step": 917
+    },
+    {
+      "epoch": 0.00918,
+      "grad_norm": 0.6455523293432982,
+      "learning_rate": 0.0027540000000000004,
+      "loss": 4.5159,
+      "step": 918
+    },
+    {
+      "epoch": 0.00919,
+      "grad_norm": 0.6153975854015812,
+      "learning_rate": 0.0027570000000000003,
+      "loss": 4.4956,
+      "step": 919
+    },
+    {
+      "epoch": 0.0092,
+      "grad_norm": 0.6747074236078813,
+      "learning_rate": 0.0027600000000000003,
+      "loss": 4.516,
+      "step": 920
+    },
+    {
+      "epoch": 0.00921,
+      "grad_norm": 0.7525938258933302,
+      "learning_rate": 0.0027630000000000003,
+      "loss": 4.5035,
+      "step": 921
+    },
+    {
+      "epoch": 0.00922,
+      "grad_norm": 0.6613679668687953,
+      "learning_rate": 0.0027660000000000002,
+      "loss": 4.5161,
+      "step": 922
+    },
+    {
+      "epoch": 0.00923,
+      "grad_norm": 0.657751627761755,
+      "learning_rate": 0.002769,
+      "loss": 4.4882,
+      "step": 923
+    },
+    {
+      "epoch": 0.00924,
+      "grad_norm": 0.6747361910557791,
+      "learning_rate": 0.002772,
+      "loss": 4.515,
+      "step": 924
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 0.6239646474738244,
+      "learning_rate": 0.002775,
+      "loss": 4.4929,
+      "step": 925
+    },
+    {
+      "epoch": 0.00926,
+      "grad_norm": 0.599067117804374,
+      "learning_rate": 0.002778,
+      "loss": 4.5097,
+      "step": 926
+    },
+    {
+      "epoch": 0.00927,
+      "grad_norm": 0.5594951339370652,
+      "learning_rate": 0.002781,
+      "loss": 4.4719,
+      "step": 927
+    },
+    {
+      "epoch": 0.00928,
+      "grad_norm": 0.6063845847981192,
+      "learning_rate": 0.002784,
+      "loss": 4.518,
+      "step": 928
+    },
+    {
+      "epoch": 0.00929,
+      "grad_norm": 0.6363503981385901,
+      "learning_rate": 0.0027870000000000004,
+      "loss": 4.5031,
+      "step": 929
+    },
+    {
+      "epoch": 0.0093,
+      "grad_norm": 0.7396258221206569,
+      "learning_rate": 0.0027900000000000004,
+      "loss": 4.4944,
+      "step": 930
+    },
+    {
+      "epoch": 0.00931,
+      "grad_norm": 0.8942550404249334,
+      "learning_rate": 0.0027930000000000003,
+      "loss": 4.517,
+      "step": 931
+    },
+    {
+      "epoch": 0.00932,
+      "grad_norm": 1.0354660733966428,
+      "learning_rate": 0.0027960000000000003,
+      "loss": 4.5402,
+      "step": 932
+    },
+    {
+      "epoch": 0.00933,
+      "grad_norm": 1.180367237398422,
+      "learning_rate": 0.0027990000000000003,
+      "loss": 4.525,
+      "step": 933
+    },
+    {
+      "epoch": 0.00934,
+      "grad_norm": 0.9502642927196222,
+      "learning_rate": 0.0028020000000000002,
+      "loss": 4.5541,
+      "step": 934
+    },
+    {
+      "epoch": 0.00935,
+      "grad_norm": 0.8858808588378486,
+      "learning_rate": 0.002805,
+      "loss": 4.5601,
+      "step": 935
+    },
+    {
+      "epoch": 0.00936,
+      "grad_norm": 0.9244572615029755,
+      "learning_rate": 0.002808,
+      "loss": 4.569,
+      "step": 936
+    },
+    {
+      "epoch": 0.00937,
+      "grad_norm": 0.9225067704838915,
+      "learning_rate": 0.002811,
+      "loss": 4.5806,
+      "step": 937
+    },
+    {
+      "epoch": 0.00938,
+      "grad_norm": 0.8534895885659719,
+      "learning_rate": 0.002814,
+      "loss": 4.5604,
+      "step": 938
+    },
+    {
+      "epoch": 0.00939,
+      "grad_norm": 0.9046137087836131,
+      "learning_rate": 0.002817,
+      "loss": 4.5554,
+      "step": 939
+    },
+    {
+      "epoch": 0.0094,
+      "grad_norm": 0.7476401969651744,
+      "learning_rate": 0.00282,
+      "loss": 4.5482,
+      "step": 940
+    },
+    {
+      "epoch": 0.00941,
+      "grad_norm": 0.7066455268972154,
+      "learning_rate": 0.002823,
+      "loss": 4.5225,
+      "step": 941
+    },
+    {
+      "epoch": 0.00942,
+      "grad_norm": 0.6459119835251312,
+      "learning_rate": 0.002826,
+      "loss": 4.544,
+      "step": 942
+    },
+    {
+      "epoch": 0.00943,
+      "grad_norm": 0.6055625869260791,
+      "learning_rate": 0.002829,
+      "loss": 4.5363,
+      "step": 943
+    },
+    {
+      "epoch": 0.00944,
+      "grad_norm": 0.5293280202161804,
+      "learning_rate": 0.002832,
+      "loss": 4.5307,
+      "step": 944
+    },
+    {
+      "epoch": 0.00945,
+      "grad_norm": 0.5438831110130244,
+      "learning_rate": 0.002835,
+      "loss": 4.5334,
+      "step": 945
+    },
+    {
+      "epoch": 0.00946,
+      "grad_norm": 0.4766420743521973,
+      "learning_rate": 0.002838,
+      "loss": 4.5433,
+      "step": 946
+    },
+    {
+      "epoch": 0.00947,
+      "grad_norm": 0.40195567884756706,
+      "learning_rate": 0.0028409999999999998,
+      "loss": 4.5081,
+      "step": 947
+    },
+    {
+      "epoch": 0.00948,
+      "grad_norm": 0.3783844921089427,
+      "learning_rate": 0.0028439999999999997,
+      "loss": 4.4865,
+      "step": 948
+    },
+    {
+      "epoch": 0.00949,
+      "grad_norm": 0.4197576569837563,
+      "learning_rate": 0.002847,
+      "loss": 4.5306,
+      "step": 949
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 0.45947372263331304,
+      "learning_rate": 0.00285,
+      "loss": 4.5193,
+      "step": 950
+    },
+    {
+      "epoch": 0.00951,
+      "grad_norm": 0.5187245758366383,
+      "learning_rate": 0.002853,
+      "loss": 4.4969,
+      "step": 951
+    },
+    {
+      "epoch": 0.00952,
+      "grad_norm": 0.5515692080168162,
+      "learning_rate": 0.002856,
+      "loss": 4.5218,
+      "step": 952
+    },
+    {
+      "epoch": 0.00953,
+      "grad_norm": 0.501582875002041,
+      "learning_rate": 0.002859,
+      "loss": 4.4606,
+      "step": 953
+    },
+    {
+      "epoch": 0.00954,
+      "grad_norm": 0.5014106294436917,
+      "learning_rate": 0.002862,
+      "loss": 4.5197,
+      "step": 954
+    },
+    {
+      "epoch": 0.00955,
+      "grad_norm": 0.6047606934565909,
+      "learning_rate": 0.002865,
+      "loss": 4.5086,
+      "step": 955
+    },
+    {
+      "epoch": 0.00956,
+      "grad_norm": 0.6661868633369662,
+      "learning_rate": 0.002868,
+      "loss": 4.4921,
+      "step": 956
+    },
+    {
+      "epoch": 0.00957,
+      "grad_norm": 0.6511713371124522,
+      "learning_rate": 0.002871,
+      "loss": 4.514,
+      "step": 957
+    },
+    {
+      "epoch": 0.00958,
+      "grad_norm": 0.5733443203887492,
+      "learning_rate": 0.002874,
+      "loss": 4.4931,
+      "step": 958
+    },
+    {
+      "epoch": 0.00959,
+      "grad_norm": 0.6024952806359369,
+      "learning_rate": 0.002877,
+      "loss": 4.4895,
+      "step": 959
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 0.6029559818977924,
+      "learning_rate": 0.0028799999999999997,
+      "loss": 4.4868,
+      "step": 960
+    },
+    {
+      "epoch": 0.00961,
+      "grad_norm": 0.5721073369283843,
+      "learning_rate": 0.002883,
+      "loss": 4.4604,
+      "step": 961
+    },
+    {
+      "epoch": 0.00962,
+      "grad_norm": 0.5737900491823522,
+      "learning_rate": 0.002886,
+      "loss": 4.4898,
+      "step": 962
+    },
+    {
+      "epoch": 0.00963,
+      "grad_norm": 0.5323481251626608,
+      "learning_rate": 0.002889,
+      "loss": 4.4867,
+      "step": 963
+    },
+    {
+      "epoch": 0.00964,
+      "grad_norm": 0.5436801325002781,
+      "learning_rate": 0.002892,
+      "loss": 4.4807,
+      "step": 964
+    },
+    {
+      "epoch": 0.00965,
+      "grad_norm": 0.60229729083351,
+      "learning_rate": 0.002895,
+      "loss": 4.4568,
+      "step": 965
+    },
+    {
+      "epoch": 0.00966,
+      "grad_norm": 0.6629818387101306,
+      "learning_rate": 0.002898,
+      "loss": 4.4766,
+      "step": 966
+    },
+    {
+      "epoch": 0.00967,
+      "grad_norm": 0.6748155978904155,
+      "learning_rate": 0.002901,
+      "loss": 4.5156,
+      "step": 967
+    },
+    {
+      "epoch": 0.00968,
+      "grad_norm": 0.7427494012599226,
+      "learning_rate": 0.002904,
+      "loss": 4.4866,
+      "step": 968
+    },
+    {
+      "epoch": 0.00969,
+      "grad_norm": 0.8794961931178971,
+      "learning_rate": 0.002907,
+      "loss": 4.5273,
+      "step": 969
+    },
+    {
+      "epoch": 0.0097,
+      "grad_norm": 0.8586008854691127,
+      "learning_rate": 0.00291,
+      "loss": 4.4896,
+      "step": 970
+    },
+    {
+      "epoch": 0.00971,
+      "grad_norm": 0.8273563438543869,
+      "learning_rate": 0.002913,
+      "loss": 4.4955,
+      "step": 971
+    },
+    {
+      "epoch": 0.00972,
+      "grad_norm": 0.7536097688784559,
+      "learning_rate": 0.002916,
+      "loss": 4.5029,
+      "step": 972
+    },
+    {
+      "epoch": 0.00973,
+      "grad_norm": 0.7541845251322323,
+      "learning_rate": 0.002919,
+      "loss": 4.4985,
+      "step": 973
+    },
+    {
+      "epoch": 0.00974,
+      "grad_norm": 0.6473607436694337,
+      "learning_rate": 0.002922,
+      "loss": 4.4796,
+      "step": 974
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 0.7361402706574074,
+      "learning_rate": 0.002925,
+      "loss": 4.5044,
+      "step": 975
+    },
+    {
+      "epoch": 0.00976,
+      "grad_norm": 0.868936228763895,
+      "learning_rate": 0.002928,
+      "loss": 4.508,
+      "step": 976
+    },
+    {
+      "epoch": 0.00977,
+      "grad_norm": 0.8813019375073942,
+      "learning_rate": 0.002931,
+      "loss": 4.5408,
+      "step": 977
+    },
+    {
+      "epoch": 0.00978,
+      "grad_norm": 0.9426880168937273,
+      "learning_rate": 0.002934,
+      "loss": 4.5239,
+      "step": 978
+    },
+    {
+      "epoch": 0.00979,
+      "grad_norm": 0.8776049562434768,
+      "learning_rate": 0.002937,
+      "loss": 4.5177,
+      "step": 979
+    },
+    {
+      "epoch": 0.0098,
+      "grad_norm": 0.7621714209410982,
+      "learning_rate": 0.00294,
+      "loss": 4.5005,
+      "step": 980
+    },
+    {
+      "epoch": 0.00981,
+      "grad_norm": 0.7607321859563556,
+      "learning_rate": 0.002943,
+      "loss": 4.5183,
+      "step": 981
+    },
+    {
+      "epoch": 0.00982,
+      "grad_norm": 0.8148690145722087,
+      "learning_rate": 0.002946,
+      "loss": 4.5101,
+      "step": 982
+    },
+    {
+      "epoch": 0.00983,
+      "grad_norm": 0.8602879224908239,
+      "learning_rate": 0.0029490000000000002,
+      "loss": 4.5186,
+      "step": 983
+    },
+    {
+      "epoch": 0.00984,
+      "grad_norm": 0.9348586711938943,
+      "learning_rate": 0.002952,
+      "loss": 4.5105,
+      "step": 984
+    },
+    {
+      "epoch": 0.00985,
+      "grad_norm": 0.9010607870226212,
+      "learning_rate": 0.002955,
+      "loss": 4.5341,
+      "step": 985
+    },
+    {
+      "epoch": 0.00986,
+      "grad_norm": 0.8225410280635316,
+      "learning_rate": 0.002958,
+      "loss": 4.497,
+      "step": 986
+    },
+    {
+      "epoch": 0.00987,
+      "grad_norm": 0.820560994458863,
+      "learning_rate": 0.002961,
+      "loss": 4.5111,
+      "step": 987
+    },
+    {
+      "epoch": 0.00988,
+      "grad_norm": 0.7430257271274537,
+      "learning_rate": 0.002964,
+      "loss": 4.5437,
+      "step": 988
+    },
+    {
+      "epoch": 0.00989,
+      "grad_norm": 0.7193873820034543,
+      "learning_rate": 0.002967,
+      "loss": 4.4838,
+      "step": 989
+    },
+    {
+      "epoch": 0.0099,
+      "grad_norm": 0.7329220852773792,
+      "learning_rate": 0.00297,
+      "loss": 4.5219,
+      "step": 990
+    },
+    {
+      "epoch": 0.00991,
+      "grad_norm": 0.7694030006138932,
+      "learning_rate": 0.002973,
+      "loss": 4.5213,
+      "step": 991
+    },
+    {
+      "epoch": 0.00992,
+      "grad_norm": 0.7726534251991994,
+      "learning_rate": 0.002976,
+      "loss": 4.5153,
+      "step": 992
+    },
+    {
+      "epoch": 0.00993,
+      "grad_norm": 0.6817474065224322,
+      "learning_rate": 0.002979,
+      "loss": 4.511,
+      "step": 993
+    },
+    {
+      "epoch": 0.00994,
+      "grad_norm": 0.6628189744120299,
+      "learning_rate": 0.002982,
+      "loss": 4.5078,
+      "step": 994
+    },
+    {
+      "epoch": 0.00995,
+      "grad_norm": 0.6249124245549155,
+      "learning_rate": 0.0029850000000000002,
+      "loss": 4.5069,
+      "step": 995
+    },
+    {
+      "epoch": 0.00996,
+      "grad_norm": 0.54278083452404,
+      "learning_rate": 0.002988,
+      "loss": 4.503,
+      "step": 996
+    },
+    {
+      "epoch": 0.00997,
+      "grad_norm": 0.5131542547273349,
+      "learning_rate": 0.002991,
+      "loss": 4.49,
+      "step": 997
+    },
+    {
+      "epoch": 0.00998,
+      "grad_norm": 0.4760166868407609,
+      "learning_rate": 0.002994,
+      "loss": 4.4895,
+      "step": 998
+    },
+    {
+      "epoch": 0.00999,
+      "grad_norm": 0.4855483581267517,
+      "learning_rate": 0.002997,
+      "loss": 4.4707,
+      "step": 999
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.44929338024832627,
+      "learning_rate": 0.003,
+      "loss": 4.4897,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 100000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.9643642855424e+16,
+  "train_batch_size": 1024,
+  "trial_name": null,
+  "trial_params": null
+}