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": 1.2723733186721802,
+      "learning_rate": 3e-06,
+      "loss": 10.8324,
+      "step": 1
+    },
+    {
+      "epoch": 2e-05,
+      "grad_norm": 1.2627283334732056,
+      "learning_rate": 6e-06,
+      "loss": 10.8325,
+      "step": 2
+    },
+    {
+      "epoch": 3e-05,
+      "grad_norm": 1.2797267436981201,
+      "learning_rate": 9e-06,
+      "loss": 10.8328,
+      "step": 3
+    },
+    {
+      "epoch": 4e-05,
+      "grad_norm": 1.2568743228912354,
+      "learning_rate": 1.2e-05,
+      "loss": 10.8307,
+      "step": 4
+    },
+    {
+      "epoch": 5e-05,
+      "grad_norm": 1.26583731174469,
+      "learning_rate": 1.5e-05,
+      "loss": 10.8273,
+      "step": 5
+    },
+    {
+      "epoch": 6e-05,
+      "grad_norm": 1.268788456916809,
+      "learning_rate": 1.8e-05,
+      "loss": 10.8268,
+      "step": 6
+    },
+    {
+      "epoch": 7e-05,
+      "grad_norm": 1.215290904045105,
+      "learning_rate": 2.1000000000000002e-05,
+      "loss": 10.8151,
+      "step": 7
+    },
+    {
+      "epoch": 8e-05,
+      "grad_norm": 1.1221675872802734,
+      "learning_rate": 2.4e-05,
+      "loss": 10.794,
+      "step": 8
+    },
+    {
+      "epoch": 9e-05,
+      "grad_norm": 1.0951769351959229,
+      "learning_rate": 2.7e-05,
+      "loss": 10.7888,
+      "step": 9
+    },
+    {
+      "epoch": 0.0001,
+      "grad_norm": 1.105851650238037,
+      "learning_rate": 3e-05,
+      "loss": 10.7765,
+      "step": 10
+    },
+    {
+      "epoch": 0.00011,
+      "grad_norm": 1.0884467363357544,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 10.763,
+      "step": 11
+    },
+    {
+      "epoch": 0.00012,
+      "grad_norm": 1.0829719305038452,
+      "learning_rate": 3.6e-05,
+      "loss": 10.7508,
+      "step": 12
+    },
+    {
+      "epoch": 0.00013,
+      "grad_norm": 1.0506291389465332,
+      "learning_rate": 3.9e-05,
+      "loss": 10.7324,
+      "step": 13
+    },
+    {
+      "epoch": 0.00014,
+      "grad_norm": 1.037864089012146,
+      "learning_rate": 4.2000000000000004e-05,
+      "loss": 10.7208,
+      "step": 14
+    },
+    {
+      "epoch": 0.00015,
+      "grad_norm": 1.0115288496017456,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 10.7117,
+      "step": 15
+    },
+    {
+      "epoch": 0.00016,
+      "grad_norm": 0.9676744341850281,
+      "learning_rate": 4.8e-05,
+      "loss": 10.6952,
+      "step": 16
+    },
+    {
+      "epoch": 0.00017,
+      "grad_norm": 0.9355509877204895,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 10.6792,
+      "step": 17
+    },
+    {
+      "epoch": 0.00018,
+      "grad_norm": 0.9286826848983765,
+      "learning_rate": 5.4e-05,
+      "loss": 10.6649,
+      "step": 18
+    },
+    {
+      "epoch": 0.00019,
+      "grad_norm": 0.9110698699951172,
+      "learning_rate": 5.7e-05,
+      "loss": 10.6512,
+      "step": 19
+    },
+    {
+      "epoch": 0.0002,
+      "grad_norm": 0.9136782288551331,
+      "learning_rate": 6e-05,
+      "loss": 10.6404,
+      "step": 20
+    },
+    {
+      "epoch": 0.00021,
+      "grad_norm": 0.8971966505050659,
+      "learning_rate": 6.3e-05,
+      "loss": 10.6274,
+      "step": 21
+    },
+    {
+      "epoch": 0.00022,
+      "grad_norm": 0.8972620368003845,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 10.6136,
+      "step": 22
+    },
+    {
+      "epoch": 0.00023,
+      "grad_norm": 0.8984483480453491,
+      "learning_rate": 6.9e-05,
+      "loss": 10.6016,
+      "step": 23
+    },
+    {
+      "epoch": 0.00024,
+      "grad_norm": 0.8967456817626953,
+      "learning_rate": 7.2e-05,
+      "loss": 10.5894,
+      "step": 24
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 0.8972211480140686,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 10.5752,
+      "step": 25
+    },
+    {
+      "epoch": 0.00026,
+      "grad_norm": 0.9028114080429077,
+      "learning_rate": 7.8e-05,
+      "loss": 10.5611,
+      "step": 26
+    },
+    {
+      "epoch": 0.00027,
+      "grad_norm": 0.8926876783370972,
+      "learning_rate": 8.1e-05,
+      "loss": 10.5491,
+      "step": 27
+    },
+    {
+      "epoch": 0.00028,
+      "grad_norm": 0.8921052813529968,
+      "learning_rate": 8.400000000000001e-05,
+      "loss": 10.536,
+      "step": 28
+    },
+    {
+      "epoch": 0.00029,
+      "grad_norm": 0.8942669034004211,
+      "learning_rate": 8.7e-05,
+      "loss": 10.5219,
+      "step": 29
+    },
+    {
+      "epoch": 0.0003,
+      "grad_norm": 0.9005073308944702,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 10.5056,
+      "step": 30
+    },
+    {
+      "epoch": 0.00031,
+      "grad_norm": 0.8994124531745911,
+      "learning_rate": 9.3e-05,
+      "loss": 10.491,
+      "step": 31
+    },
+    {
+      "epoch": 0.00032,
+      "grad_norm": 0.8968571424484253,
+      "learning_rate": 9.6e-05,
+      "loss": 10.4763,
+      "step": 32
+    },
+    {
+      "epoch": 0.00033,
+      "grad_norm": 0.8976972103118896,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 10.4597,
+      "step": 33
+    },
+    {
+      "epoch": 0.00034,
+      "grad_norm": 0.8977769017219543,
+      "learning_rate": 0.00010200000000000001,
+      "loss": 10.4427,
+      "step": 34
+    },
+    {
+      "epoch": 0.00035,
+      "grad_norm": 0.902169406414032,
+      "learning_rate": 0.00010500000000000002,
+      "loss": 10.4252,
+      "step": 35
+    },
+    {
+      "epoch": 0.00036,
+      "grad_norm": 0.8990501165390015,
+      "learning_rate": 0.000108,
+      "loss": 10.4079,
+      "step": 36
+    },
+    {
+      "epoch": 0.00037,
+      "grad_norm": 0.8933607935905457,
+      "learning_rate": 0.000111,
+      "loss": 10.39,
+      "step": 37
+    },
+    {
+      "epoch": 0.00038,
+      "grad_norm": 0.8925058245658875,
+      "learning_rate": 0.000114,
+      "loss": 10.3704,
+      "step": 38
+    },
+    {
+      "epoch": 0.00039,
+      "grad_norm": 0.8942745923995972,
+      "learning_rate": 0.000117,
+      "loss": 10.3512,
+      "step": 39
+    },
+    {
+      "epoch": 0.0004,
+      "grad_norm": 0.8984111547470093,
+      "learning_rate": 0.00012,
+      "loss": 10.3286,
+      "step": 40
+    },
+    {
+      "epoch": 0.00041,
+      "grad_norm": 0.8943851590156555,
+      "learning_rate": 0.000123,
+      "loss": 10.3097,
+      "step": 41
+    },
+    {
+      "epoch": 0.00042,
+      "grad_norm": 0.8935915231704712,
+      "learning_rate": 0.000126,
+      "loss": 10.2894,
+      "step": 42
+    },
+    {
+      "epoch": 0.00043,
+      "grad_norm": 0.8975799679756165,
+      "learning_rate": 0.000129,
+      "loss": 10.2654,
+      "step": 43
+    },
+    {
+      "epoch": 0.00044,
+      "grad_norm": 0.8982045650482178,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 10.2433,
+      "step": 44
+    },
+    {
+      "epoch": 0.00045,
+      "grad_norm": 0.9000449180603027,
+      "learning_rate": 0.000135,
+      "loss": 10.2204,
+      "step": 45
+    },
+    {
+      "epoch": 0.00046,
+      "grad_norm": 0.8900250792503357,
+      "learning_rate": 0.000138,
+      "loss": 10.1983,
+      "step": 46
+    },
+    {
+      "epoch": 0.00047,
+      "grad_norm": 0.8965498805046082,
+      "learning_rate": 0.000141,
+      "loss": 10.1723,
+      "step": 47
+    },
+    {
+      "epoch": 0.00048,
+      "grad_norm": 0.8975719213485718,
+      "learning_rate": 0.000144,
+      "loss": 10.149,
+      "step": 48
+    },
+    {
+      "epoch": 0.00049,
+      "grad_norm": 0.8933398127555847,
+      "learning_rate": 0.000147,
+      "loss": 10.1239,
+      "step": 49
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 0.8988479375839233,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 10.0972,
+      "step": 50
+    },
+    {
+      "epoch": 0.00051,
+      "grad_norm": 0.8983374834060669,
+      "learning_rate": 0.000153,
+      "loss": 10.0711,
+      "step": 51
+    },
+    {
+      "epoch": 0.00052,
+      "grad_norm": 0.8959178328514099,
+      "learning_rate": 0.000156,
+      "loss": 10.0437,
+      "step": 52
+    },
+    {
+      "epoch": 0.00053,
+      "grad_norm": 0.8871957063674927,
+      "learning_rate": 0.000159,
+      "loss": 10.0204,
+      "step": 53
+    },
+    {
+      "epoch": 0.00054,
+      "grad_norm": 0.9051761627197266,
+      "learning_rate": 0.000162,
+      "loss": 9.9878,
+      "step": 54
+    },
+    {
+      "epoch": 0.00055,
+      "grad_norm": 0.8952219486236572,
+      "learning_rate": 0.000165,
+      "loss": 9.963,
+      "step": 55
+    },
+    {
+      "epoch": 0.00056,
+      "grad_norm": 0.890164315700531,
+      "learning_rate": 0.00016800000000000002,
+      "loss": 9.9341,
+      "step": 56
+    },
+    {
+      "epoch": 0.00057,
+      "grad_norm": 0.8922548890113831,
+      "learning_rate": 0.000171,
+      "loss": 9.9069,
+      "step": 57
+    },
+    {
+      "epoch": 0.00058,
+      "grad_norm": 0.8901249766349792,
+      "learning_rate": 0.000174,
+      "loss": 9.882,
+      "step": 58
+    },
+    {
+      "epoch": 0.00059,
+      "grad_norm": 0.8989579677581787,
+      "learning_rate": 0.000177,
+      "loss": 9.8502,
+      "step": 59
+    },
+    {
+      "epoch": 0.0006,
+      "grad_norm": 0.8829832673072815,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 9.8242,
+      "step": 60
+    },
+    {
+      "epoch": 0.00061,
+      "grad_norm": 0.8862175345420837,
+      "learning_rate": 0.000183,
+      "loss": 9.7955,
+      "step": 61
+    },
+    {
+      "epoch": 0.00062,
+      "grad_norm": 0.8893216848373413,
+      "learning_rate": 0.000186,
+      "loss": 9.7648,
+      "step": 62
+    },
+    {
+      "epoch": 0.00063,
+      "grad_norm": 0.8881028294563293,
+      "learning_rate": 0.000189,
+      "loss": 9.7373,
+      "step": 63
+    },
+    {
+      "epoch": 0.00064,
+      "grad_norm": 0.8868633508682251,
+      "learning_rate": 0.000192,
+      "loss": 9.7068,
+      "step": 64
+    },
+    {
+      "epoch": 0.00065,
+      "grad_norm": 0.8924434185028076,
+      "learning_rate": 0.00019500000000000002,
+      "loss": 9.6743,
+      "step": 65
+    },
+    {
+      "epoch": 0.00066,
+      "grad_norm": 0.8872416019439697,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 9.6503,
+      "step": 66
+    },
+    {
+      "epoch": 0.00067,
+      "grad_norm": 0.8866419196128845,
+      "learning_rate": 0.000201,
+      "loss": 9.62,
+      "step": 67
+    },
+    {
+      "epoch": 0.00068,
+      "grad_norm": 0.8931677937507629,
+      "learning_rate": 0.00020400000000000003,
+      "loss": 9.5881,
+      "step": 68
+    },
+    {
+      "epoch": 0.00069,
+      "grad_norm": 0.879610002040863,
+      "learning_rate": 0.00020700000000000002,
+      "loss": 9.5597,
+      "step": 69
+    },
+    {
+      "epoch": 0.0007,
+      "grad_norm": 0.8912403583526611,
+      "learning_rate": 0.00021000000000000004,
+      "loss": 9.5236,
+      "step": 70
+    },
+    {
+      "epoch": 0.00071,
+      "grad_norm": 0.8847392797470093,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 9.4965,
+      "step": 71
+    },
+    {
+      "epoch": 0.00072,
+      "grad_norm": 0.8860267996788025,
+      "learning_rate": 0.000216,
+      "loss": 9.4698,
+      "step": 72
+    },
+    {
+      "epoch": 0.00073,
+      "grad_norm": 0.886963963508606,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 9.4354,
+      "step": 73
+    },
+    {
+      "epoch": 0.00074,
+      "grad_norm": 0.8829045295715332,
+      "learning_rate": 0.000222,
+      "loss": 9.4089,
+      "step": 74
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 0.8822581768035889,
+      "learning_rate": 0.000225,
+      "loss": 9.3746,
+      "step": 75
+    },
+    {
+      "epoch": 0.00076,
+      "grad_norm": 0.8918945789337158,
+      "learning_rate": 0.000228,
+      "loss": 9.3485,
+      "step": 76
+    },
+    {
+      "epoch": 0.00077,
+      "grad_norm": 0.8851014375686646,
+      "learning_rate": 0.000231,
+      "loss": 9.3206,
+      "step": 77
+    },
+    {
+      "epoch": 0.00078,
+      "grad_norm": 0.8782386183738708,
+      "learning_rate": 0.000234,
+      "loss": 9.2909,
+      "step": 78
+    },
+    {
+      "epoch": 0.00079,
+      "grad_norm": 0.8847852349281311,
+      "learning_rate": 0.00023700000000000001,
+      "loss": 9.2539,
+      "step": 79
+    },
+    {
+      "epoch": 0.0008,
+      "grad_norm": 0.8866625428199768,
+      "learning_rate": 0.00024,
+      "loss": 9.2184,
+      "step": 80
+    },
+    {
+      "epoch": 0.00081,
+      "grad_norm": 0.8906494975090027,
+      "learning_rate": 0.00024300000000000002,
+      "loss": 9.1857,
+      "step": 81
+    },
+    {
+      "epoch": 0.00082,
+      "grad_norm": 0.8960816264152527,
+      "learning_rate": 0.000246,
+      "loss": 9.1611,
+      "step": 82
+    },
+    {
+      "epoch": 0.00083,
+      "grad_norm": 0.8946034908294678,
+      "learning_rate": 0.00024900000000000004,
+      "loss": 9.1247,
+      "step": 83
+    },
+    {
+      "epoch": 0.00084,
+      "grad_norm": 0.8957815170288086,
+      "learning_rate": 0.000252,
+      "loss": 9.0939,
+      "step": 84
+    },
+    {
+      "epoch": 0.00085,
+      "grad_norm": 0.8891732692718506,
+      "learning_rate": 0.000255,
+      "loss": 9.0702,
+      "step": 85
+    },
+    {
+      "epoch": 0.00086,
+      "grad_norm": 0.8951762318611145,
+      "learning_rate": 0.000258,
+      "loss": 9.0346,
+      "step": 86
+    },
+    {
+      "epoch": 0.00087,
+      "grad_norm": 0.8848313093185425,
+      "learning_rate": 0.000261,
+      "loss": 9.0125,
+      "step": 87
+    },
+    {
+      "epoch": 0.00088,
+      "grad_norm": 0.8860167860984802,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 8.9751,
+      "step": 88
+    },
+    {
+      "epoch": 0.00089,
+      "grad_norm": 0.8798444271087646,
+      "learning_rate": 0.000267,
+      "loss": 8.9486,
+      "step": 89
+    },
+    {
+      "epoch": 0.0009,
+      "grad_norm": 0.8837233185768127,
+      "learning_rate": 0.00027,
+      "loss": 8.9221,
+      "step": 90
+    },
+    {
+      "epoch": 0.00091,
+      "grad_norm": 0.879225492477417,
+      "learning_rate": 0.000273,
+      "loss": 8.8916,
+      "step": 91
+    },
+    {
+      "epoch": 0.00092,
+      "grad_norm": 0.8843948245048523,
+      "learning_rate": 0.000276,
+      "loss": 8.8596,
+      "step": 92
+    },
+    {
+      "epoch": 0.00093,
+      "grad_norm": 0.882838785648346,
+      "learning_rate": 0.000279,
+      "loss": 8.8297,
+      "step": 93
+    },
+    {
+      "epoch": 0.00094,
+      "grad_norm": 0.8830418586730957,
+      "learning_rate": 0.000282,
+      "loss": 8.8034,
+      "step": 94
+    },
+    {
+      "epoch": 0.00095,
+      "grad_norm": 0.8770962357521057,
+      "learning_rate": 0.000285,
+      "loss": 8.7743,
+      "step": 95
+    },
+    {
+      "epoch": 0.00096,
+      "grad_norm": 0.8804563879966736,
+      "learning_rate": 0.000288,
+      "loss": 8.7444,
+      "step": 96
+    },
+    {
+      "epoch": 0.00097,
+      "grad_norm": 0.8753241300582886,
+      "learning_rate": 0.000291,
+      "loss": 8.7211,
+      "step": 97
+    },
+    {
+      "epoch": 0.00098,
+      "grad_norm": 0.8762865662574768,
+      "learning_rate": 0.000294,
+      "loss": 8.6826,
+      "step": 98
+    },
+    {
+      "epoch": 0.00099,
+      "grad_norm": 0.8762408494949341,
+      "learning_rate": 0.000297,
+      "loss": 8.6601,
+      "step": 99
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 0.8741625547409058,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 8.6324,
+      "step": 100
+    },
+    {
+      "epoch": 0.00101,
+      "grad_norm": 0.8789051175117493,
+      "learning_rate": 0.00030300000000000005,
+      "loss": 8.5981,
+      "step": 101
+    },
+    {
+      "epoch": 0.00102,
+      "grad_norm": 0.8656250834465027,
+      "learning_rate": 0.000306,
+      "loss": 8.5772,
+      "step": 102
+    },
+    {
+      "epoch": 0.00103,
+      "grad_norm": 0.8743636012077332,
+      "learning_rate": 0.000309,
+      "loss": 8.5519,
+      "step": 103
+    },
+    {
+      "epoch": 0.00104,
+      "grad_norm": 0.8708215951919556,
+      "learning_rate": 0.000312,
+      "loss": 8.5298,
+      "step": 104
+    },
+    {
+      "epoch": 0.00105,
+      "grad_norm": 0.8859707713127136,
+      "learning_rate": 0.000315,
+      "loss": 8.5033,
+      "step": 105
+    },
+    {
+      "epoch": 0.00106,
+      "grad_norm": 0.9068981409072876,
+      "learning_rate": 0.000318,
+      "loss": 8.4736,
+      "step": 106
+    },
+    {
+      "epoch": 0.00107,
+      "grad_norm": 0.9245584011077881,
+      "learning_rate": 0.000321,
+      "loss": 8.4478,
+      "step": 107
+    },
+    {
+      "epoch": 0.00108,
+      "grad_norm": 0.9128947257995605,
+      "learning_rate": 0.000324,
+      "loss": 8.4188,
+      "step": 108
+    },
+    {
+      "epoch": 0.00109,
+      "grad_norm": 0.853670060634613,
+      "learning_rate": 0.000327,
+      "loss": 8.3963,
+      "step": 109
+    },
+    {
+      "epoch": 0.0011,
+      "grad_norm": 0.8760496377944946,
+      "learning_rate": 0.00033,
+      "loss": 8.3734,
+      "step": 110
+    },
+    {
+      "epoch": 0.00111,
+      "grad_norm": 0.9078761339187622,
+      "learning_rate": 0.000333,
+      "loss": 8.3444,
+      "step": 111
+    },
+    {
+      "epoch": 0.00112,
+      "grad_norm": 0.866322934627533,
+      "learning_rate": 0.00033600000000000004,
+      "loss": 8.3207,
+      "step": 112
+    },
+    {
+      "epoch": 0.00113,
+      "grad_norm": 0.8490086197853088,
+      "learning_rate": 0.000339,
+      "loss": 8.2796,
+      "step": 113
+    },
+    {
+      "epoch": 0.00114,
+      "grad_norm": 0.8713237047195435,
+      "learning_rate": 0.000342,
+      "loss": 8.2694,
+      "step": 114
+    },
+    {
+      "epoch": 0.00115,
+      "grad_norm": 0.8709179162979126,
+      "learning_rate": 0.00034500000000000004,
+      "loss": 8.2404,
+      "step": 115
+    },
+    {
+      "epoch": 0.00116,
+      "grad_norm": 0.8300504684448242,
+      "learning_rate": 0.000348,
+      "loss": 8.214,
+      "step": 116
+    },
+    {
+      "epoch": 0.00117,
+      "grad_norm": 0.8302497863769531,
+      "learning_rate": 0.000351,
+      "loss": 8.1818,
+      "step": 117
+    },
+    {
+      "epoch": 0.00118,
+      "grad_norm": 0.8652266263961792,
+      "learning_rate": 0.000354,
+      "loss": 8.1647,
+      "step": 118
+    },
+    {
+      "epoch": 0.00119,
+      "grad_norm": 0.9069057106971741,
+      "learning_rate": 0.000357,
+      "loss": 8.1514,
+      "step": 119
+    },
+    {
+      "epoch": 0.0012,
+      "grad_norm": 0.9525896906852722,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 8.1239,
+      "step": 120
+    },
+    {
+      "epoch": 0.00121,
+      "grad_norm": 0.9400830268859863,
+      "learning_rate": 0.000363,
+      "loss": 8.095,
+      "step": 121
+    },
+    {
+      "epoch": 0.00122,
+      "grad_norm": 0.8189533948898315,
+      "learning_rate": 0.000366,
+      "loss": 8.0717,
+      "step": 122
+    },
+    {
+      "epoch": 0.00123,
+      "grad_norm": 0.8198633790016174,
+      "learning_rate": 0.000369,
+      "loss": 8.0459,
+      "step": 123
+    },
+    {
+      "epoch": 0.00124,
+      "grad_norm": 0.9460069537162781,
+      "learning_rate": 0.000372,
+      "loss": 8.0261,
+      "step": 124
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 1.0734294652938843,
+      "learning_rate": 0.000375,
+      "loss": 8.002,
+      "step": 125
+    },
+    {
+      "epoch": 0.00126,
+      "grad_norm": 0.9635769724845886,
+      "learning_rate": 0.000378,
+      "loss": 7.9871,
+      "step": 126
+    },
+    {
+      "epoch": 0.00127,
+      "grad_norm": 0.7875692844390869,
+      "learning_rate": 0.000381,
+      "loss": 7.9581,
+      "step": 127
+    },
+    {
+      "epoch": 0.00128,
+      "grad_norm": 0.8865201473236084,
+      "learning_rate": 0.000384,
+      "loss": 7.9374,
+      "step": 128
+    },
+    {
+      "epoch": 0.00129,
+      "grad_norm": 0.9998716115951538,
+      "learning_rate": 0.00038700000000000003,
+      "loss": 7.9265,
+      "step": 129
+    },
+    {
+      "epoch": 0.0013,
+      "grad_norm": 0.8098431825637817,
+      "learning_rate": 0.00039000000000000005,
+      "loss": 7.8932,
+      "step": 130
+    },
+    {
+      "epoch": 0.00131,
+      "grad_norm": 0.8202345967292786,
+      "learning_rate": 0.000393,
+      "loss": 7.8568,
+      "step": 131
+    },
+    {
+      "epoch": 0.00132,
+      "grad_norm": 0.9445962905883789,
+      "learning_rate": 0.00039600000000000003,
+      "loss": 7.8481,
+      "step": 132
+    },
+    {
+      "epoch": 0.00133,
+      "grad_norm": 0.8225625157356262,
+      "learning_rate": 0.00039900000000000005,
+      "loss": 7.8198,
+      "step": 133
+    },
+    {
+      "epoch": 0.00134,
+      "grad_norm": 0.8087729811668396,
+      "learning_rate": 0.000402,
+      "loss": 7.8072,
+      "step": 134
+    },
+    {
+      "epoch": 0.00135,
+      "grad_norm": 0.7232753038406372,
+      "learning_rate": 0.00040500000000000003,
+      "loss": 7.7727,
+      "step": 135
+    },
+    {
+      "epoch": 0.00136,
+      "grad_norm": 0.8383559584617615,
+      "learning_rate": 0.00040800000000000005,
+      "loss": 7.7588,
+      "step": 136
+    },
+    {
+      "epoch": 0.00137,
+      "grad_norm": 0.8329391479492188,
+      "learning_rate": 0.000411,
+      "loss": 7.738,
+      "step": 137
+    },
+    {
+      "epoch": 0.00138,
+      "grad_norm": 0.8072112202644348,
+      "learning_rate": 0.00041400000000000003,
+      "loss": 7.7102,
+      "step": 138
+    },
+    {
+      "epoch": 0.00139,
+      "grad_norm": 0.7906625270843506,
+      "learning_rate": 0.00041700000000000005,
+      "loss": 7.6947,
+      "step": 139
+    },
+    {
+      "epoch": 0.0014,
+      "grad_norm": 0.7997886538505554,
+      "learning_rate": 0.00042000000000000007,
+      "loss": 7.6753,
+      "step": 140
+    },
+    {
+      "epoch": 0.00141,
+      "grad_norm": 0.9642479419708252,
+      "learning_rate": 0.000423,
+      "loss": 7.6647,
+      "step": 141
+    },
+    {
+      "epoch": 0.00142,
+      "grad_norm": 0.8051616549491882,
+      "learning_rate": 0.00042599999999999995,
+      "loss": 7.6463,
+      "step": 142
+    },
+    {
+      "epoch": 0.00143,
+      "grad_norm": 0.7901502251625061,
+      "learning_rate": 0.00042899999999999997,
+      "loss": 7.6185,
+      "step": 143
+    },
+    {
+      "epoch": 0.00144,
+      "grad_norm": 0.6530913710594177,
+      "learning_rate": 0.000432,
+      "loss": 7.5947,
+      "step": 144
+    },
+    {
+      "epoch": 0.00145,
+      "grad_norm": 0.6823164820671082,
+      "learning_rate": 0.000435,
+      "loss": 7.5887,
+      "step": 145
+    },
+    {
+      "epoch": 0.00146,
+      "grad_norm": 0.7043561935424805,
+      "learning_rate": 0.00043799999999999997,
+      "loss": 7.5506,
+      "step": 146
+    },
+    {
+      "epoch": 0.00147,
+      "grad_norm": 0.660875141620636,
+      "learning_rate": 0.000441,
+      "loss": 7.5403,
+      "step": 147
+    },
+    {
+      "epoch": 0.00148,
+      "grad_norm": 0.6651095747947693,
+      "learning_rate": 0.000444,
+      "loss": 7.527,
+      "step": 148
+    },
+    {
+      "epoch": 0.00149,
+      "grad_norm": 0.594466507434845,
+      "learning_rate": 0.00044699999999999997,
+      "loss": 7.5177,
+      "step": 149
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 0.640634298324585,
+      "learning_rate": 0.00045,
+      "loss": 7.4872,
+      "step": 150
+    },
+    {
+      "epoch": 0.00151,
+      "grad_norm": 0.6399310231208801,
+      "learning_rate": 0.000453,
+      "loss": 7.4669,
+      "step": 151
+    },
+    {
+      "epoch": 0.00152,
+      "grad_norm": 0.6032711863517761,
+      "learning_rate": 0.000456,
+      "loss": 7.4595,
+      "step": 152
+    },
+    {
+      "epoch": 0.00153,
+      "grad_norm": 0.8105739951133728,
+      "learning_rate": 0.000459,
+      "loss": 7.4504,
+      "step": 153
+    },
+    {
+      "epoch": 0.00154,
+      "grad_norm": 0.9096337556838989,
+      "learning_rate": 0.000462,
+      "loss": 7.4401,
+      "step": 154
+    },
+    {
+      "epoch": 0.00155,
+      "grad_norm": 0.8464334607124329,
+      "learning_rate": 0.000465,
+      "loss": 7.4179,
+      "step": 155
+    },
+    {
+      "epoch": 0.00156,
+      "grad_norm": 0.8338698148727417,
+      "learning_rate": 0.000468,
+      "loss": 7.392,
+      "step": 156
+    },
+    {
+      "epoch": 0.00157,
+      "grad_norm": 0.6862301230430603,
+      "learning_rate": 0.000471,
+      "loss": 7.3898,
+      "step": 157
+    },
+    {
+      "epoch": 0.00158,
+      "grad_norm": 0.6174972057342529,
+      "learning_rate": 0.00047400000000000003,
+      "loss": 7.3642,
+      "step": 158
+    },
+    {
+      "epoch": 0.00159,
+      "grad_norm": 0.7215908765792847,
+      "learning_rate": 0.000477,
+      "loss": 7.3548,
+      "step": 159
+    },
+    {
+      "epoch": 0.0016,
+      "grad_norm": 0.7243570685386658,
+      "learning_rate": 0.00048,
+      "loss": 7.3363,
+      "step": 160
+    },
+    {
+      "epoch": 0.00161,
+      "grad_norm": 0.5998020172119141,
+      "learning_rate": 0.00048300000000000003,
+      "loss": 7.3105,
+      "step": 161
+    },
+    {
+      "epoch": 0.00162,
+      "grad_norm": 0.6509896516799927,
+      "learning_rate": 0.00048600000000000005,
+      "loss": 7.3157,
+      "step": 162
+    },
+    {
+      "epoch": 0.00163,
+      "grad_norm": 0.645023763179779,
+      "learning_rate": 0.0004890000000000001,
+      "loss": 7.2995,
+      "step": 163
+    },
+    {
+      "epoch": 0.00164,
+      "grad_norm": 0.47143545746803284,
+      "learning_rate": 0.000492,
+      "loss": 7.2948,
+      "step": 164
+    },
+    {
+      "epoch": 0.00165,
+      "grad_norm": 0.6931191086769104,
+      "learning_rate": 0.000495,
+      "loss": 7.2744,
+      "step": 165
+    },
+    {
+      "epoch": 0.00166,
+      "grad_norm": 0.5705839991569519,
+      "learning_rate": 0.0004980000000000001,
+      "loss": 7.2462,
+      "step": 166
+    },
+    {
+      "epoch": 0.00167,
+      "grad_norm": 0.5678435564041138,
+      "learning_rate": 0.000501,
+      "loss": 7.2388,
+      "step": 167
+    },
+    {
+      "epoch": 0.00168,
+      "grad_norm": 0.4481411278247833,
+      "learning_rate": 0.000504,
+      "loss": 7.2152,
+      "step": 168
+    },
+    {
+      "epoch": 0.00169,
+      "grad_norm": 0.5297079086303711,
+      "learning_rate": 0.0005070000000000001,
+      "loss": 7.2111,
+      "step": 169
+    },
+    {
+      "epoch": 0.0017,
+      "grad_norm": 0.5522683262825012,
+      "learning_rate": 0.00051,
+      "loss": 7.1957,
+      "step": 170
+    },
+    {
+      "epoch": 0.00171,
+      "grad_norm": 0.4941532611846924,
+      "learning_rate": 0.000513,
+      "loss": 7.1764,
+      "step": 171
+    },
+    {
+      "epoch": 0.00172,
+      "grad_norm": 0.40358924865722656,
+      "learning_rate": 0.000516,
+      "loss": 7.1752,
+      "step": 172
+    },
+    {
+      "epoch": 0.00173,
+      "grad_norm": 0.43254554271698,
+      "learning_rate": 0.0005189999999999999,
+      "loss": 7.1608,
+      "step": 173
+    },
+    {
+      "epoch": 0.00174,
+      "grad_norm": 0.49833717942237854,
+      "learning_rate": 0.000522,
+      "loss": 7.1779,
+      "step": 174
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 0.469341903924942,
+      "learning_rate": 0.000525,
+      "loss": 7.1406,
+      "step": 175
+    },
+    {
+      "epoch": 0.00176,
+      "grad_norm": 0.46195128560066223,
+      "learning_rate": 0.0005279999999999999,
+      "loss": 7.1165,
+      "step": 176
+    },
+    {
+      "epoch": 0.00177,
+      "grad_norm": 0.4399634003639221,
+      "learning_rate": 0.000531,
+      "loss": 7.1206,
+      "step": 177
+    },
+    {
+      "epoch": 0.00178,
+      "grad_norm": 0.42299556732177734,
+      "learning_rate": 0.000534,
+      "loss": 7.1134,
+      "step": 178
+    },
+    {
+      "epoch": 0.00179,
+      "grad_norm": 0.3987540602684021,
+      "learning_rate": 0.000537,
+      "loss": 7.1017,
+      "step": 179
+    },
+    {
+      "epoch": 0.0018,
+      "grad_norm": 0.37395715713500977,
+      "learning_rate": 0.00054,
+      "loss": 7.074,
+      "step": 180
+    },
+    {
+      "epoch": 0.00181,
+      "grad_norm": 0.3351408541202545,
+      "learning_rate": 0.000543,
+      "loss": 7.0613,
+      "step": 181
+    },
+    {
+      "epoch": 0.00182,
+      "grad_norm": 0.3589305281639099,
+      "learning_rate": 0.000546,
+      "loss": 7.0589,
+      "step": 182
+    },
+    {
+      "epoch": 0.00183,
+      "grad_norm": 0.38710907101631165,
+      "learning_rate": 0.000549,
+      "loss": 7.0486,
+      "step": 183
+    },
+    {
+      "epoch": 0.00184,
+      "grad_norm": 0.4560106098651886,
+      "learning_rate": 0.000552,
+      "loss": 7.0384,
+      "step": 184
+    },
+    {
+      "epoch": 0.00185,
+      "grad_norm": 0.8466277718544006,
+      "learning_rate": 0.000555,
+      "loss": 7.0199,
+      "step": 185
+    },
+    {
+      "epoch": 0.00186,
+      "grad_norm": 1.4107517004013062,
+      "learning_rate": 0.000558,
+      "loss": 7.0664,
+      "step": 186
+    },
+    {
+      "epoch": 0.00187,
+      "grad_norm": 0.5632089376449585,
+      "learning_rate": 0.000561,
+      "loss": 7.004,
+      "step": 187
+    },
+    {
+      "epoch": 0.00188,
+      "grad_norm": 1.07405686378479,
+      "learning_rate": 0.000564,
+      "loss": 6.9945,
+      "step": 188
+    },
+    {
+      "epoch": 0.00189,
+      "grad_norm": 1.1665420532226562,
+      "learning_rate": 0.000567,
+      "loss": 7.0095,
+      "step": 189
+    },
+    {
+      "epoch": 0.0019,
+      "grad_norm": 0.4235672950744629,
+      "learning_rate": 0.00057,
+      "loss": 6.962,
+      "step": 190
+    },
+    {
+      "epoch": 0.00191,
+      "grad_norm": 1.2953448295593262,
+      "learning_rate": 0.000573,
+      "loss": 6.9853,
+      "step": 191
+    },
+    {
+      "epoch": 0.00192,
+      "grad_norm": 0.5110867023468018,
+      "learning_rate": 0.000576,
+      "loss": 6.9512,
+      "step": 192
+    },
+    {
+      "epoch": 0.00193,
+      "grad_norm": 0.7966066002845764,
+      "learning_rate": 0.000579,
+      "loss": 6.9656,
+      "step": 193
+    },
+    {
+      "epoch": 0.00194,
+      "grad_norm": 0.5008851289749146,
+      "learning_rate": 0.000582,
+      "loss": 6.9458,
+      "step": 194
+    },
+    {
+      "epoch": 0.00195,
+      "grad_norm": 0.714582085609436,
+      "learning_rate": 0.000585,
+      "loss": 6.9325,
+      "step": 195
+    },
+    {
+      "epoch": 0.00196,
+      "grad_norm": 0.48010018467903137,
+      "learning_rate": 0.000588,
+      "loss": 6.917,
+      "step": 196
+    },
+    {
+      "epoch": 0.00197,
+      "grad_norm": 0.5283955335617065,
+      "learning_rate": 0.000591,
+      "loss": 6.9149,
+      "step": 197
+    },
+    {
+      "epoch": 0.00198,
+      "grad_norm": 0.5033705830574036,
+      "learning_rate": 0.000594,
+      "loss": 6.9041,
+      "step": 198
+    },
+    {
+      "epoch": 0.00199,
+      "grad_norm": 0.40711161494255066,
+      "learning_rate": 0.0005970000000000001,
+      "loss": 6.8819,
+      "step": 199
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 0.4253259003162384,
+      "learning_rate": 0.0006000000000000001,
+      "loss": 6.8842,
+      "step": 200
+    },
+    {
+      "epoch": 0.00201,
+      "grad_norm": 0.347766637802124,
+      "learning_rate": 0.000603,
+      "loss": 6.8566,
+      "step": 201
+    },
+    {
+      "epoch": 0.00202,
+      "grad_norm": 0.4044833779335022,
+      "learning_rate": 0.0006060000000000001,
+      "loss": 6.8448,
+      "step": 202
+    },
+    {
+      "epoch": 0.00203,
+      "grad_norm": 0.3598291575908661,
+      "learning_rate": 0.0006090000000000001,
+      "loss": 6.8489,
+      "step": 203
+    },
+    {
+      "epoch": 0.00204,
+      "grad_norm": 0.35803648829460144,
+      "learning_rate": 0.000612,
+      "loss": 6.8337,
+      "step": 204
+    },
+    {
+      "epoch": 0.00205,
+      "grad_norm": 0.3630695044994354,
+      "learning_rate": 0.000615,
+      "loss": 6.82,
+      "step": 205
+    },
+    {
+      "epoch": 0.00206,
+      "grad_norm": 0.3439967930316925,
+      "learning_rate": 0.000618,
+      "loss": 6.8125,
+      "step": 206
+    },
+    {
+      "epoch": 0.00207,
+      "grad_norm": 0.3219742476940155,
+      "learning_rate": 0.000621,
+      "loss": 6.8026,
+      "step": 207
+    },
+    {
+      "epoch": 0.00208,
+      "grad_norm": 0.42236459255218506,
+      "learning_rate": 0.000624,
+      "loss": 6.8006,
+      "step": 208
+    },
+    {
+      "epoch": 0.00209,
+      "grad_norm": 0.44003602862358093,
+      "learning_rate": 0.000627,
+      "loss": 6.7802,
+      "step": 209
+    },
+    {
+      "epoch": 0.0021,
+      "grad_norm": 0.6629877686500549,
+      "learning_rate": 0.00063,
+      "loss": 6.7737,
+      "step": 210
+    },
+    {
+      "epoch": 0.00211,
+      "grad_norm": 0.9569016695022583,
+      "learning_rate": 0.000633,
+      "loss": 6.7754,
+      "step": 211
+    },
+    {
+      "epoch": 0.00212,
+      "grad_norm": 1.1293443441390991,
+      "learning_rate": 0.000636,
+      "loss": 6.7746,
+      "step": 212
+    },
+    {
+      "epoch": 0.00213,
+      "grad_norm": 0.43011096119880676,
+      "learning_rate": 0.000639,
+      "loss": 6.7385,
+      "step": 213
+    },
+    {
+      "epoch": 0.00214,
+      "grad_norm": 0.6478229761123657,
+      "learning_rate": 0.000642,
+      "loss": 6.7415,
+      "step": 214
+    },
+    {
+      "epoch": 0.00215,
+      "grad_norm": 0.6323032975196838,
+      "learning_rate": 0.000645,
+      "loss": 6.7384,
+      "step": 215
+    },
+    {
+      "epoch": 0.00216,
+      "grad_norm": 0.441693514585495,
+      "learning_rate": 0.000648,
+      "loss": 6.7285,
+      "step": 216
+    },
+    {
+      "epoch": 0.00217,
+      "grad_norm": 0.5594473481178284,
+      "learning_rate": 0.000651,
+      "loss": 6.7033,
+      "step": 217
+    },
+    {
+      "epoch": 0.00218,
+      "grad_norm": 0.5135915279388428,
+      "learning_rate": 0.000654,
+      "loss": 6.7073,
+      "step": 218
+    },
+    {
+      "epoch": 0.00219,
+      "grad_norm": 0.4307027757167816,
+      "learning_rate": 0.000657,
+      "loss": 6.6782,
+      "step": 219
+    },
+    {
+      "epoch": 0.0022,
+      "grad_norm": 0.45137229561805725,
+      "learning_rate": 0.00066,
+      "loss": 6.6953,
+      "step": 220
+    },
+    {
+      "epoch": 0.00221,
+      "grad_norm": 0.4729914963245392,
+      "learning_rate": 0.0006630000000000001,
+      "loss": 6.6737,
+      "step": 221
+    },
+    {
+      "epoch": 0.00222,
+      "grad_norm": 0.47246506810188293,
+      "learning_rate": 0.000666,
+      "loss": 6.6615,
+      "step": 222
+    },
+    {
+      "epoch": 0.00223,
+      "grad_norm": 0.36316192150115967,
+      "learning_rate": 0.000669,
+      "loss": 6.6543,
+      "step": 223
+    },
+    {
+      "epoch": 0.00224,
+      "grad_norm": 0.4332623779773712,
+      "learning_rate": 0.0006720000000000001,
+      "loss": 6.6525,
+      "step": 224
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 0.41814228892326355,
+      "learning_rate": 0.000675,
+      "loss": 6.639,
+      "step": 225
+    },
+    {
+      "epoch": 0.00226,
+      "grad_norm": 0.42956992983818054,
+      "learning_rate": 0.000678,
+      "loss": 6.634,
+      "step": 226
+    },
+    {
+      "epoch": 0.00227,
+      "grad_norm": 0.40267884731292725,
+      "learning_rate": 0.0006810000000000001,
+      "loss": 6.6072,
+      "step": 227
+    },
+    {
+      "epoch": 0.00228,
+      "grad_norm": 0.4361991882324219,
+      "learning_rate": 0.000684,
+      "loss": 6.6099,
+      "step": 228
+    },
+    {
+      "epoch": 0.00229,
+      "grad_norm": 0.47655290365219116,
+      "learning_rate": 0.000687,
+      "loss": 6.6086,
+      "step": 229
+    },
+    {
+      "epoch": 0.0023,
+      "grad_norm": 0.5011177659034729,
+      "learning_rate": 0.0006900000000000001,
+      "loss": 6.5846,
+      "step": 230
+    },
+    {
+      "epoch": 0.00231,
+      "grad_norm": 0.5389447212219238,
+      "learning_rate": 0.000693,
+      "loss": 6.5719,
+      "step": 231
+    },
+    {
+      "epoch": 0.00232,
+      "grad_norm": 0.5394959449768066,
+      "learning_rate": 0.000696,
+      "loss": 6.5809,
+      "step": 232
+    },
+    {
+      "epoch": 0.00233,
+      "grad_norm": 0.49784839153289795,
+      "learning_rate": 0.0006990000000000001,
+      "loss": 6.5675,
+      "step": 233
+    },
+    {
+      "epoch": 0.00234,
+      "grad_norm": 0.42049404978752136,
+      "learning_rate": 0.000702,
+      "loss": 6.5581,
+      "step": 234
+    },
+    {
+      "epoch": 0.00235,
+      "grad_norm": 0.5810425281524658,
+      "learning_rate": 0.000705,
+      "loss": 6.5463,
+      "step": 235
+    },
+    {
+      "epoch": 0.00236,
+      "grad_norm": 0.6429721117019653,
+      "learning_rate": 0.000708,
+      "loss": 6.5306,
+      "step": 236
+    },
+    {
+      "epoch": 0.00237,
+      "grad_norm": 0.6626091003417969,
+      "learning_rate": 0.0007109999999999999,
+      "loss": 6.5402,
+      "step": 237
+    },
+    {
+      "epoch": 0.00238,
+      "grad_norm": 0.5873957872390747,
+      "learning_rate": 0.000714,
+      "loss": 6.5299,
+      "step": 238
+    },
+    {
+      "epoch": 0.00239,
+      "grad_norm": 0.4890768527984619,
+      "learning_rate": 0.000717,
+      "loss": 6.5085,
+      "step": 239
+    },
+    {
+      "epoch": 0.0024,
+      "grad_norm": 0.6101468801498413,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 6.5266,
+      "step": 240
+    },
+    {
+      "epoch": 0.00241,
+      "grad_norm": 0.5837545394897461,
+      "learning_rate": 0.000723,
+      "loss": 6.5125,
+      "step": 241
+    },
+    {
+      "epoch": 0.00242,
+      "grad_norm": 0.38101619482040405,
+      "learning_rate": 0.000726,
+      "loss": 6.481,
+      "step": 242
+    },
+    {
+      "epoch": 0.00243,
+      "grad_norm": 0.5516716837882996,
+      "learning_rate": 0.000729,
+      "loss": 6.4712,
+      "step": 243
+    },
+    {
+      "epoch": 0.00244,
+      "grad_norm": 0.6402163505554199,
+      "learning_rate": 0.000732,
+      "loss": 6.4787,
+      "step": 244
+    },
+    {
+      "epoch": 0.00245,
+      "grad_norm": 0.661472737789154,
+      "learning_rate": 0.000735,
+      "loss": 6.4666,
+      "step": 245
+    },
+    {
+      "epoch": 0.00246,
+      "grad_norm": 0.8242950439453125,
+      "learning_rate": 0.000738,
+      "loss": 6.457,
+      "step": 246
+    },
+    {
+      "epoch": 0.00247,
+      "grad_norm": 0.8979260921478271,
+      "learning_rate": 0.000741,
+      "loss": 6.4532,
+      "step": 247
+    },
+    {
+      "epoch": 0.00248,
+      "grad_norm": 0.7822521924972534,
+      "learning_rate": 0.000744,
+      "loss": 6.4595,
+      "step": 248
+    },
+    {
+      "epoch": 0.00249,
+      "grad_norm": 0.5830389261245728,
+      "learning_rate": 0.000747,
+      "loss": 6.4336,
+      "step": 249
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 0.5053834319114685,
+      "learning_rate": 0.00075,
+      "loss": 6.436,
+      "step": 250
+    },
+    {
+      "epoch": 0.00251,
+      "grad_norm": 0.6169653534889221,
+      "learning_rate": 0.000753,
+      "loss": 6.4018,
+      "step": 251
+    },
+    {
+      "epoch": 0.00252,
+      "grad_norm": 0.693821370601654,
+      "learning_rate": 0.000756,
+      "loss": 6.411,
+      "step": 252
+    },
+    {
+      "epoch": 0.00253,
+      "grad_norm": 0.7529676556587219,
+      "learning_rate": 0.000759,
+      "loss": 6.4053,
+      "step": 253
+    },
+    {
+      "epoch": 0.00254,
+      "grad_norm": 0.7882714867591858,
+      "learning_rate": 0.000762,
+      "loss": 6.3993,
+      "step": 254
+    },
+    {
+      "epoch": 0.00255,
+      "grad_norm": 0.8540387153625488,
+      "learning_rate": 0.0007650000000000001,
+      "loss": 6.399,
+      "step": 255
+    },
+    {
+      "epoch": 0.00256,
+      "grad_norm": 0.8460838198661804,
+      "learning_rate": 0.000768,
+      "loss": 6.3805,
+      "step": 256
+    },
+    {
+      "epoch": 0.00257,
+      "grad_norm": 0.7798411250114441,
+      "learning_rate": 0.000771,
+      "loss": 6.383,
+      "step": 257
+    },
+    {
+      "epoch": 0.00258,
+      "grad_norm": 0.7431406378746033,
+      "learning_rate": 0.0007740000000000001,
+      "loss": 6.362,
+      "step": 258
+    },
+    {
+      "epoch": 0.00259,
+      "grad_norm": 0.9195736646652222,
+      "learning_rate": 0.000777,
+      "loss": 6.3634,
+      "step": 259
+    },
+    {
+      "epoch": 0.0026,
+      "grad_norm": 0.9989089965820312,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 6.3605,
+      "step": 260
+    },
+    {
+      "epoch": 0.00261,
+      "grad_norm": 0.8553241491317749,
+      "learning_rate": 0.0007830000000000001,
+      "loss": 6.3627,
+      "step": 261
+    },
+    {
+      "epoch": 0.00262,
+      "grad_norm": 0.8126974701881409,
+      "learning_rate": 0.000786,
+      "loss": 6.3448,
+      "step": 262
+    },
+    {
+      "epoch": 0.00263,
+      "grad_norm": 0.5453911423683167,
+      "learning_rate": 0.0007890000000000001,
+      "loss": 6.3416,
+      "step": 263
+    },
+    {
+      "epoch": 0.00264,
+      "grad_norm": 0.5339396595954895,
+      "learning_rate": 0.0007920000000000001,
+      "loss": 6.3252,
+      "step": 264
+    },
+    {
+      "epoch": 0.00265,
+      "grad_norm": 0.561769425868988,
+      "learning_rate": 0.000795,
+      "loss": 6.3181,
+      "step": 265
+    },
+    {
+      "epoch": 0.00266,
+      "grad_norm": 0.47888532280921936,
+      "learning_rate": 0.0007980000000000001,
+      "loss": 6.3105,
+      "step": 266
+    },
+    {
+      "epoch": 0.00267,
+      "grad_norm": 0.4935484528541565,
+      "learning_rate": 0.0008010000000000001,
+      "loss": 6.3049,
+      "step": 267
+    },
+    {
+      "epoch": 0.00268,
+      "grad_norm": 0.4065157175064087,
+      "learning_rate": 0.000804,
+      "loss": 6.2969,
+      "step": 268
+    },
+    {
+      "epoch": 0.00269,
+      "grad_norm": 0.5361817479133606,
+      "learning_rate": 0.0008070000000000001,
+      "loss": 6.2818,
+      "step": 269
+    },
+    {
+      "epoch": 0.0027,
+      "grad_norm": 0.6360214352607727,
+      "learning_rate": 0.0008100000000000001,
+      "loss": 6.2858,
+      "step": 270
+    },
+    {
+      "epoch": 0.00271,
+      "grad_norm": 0.6580653190612793,
+      "learning_rate": 0.000813,
+      "loss": 6.2898,
+      "step": 271
+    },
+    {
+      "epoch": 0.00272,
+      "grad_norm": 0.719866931438446,
+      "learning_rate": 0.0008160000000000001,
+      "loss": 6.2764,
+      "step": 272
+    },
+    {
+      "epoch": 0.00273,
+      "grad_norm": 0.726635754108429,
+      "learning_rate": 0.0008190000000000001,
+      "loss": 6.2654,
+      "step": 273
+    },
+    {
+      "epoch": 0.00274,
+      "grad_norm": 0.5728192925453186,
+      "learning_rate": 0.000822,
+      "loss": 6.2437,
+      "step": 274
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 0.469969242811203,
+      "learning_rate": 0.0008250000000000001,
+      "loss": 6.2475,
+      "step": 275
+    },
+    {
+      "epoch": 0.00276,
+      "grad_norm": 0.5783148407936096,
+      "learning_rate": 0.0008280000000000001,
+      "loss": 6.2405,
+      "step": 276
+    },
+    {
+      "epoch": 0.00277,
+      "grad_norm": 0.5995691418647766,
+      "learning_rate": 0.0008310000000000001,
+      "loss": 6.2356,
+      "step": 277
+    },
+    {
+      "epoch": 0.00278,
+      "grad_norm": 0.5173709988594055,
+      "learning_rate": 0.0008340000000000001,
+      "loss": 6.2181,
+      "step": 278
+    },
+    {
+      "epoch": 0.00279,
+      "grad_norm": 0.47013920545578003,
+      "learning_rate": 0.0008370000000000001,
+      "loss": 6.2175,
+      "step": 279
+    },
+    {
+      "epoch": 0.0028,
+      "grad_norm": 0.38072702288627625,
+      "learning_rate": 0.0008400000000000001,
+      "loss": 6.1988,
+      "step": 280
+    },
+    {
+      "epoch": 0.00281,
+      "grad_norm": 0.44907790422439575,
+      "learning_rate": 0.0008430000000000001,
+      "loss": 6.1893,
+      "step": 281
+    },
+    {
+      "epoch": 0.00282,
+      "grad_norm": 0.40965142846107483,
+      "learning_rate": 0.000846,
+      "loss": 6.1869,
+      "step": 282
+    },
+    {
+      "epoch": 0.00283,
+      "grad_norm": 0.48822489380836487,
+      "learning_rate": 0.0008489999999999999,
+      "loss": 6.1783,
+      "step": 283
+    },
+    {
+      "epoch": 0.00284,
+      "grad_norm": 0.726660966873169,
+      "learning_rate": 0.0008519999999999999,
+      "loss": 6.1771,
+      "step": 284
+    },
+    {
+      "epoch": 0.00285,
+      "grad_norm": 1.0991517305374146,
+      "learning_rate": 0.000855,
+      "loss": 6.1879,
+      "step": 285
+    },
+    {
+      "epoch": 0.00286,
+      "grad_norm": 0.9898068904876709,
+      "learning_rate": 0.0008579999999999999,
+      "loss": 6.1694,
+      "step": 286
+    },
+    {
+      "epoch": 0.00287,
+      "grad_norm": 0.9177366495132446,
+      "learning_rate": 0.000861,
+      "loss": 6.1687,
+      "step": 287
+    },
+    {
+      "epoch": 0.00288,
+      "grad_norm": 1.3172835111618042,
+      "learning_rate": 0.000864,
+      "loss": 6.175,
+      "step": 288
+    },
+    {
+      "epoch": 0.00289,
+      "grad_norm": 1.0531185865402222,
+      "learning_rate": 0.0008669999999999999,
+      "loss": 6.1733,
+      "step": 289
+    },
+    {
+      "epoch": 0.0029,
+      "grad_norm": 0.9814063310623169,
+      "learning_rate": 0.00087,
+      "loss": 6.1665,
+      "step": 290
+    },
+    {
+      "epoch": 0.00291,
+      "grad_norm": 0.7696391344070435,
+      "learning_rate": 0.000873,
+      "loss": 6.1455,
+      "step": 291
+    },
+    {
+      "epoch": 0.00292,
+      "grad_norm": 0.8531065583229065,
+      "learning_rate": 0.0008759999999999999,
+      "loss": 6.15,
+      "step": 292
+    },
+    {
+      "epoch": 0.00293,
+      "grad_norm": 0.7760049700737,
+      "learning_rate": 0.000879,
+      "loss": 6.1273,
+      "step": 293
+    },
+    {
+      "epoch": 0.00294,
+      "grad_norm": 0.7517282366752625,
+      "learning_rate": 0.000882,
+      "loss": 6.1432,
+      "step": 294
+    },
+    {
+      "epoch": 0.00295,
+      "grad_norm": 0.5758442282676697,
+      "learning_rate": 0.0008849999999999999,
+      "loss": 6.106,
+      "step": 295
+    },
+    {
+      "epoch": 0.00296,
+      "grad_norm": 0.5470280647277832,
+      "learning_rate": 0.000888,
+      "loss": 6.1152,
+      "step": 296
+    },
+    {
+      "epoch": 0.00297,
+      "grad_norm": 0.46315857768058777,
+      "learning_rate": 0.000891,
+      "loss": 6.0848,
+      "step": 297
+    },
+    {
+      "epoch": 0.00298,
+      "grad_norm": 0.52577805519104,
+      "learning_rate": 0.0008939999999999999,
+      "loss": 6.0996,
+      "step": 298
+    },
+    {
+      "epoch": 0.00299,
+      "grad_norm": 0.5289214253425598,
+      "learning_rate": 0.000897,
+      "loss": 6.0753,
+      "step": 299
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 0.39721718430519104,
+      "learning_rate": 0.0009,
+      "loss": 6.0729,
+      "step": 300
+    },
+    {
+      "epoch": 0.00301,
+      "grad_norm": 0.42188870906829834,
+      "learning_rate": 0.0009029999999999999,
+      "loss": 6.0649,
+      "step": 301
+    },
+    {
+      "epoch": 0.00302,
+      "grad_norm": 0.49844104051589966,
+      "learning_rate": 0.000906,
+      "loss": 6.0504,
+      "step": 302
+    },
+    {
+      "epoch": 0.00303,
+      "grad_norm": 0.5113502144813538,
+      "learning_rate": 0.000909,
+      "loss": 6.0629,
+      "step": 303
+    },
+    {
+      "epoch": 0.00304,
+      "grad_norm": 0.6390882730484009,
+      "learning_rate": 0.000912,
+      "loss": 6.0593,
+      "step": 304
+    },
+    {
+      "epoch": 0.00305,
+      "grad_norm": 0.8851528763771057,
+      "learning_rate": 0.000915,
+      "loss": 6.0668,
+      "step": 305
+    },
+    {
+      "epoch": 0.00306,
+      "grad_norm": 0.9017530083656311,
+      "learning_rate": 0.000918,
+      "loss": 6.0552,
+      "step": 306
+    },
+    {
+      "epoch": 0.00307,
+      "grad_norm": 0.563444197177887,
+      "learning_rate": 0.000921,
+      "loss": 6.0386,
+      "step": 307
+    },
+    {
+      "epoch": 0.00308,
+      "grad_norm": 0.6002116203308105,
+      "learning_rate": 0.000924,
+      "loss": 6.0248,
+      "step": 308
+    },
+    {
+      "epoch": 0.00309,
+      "grad_norm": 0.7035393118858337,
+      "learning_rate": 0.000927,
+      "loss": 6.0412,
+      "step": 309
+    },
+    {
+      "epoch": 0.0031,
+      "grad_norm": 1.01050865650177,
+      "learning_rate": 0.00093,
+      "loss": 6.0314,
+      "step": 310
+    },
+    {
+      "epoch": 0.00311,
+      "grad_norm": 1.16908860206604,
+      "learning_rate": 0.000933,
+      "loss": 6.0397,
+      "step": 311
+    },
+    {
+      "epoch": 0.00312,
+      "grad_norm": 0.6785458922386169,
+      "learning_rate": 0.000936,
+      "loss": 6.0006,
+      "step": 312
+    },
+    {
+      "epoch": 0.00313,
+      "grad_norm": 0.7975029349327087,
+      "learning_rate": 0.0009390000000000001,
+      "loss": 6.0267,
+      "step": 313
+    },
+    {
+      "epoch": 0.00314,
+      "grad_norm": 0.6784241795539856,
+      "learning_rate": 0.000942,
+      "loss": 6.0253,
+      "step": 314
+    },
+    {
+      "epoch": 0.00315,
+      "grad_norm": 0.5287242531776428,
+      "learning_rate": 0.000945,
+      "loss": 5.9989,
+      "step": 315
+    },
+    {
+      "epoch": 0.00316,
+      "grad_norm": 0.5889810919761658,
+      "learning_rate": 0.0009480000000000001,
+      "loss": 5.9738,
+      "step": 316
+    },
+    {
+      "epoch": 0.00317,
+      "grad_norm": 0.8596201539039612,
+      "learning_rate": 0.000951,
+      "loss": 5.9963,
+      "step": 317
+    },
+    {
+      "epoch": 0.00318,
+      "grad_norm": 1.220719575881958,
+      "learning_rate": 0.000954,
+      "loss": 6.0052,
+      "step": 318
+    },
+    {
+      "epoch": 0.00319,
+      "grad_norm": 0.7490801215171814,
+      "learning_rate": 0.0009570000000000001,
+      "loss": 5.9868,
+      "step": 319
+    },
+    {
+      "epoch": 0.0032,
+      "grad_norm": 0.6210083365440369,
+      "learning_rate": 0.00096,
+      "loss": 5.9801,
+      "step": 320
+    },
+    {
+      "epoch": 0.00321,
+      "grad_norm": 0.5945920944213867,
+      "learning_rate": 0.000963,
+      "loss": 5.97,
+      "step": 321
+    },
+    {
+      "epoch": 0.00322,
+      "grad_norm": 0.6791667342185974,
+      "learning_rate": 0.0009660000000000001,
+      "loss": 5.9544,
+      "step": 322
+    },
+    {
+      "epoch": 0.00323,
+      "grad_norm": 0.9637515544891357,
+      "learning_rate": 0.000969,
+      "loss": 5.9627,
+      "step": 323
+    },
+    {
+      "epoch": 0.00324,
+      "grad_norm": 1.1402119398117065,
+      "learning_rate": 0.0009720000000000001,
+      "loss": 5.972,
+      "step": 324
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 1.0057023763656616,
+      "learning_rate": 0.0009750000000000001,
+      "loss": 5.9423,
+      "step": 325
+    },
+    {
+      "epoch": 0.00326,
+      "grad_norm": 0.5953328609466553,
+      "learning_rate": 0.0009780000000000001,
+      "loss": 5.942,
+      "step": 326
+    },
+    {
+      "epoch": 0.00327,
+      "grad_norm": 0.7124008536338806,
+      "learning_rate": 0.000981,
+      "loss": 5.9465,
+      "step": 327
+    },
+    {
+      "epoch": 0.00328,
+      "grad_norm": 0.7318410277366638,
+      "learning_rate": 0.000984,
+      "loss": 5.9241,
+      "step": 328
+    },
+    {
+      "epoch": 0.00329,
+      "grad_norm": 0.6503687500953674,
+      "learning_rate": 0.000987,
+      "loss": 5.922,
+      "step": 329
+    },
+    {
+      "epoch": 0.0033,
+      "grad_norm": 0.6151977181434631,
+      "learning_rate": 0.00099,
+      "loss": 5.9159,
+      "step": 330
+    },
+    {
+      "epoch": 0.00331,
+      "grad_norm": 0.49422070384025574,
+      "learning_rate": 0.0009930000000000002,
+      "loss": 5.9144,
+      "step": 331
+    },
+    {
+      "epoch": 0.00332,
+      "grad_norm": 0.563934326171875,
+      "learning_rate": 0.0009960000000000001,
+      "loss": 5.9008,
+      "step": 332
+    },
+    {
+      "epoch": 0.00333,
+      "grad_norm": 0.5146680474281311,
+      "learning_rate": 0.000999,
+      "loss": 5.8812,
+      "step": 333
+    },
+    {
+      "epoch": 0.00334,
+      "grad_norm": 0.5699781775474548,
+      "learning_rate": 0.001002,
+      "loss": 5.8922,
+      "step": 334
+    },
+    {
+      "epoch": 0.00335,
+      "grad_norm": 0.628279983997345,
+      "learning_rate": 0.001005,
+      "loss": 5.8934,
+      "step": 335
+    },
+    {
+      "epoch": 0.00336,
+      "grad_norm": 0.638155996799469,
+      "learning_rate": 0.001008,
+      "loss": 5.8854,
+      "step": 336
+    },
+    {
+      "epoch": 0.00337,
+      "grad_norm": 0.5850276947021484,
+      "learning_rate": 0.0010110000000000002,
+      "loss": 5.8631,
+      "step": 337
+    },
+    {
+      "epoch": 0.00338,
+      "grad_norm": 0.5985286831855774,
+      "learning_rate": 0.0010140000000000001,
+      "loss": 5.879,
+      "step": 338
+    },
+    {
+      "epoch": 0.00339,
+      "grad_norm": 0.9502546787261963,
+      "learning_rate": 0.0010170000000000001,
+      "loss": 5.8893,
+      "step": 339
+    },
+    {
+      "epoch": 0.0034,
+      "grad_norm": 1.3471951484680176,
+      "learning_rate": 0.00102,
+      "loss": 5.8789,
+      "step": 340
+    },
+    {
+      "epoch": 0.00341,
+      "grad_norm": 0.6621291041374207,
+      "learning_rate": 0.001023,
+      "loss": 5.8449,
+      "step": 341
+    },
+    {
+      "epoch": 0.00342,
+      "grad_norm": 0.8053567409515381,
+      "learning_rate": 0.001026,
+      "loss": 5.8551,
+      "step": 342
+    },
+    {
+      "epoch": 0.00343,
+      "grad_norm": 1.0873545408248901,
+      "learning_rate": 0.0010290000000000002,
+      "loss": 5.8656,
+      "step": 343
+    },
+    {
+      "epoch": 0.00344,
+      "grad_norm": 1.1315717697143555,
+      "learning_rate": 0.001032,
+      "loss": 5.8613,
+      "step": 344
+    },
+    {
+      "epoch": 0.00345,
+      "grad_norm": 0.7909812927246094,
+      "learning_rate": 0.001035,
+      "loss": 5.8449,
+      "step": 345
+    },
+    {
+      "epoch": 0.00346,
+      "grad_norm": 0.8379076719284058,
+      "learning_rate": 0.0010379999999999999,
+      "loss": 5.8518,
+      "step": 346
+    },
+    {
+      "epoch": 0.00347,
+      "grad_norm": 0.5857303738594055,
+      "learning_rate": 0.001041,
+      "loss": 5.8133,
+      "step": 347
+    },
+    {
+      "epoch": 0.00348,
+      "grad_norm": 0.6455392241477966,
+      "learning_rate": 0.001044,
+      "loss": 5.8262,
+      "step": 348
+    },
+    {
+      "epoch": 0.00349,
+      "grad_norm": 0.6315843462944031,
+      "learning_rate": 0.001047,
+      "loss": 5.8046,
+      "step": 349
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 0.6185011863708496,
+      "learning_rate": 0.00105,
+      "loss": 5.8282,
+      "step": 350
+    },
+    {
+      "epoch": 0.00351,
+      "grad_norm": 0.49840763211250305,
+      "learning_rate": 0.001053,
+      "loss": 5.8234,
+      "step": 351
+    },
+    {
+      "epoch": 0.00352,
+      "grad_norm": 0.43951740860939026,
+      "learning_rate": 0.0010559999999999999,
+      "loss": 5.7984,
+      "step": 352
+    },
+    {
+      "epoch": 0.00353,
+      "grad_norm": 0.4452185034751892,
+      "learning_rate": 0.001059,
+      "loss": 5.7993,
+      "step": 353
+    },
+    {
+      "epoch": 0.00354,
+      "grad_norm": 0.43185603618621826,
+      "learning_rate": 0.001062,
+      "loss": 5.7703,
+      "step": 354
+    },
+    {
+      "epoch": 0.00355,
+      "grad_norm": 0.4053448736667633,
+      "learning_rate": 0.001065,
+      "loss": 5.7848,
+      "step": 355
+    },
+    {
+      "epoch": 0.00356,
+      "grad_norm": 0.48363247513771057,
+      "learning_rate": 0.001068,
+      "loss": 5.777,
+      "step": 356
+    },
+    {
+      "epoch": 0.00357,
+      "grad_norm": 0.5601730346679688,
+      "learning_rate": 0.001071,
+      "loss": 5.7693,
+      "step": 357
+    },
+    {
+      "epoch": 0.00358,
+      "grad_norm": 0.7239671349525452,
+      "learning_rate": 0.001074,
+      "loss": 5.7638,
+      "step": 358
+    },
+    {
+      "epoch": 0.00359,
+      "grad_norm": 0.8531132936477661,
+      "learning_rate": 0.001077,
+      "loss": 5.76,
+      "step": 359
+    },
+    {
+      "epoch": 0.0036,
+      "grad_norm": 1.0332695245742798,
+      "learning_rate": 0.00108,
+      "loss": 5.7701,
+      "step": 360
+    },
+    {
+      "epoch": 0.00361,
+      "grad_norm": 1.3098387718200684,
+      "learning_rate": 0.001083,
+      "loss": 5.7755,
+      "step": 361
+    },
+    {
+      "epoch": 0.00362,
+      "grad_norm": 0.7546947598457336,
+      "learning_rate": 0.001086,
+      "loss": 5.7474,
+      "step": 362
+    },
+    {
+      "epoch": 0.00363,
+      "grad_norm": 0.6924042701721191,
+      "learning_rate": 0.001089,
+      "loss": 5.7394,
+      "step": 363
+    },
+    {
+      "epoch": 0.00364,
+      "grad_norm": 0.87959223985672,
+      "learning_rate": 0.001092,
+      "loss": 5.7603,
+      "step": 364
+    },
+    {
+      "epoch": 0.00365,
+      "grad_norm": 1.037275791168213,
+      "learning_rate": 0.001095,
+      "loss": 5.7581,
+      "step": 365
+    },
+    {
+      "epoch": 0.00366,
+      "grad_norm": 1.056171178817749,
+      "learning_rate": 0.001098,
+      "loss": 5.7464,
+      "step": 366
+    },
+    {
+      "epoch": 0.00367,
+      "grad_norm": 1.0635496377944946,
+      "learning_rate": 0.001101,
+      "loss": 5.7478,
+      "step": 367
+    },
+    {
+      "epoch": 0.00368,
+      "grad_norm": 0.9701796174049377,
+      "learning_rate": 0.001104,
+      "loss": 5.7478,
+      "step": 368
+    },
+    {
+      "epoch": 0.00369,
+      "grad_norm": 0.7430213689804077,
+      "learning_rate": 0.001107,
+      "loss": 5.745,
+      "step": 369
+    },
+    {
+      "epoch": 0.0037,
+      "grad_norm": 0.7348084449768066,
+      "learning_rate": 0.00111,
+      "loss": 5.7264,
+      "step": 370
+    },
+    {
+      "epoch": 0.00371,
+      "grad_norm": 0.8778790831565857,
+      "learning_rate": 0.001113,
+      "loss": 5.6986,
+      "step": 371
+    },
+    {
+      "epoch": 0.00372,
+      "grad_norm": 1.160132884979248,
+      "learning_rate": 0.001116,
+      "loss": 5.7343,
+      "step": 372
+    },
+    {
+      "epoch": 0.00373,
+      "grad_norm": 0.8288450241088867,
+      "learning_rate": 0.001119,
+      "loss": 5.7145,
+      "step": 373
+    },
+    {
+      "epoch": 0.00374,
+      "grad_norm": 0.7645081281661987,
+      "learning_rate": 0.001122,
+      "loss": 5.7142,
+      "step": 374
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 0.8342962265014648,
+      "learning_rate": 0.0011250000000000001,
+      "loss": 5.7215,
+      "step": 375
+    },
+    {
+      "epoch": 0.00376,
+      "grad_norm": 0.8966416716575623,
+      "learning_rate": 0.001128,
+      "loss": 5.7142,
+      "step": 376
+    },
+    {
+      "epoch": 0.00377,
+      "grad_norm": 1.1411352157592773,
+      "learning_rate": 0.001131,
+      "loss": 5.7202,
+      "step": 377
+    },
+    {
+      "epoch": 0.00378,
+      "grad_norm": 1.0639731884002686,
+      "learning_rate": 0.001134,
+      "loss": 5.7166,
+      "step": 378
+    },
+    {
+      "epoch": 0.00379,
+      "grad_norm": 1.0386251211166382,
+      "learning_rate": 0.001137,
+      "loss": 5.701,
+      "step": 379
+    },
+    {
+      "epoch": 0.0038,
+      "grad_norm": 0.8551567196846008,
+      "learning_rate": 0.00114,
+      "loss": 5.7166,
+      "step": 380
+    },
+    {
+      "epoch": 0.00381,
+      "grad_norm": 1.171457290649414,
+      "learning_rate": 0.0011430000000000001,
+      "loss": 5.693,
+      "step": 381
+    },
+    {
+      "epoch": 0.00382,
+      "grad_norm": 0.9382472634315491,
+      "learning_rate": 0.001146,
+      "loss": 5.6768,
+      "step": 382
+    },
+    {
+      "epoch": 0.00383,
+      "grad_norm": 1.011130452156067,
+      "learning_rate": 0.001149,
+      "loss": 5.6945,
+      "step": 383
+    },
+    {
+      "epoch": 0.00384,
+      "grad_norm": 0.7897657155990601,
+      "learning_rate": 0.001152,
+      "loss": 5.6883,
+      "step": 384
+    },
+    {
+      "epoch": 0.00385,
+      "grad_norm": 0.6210044622421265,
+      "learning_rate": 0.001155,
+      "loss": 5.673,
+      "step": 385
+    },
+    {
+      "epoch": 0.00386,
+      "grad_norm": 0.5515205264091492,
+      "learning_rate": 0.001158,
+      "loss": 5.6723,
+      "step": 386
+    },
+    {
+      "epoch": 0.00387,
+      "grad_norm": 0.5881383419036865,
+      "learning_rate": 0.0011610000000000001,
+      "loss": 5.6493,
+      "step": 387
+    },
+    {
+      "epoch": 0.00388,
+      "grad_norm": 0.4862520396709442,
+      "learning_rate": 0.001164,
+      "loss": 5.6435,
+      "step": 388
+    },
+    {
+      "epoch": 0.00389,
+      "grad_norm": 0.4490566551685333,
+      "learning_rate": 0.001167,
+      "loss": 5.649,
+      "step": 389
+    },
+    {
+      "epoch": 0.0039,
+      "grad_norm": 0.4499252140522003,
+      "learning_rate": 0.00117,
+      "loss": 5.6299,
+      "step": 390
+    },
+    {
+      "epoch": 0.00391,
+      "grad_norm": 0.3959794044494629,
+      "learning_rate": 0.001173,
+      "loss": 5.6342,
+      "step": 391
+    },
+    {
+      "epoch": 0.00392,
+      "grad_norm": 0.41570809483528137,
+      "learning_rate": 0.001176,
+      "loss": 5.6441,
+      "step": 392
+    },
+    {
+      "epoch": 0.00393,
+      "grad_norm": 0.4579019248485565,
+      "learning_rate": 0.0011790000000000001,
+      "loss": 5.615,
+      "step": 393
+    },
+    {
+      "epoch": 0.00394,
+      "grad_norm": 0.5718971490859985,
+      "learning_rate": 0.001182,
+      "loss": 5.6031,
+      "step": 394
+    },
+    {
+      "epoch": 0.00395,
+      "grad_norm": 0.8492469191551208,
+      "learning_rate": 0.001185,
+      "loss": 5.5961,
+      "step": 395
+    },
+    {
+      "epoch": 0.00396,
+      "grad_norm": 1.173663854598999,
+      "learning_rate": 0.001188,
+      "loss": 5.6242,
+      "step": 396
+    },
+    {
+      "epoch": 0.00397,
+      "grad_norm": 0.828730046749115,
+      "learning_rate": 0.001191,
+      "loss": 5.6071,
+      "step": 397
+    },
+    {
+      "epoch": 0.00398,
+      "grad_norm": 1.2022807598114014,
+      "learning_rate": 0.0011940000000000002,
+      "loss": 5.6278,
+      "step": 398
+    },
+    {
+      "epoch": 0.00399,
+      "grad_norm": 0.9577529430389404,
+      "learning_rate": 0.0011970000000000001,
+      "loss": 5.6234,
+      "step": 399
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 1.0115303993225098,
+      "learning_rate": 0.0012000000000000001,
+      "loss": 5.6174,
+      "step": 400
+    },
+    {
+      "epoch": 0.00401,
+      "grad_norm": 0.9447324872016907,
+      "learning_rate": 0.001203,
+      "loss": 5.594,
+      "step": 401
+    },
+    {
+      "epoch": 0.00402,
+      "grad_norm": 0.8248692154884338,
+      "learning_rate": 0.001206,
+      "loss": 5.5987,
+      "step": 402
+    },
+    {
+      "epoch": 0.00403,
+      "grad_norm": 1.0206302404403687,
+      "learning_rate": 0.001209,
+      "loss": 5.6026,
+      "step": 403
+    },
+    {
+      "epoch": 0.00404,
+      "grad_norm": 1.090654730796814,
+      "learning_rate": 0.0012120000000000002,
+      "loss": 5.6036,
+      "step": 404
+    },
+    {
+      "epoch": 0.00405,
+      "grad_norm": 0.84739750623703,
+      "learning_rate": 0.0012150000000000002,
+      "loss": 5.5901,
+      "step": 405
+    },
+    {
+      "epoch": 0.00406,
+      "grad_norm": 0.9469770193099976,
+      "learning_rate": 0.0012180000000000001,
+      "loss": 5.5817,
+      "step": 406
+    },
+    {
+      "epoch": 0.00407,
+      "grad_norm": 0.9283969402313232,
+      "learning_rate": 0.0012209999999999999,
+      "loss": 5.5746,
+      "step": 407
+    },
+    {
+      "epoch": 0.00408,
+      "grad_norm": 0.8949937224388123,
+      "learning_rate": 0.001224,
+      "loss": 5.5962,
+      "step": 408
+    },
+    {
+      "epoch": 0.00409,
+      "grad_norm": 0.8844306468963623,
+      "learning_rate": 0.001227,
+      "loss": 5.573,
+      "step": 409
+    },
+    {
+      "epoch": 0.0041,
+      "grad_norm": 0.8062122464179993,
+      "learning_rate": 0.00123,
+      "loss": 5.5618,
+      "step": 410
+    },
+    {
+      "epoch": 0.00411,
+      "grad_norm": 0.8730618357658386,
+      "learning_rate": 0.001233,
+      "loss": 5.5872,
+      "step": 411
+    },
+    {
+      "epoch": 0.00412,
+      "grad_norm": 0.665547788143158,
+      "learning_rate": 0.001236,
+      "loss": 5.5644,
+      "step": 412
+    },
+    {
+      "epoch": 0.00413,
+      "grad_norm": 0.8426138758659363,
+      "learning_rate": 0.0012389999999999999,
+      "loss": 5.5674,
+      "step": 413
+    },
+    {
+      "epoch": 0.00414,
+      "grad_norm": 1.051952838897705,
+      "learning_rate": 0.001242,
+      "loss": 5.5591,
+      "step": 414
+    },
+    {
+      "epoch": 0.00415,
+      "grad_norm": 0.7785534858703613,
+      "learning_rate": 0.001245,
+      "loss": 5.5542,
+      "step": 415
+    },
+    {
+      "epoch": 0.00416,
+      "grad_norm": 0.5227160453796387,
+      "learning_rate": 0.001248,
+      "loss": 5.5279,
+      "step": 416
+    },
+    {
+      "epoch": 0.00417,
+      "grad_norm": 0.5770328044891357,
+      "learning_rate": 0.001251,
+      "loss": 5.5611,
+      "step": 417
+    },
+    {
+      "epoch": 0.00418,
+      "grad_norm": 0.4929839074611664,
+      "learning_rate": 0.001254,
+      "loss": 5.5305,
+      "step": 418
+    },
+    {
+      "epoch": 0.00419,
+      "grad_norm": 0.4660792648792267,
+      "learning_rate": 0.0012569999999999999,
+      "loss": 5.5168,
+      "step": 419
+    },
+    {
+      "epoch": 0.0042,
+      "grad_norm": 0.5160586833953857,
+      "learning_rate": 0.00126,
+      "loss": 5.5326,
+      "step": 420
+    },
+    {
+      "epoch": 0.00421,
+      "grad_norm": 0.5846797823905945,
+      "learning_rate": 0.001263,
+      "loss": 5.5249,
+      "step": 421
+    },
+    {
+      "epoch": 0.00422,
+      "grad_norm": 0.6270997524261475,
+      "learning_rate": 0.001266,
+      "loss": 5.5159,
+      "step": 422
+    },
+    {
+      "epoch": 0.00423,
+      "grad_norm": 0.6081735491752625,
+      "learning_rate": 0.001269,
+      "loss": 5.5118,
+      "step": 423
+    },
+    {
+      "epoch": 0.00424,
+      "grad_norm": 0.557420551776886,
+      "learning_rate": 0.001272,
+      "loss": 5.5105,
+      "step": 424
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 0.821638286113739,
+      "learning_rate": 0.001275,
+      "loss": 5.5176,
+      "step": 425
+    },
+    {
+      "epoch": 0.00426,
+      "grad_norm": 1.0497279167175293,
+      "learning_rate": 0.001278,
+      "loss": 5.5294,
+      "step": 426
+    },
+    {
+      "epoch": 0.00427,
+      "grad_norm": 0.8568355441093445,
+      "learning_rate": 0.001281,
+      "loss": 5.5072,
+      "step": 427
+    },
+    {
+      "epoch": 0.00428,
+      "grad_norm": 0.9392327070236206,
+      "learning_rate": 0.001284,
+      "loss": 5.488,
+      "step": 428
+    },
+    {
+      "epoch": 0.00429,
+      "grad_norm": 0.8972091674804688,
+      "learning_rate": 0.001287,
+      "loss": 5.5041,
+      "step": 429
+    },
+    {
+      "epoch": 0.0043,
+      "grad_norm": 0.8478754162788391,
+      "learning_rate": 0.00129,
+      "loss": 5.4993,
+      "step": 430
+    },
+    {
+      "epoch": 0.00431,
+      "grad_norm": 1.0406945943832397,
+      "learning_rate": 0.001293,
+      "loss": 5.4959,
+      "step": 431
+    },
+    {
+      "epoch": 0.00432,
+      "grad_norm": 0.8937470316886902,
+      "learning_rate": 0.001296,
+      "loss": 5.5023,
+      "step": 432
+    },
+    {
+      "epoch": 0.00433,
+      "grad_norm": 0.7159745693206787,
+      "learning_rate": 0.001299,
+      "loss": 5.4892,
+      "step": 433
+    },
+    {
+      "epoch": 0.00434,
+      "grad_norm": 0.6872638463973999,
+      "learning_rate": 0.001302,
+      "loss": 5.4897,
+      "step": 434
+    },
+    {
+      "epoch": 0.00435,
+      "grad_norm": 0.7336323857307434,
+      "learning_rate": 0.001305,
+      "loss": 5.4832,
+      "step": 435
+    },
+    {
+      "epoch": 0.00436,
+      "grad_norm": 0.727497935295105,
+      "learning_rate": 0.001308,
+      "loss": 5.4615,
+      "step": 436
+    },
+    {
+      "epoch": 0.00437,
+      "grad_norm": 0.8729015588760376,
+      "learning_rate": 0.001311,
+      "loss": 5.4821,
+      "step": 437
+    },
+    {
+      "epoch": 0.00438,
+      "grad_norm": 1.0639538764953613,
+      "learning_rate": 0.001314,
+      "loss": 5.4733,
+      "step": 438
+    },
+    {
+      "epoch": 0.00439,
+      "grad_norm": 0.9613514542579651,
+      "learning_rate": 0.001317,
+      "loss": 5.5005,
+      "step": 439
+    },
+    {
+      "epoch": 0.0044,
+      "grad_norm": 0.7993902564048767,
+      "learning_rate": 0.00132,
+      "loss": 5.4683,
+      "step": 440
+    },
+    {
+      "epoch": 0.00441,
+      "grad_norm": 0.6981948018074036,
+      "learning_rate": 0.001323,
+      "loss": 5.4617,
+      "step": 441
+    },
+    {
+      "epoch": 0.00442,
+      "grad_norm": 0.5610657334327698,
+      "learning_rate": 0.0013260000000000001,
+      "loss": 5.4493,
+      "step": 442
+    },
+    {
+      "epoch": 0.00443,
+      "grad_norm": 0.6162020564079285,
+      "learning_rate": 0.001329,
+      "loss": 5.4545,
+      "step": 443
+    },
+    {
+      "epoch": 0.00444,
+      "grad_norm": 0.5797529816627502,
+      "learning_rate": 0.001332,
+      "loss": 5.4538,
+      "step": 444
+    },
+    {
+      "epoch": 0.00445,
+      "grad_norm": 0.551798939704895,
+      "learning_rate": 0.001335,
+      "loss": 5.4358,
+      "step": 445
+    },
+    {
+      "epoch": 0.00446,
+      "grad_norm": 0.48300743103027344,
+      "learning_rate": 0.001338,
+      "loss": 5.4369,
+      "step": 446
+    },
+    {
+      "epoch": 0.00447,
+      "grad_norm": 0.5713039040565491,
+      "learning_rate": 0.001341,
+      "loss": 5.4366,
+      "step": 447
+    },
+    {
+      "epoch": 0.00448,
+      "grad_norm": 0.7566826939582825,
+      "learning_rate": 0.0013440000000000001,
+      "loss": 5.4234,
+      "step": 448
+    },
+    {
+      "epoch": 0.00449,
+      "grad_norm": 1.1563501358032227,
+      "learning_rate": 0.001347,
+      "loss": 5.4418,
+      "step": 449
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 1.132352590560913,
+      "learning_rate": 0.00135,
+      "loss": 5.4433,
+      "step": 450
+    },
+    {
+      "epoch": 0.00451,
+      "grad_norm": 1.0986182689666748,
+      "learning_rate": 0.001353,
+      "loss": 5.4341,
+      "step": 451
+    },
+    {
+      "epoch": 0.00452,
+      "grad_norm": 1.066072702407837,
+      "learning_rate": 0.001356,
+      "loss": 5.4212,
+      "step": 452
+    },
+    {
+      "epoch": 0.00453,
+      "grad_norm": 0.9297358989715576,
+      "learning_rate": 0.001359,
+      "loss": 5.4103,
+      "step": 453
+    },
+    {
+      "epoch": 0.00454,
+      "grad_norm": 0.9204379320144653,
+      "learning_rate": 0.0013620000000000001,
+      "loss": 5.4294,
+      "step": 454
+    },
+    {
+      "epoch": 0.00455,
+      "grad_norm": 1.0156644582748413,
+      "learning_rate": 0.0013650000000000001,
+      "loss": 5.4283,
+      "step": 455
+    },
+    {
+      "epoch": 0.00456,
+      "grad_norm": 1.070080041885376,
+      "learning_rate": 0.001368,
+      "loss": 5.4307,
+      "step": 456
+    },
+    {
+      "epoch": 0.00457,
+      "grad_norm": 0.7985509634017944,
+      "learning_rate": 0.001371,
+      "loss": 5.4131,
+      "step": 457
+    },
+    {
+      "epoch": 0.00458,
+      "grad_norm": 0.8899184465408325,
+      "learning_rate": 0.001374,
+      "loss": 5.4253,
+      "step": 458
+    },
+    {
+      "epoch": 0.00459,
+      "grad_norm": 1.0247424840927124,
+      "learning_rate": 0.0013770000000000002,
+      "loss": 5.4196,
+      "step": 459
+    },
+    {
+      "epoch": 0.0046,
+      "grad_norm": 0.8818691968917847,
+      "learning_rate": 0.0013800000000000002,
+      "loss": 5.4023,
+      "step": 460
+    },
+    {
+      "epoch": 0.00461,
+      "grad_norm": 0.9398977160453796,
+      "learning_rate": 0.0013830000000000001,
+      "loss": 5.4028,
+      "step": 461
+    },
+    {
+      "epoch": 0.00462,
+      "grad_norm": 0.9095609784126282,
+      "learning_rate": 0.001386,
+      "loss": 5.4141,
+      "step": 462
+    },
+    {
+      "epoch": 0.00463,
+      "grad_norm": 0.7167875170707703,
+      "learning_rate": 0.001389,
+      "loss": 5.3816,
+      "step": 463
+    },
+    {
+      "epoch": 0.00464,
+      "grad_norm": 0.8864797353744507,
+      "learning_rate": 0.001392,
+      "loss": 5.4115,
+      "step": 464
+    },
+    {
+      "epoch": 0.00465,
+      "grad_norm": 0.8739372491836548,
+      "learning_rate": 0.0013950000000000002,
+      "loss": 5.3819,
+      "step": 465
+    },
+    {
+      "epoch": 0.00466,
+      "grad_norm": 0.808113157749176,
+      "learning_rate": 0.0013980000000000002,
+      "loss": 5.3947,
+      "step": 466
+    },
+    {
+      "epoch": 0.00467,
+      "grad_norm": 0.878140389919281,
+      "learning_rate": 0.0014010000000000001,
+      "loss": 5.3861,
+      "step": 467
+    },
+    {
+      "epoch": 0.00468,
+      "grad_norm": 1.0618577003479004,
+      "learning_rate": 0.001404,
+      "loss": 5.3874,
+      "step": 468
+    },
+    {
+      "epoch": 0.00469,
+      "grad_norm": 0.8678603172302246,
+      "learning_rate": 0.001407,
+      "loss": 5.3793,
+      "step": 469
+    },
+    {
+      "epoch": 0.0047,
+      "grad_norm": 1.0598393678665161,
+      "learning_rate": 0.00141,
+      "loss": 5.397,
+      "step": 470
+    },
+    {
+      "epoch": 0.00471,
+      "grad_norm": 0.9507164359092712,
+      "learning_rate": 0.001413,
+      "loss": 5.3744,
+      "step": 471
+    },
+    {
+      "epoch": 0.00472,
+      "grad_norm": 0.891309916973114,
+      "learning_rate": 0.001416,
+      "loss": 5.3876,
+      "step": 472
+    },
+    {
+      "epoch": 0.00473,
+      "grad_norm": 0.9032427072525024,
+      "learning_rate": 0.001419,
+      "loss": 5.3933,
+      "step": 473
+    },
+    {
+      "epoch": 0.00474,
+      "grad_norm": 1.2588310241699219,
+      "learning_rate": 0.0014219999999999999,
+      "loss": 5.3882,
+      "step": 474
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 0.8014562129974365,
+      "learning_rate": 0.001425,
+      "loss": 5.367,
+      "step": 475
+    },
+    {
+      "epoch": 0.00476,
+      "grad_norm": 0.7612058520317078,
+      "learning_rate": 0.001428,
+      "loss": 5.3988,
+      "step": 476
+    },
+    {
+      "epoch": 0.00477,
+      "grad_norm": 0.6699860095977783,
+      "learning_rate": 0.001431,
+      "loss": 5.3462,
+      "step": 477
+    },
+    {
+      "epoch": 0.00478,
+      "grad_norm": 0.7476372718811035,
+      "learning_rate": 0.001434,
+      "loss": 5.3761,
+      "step": 478
+    },
+    {
+      "epoch": 0.00479,
+      "grad_norm": 0.7134982347488403,
+      "learning_rate": 0.001437,
+      "loss": 5.3522,
+      "step": 479
+    },
+    {
+      "epoch": 0.0048,
+      "grad_norm": 0.5948371887207031,
+      "learning_rate": 0.0014399999999999999,
+      "loss": 5.343,
+      "step": 480
+    },
+    {
+      "epoch": 0.00481,
+      "grad_norm": 0.4980184733867645,
+      "learning_rate": 0.001443,
+      "loss": 5.3421,
+      "step": 481
+    },
+    {
+      "epoch": 0.00482,
+      "grad_norm": 0.49235713481903076,
+      "learning_rate": 0.001446,
+      "loss": 5.3164,
+      "step": 482
+    },
+    {
+      "epoch": 0.00483,
+      "grad_norm": 0.5064442157745361,
+      "learning_rate": 0.001449,
+      "loss": 5.327,
+      "step": 483
+    },
+    {
+      "epoch": 0.00484,
+      "grad_norm": 0.506096601486206,
+      "learning_rate": 0.001452,
+      "loss": 5.3265,
+      "step": 484
+    },
+    {
+      "epoch": 0.00485,
+      "grad_norm": 0.5634677410125732,
+      "learning_rate": 0.001455,
+      "loss": 5.3337,
+      "step": 485
+    },
+    {
+      "epoch": 0.00486,
+      "grad_norm": 0.5971659421920776,
+      "learning_rate": 0.001458,
+      "loss": 5.306,
+      "step": 486
+    },
+    {
+      "epoch": 0.00487,
+      "grad_norm": 0.5582761168479919,
+      "learning_rate": 0.001461,
+      "loss": 5.3211,
+      "step": 487
+    },
+    {
+      "epoch": 0.00488,
+      "grad_norm": 0.5640081763267517,
+      "learning_rate": 0.001464,
+      "loss": 5.2971,
+      "step": 488
+    },
+    {
+      "epoch": 0.00489,
+      "grad_norm": 0.6793756484985352,
+      "learning_rate": 0.001467,
+      "loss": 5.2905,
+      "step": 489
+    },
+    {
+      "epoch": 0.0049,
+      "grad_norm": 0.7497550249099731,
+      "learning_rate": 0.00147,
+      "loss": 5.3082,
+      "step": 490
+    },
+    {
+      "epoch": 0.00491,
+      "grad_norm": 0.7264507412910461,
+      "learning_rate": 0.001473,
+      "loss": 5.3062,
+      "step": 491
+    },
+    {
+      "epoch": 0.00492,
+      "grad_norm": 0.5965330004692078,
+      "learning_rate": 0.001476,
+      "loss": 5.3011,
+      "step": 492
+    },
+    {
+      "epoch": 0.00493,
+      "grad_norm": 0.8212659358978271,
+      "learning_rate": 0.001479,
+      "loss": 5.3006,
+      "step": 493
+    },
+    {
+      "epoch": 0.00494,
+      "grad_norm": 1.2706849575042725,
+      "learning_rate": 0.001482,
+      "loss": 5.3221,
+      "step": 494
+    },
+    {
+      "epoch": 0.00495,
+      "grad_norm": 0.9726585149765015,
+      "learning_rate": 0.001485,
+      "loss": 5.3189,
+      "step": 495
+    },
+    {
+      "epoch": 0.00496,
+      "grad_norm": 1.0229647159576416,
+      "learning_rate": 0.001488,
+      "loss": 5.3095,
+      "step": 496
+    },
+    {
+      "epoch": 0.00497,
+      "grad_norm": 0.9751450419425964,
+      "learning_rate": 0.001491,
+      "loss": 5.294,
+      "step": 497
+    },
+    {
+      "epoch": 0.00498,
+      "grad_norm": 0.9788212776184082,
+      "learning_rate": 0.001494,
+      "loss": 5.3219,
+      "step": 498
+    },
+    {
+      "epoch": 0.00499,
+      "grad_norm": 0.897365391254425,
+      "learning_rate": 0.001497,
+      "loss": 5.2943,
+      "step": 499
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 0.8972038626670837,
+      "learning_rate": 0.0015,
+      "loss": 5.2998,
+      "step": 500
+    },
+    {
+      "epoch": 0.00501,
+      "grad_norm": 1.123322606086731,
+      "learning_rate": 0.001503,
+      "loss": 5.3107,
+      "step": 501
+    },
+    {
+      "epoch": 0.00502,
+      "grad_norm": 1.085119366645813,
+      "learning_rate": 0.001506,
+      "loss": 5.3147,
+      "step": 502
+    },
+    {
+      "epoch": 0.00503,
+      "grad_norm": 0.9612423181533813,
+      "learning_rate": 0.0015090000000000001,
+      "loss": 5.3162,
+      "step": 503
+    },
+    {
+      "epoch": 0.00504,
+      "grad_norm": 1.0992624759674072,
+      "learning_rate": 0.001512,
+      "loss": 5.3083,
+      "step": 504
+    },
+    {
+      "epoch": 0.00505,
+      "grad_norm": 0.9857581257820129,
+      "learning_rate": 0.001515,
+      "loss": 5.2946,
+      "step": 505
+    },
+    {
+      "epoch": 0.00506,
+      "grad_norm": 1.1049542427062988,
+      "learning_rate": 0.001518,
+      "loss": 5.3003,
+      "step": 506
+    },
+    {
+      "epoch": 0.00507,
+      "grad_norm": 0.8998914957046509,
+      "learning_rate": 0.001521,
+      "loss": 5.298,
+      "step": 507
+    },
+    {
+      "epoch": 0.00508,
+      "grad_norm": 0.6991334557533264,
+      "learning_rate": 0.001524,
+      "loss": 5.2862,
+      "step": 508
+    },
+    {
+      "epoch": 0.00509,
+      "grad_norm": 0.7653549313545227,
+      "learning_rate": 0.0015270000000000001,
+      "loss": 5.2723,
+      "step": 509
+    },
+    {
+      "epoch": 0.0051,
+      "grad_norm": 0.7315691113471985,
+      "learning_rate": 0.0015300000000000001,
+      "loss": 5.288,
+      "step": 510
+    },
+    {
+      "epoch": 0.00511,
+      "grad_norm": 0.7975103855133057,
+      "learning_rate": 0.001533,
+      "loss": 5.2648,
+      "step": 511
+    },
+    {
+      "epoch": 0.00512,
+      "grad_norm": 0.9781049489974976,
+      "learning_rate": 0.001536,
+      "loss": 5.2672,
+      "step": 512
+    },
+    {
+      "epoch": 0.00513,
+      "grad_norm": 1.084666132926941,
+      "learning_rate": 0.001539,
+      "loss": 5.2896,
+      "step": 513
+    },
+    {
+      "epoch": 0.00514,
+      "grad_norm": 0.9010921120643616,
+      "learning_rate": 0.001542,
+      "loss": 5.2706,
+      "step": 514
+    },
+    {
+      "epoch": 0.00515,
+      "grad_norm": 0.96586012840271,
+      "learning_rate": 0.0015450000000000001,
+      "loss": 5.2764,
+      "step": 515
+    },
+    {
+      "epoch": 0.00516,
+      "grad_norm": 0.9655681848526001,
+      "learning_rate": 0.0015480000000000001,
+      "loss": 5.2769,
+      "step": 516
+    },
+    {
+      "epoch": 0.00517,
+      "grad_norm": 0.8448560833930969,
+      "learning_rate": 0.001551,
+      "loss": 5.2634,
+      "step": 517
+    },
+    {
+      "epoch": 0.00518,
+      "grad_norm": 0.7809770703315735,
+      "learning_rate": 0.001554,
+      "loss": 5.2515,
+      "step": 518
+    },
+    {
+      "epoch": 0.00519,
+      "grad_norm": 0.913107693195343,
+      "learning_rate": 0.001557,
+      "loss": 5.2572,
+      "step": 519
+    },
+    {
+      "epoch": 0.0052,
+      "grad_norm": 0.9221185445785522,
+      "learning_rate": 0.0015600000000000002,
+      "loss": 5.2658,
+      "step": 520
+    },
+    {
+      "epoch": 0.00521,
+      "grad_norm": 0.8511962294578552,
+      "learning_rate": 0.0015630000000000002,
+      "loss": 5.2423,
+      "step": 521
+    },
+    {
+      "epoch": 0.00522,
+      "grad_norm": 0.7266805171966553,
+      "learning_rate": 0.0015660000000000001,
+      "loss": 5.2383,
+      "step": 522
+    },
+    {
+      "epoch": 0.00523,
+      "grad_norm": 0.8215247392654419,
+      "learning_rate": 0.001569,
+      "loss": 5.2547,
+      "step": 523
+    },
+    {
+      "epoch": 0.00524,
+      "grad_norm": 0.8742693066596985,
+      "learning_rate": 0.001572,
+      "loss": 5.2653,
+      "step": 524
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 0.6882407069206238,
+      "learning_rate": 0.001575,
+      "loss": 5.2501,
+      "step": 525
+    },
+    {
+      "epoch": 0.00526,
+      "grad_norm": 0.7224147319793701,
+      "learning_rate": 0.0015780000000000002,
+      "loss": 5.2254,
+      "step": 526
+    },
+    {
+      "epoch": 0.00527,
+      "grad_norm": 0.6562958359718323,
+      "learning_rate": 0.0015810000000000002,
+      "loss": 5.2385,
+      "step": 527
+    },
+    {
+      "epoch": 0.00528,
+      "grad_norm": 0.6051112413406372,
+      "learning_rate": 0.0015840000000000001,
+      "loss": 5.2165,
+      "step": 528
+    },
+    {
+      "epoch": 0.00529,
+      "grad_norm": 0.647803008556366,
+      "learning_rate": 0.001587,
+      "loss": 5.2244,
+      "step": 529
+    },
+    {
+      "epoch": 0.0053,
+      "grad_norm": 0.6608071327209473,
+      "learning_rate": 0.00159,
+      "loss": 5.2339,
+      "step": 530
+    },
+    {
+      "epoch": 0.00531,
+      "grad_norm": 0.6765715479850769,
+      "learning_rate": 0.001593,
+      "loss": 5.2181,
+      "step": 531
+    },
+    {
+      "epoch": 0.00532,
+      "grad_norm": 0.7380223870277405,
+      "learning_rate": 0.0015960000000000002,
+      "loss": 5.2256,
+      "step": 532
+    },
+    {
+      "epoch": 0.00533,
+      "grad_norm": 0.7506837248802185,
+      "learning_rate": 0.0015990000000000002,
+      "loss": 5.2236,
+      "step": 533
+    },
+    {
+      "epoch": 0.00534,
+      "grad_norm": 0.577497661113739,
+      "learning_rate": 0.0016020000000000001,
+      "loss": 5.2066,
+      "step": 534
+    },
+    {
+      "epoch": 0.00535,
+      "grad_norm": 0.4974853992462158,
+      "learning_rate": 0.001605,
+      "loss": 5.2096,
+      "step": 535
+    },
+    {
+      "epoch": 0.00536,
+      "grad_norm": 0.5754765272140503,
+      "learning_rate": 0.001608,
+      "loss": 5.2121,
+      "step": 536
+    },
+    {
+      "epoch": 0.00537,
+      "grad_norm": 0.6681102514266968,
+      "learning_rate": 0.0016110000000000002,
+      "loss": 5.2067,
+      "step": 537
+    },
+    {
+      "epoch": 0.00538,
+      "grad_norm": 0.8286970257759094,
+      "learning_rate": 0.0016140000000000002,
+      "loss": 5.1882,
+      "step": 538
+    },
+    {
+      "epoch": 0.00539,
+      "grad_norm": 1.0212045907974243,
+      "learning_rate": 0.0016170000000000002,
+      "loss": 5.193,
+      "step": 539
+    },
+    {
+      "epoch": 0.0054,
+      "grad_norm": 1.0495171546936035,
+      "learning_rate": 0.0016200000000000001,
+      "loss": 5.2061,
+      "step": 540
+    },
+    {
+      "epoch": 0.00541,
+      "grad_norm": 0.9756328463554382,
+      "learning_rate": 0.001623,
+      "loss": 5.1953,
+      "step": 541
+    },
+    {
+      "epoch": 0.00542,
+      "grad_norm": 1.024538278579712,
+      "learning_rate": 0.001626,
+      "loss": 5.2247,
+      "step": 542
+    },
+    {
+      "epoch": 0.00543,
+      "grad_norm": 1.005081057548523,
+      "learning_rate": 0.0016290000000000002,
+      "loss": 5.1883,
+      "step": 543
+    },
+    {
+      "epoch": 0.00544,
+      "grad_norm": 1.1540062427520752,
+      "learning_rate": 0.0016320000000000002,
+      "loss": 5.2159,
+      "step": 544
+    },
+    {
+      "epoch": 0.00545,
+      "grad_norm": 1.0779460668563843,
+      "learning_rate": 0.0016350000000000002,
+      "loss": 5.2271,
+      "step": 545
+    },
+    {
+      "epoch": 0.00546,
+      "grad_norm": 0.9231882691383362,
+      "learning_rate": 0.0016380000000000001,
+      "loss": 5.1917,
+      "step": 546
+    },
+    {
+      "epoch": 0.00547,
+      "grad_norm": 0.9463688731193542,
+      "learning_rate": 0.001641,
+      "loss": 5.1943,
+      "step": 547
+    },
+    {
+      "epoch": 0.00548,
+      "grad_norm": 0.8151195645332336,
+      "learning_rate": 0.001644,
+      "loss": 5.1898,
+      "step": 548
+    },
+    {
+      "epoch": 0.00549,
+      "grad_norm": 0.6324855089187622,
+      "learning_rate": 0.0016470000000000002,
+      "loss": 5.1802,
+      "step": 549
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 0.7012510299682617,
+      "learning_rate": 0.0016500000000000002,
+      "loss": 5.1963,
+      "step": 550
+    },
+    {
+      "epoch": 0.00551,
+      "grad_norm": 0.6119561195373535,
+      "learning_rate": 0.0016530000000000002,
+      "loss": 5.1867,
+      "step": 551
+    },
+    {
+      "epoch": 0.00552,
+      "grad_norm": 0.6565516591072083,
+      "learning_rate": 0.0016560000000000001,
+      "loss": 5.1814,
+      "step": 552
+    },
+    {
+      "epoch": 0.00553,
+      "grad_norm": 0.60141921043396,
+      "learning_rate": 0.001659,
+      "loss": 5.1716,
+      "step": 553
+    },
+    {
+      "epoch": 0.00554,
+      "grad_norm": 0.5632038116455078,
+      "learning_rate": 0.0016620000000000003,
+      "loss": 5.1664,
+      "step": 554
+    },
+    {
+      "epoch": 0.00555,
+      "grad_norm": 0.7644810080528259,
+      "learning_rate": 0.0016650000000000002,
+      "loss": 5.1782,
+      "step": 555
+    },
+    {
+      "epoch": 0.00556,
+      "grad_norm": 0.8742493987083435,
+      "learning_rate": 0.0016680000000000002,
+      "loss": 5.1659,
+      "step": 556
+    },
+    {
+      "epoch": 0.00557,
+      "grad_norm": 0.8544741272926331,
+      "learning_rate": 0.0016710000000000002,
+      "loss": 5.1502,
+      "step": 557
+    },
+    {
+      "epoch": 0.00558,
+      "grad_norm": 0.9127110838890076,
+      "learning_rate": 0.0016740000000000001,
+      "loss": 5.182,
+      "step": 558
+    },
+    {
+      "epoch": 0.00559,
+      "grad_norm": 0.9648676514625549,
+      "learning_rate": 0.001677,
+      "loss": 5.1678,
+      "step": 559
+    },
+    {
+      "epoch": 0.0056,
+      "grad_norm": 0.8821786642074585,
+      "learning_rate": 0.0016800000000000003,
+      "loss": 5.1427,
+      "step": 560
+    },
+    {
+      "epoch": 0.00561,
+      "grad_norm": 0.891691267490387,
+      "learning_rate": 0.0016830000000000003,
+      "loss": 5.1733,
+      "step": 561
+    },
+    {
+      "epoch": 0.00562,
+      "grad_norm": 0.9146907925605774,
+      "learning_rate": 0.0016860000000000002,
+      "loss": 5.1636,
+      "step": 562
+    },
+    {
+      "epoch": 0.00563,
+      "grad_norm": 0.9244825839996338,
+      "learning_rate": 0.001689,
+      "loss": 5.1722,
+      "step": 563
+    },
+    {
+      "epoch": 0.00564,
+      "grad_norm": 0.7839826345443726,
+      "learning_rate": 0.001692,
+      "loss": 5.1431,
+      "step": 564
+    },
+    {
+      "epoch": 0.00565,
+      "grad_norm": 1.062433123588562,
+      "learning_rate": 0.001695,
+      "loss": 5.1564,
+      "step": 565
+    },
+    {
+      "epoch": 0.00566,
+      "grad_norm": 0.8471325635910034,
+      "learning_rate": 0.0016979999999999999,
+      "loss": 5.1512,
+      "step": 566
+    },
+    {
+      "epoch": 0.00567,
+      "grad_norm": 0.8534058928489685,
+      "learning_rate": 0.0017009999999999998,
+      "loss": 5.1473,
+      "step": 567
+    },
+    {
+      "epoch": 0.00568,
+      "grad_norm": 0.8670461177825928,
+      "learning_rate": 0.0017039999999999998,
+      "loss": 5.1567,
+      "step": 568
+    },
+    {
+      "epoch": 0.00569,
+      "grad_norm": 0.9165869951248169,
+      "learning_rate": 0.001707,
+      "loss": 5.1375,
+      "step": 569
+    },
+    {
+      "epoch": 0.0057,
+      "grad_norm": 1.1377770900726318,
+      "learning_rate": 0.00171,
+      "loss": 5.1664,
+      "step": 570
+    },
+    {
+      "epoch": 0.00571,
+      "grad_norm": 0.8107508420944214,
+      "learning_rate": 0.001713,
+      "loss": 5.1589,
+      "step": 571
+    },
+    {
+      "epoch": 0.00572,
+      "grad_norm": 0.7913006544113159,
+      "learning_rate": 0.0017159999999999999,
+      "loss": 5.1345,
+      "step": 572
+    },
+    {
+      "epoch": 0.00573,
+      "grad_norm": 0.7625595927238464,
+      "learning_rate": 0.0017189999999999998,
+      "loss": 5.1388,
+      "step": 573
+    },
+    {
+      "epoch": 0.00574,
+      "grad_norm": 0.8708691596984863,
+      "learning_rate": 0.001722,
+      "loss": 5.1477,
+      "step": 574
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 0.8358116149902344,
+      "learning_rate": 0.001725,
+      "loss": 5.126,
+      "step": 575
+    },
+    {
+      "epoch": 0.00576,
+      "grad_norm": 0.7503964900970459,
+      "learning_rate": 0.001728,
+      "loss": 5.1165,
+      "step": 576
+    },
+    {
+      "epoch": 0.00577,
+      "grad_norm": 0.743698000907898,
+      "learning_rate": 0.001731,
+      "loss": 5.1434,
+      "step": 577
+    },
+    {
+      "epoch": 0.00578,
+      "grad_norm": 0.7937496900558472,
+      "learning_rate": 0.0017339999999999999,
+      "loss": 5.1222,
+      "step": 578
+    },
+    {
+      "epoch": 0.00579,
+      "grad_norm": 0.7887423038482666,
+      "learning_rate": 0.0017369999999999998,
+      "loss": 5.1084,
+      "step": 579
+    },
+    {
+      "epoch": 0.0058,
+      "grad_norm": 0.7772188186645508,
+      "learning_rate": 0.00174,
+      "loss": 5.1272,
+      "step": 580
+    },
+    {
+      "epoch": 0.00581,
+      "grad_norm": 0.8483501076698303,
+      "learning_rate": 0.001743,
+      "loss": 5.1251,
+      "step": 581
+    },
+    {
+      "epoch": 0.00582,
+      "grad_norm": 1.0748672485351562,
+      "learning_rate": 0.001746,
+      "loss": 5.1133,
+      "step": 582
+    },
+    {
+      "epoch": 0.00583,
+      "grad_norm": 0.9845912456512451,
+      "learning_rate": 0.001749,
+      "loss": 5.1338,
+      "step": 583
+    },
+    {
+      "epoch": 0.00584,
+      "grad_norm": 1.0171496868133545,
+      "learning_rate": 0.0017519999999999999,
+      "loss": 5.1328,
+      "step": 584
+    },
+    {
+      "epoch": 0.00585,
+      "grad_norm": 0.932063102722168,
+      "learning_rate": 0.0017549999999999998,
+      "loss": 5.1125,
+      "step": 585
+    },
+    {
+      "epoch": 0.00586,
+      "grad_norm": 1.0053131580352783,
+      "learning_rate": 0.001758,
+      "loss": 5.1277,
+      "step": 586
+    },
+    {
+      "epoch": 0.00587,
+      "grad_norm": 1.0553542375564575,
+      "learning_rate": 0.001761,
+      "loss": 5.1113,
+      "step": 587
+    },
+    {
+      "epoch": 0.00588,
+      "grad_norm": 0.9641870260238647,
+      "learning_rate": 0.001764,
+      "loss": 5.1209,
+      "step": 588
+    },
+    {
+      "epoch": 0.00589,
+      "grad_norm": 0.7223602533340454,
+      "learning_rate": 0.001767,
+      "loss": 5.1139,
+      "step": 589
+    },
+    {
+      "epoch": 0.0059,
+      "grad_norm": 0.7580032348632812,
+      "learning_rate": 0.0017699999999999999,
+      "loss": 5.1016,
+      "step": 590
+    },
+    {
+      "epoch": 0.00591,
+      "grad_norm": 0.7553709149360657,
+      "learning_rate": 0.001773,
+      "loss": 5.097,
+      "step": 591
+    },
+    {
+      "epoch": 0.00592,
+      "grad_norm": 0.7395292520523071,
+      "learning_rate": 0.001776,
+      "loss": 5.122,
+      "step": 592
+    },
+    {
+      "epoch": 0.00593,
+      "grad_norm": 0.6007040739059448,
+      "learning_rate": 0.001779,
+      "loss": 5.1118,
+      "step": 593
+    },
+    {
+      "epoch": 0.00594,
+      "grad_norm": 0.6126047372817993,
+      "learning_rate": 0.001782,
+      "loss": 5.1081,
+      "step": 594
+    },
+    {
+      "epoch": 0.00595,
+      "grad_norm": 0.607721745967865,
+      "learning_rate": 0.001785,
+      "loss": 5.0957,
+      "step": 595
+    },
+    {
+      "epoch": 0.00596,
+      "grad_norm": 0.6556451916694641,
+      "learning_rate": 0.0017879999999999999,
+      "loss": 5.0948,
+      "step": 596
+    },
+    {
+      "epoch": 0.00597,
+      "grad_norm": 0.7172878384590149,
+      "learning_rate": 0.001791,
+      "loss": 5.0729,
+      "step": 597
+    },
+    {
+      "epoch": 0.00598,
+      "grad_norm": 0.6043835878372192,
+      "learning_rate": 0.001794,
+      "loss": 5.0805,
+      "step": 598
+    },
+    {
+      "epoch": 0.00599,
+      "grad_norm": 0.5666232109069824,
+      "learning_rate": 0.001797,
+      "loss": 5.0796,
+      "step": 599
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 0.5673431754112244,
+      "learning_rate": 0.0018,
+      "loss": 5.0764,
+      "step": 600
+    },
+    {
+      "epoch": 0.00601,
+      "grad_norm": 0.5798671841621399,
+      "learning_rate": 0.001803,
+      "loss": 5.0708,
+      "step": 601
+    },
+    {
+      "epoch": 0.00602,
+      "grad_norm": 0.520574152469635,
+      "learning_rate": 0.0018059999999999999,
+      "loss": 5.0544,
+      "step": 602
+    },
+    {
+      "epoch": 0.00603,
+      "grad_norm": 0.5210540294647217,
+      "learning_rate": 0.001809,
+      "loss": 5.0732,
+      "step": 603
+    },
+    {
+      "epoch": 0.00604,
+      "grad_norm": 0.6759857535362244,
+      "learning_rate": 0.001812,
+      "loss": 5.0576,
+      "step": 604
+    },
+    {
+      "epoch": 0.00605,
+      "grad_norm": 0.7568657994270325,
+      "learning_rate": 0.001815,
+      "loss": 5.0701,
+      "step": 605
+    },
+    {
+      "epoch": 0.00606,
+      "grad_norm": 0.7632762789726257,
+      "learning_rate": 0.001818,
+      "loss": 5.0711,
+      "step": 606
+    },
+    {
+      "epoch": 0.00607,
+      "grad_norm": 0.788451611995697,
+      "learning_rate": 0.001821,
+      "loss": 5.073,
+      "step": 607
+    },
+    {
+      "epoch": 0.00608,
+      "grad_norm": 0.763842761516571,
+      "learning_rate": 0.001824,
+      "loss": 5.0575,
+      "step": 608
+    },
+    {
+      "epoch": 0.00609,
+      "grad_norm": 0.825861930847168,
+      "learning_rate": 0.001827,
+      "loss": 5.0665,
+      "step": 609
+    },
+    {
+      "epoch": 0.0061,
+      "grad_norm": 1.0143935680389404,
+      "learning_rate": 0.00183,
+      "loss": 5.0488,
+      "step": 610
+    },
+    {
+      "epoch": 0.00611,
+      "grad_norm": 1.1116124391555786,
+      "learning_rate": 0.001833,
+      "loss": 5.0562,
+      "step": 611
+    },
+    {
+      "epoch": 0.00612,
+      "grad_norm": 1.0579830408096313,
+      "learning_rate": 0.001836,
+      "loss": 5.0621,
+      "step": 612
+    },
+    {
+      "epoch": 0.00613,
+      "grad_norm": 1.2180272340774536,
+      "learning_rate": 0.001839,
+      "loss": 5.069,
+      "step": 613
+    },
+    {
+      "epoch": 0.00614,
+      "grad_norm": 0.7525346875190735,
+      "learning_rate": 0.001842,
+      "loss": 5.0757,
+      "step": 614
+    },
+    {
+      "epoch": 0.00615,
+      "grad_norm": 0.833108127117157,
+      "learning_rate": 0.001845,
+      "loss": 5.0659,
+      "step": 615
+    },
+    {
+      "epoch": 0.00616,
+      "grad_norm": 0.7170072793960571,
+      "learning_rate": 0.001848,
+      "loss": 5.0473,
+      "step": 616
+    },
+    {
+      "epoch": 0.00617,
+      "grad_norm": 0.718910276889801,
+      "learning_rate": 0.001851,
+      "loss": 5.0608,
+      "step": 617
+    },
+    {
+      "epoch": 0.00618,
+      "grad_norm": 0.6572015881538391,
+      "learning_rate": 0.001854,
+      "loss": 5.0425,
+      "step": 618
+    },
+    {
+      "epoch": 0.00619,
+      "grad_norm": 0.7236103415489197,
+      "learning_rate": 0.001857,
+      "loss": 5.0504,
+      "step": 619
+    },
+    {
+      "epoch": 0.0062,
+      "grad_norm": 0.8058017492294312,
+      "learning_rate": 0.00186,
+      "loss": 5.0386,
+      "step": 620
+    },
+    {
+      "epoch": 0.00621,
+      "grad_norm": 0.861880898475647,
+      "learning_rate": 0.001863,
+      "loss": 5.0409,
+      "step": 621
+    },
+    {
+      "epoch": 0.00622,
+      "grad_norm": 1.0328248739242554,
+      "learning_rate": 0.001866,
+      "loss": 5.0489,
+      "step": 622
+    },
+    {
+      "epoch": 0.00623,
+      "grad_norm": 0.8494102358818054,
+      "learning_rate": 0.001869,
+      "loss": 5.0318,
+      "step": 623
+    },
+    {
+      "epoch": 0.00624,
+      "grad_norm": 0.9827755093574524,
+      "learning_rate": 0.001872,
+      "loss": 5.0476,
+      "step": 624
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 1.0741342306137085,
+      "learning_rate": 0.001875,
+      "loss": 5.0686,
+      "step": 625
+    },
+    {
+      "epoch": 0.00626,
+      "grad_norm": 0.7305473685264587,
+      "learning_rate": 0.0018780000000000001,
+      "loss": 5.0304,
+      "step": 626
+    },
+    {
+      "epoch": 0.00627,
+      "grad_norm": 0.8084409832954407,
+      "learning_rate": 0.001881,
+      "loss": 5.0255,
+      "step": 627
+    },
+    {
+      "epoch": 0.00628,
+      "grad_norm": 1.0902513265609741,
+      "learning_rate": 0.001884,
+      "loss": 5.0666,
+      "step": 628
+    },
+    {
+      "epoch": 0.00629,
+      "grad_norm": 1.036152958869934,
+      "learning_rate": 0.001887,
+      "loss": 5.0439,
+      "step": 629
+    },
+    {
+      "epoch": 0.0063,
+      "grad_norm": 1.0308260917663574,
+      "learning_rate": 0.00189,
+      "loss": 5.0356,
+      "step": 630
+    },
+    {
+      "epoch": 0.00631,
+      "grad_norm": 0.984308123588562,
+      "learning_rate": 0.0018930000000000002,
+      "loss": 5.0356,
+      "step": 631
+    },
+    {
+      "epoch": 0.00632,
+      "grad_norm": 0.8649469017982483,
+      "learning_rate": 0.0018960000000000001,
+      "loss": 5.0479,
+      "step": 632
+    },
+    {
+      "epoch": 0.00633,
+      "grad_norm": 0.8776420950889587,
+      "learning_rate": 0.001899,
+      "loss": 5.0459,
+      "step": 633
+    },
+    {
+      "epoch": 0.00634,
+      "grad_norm": 0.8304409980773926,
+      "learning_rate": 0.001902,
+      "loss": 5.037,
+      "step": 634
+    },
+    {
+      "epoch": 0.00635,
+      "grad_norm": 0.7966147065162659,
+      "learning_rate": 0.001905,
+      "loss": 5.0227,
+      "step": 635
+    },
+    {
+      "epoch": 0.00636,
+      "grad_norm": 0.9172542095184326,
+      "learning_rate": 0.001908,
+      "loss": 5.0337,
+      "step": 636
+    },
+    {
+      "epoch": 0.00637,
+      "grad_norm": 1.3219475746154785,
+      "learning_rate": 0.0019110000000000002,
+      "loss": 5.0715,
+      "step": 637
+    },
+    {
+      "epoch": 0.00638,
+      "grad_norm": 0.7924789190292358,
+      "learning_rate": 0.0019140000000000001,
+      "loss": 5.0347,
+      "step": 638
+    },
+    {
+      "epoch": 0.00639,
+      "grad_norm": 0.8469759225845337,
+      "learning_rate": 0.001917,
+      "loss": 5.0139,
+      "step": 639
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 0.9544380307197571,
+      "learning_rate": 0.00192,
+      "loss": 5.0191,
+      "step": 640
+    },
+    {
+      "epoch": 0.00641,
+      "grad_norm": 1.0876184701919556,
+      "learning_rate": 0.001923,
+      "loss": 5.0379,
+      "step": 641
+    },
+    {
+      "epoch": 0.00642,
+      "grad_norm": 0.8299298286437988,
+      "learning_rate": 0.001926,
+      "loss": 5.0146,
+      "step": 642
+    },
+    {
+      "epoch": 0.00643,
+      "grad_norm": 0.9603999257087708,
+      "learning_rate": 0.0019290000000000002,
+      "loss": 5.0399,
+      "step": 643
+    },
+    {
+      "epoch": 0.00644,
+      "grad_norm": 0.7977001070976257,
+      "learning_rate": 0.0019320000000000001,
+      "loss": 5.0094,
+      "step": 644
+    },
+    {
+      "epoch": 0.00645,
+      "grad_norm": 0.7244200706481934,
+      "learning_rate": 0.001935,
+      "loss": 5.0161,
+      "step": 645
+    },
+    {
+      "epoch": 0.00646,
+      "grad_norm": 0.7832402586936951,
+      "learning_rate": 0.001938,
+      "loss": 5.0193,
+      "step": 646
+    },
+    {
+      "epoch": 0.00647,
+      "grad_norm": 0.8586620092391968,
+      "learning_rate": 0.001941,
+      "loss": 5.0299,
+      "step": 647
+    },
+    {
+      "epoch": 0.00648,
+      "grad_norm": 0.8153418302536011,
+      "learning_rate": 0.0019440000000000002,
+      "loss": 5.0163,
+      "step": 648
+    },
+    {
+      "epoch": 0.00649,
+      "grad_norm": 0.766000509262085,
+      "learning_rate": 0.0019470000000000002,
+      "loss": 4.9982,
+      "step": 649
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 0.7875446677207947,
+      "learning_rate": 0.0019500000000000001,
+      "loss": 5.0238,
+      "step": 650
+    },
+    {
+      "epoch": 0.00651,
+      "grad_norm": 0.7245673537254333,
+      "learning_rate": 0.001953,
+      "loss": 5.0037,
+      "step": 651
+    },
+    {
+      "epoch": 0.00652,
+      "grad_norm": 0.7840576767921448,
+      "learning_rate": 0.0019560000000000003,
+      "loss": 4.9987,
+      "step": 652
+    },
+    {
+      "epoch": 0.00653,
+      "grad_norm": 0.8112754225730896,
+      "learning_rate": 0.0019590000000000002,
+      "loss": 4.9969,
+      "step": 653
+    },
+    {
+      "epoch": 0.00654,
+      "grad_norm": 0.6785946488380432,
+      "learning_rate": 0.001962,
+      "loss": 4.9786,
+      "step": 654
+    },
+    {
+      "epoch": 0.00655,
+      "grad_norm": 0.6141355633735657,
+      "learning_rate": 0.001965,
+      "loss": 4.9897,
+      "step": 655
+    },
+    {
+      "epoch": 0.00656,
+      "grad_norm": 0.647098958492279,
+      "learning_rate": 0.001968,
+      "loss": 4.982,
+      "step": 656
+    },
+    {
+      "epoch": 0.00657,
+      "grad_norm": 0.6601396203041077,
+      "learning_rate": 0.001971,
+      "loss": 4.9729,
+      "step": 657
+    },
+    {
+      "epoch": 0.00658,
+      "grad_norm": 0.5589995384216309,
+      "learning_rate": 0.001974,
+      "loss": 4.9927,
+      "step": 658
+    },
+    {
+      "epoch": 0.00659,
+      "grad_norm": 0.551213264465332,
+      "learning_rate": 0.001977,
+      "loss": 4.972,
+      "step": 659
+    },
+    {
+      "epoch": 0.0066,
+      "grad_norm": 0.5106261968612671,
+      "learning_rate": 0.00198,
+      "loss": 4.963,
+      "step": 660
+    },
+    {
+      "epoch": 0.00661,
+      "grad_norm": 0.49552980065345764,
+      "learning_rate": 0.001983,
+      "loss": 4.9449,
+      "step": 661
+    },
+    {
+      "epoch": 0.00662,
+      "grad_norm": 0.4502437114715576,
+      "learning_rate": 0.0019860000000000004,
+      "loss": 4.9438,
+      "step": 662
+    },
+    {
+      "epoch": 0.00663,
+      "grad_norm": 0.4717095196247101,
+      "learning_rate": 0.0019890000000000003,
+      "loss": 4.9599,
+      "step": 663
+    },
+    {
+      "epoch": 0.00664,
+      "grad_norm": 0.5179165005683899,
+      "learning_rate": 0.0019920000000000003,
+      "loss": 4.9431,
+      "step": 664
+    },
+    {
+      "epoch": 0.00665,
+      "grad_norm": 0.6156288385391235,
+      "learning_rate": 0.0019950000000000002,
+      "loss": 4.9222,
+      "step": 665
+    },
+    {
+      "epoch": 0.00666,
+      "grad_norm": 0.6403276920318604,
+      "learning_rate": 0.001998,
+      "loss": 4.9456,
+      "step": 666
+    },
+    {
+      "epoch": 0.00667,
+      "grad_norm": 0.6612551808357239,
+      "learning_rate": 0.002001,
+      "loss": 4.9436,
+      "step": 667
+    },
+    {
+      "epoch": 0.00668,
+      "grad_norm": 0.6370317339897156,
+      "learning_rate": 0.002004,
+      "loss": 4.9618,
+      "step": 668
+    },
+    {
+      "epoch": 0.00669,
+      "grad_norm": 0.6632084250450134,
+      "learning_rate": 0.002007,
+      "loss": 4.9618,
+      "step": 669
+    },
+    {
+      "epoch": 0.0067,
+      "grad_norm": 0.6647160649299622,
+      "learning_rate": 0.00201,
+      "loss": 4.9534,
+      "step": 670
+    },
+    {
+      "epoch": 0.00671,
+      "grad_norm": 0.6171524524688721,
+      "learning_rate": 0.002013,
+      "loss": 4.9269,
+      "step": 671
+    },
+    {
+      "epoch": 0.00672,
+      "grad_norm": 0.6804357171058655,
+      "learning_rate": 0.002016,
+      "loss": 4.9362,
+      "step": 672
+    },
+    {
+      "epoch": 0.00673,
+      "grad_norm": 0.7436751127243042,
+      "learning_rate": 0.002019,
+      "loss": 4.9351,
+      "step": 673
+    },
+    {
+      "epoch": 0.00674,
+      "grad_norm": 0.8610292077064514,
+      "learning_rate": 0.0020220000000000004,
+      "loss": 4.9433,
+      "step": 674
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 1.046964168548584,
+      "learning_rate": 0.0020250000000000003,
+      "loss": 4.9648,
+      "step": 675
+    },
+    {
+      "epoch": 0.00676,
+      "grad_norm": 0.9578864574432373,
+      "learning_rate": 0.0020280000000000003,
+      "loss": 4.9492,
+      "step": 676
+    },
+    {
+      "epoch": 0.00677,
+      "grad_norm": 1.1337146759033203,
+      "learning_rate": 0.0020310000000000003,
+      "loss": 4.9695,
+      "step": 677
+    },
+    {
+      "epoch": 0.00678,
+      "grad_norm": 1.0358091592788696,
+      "learning_rate": 0.0020340000000000002,
+      "loss": 4.9763,
+      "step": 678
+    },
+    {
+      "epoch": 0.00679,
+      "grad_norm": 1.230987787246704,
+      "learning_rate": 0.002037,
+      "loss": 4.9735,
+      "step": 679
+    },
+    {
+      "epoch": 0.0068,
+      "grad_norm": 0.9104715585708618,
+      "learning_rate": 0.00204,
+      "loss": 4.9643,
+      "step": 680
+    },
+    {
+      "epoch": 0.00681,
+      "grad_norm": 1.1940183639526367,
+      "learning_rate": 0.002043,
+      "loss": 4.9659,
+      "step": 681
+    },
+    {
+      "epoch": 0.00682,
+      "grad_norm": 1.0067143440246582,
+      "learning_rate": 0.002046,
+      "loss": 4.9735,
+      "step": 682
+    },
+    {
+      "epoch": 0.00683,
+      "grad_norm": 1.224305510520935,
+      "learning_rate": 0.002049,
+      "loss": 4.9612,
+      "step": 683
+    },
+    {
+      "epoch": 0.00684,
+      "grad_norm": 0.9917755126953125,
+      "learning_rate": 0.002052,
+      "loss": 4.9694,
+      "step": 684
+    },
+    {
+      "epoch": 0.00685,
+      "grad_norm": 0.9187195897102356,
+      "learning_rate": 0.0020550000000000004,
+      "loss": 4.964,
+      "step": 685
+    },
+    {
+      "epoch": 0.00686,
+      "grad_norm": 1.035937786102295,
+      "learning_rate": 0.0020580000000000004,
+      "loss": 4.9577,
+      "step": 686
+    },
+    {
+      "epoch": 0.00687,
+      "grad_norm": 1.0923206806182861,
+      "learning_rate": 0.0020610000000000003,
+      "loss": 4.9579,
+      "step": 687
+    },
+    {
+      "epoch": 0.00688,
+      "grad_norm": 0.8355166912078857,
+      "learning_rate": 0.002064,
+      "loss": 4.9524,
+      "step": 688
+    },
+    {
+      "epoch": 0.00689,
+      "grad_norm": 0.7577447891235352,
+      "learning_rate": 0.002067,
+      "loss": 4.9417,
+      "step": 689
+    },
+    {
+      "epoch": 0.0069,
+      "grad_norm": 0.7621678709983826,
+      "learning_rate": 0.00207,
+      "loss": 4.9421,
+      "step": 690
+    },
+    {
+      "epoch": 0.00691,
+      "grad_norm": 0.6061983108520508,
+      "learning_rate": 0.0020729999999999998,
+      "loss": 4.9082,
+      "step": 691
+    },
+    {
+      "epoch": 0.00692,
+      "grad_norm": 0.591027021408081,
+      "learning_rate": 0.0020759999999999997,
+      "loss": 4.9292,
+      "step": 692
+    },
+    {
+      "epoch": 0.00693,
+      "grad_norm": 0.5834758281707764,
+      "learning_rate": 0.0020789999999999997,
+      "loss": 4.9222,
+      "step": 693
+    },
+    {
+      "epoch": 0.00694,
+      "grad_norm": 0.7208871841430664,
+      "learning_rate": 0.002082,
+      "loss": 4.9071,
+      "step": 694
+    },
+    {
+      "epoch": 0.00695,
+      "grad_norm": 0.8771729469299316,
+      "learning_rate": 0.002085,
+      "loss": 4.9171,
+      "step": 695
+    },
+    {
+      "epoch": 0.00696,
+      "grad_norm": 0.9051836133003235,
+      "learning_rate": 0.002088,
+      "loss": 4.9182,
+      "step": 696
+    },
+    {
+      "epoch": 0.00697,
+      "grad_norm": 1.1665294170379639,
+      "learning_rate": 0.002091,
+      "loss": 4.9339,
+      "step": 697
+    },
+    {
+      "epoch": 0.00698,
+      "grad_norm": 1.092050313949585,
+      "learning_rate": 0.002094,
+      "loss": 4.9261,
+      "step": 698
+    },
+    {
+      "epoch": 0.00699,
+      "grad_norm": 0.8412545323371887,
+      "learning_rate": 0.002097,
+      "loss": 4.9128,
+      "step": 699
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 0.6720849871635437,
+      "learning_rate": 0.0021,
+      "loss": 4.9176,
+      "step": 700
+    },
+    {
+      "epoch": 0.00701,
+      "grad_norm": 0.563408613204956,
+      "learning_rate": 0.002103,
+      "loss": 4.8944,
+      "step": 701
+    },
+    {
+      "epoch": 0.00702,
+      "grad_norm": 0.5357261300086975,
+      "learning_rate": 0.002106,
+      "loss": 4.893,
+      "step": 702
+    },
+    {
+      "epoch": 0.00703,
+      "grad_norm": 0.5667074918746948,
+      "learning_rate": 0.0021089999999999998,
+      "loss": 4.8766,
+      "step": 703
+    },
+    {
+      "epoch": 0.00704,
+      "grad_norm": 0.597253143787384,
+      "learning_rate": 0.0021119999999999997,
+      "loss": 4.8825,
+      "step": 704
+    },
+    {
+      "epoch": 0.00705,
+      "grad_norm": 0.6211616396903992,
+      "learning_rate": 0.002115,
+      "loss": 4.8877,
+      "step": 705
+    },
+    {
+      "epoch": 0.00706,
+      "grad_norm": 0.687544047832489,
+      "learning_rate": 0.002118,
+      "loss": 4.8929,
+      "step": 706
+    },
+    {
+      "epoch": 0.00707,
+      "grad_norm": 0.7621776461601257,
+      "learning_rate": 0.002121,
+      "loss": 4.8688,
+      "step": 707
+    },
+    {
+      "epoch": 0.00708,
+      "grad_norm": 0.9027195572853088,
+      "learning_rate": 0.002124,
+      "loss": 4.8567,
+      "step": 708
+    },
+    {
+      "epoch": 0.00709,
+      "grad_norm": 0.9446965456008911,
+      "learning_rate": 0.002127,
+      "loss": 4.8935,
+      "step": 709
+    },
+    {
+      "epoch": 0.0071,
+      "grad_norm": 0.8206554055213928,
+      "learning_rate": 0.00213,
+      "loss": 4.8706,
+      "step": 710
+    },
+    {
+      "epoch": 0.00711,
+      "grad_norm": 0.8629757165908813,
+      "learning_rate": 0.002133,
+      "loss": 4.8565,
+      "step": 711
+    },
+    {
+      "epoch": 0.00712,
+      "grad_norm": 0.7438434362411499,
+      "learning_rate": 0.002136,
+      "loss": 4.869,
+      "step": 712
+    },
+    {
+      "epoch": 0.00713,
+      "grad_norm": 0.7951372861862183,
+      "learning_rate": 0.002139,
+      "loss": 4.8846,
+      "step": 713
+    },
+    {
+      "epoch": 0.00714,
+      "grad_norm": 0.9020676016807556,
+      "learning_rate": 0.002142,
+      "loss": 4.8773,
+      "step": 714
+    },
+    {
+      "epoch": 0.00715,
+      "grad_norm": 1.064352035522461,
+      "learning_rate": 0.0021449999999999998,
+      "loss": 4.8624,
+      "step": 715
+    },
+    {
+      "epoch": 0.00716,
+      "grad_norm": 0.7318432927131653,
+      "learning_rate": 0.002148,
+      "loss": 4.8409,
+      "step": 716
+    },
+    {
+      "epoch": 0.00717,
+      "grad_norm": 0.886417031288147,
+      "learning_rate": 0.002151,
+      "loss": 4.8705,
+      "step": 717
+    },
+    {
+      "epoch": 0.00718,
+      "grad_norm": 0.8509985208511353,
+      "learning_rate": 0.002154,
+      "loss": 4.8473,
+      "step": 718
+    },
+    {
+      "epoch": 0.00719,
+      "grad_norm": 0.8979188203811646,
+      "learning_rate": 0.002157,
+      "loss": 4.8656,
+      "step": 719
+    },
+    {
+      "epoch": 0.0072,
+      "grad_norm": 1.0766488313674927,
+      "learning_rate": 0.00216,
+      "loss": 4.8622,
+      "step": 720
+    },
+    {
+      "epoch": 0.00721,
+      "grad_norm": 1.0330792665481567,
+      "learning_rate": 0.002163,
+      "loss": 4.8716,
+      "step": 721
+    },
+    {
+      "epoch": 0.00722,
+      "grad_norm": 1.097432255744934,
+      "learning_rate": 0.002166,
+      "loss": 4.8541,
+      "step": 722
+    },
+    {
+      "epoch": 0.00723,
+      "grad_norm": 1.0046334266662598,
+      "learning_rate": 0.002169,
+      "loss": 4.8596,
+      "step": 723
+    },
+    {
+      "epoch": 0.00724,
+      "grad_norm": 1.0316451787948608,
+      "learning_rate": 0.002172,
+      "loss": 4.8595,
+      "step": 724
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 0.7518482804298401,
+      "learning_rate": 0.002175,
+      "loss": 4.8464,
+      "step": 725
+    },
+    {
+      "epoch": 0.00726,
+      "grad_norm": 0.8229779005050659,
+      "learning_rate": 0.002178,
+      "loss": 4.8512,
+      "step": 726
+    },
+    {
+      "epoch": 0.00727,
+      "grad_norm": 0.9577723741531372,
+      "learning_rate": 0.0021809999999999998,
+      "loss": 4.841,
+      "step": 727
+    },
+    {
+      "epoch": 0.00728,
+      "grad_norm": 0.9698926210403442,
+      "learning_rate": 0.002184,
+      "loss": 4.8594,
+      "step": 728
+    },
+    {
+      "epoch": 0.00729,
+      "grad_norm": 1.1789532899856567,
+      "learning_rate": 0.002187,
+      "loss": 4.8599,
+      "step": 729
+    },
+    {
+      "epoch": 0.0073,
+      "grad_norm": 1.0923309326171875,
+      "learning_rate": 0.00219,
+      "loss": 4.8656,
+      "step": 730
+    },
+    {
+      "epoch": 0.00731,
+      "grad_norm": 1.0404242277145386,
+      "learning_rate": 0.002193,
+      "loss": 4.8544,
+      "step": 731
+    },
+    {
+      "epoch": 0.00732,
+      "grad_norm": 0.8924814462661743,
+      "learning_rate": 0.002196,
+      "loss": 4.8537,
+      "step": 732
+    },
+    {
+      "epoch": 0.00733,
+      "grad_norm": 0.7918877005577087,
+      "learning_rate": 0.002199,
+      "loss": 4.8326,
+      "step": 733
+    },
+    {
+      "epoch": 0.00734,
+      "grad_norm": 0.8231402635574341,
+      "learning_rate": 0.002202,
+      "loss": 4.8485,
+      "step": 734
+    },
+    {
+      "epoch": 0.00735,
+      "grad_norm": 1.1126084327697754,
+      "learning_rate": 0.002205,
+      "loss": 4.8661,
+      "step": 735
+    },
+    {
+      "epoch": 0.00736,
+      "grad_norm": 1.1928813457489014,
+      "learning_rate": 0.002208,
+      "loss": 4.857,
+      "step": 736
+    },
+    {
+      "epoch": 0.00737,
+      "grad_norm": 0.9305274486541748,
+      "learning_rate": 0.002211,
+      "loss": 4.8183,
+      "step": 737
+    },
+    {
+      "epoch": 0.00738,
+      "grad_norm": 1.2922184467315674,
+      "learning_rate": 0.002214,
+      "loss": 4.8313,
+      "step": 738
+    },
+    {
+      "epoch": 0.00739,
+      "grad_norm": 1.0224477052688599,
+      "learning_rate": 0.0022170000000000002,
+      "loss": 4.8302,
+      "step": 739
+    },
+    {
+      "epoch": 0.0074,
+      "grad_norm": 0.8254541158676147,
+      "learning_rate": 0.00222,
+      "loss": 4.8183,
+      "step": 740
+    },
+    {
+      "epoch": 0.00741,
+      "grad_norm": 0.8494399785995483,
+      "learning_rate": 0.002223,
+      "loss": 4.7936,
+      "step": 741
+    },
+    {
+      "epoch": 0.00742,
+      "grad_norm": 0.8097528219223022,
+      "learning_rate": 0.002226,
+      "loss": 4.8203,
+      "step": 742
+    },
+    {
+      "epoch": 0.00743,
+      "grad_norm": 0.7318201065063477,
+      "learning_rate": 0.002229,
+      "loss": 4.782,
+      "step": 743
+    },
+    {
+      "epoch": 0.00744,
+      "grad_norm": 0.770041286945343,
+      "learning_rate": 0.002232,
+      "loss": 4.7897,
+      "step": 744
+    },
+    {
+      "epoch": 0.00745,
+      "grad_norm": 0.8400176167488098,
+      "learning_rate": 0.002235,
+      "loss": 4.793,
+      "step": 745
+    },
+    {
+      "epoch": 0.00746,
+      "grad_norm": 0.8187500834465027,
+      "learning_rate": 0.002238,
+      "loss": 4.7927,
+      "step": 746
+    },
+    {
+      "epoch": 0.00747,
+      "grad_norm": 0.8396742343902588,
+      "learning_rate": 0.002241,
+      "loss": 4.7802,
+      "step": 747
+    },
+    {
+      "epoch": 0.00748,
+      "grad_norm": 0.8425725698471069,
+      "learning_rate": 0.002244,
+      "loss": 4.7863,
+      "step": 748
+    },
+    {
+      "epoch": 0.00749,
+      "grad_norm": 0.9793819189071655,
+      "learning_rate": 0.002247,
+      "loss": 4.771,
+      "step": 749
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 0.9097406268119812,
+      "learning_rate": 0.0022500000000000003,
+      "loss": 4.8036,
+      "step": 750
+    },
+    {
+      "epoch": 0.00751,
+      "grad_norm": 1.1564972400665283,
+      "learning_rate": 0.0022530000000000002,
+      "loss": 4.8008,
+      "step": 751
+    },
+    {
+      "epoch": 0.00752,
+      "grad_norm": 1.1269639730453491,
+      "learning_rate": 0.002256,
+      "loss": 4.7666,
+      "step": 752
+    },
+    {
+      "epoch": 0.00753,
+      "grad_norm": 0.7554891705513,
+      "learning_rate": 0.002259,
+      "loss": 4.7895,
+      "step": 753
+    },
+    {
+      "epoch": 0.00754,
+      "grad_norm": 0.6434245705604553,
+      "learning_rate": 0.002262,
+      "loss": 4.7936,
+      "step": 754
+    },
+    {
+      "epoch": 0.00755,
+      "grad_norm": 0.6667845249176025,
+      "learning_rate": 0.002265,
+      "loss": 4.7629,
+      "step": 755
+    },
+    {
+      "epoch": 0.00756,
+      "grad_norm": 0.6454316973686218,
+      "learning_rate": 0.002268,
+      "loss": 4.7431,
+      "step": 756
+    },
+    {
+      "epoch": 0.00757,
+      "grad_norm": 0.5992982387542725,
+      "learning_rate": 0.002271,
+      "loss": 4.7549,
+      "step": 757
+    },
+    {
+      "epoch": 0.00758,
+      "grad_norm": 0.5403345227241516,
+      "learning_rate": 0.002274,
+      "loss": 4.7325,
+      "step": 758
+    },
+    {
+      "epoch": 0.00759,
+      "grad_norm": 0.6695131063461304,
+      "learning_rate": 0.002277,
+      "loss": 4.754,
+      "step": 759
+    },
+    {
+      "epoch": 0.0076,
+      "grad_norm": 0.8114665150642395,
+      "learning_rate": 0.00228,
+      "loss": 4.7449,
+      "step": 760
+    },
+    {
+      "epoch": 0.00761,
+      "grad_norm": 0.8682258129119873,
+      "learning_rate": 0.002283,
+      "loss": 4.7184,
+      "step": 761
+    },
+    {
+      "epoch": 0.00762,
+      "grad_norm": 0.9619300961494446,
+      "learning_rate": 0.0022860000000000003,
+      "loss": 4.7088,
+      "step": 762
+    },
+    {
+      "epoch": 0.00763,
+      "grad_norm": 1.0561281442642212,
+      "learning_rate": 0.0022890000000000002,
+      "loss": 4.7527,
+      "step": 763
+    },
+    {
+      "epoch": 0.00764,
+      "grad_norm": 1.0605075359344482,
+      "learning_rate": 0.002292,
+      "loss": 4.7609,
+      "step": 764
+    },
+    {
+      "epoch": 0.00765,
+      "grad_norm": 0.8591569066047668,
+      "learning_rate": 0.002295,
+      "loss": 4.7468,
+      "step": 765
+    },
+    {
+      "epoch": 0.00766,
+      "grad_norm": 0.8702619075775146,
+      "learning_rate": 0.002298,
+      "loss": 4.7378,
+      "step": 766
+    },
+    {
+      "epoch": 0.00767,
+      "grad_norm": 0.9817199110984802,
+      "learning_rate": 0.002301,
+      "loss": 4.7481,
+      "step": 767
+    },
+    {
+      "epoch": 0.00768,
+      "grad_norm": 1.0741162300109863,
+      "learning_rate": 0.002304,
+      "loss": 4.7612,
+      "step": 768
+    },
+    {
+      "epoch": 0.00769,
+      "grad_norm": 0.8188871145248413,
+      "learning_rate": 0.002307,
+      "loss": 4.7592,
+      "step": 769
+    },
+    {
+      "epoch": 0.0077,
+      "grad_norm": 0.7658451795578003,
+      "learning_rate": 0.00231,
+      "loss": 4.7148,
+      "step": 770
+    },
+    {
+      "epoch": 0.00771,
+      "grad_norm": 0.8762615323066711,
+      "learning_rate": 0.002313,
+      "loss": 4.754,
+      "step": 771
+    },
+    {
+      "epoch": 0.00772,
+      "grad_norm": 0.9279188513755798,
+      "learning_rate": 0.002316,
+      "loss": 4.7389,
+      "step": 772
+    },
+    {
+      "epoch": 0.00773,
+      "grad_norm": 0.8405673503875732,
+      "learning_rate": 0.0023190000000000003,
+      "loss": 4.737,
+      "step": 773
+    },
+    {
+      "epoch": 0.00774,
+      "grad_norm": 0.777439296245575,
+      "learning_rate": 0.0023220000000000003,
+      "loss": 4.7452,
+      "step": 774
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 0.8980410099029541,
+      "learning_rate": 0.0023250000000000002,
+      "loss": 4.7436,
+      "step": 775
+    },
+    {
+      "epoch": 0.00776,
+      "grad_norm": 1.0412962436676025,
+      "learning_rate": 0.002328,
+      "loss": 4.7539,
+      "step": 776
+    },
+    {
+      "epoch": 0.00777,
+      "grad_norm": 0.9281516075134277,
+      "learning_rate": 0.002331,
+      "loss": 4.7268,
+      "step": 777
+    },
+    {
+      "epoch": 0.00778,
+      "grad_norm": 0.7459467649459839,
+      "learning_rate": 0.002334,
+      "loss": 4.7271,
+      "step": 778
+    },
+    {
+      "epoch": 0.00779,
+      "grad_norm": 0.7607200145721436,
+      "learning_rate": 0.002337,
+      "loss": 4.689,
+      "step": 779
+    },
+    {
+      "epoch": 0.0078,
+      "grad_norm": 0.7529038190841675,
+      "learning_rate": 0.00234,
+      "loss": 4.706,
+      "step": 780
+    },
+    {
+      "epoch": 0.00781,
+      "grad_norm": 0.7776694297790527,
+      "learning_rate": 0.002343,
+      "loss": 4.7072,
+      "step": 781
+    },
+    {
+      "epoch": 0.00782,
+      "grad_norm": 0.7648219466209412,
+      "learning_rate": 0.002346,
+      "loss": 4.6998,
+      "step": 782
+    },
+    {
+      "epoch": 0.00783,
+      "grad_norm": 0.8795627355575562,
+      "learning_rate": 0.002349,
+      "loss": 4.6893,
+      "step": 783
+    },
+    {
+      "epoch": 0.00784,
+      "grad_norm": 0.9756646156311035,
+      "learning_rate": 0.002352,
+      "loss": 4.706,
+      "step": 784
+    },
+    {
+      "epoch": 0.00785,
+      "grad_norm": 1.03944993019104,
+      "learning_rate": 0.0023550000000000003,
+      "loss": 4.7094,
+      "step": 785
+    },
+    {
+      "epoch": 0.00786,
+      "grad_norm": 0.9842208027839661,
+      "learning_rate": 0.0023580000000000003,
+      "loss": 4.7094,
+      "step": 786
+    },
+    {
+      "epoch": 0.00787,
+      "grad_norm": 0.9913274049758911,
+      "learning_rate": 0.0023610000000000003,
+      "loss": 4.7124,
+      "step": 787
+    },
+    {
+      "epoch": 0.00788,
+      "grad_norm": 1.0203496217727661,
+      "learning_rate": 0.002364,
+      "loss": 4.711,
+      "step": 788
+    },
+    {
+      "epoch": 0.00789,
+      "grad_norm": 1.0524691343307495,
+      "learning_rate": 0.002367,
+      "loss": 4.7436,
+      "step": 789
+    },
+    {
+      "epoch": 0.0079,
+      "grad_norm": 0.8448042273521423,
+      "learning_rate": 0.00237,
+      "loss": 4.7139,
+      "step": 790
+    },
+    {
+      "epoch": 0.00791,
+      "grad_norm": 0.737777054309845,
+      "learning_rate": 0.002373,
+      "loss": 4.6607,
+      "step": 791
+    },
+    {
+      "epoch": 0.00792,
+      "grad_norm": 0.8730551600456238,
+      "learning_rate": 0.002376,
+      "loss": 4.6988,
+      "step": 792
+    },
+    {
+      "epoch": 0.00793,
+      "grad_norm": 1.0680596828460693,
+      "learning_rate": 0.002379,
+      "loss": 4.6938,
+      "step": 793
+    },
+    {
+      "epoch": 0.00794,
+      "grad_norm": 0.876390814781189,
+      "learning_rate": 0.002382,
+      "loss": 4.7016,
+      "step": 794
+    },
+    {
+      "epoch": 0.00795,
+      "grad_norm": 0.8010908961296082,
+      "learning_rate": 0.002385,
+      "loss": 4.6743,
+      "step": 795
+    },
+    {
+      "epoch": 0.00796,
+      "grad_norm": 0.8452677130699158,
+      "learning_rate": 0.0023880000000000004,
+      "loss": 4.6712,
+      "step": 796
+    },
+    {
+      "epoch": 0.00797,
+      "grad_norm": 0.7743445038795471,
+      "learning_rate": 0.0023910000000000003,
+      "loss": 4.7081,
+      "step": 797
+    },
+    {
+      "epoch": 0.00798,
+      "grad_norm": 0.7820720076560974,
+      "learning_rate": 0.0023940000000000003,
+      "loss": 4.6744,
+      "step": 798
+    },
+    {
+      "epoch": 0.00799,
+      "grad_norm": 0.8602663278579712,
+      "learning_rate": 0.0023970000000000003,
+      "loss": 4.703,
+      "step": 799
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 0.83580082654953,
+      "learning_rate": 0.0024000000000000002,
+      "loss": 4.6615,
+      "step": 800
+    },
+    {
+      "epoch": 0.00801,
+      "grad_norm": 0.8171262145042419,
+      "learning_rate": 0.002403,
+      "loss": 4.6877,
+      "step": 801
+    },
+    {
+      "epoch": 0.00802,
+      "grad_norm": 0.8578851819038391,
+      "learning_rate": 0.002406,
+      "loss": 4.6772,
+      "step": 802
+    },
+    {
+      "epoch": 0.00803,
+      "grad_norm": 0.9747350215911865,
+      "learning_rate": 0.002409,
+      "loss": 4.662,
+      "step": 803
+    },
+    {
+      "epoch": 0.00804,
+      "grad_norm": 0.9363577961921692,
+      "learning_rate": 0.002412,
+      "loss": 4.6595,
+      "step": 804
+    },
+    {
+      "epoch": 0.00805,
+      "grad_norm": 0.9166631102561951,
+      "learning_rate": 0.002415,
+      "loss": 4.6601,
+      "step": 805
+    },
+    {
+      "epoch": 0.00806,
+      "grad_norm": 0.860556960105896,
+      "learning_rate": 0.002418,
+      "loss": 4.681,
+      "step": 806
+    },
+    {
+      "epoch": 0.00807,
+      "grad_norm": 0.8737242817878723,
+      "learning_rate": 0.0024210000000000004,
+      "loss": 4.6779,
+      "step": 807
+    },
+    {
+      "epoch": 0.00808,
+      "grad_norm": 0.8125926852226257,
+      "learning_rate": 0.0024240000000000004,
+      "loss": 4.6555,
+      "step": 808
+    },
+    {
+      "epoch": 0.00809,
+      "grad_norm": 0.9335473775863647,
+      "learning_rate": 0.0024270000000000003,
+      "loss": 4.6575,
+      "step": 809
+    },
+    {
+      "epoch": 0.0081,
+      "grad_norm": 0.7656557559967041,
+      "learning_rate": 0.0024300000000000003,
+      "loss": 4.6752,
+      "step": 810
+    },
+    {
+      "epoch": 0.00811,
+      "grad_norm": 0.8129197359085083,
+      "learning_rate": 0.0024330000000000003,
+      "loss": 4.6545,
+      "step": 811
+    },
+    {
+      "epoch": 0.00812,
+      "grad_norm": 0.7713532447814941,
+      "learning_rate": 0.0024360000000000002,
+      "loss": 4.6279,
+      "step": 812
+    },
+    {
+      "epoch": 0.00813,
+      "grad_norm": 0.8149270415306091,
+      "learning_rate": 0.0024389999999999998,
+      "loss": 4.6613,
+      "step": 813
+    },
+    {
+      "epoch": 0.00814,
+      "grad_norm": 0.866010308265686,
+      "learning_rate": 0.0024419999999999997,
+      "loss": 4.6696,
+      "step": 814
+    },
+    {
+      "epoch": 0.00815,
+      "grad_norm": 0.7402296662330627,
+      "learning_rate": 0.0024449999999999997,
+      "loss": 4.6746,
+      "step": 815
+    },
+    {
+      "epoch": 0.00816,
+      "grad_norm": 0.6451212167739868,
+      "learning_rate": 0.002448,
+      "loss": 4.6403,
+      "step": 816
+    },
+    {
+      "epoch": 0.00817,
+      "grad_norm": 0.7935466170310974,
+      "learning_rate": 0.002451,
+      "loss": 4.6538,
+      "step": 817
+    },
+    {
+      "epoch": 0.00818,
+      "grad_norm": 1.0368677377700806,
+      "learning_rate": 0.002454,
+      "loss": 4.6444,
+      "step": 818
+    },
+    {
+      "epoch": 0.00819,
+      "grad_norm": 1.1921635866165161,
+      "learning_rate": 0.002457,
+      "loss": 4.6695,
+      "step": 819
+    },
+    {
+      "epoch": 0.0082,
+      "grad_norm": 0.9146779179573059,
+      "learning_rate": 0.00246,
+      "loss": 4.6473,
+      "step": 820
+    },
+    {
+      "epoch": 0.00821,
+      "grad_norm": 0.8097479939460754,
+      "learning_rate": 0.002463,
+      "loss": 4.6216,
+      "step": 821
+    },
+    {
+      "epoch": 0.00822,
+      "grad_norm": 0.8611756563186646,
+      "learning_rate": 0.002466,
+      "loss": 4.6241,
+      "step": 822
+    },
+    {
+      "epoch": 0.00823,
+      "grad_norm": 0.9131811857223511,
+      "learning_rate": 0.002469,
+      "loss": 4.6307,
+      "step": 823
+    },
+    {
+      "epoch": 0.00824,
+      "grad_norm": 0.9472024440765381,
+      "learning_rate": 0.002472,
+      "loss": 4.6551,
+      "step": 824
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 1.1296700239181519,
+      "learning_rate": 0.0024749999999999998,
+      "loss": 4.6738,
+      "step": 825
+    },
+    {
+      "epoch": 0.00826,
+      "grad_norm": 0.9912372827529907,
+      "learning_rate": 0.0024779999999999997,
+      "loss": 4.6298,
+      "step": 826
+    },
+    {
+      "epoch": 0.00827,
+      "grad_norm": 0.9669742584228516,
+      "learning_rate": 0.002481,
+      "loss": 4.6558,
+      "step": 827
+    },
+    {
+      "epoch": 0.00828,
+      "grad_norm": 0.7501344680786133,
+      "learning_rate": 0.002484,
+      "loss": 4.6425,
+      "step": 828
+    },
+    {
+      "epoch": 0.00829,
+      "grad_norm": 0.6985933184623718,
+      "learning_rate": 0.002487,
+      "loss": 4.6351,
+      "step": 829
+    },
+    {
+      "epoch": 0.0083,
+      "grad_norm": 0.6681656837463379,
+      "learning_rate": 0.00249,
+      "loss": 4.6406,
+      "step": 830
+    },
+    {
+      "epoch": 0.00831,
+      "grad_norm": 0.7142512202262878,
+      "learning_rate": 0.002493,
+      "loss": 4.6218,
+      "step": 831
+    },
+    {
+      "epoch": 0.00832,
+      "grad_norm": 0.7265256643295288,
+      "learning_rate": 0.002496,
+      "loss": 4.6212,
+      "step": 832
+    },
+    {
+      "epoch": 0.00833,
+      "grad_norm": 0.7133427262306213,
+      "learning_rate": 0.002499,
+      "loss": 4.6125,
+      "step": 833
+    },
+    {
+      "epoch": 0.00834,
+      "grad_norm": 0.7037473917007446,
+      "learning_rate": 0.002502,
+      "loss": 4.6151,
+      "step": 834
+    },
+    {
+      "epoch": 0.00835,
+      "grad_norm": 0.9004167914390564,
+      "learning_rate": 0.002505,
+      "loss": 4.6154,
+      "step": 835
+    },
+    {
+      "epoch": 0.00836,
+      "grad_norm": 0.9459953904151917,
+      "learning_rate": 0.002508,
+      "loss": 4.6497,
+      "step": 836
+    },
+    {
+      "epoch": 0.00837,
+      "grad_norm": 0.9802148342132568,
+      "learning_rate": 0.0025109999999999998,
+      "loss": 4.5975,
+      "step": 837
+    },
+    {
+      "epoch": 0.00838,
+      "grad_norm": 0.8809778690338135,
+      "learning_rate": 0.0025139999999999997,
+      "loss": 4.6166,
+      "step": 838
+    },
+    {
+      "epoch": 0.00839,
+      "grad_norm": 0.792102038860321,
+      "learning_rate": 0.002517,
+      "loss": 4.6155,
+      "step": 839
+    },
+    {
+      "epoch": 0.0084,
+      "grad_norm": 0.7744638323783875,
+      "learning_rate": 0.00252,
+      "loss": 4.5999,
+      "step": 840
+    },
+    {
+      "epoch": 0.00841,
+      "grad_norm": 0.791641116142273,
+      "learning_rate": 0.002523,
+      "loss": 4.5943,
+      "step": 841
+    },
+    {
+      "epoch": 0.00842,
+      "grad_norm": 0.8285142183303833,
+      "learning_rate": 0.002526,
+      "loss": 4.5937,
+      "step": 842
+    },
+    {
+      "epoch": 0.00843,
+      "grad_norm": 0.9900093078613281,
+      "learning_rate": 0.002529,
+      "loss": 4.6057,
+      "step": 843
+    },
+    {
+      "epoch": 0.00844,
+      "grad_norm": 0.984833300113678,
+      "learning_rate": 0.002532,
+      "loss": 4.6269,
+      "step": 844
+    },
+    {
+      "epoch": 0.00845,
+      "grad_norm": 0.8697680234909058,
+      "learning_rate": 0.002535,
+      "loss": 4.5878,
+      "step": 845
+    },
+    {
+      "epoch": 0.00846,
+      "grad_norm": 0.8182123899459839,
+      "learning_rate": 0.002538,
+      "loss": 4.6097,
+      "step": 846
+    },
+    {
+      "epoch": 0.00847,
+      "grad_norm": 0.7307525277137756,
+      "learning_rate": 0.002541,
+      "loss": 4.5888,
+      "step": 847
+    },
+    {
+      "epoch": 0.00848,
+      "grad_norm": 0.7021674513816833,
+      "learning_rate": 0.002544,
+      "loss": 4.5844,
+      "step": 848
+    },
+    {
+      "epoch": 0.00849,
+      "grad_norm": 0.6287952661514282,
+      "learning_rate": 0.002547,
+      "loss": 4.5759,
+      "step": 849
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 0.5557106733322144,
+      "learning_rate": 0.00255,
+      "loss": 4.5834,
+      "step": 850
+    },
+    {
+      "epoch": 0.00851,
+      "grad_norm": 0.5636370778083801,
+      "learning_rate": 0.002553,
+      "loss": 4.5977,
+      "step": 851
+    },
+    {
+      "epoch": 0.00852,
+      "grad_norm": 0.6509172320365906,
+      "learning_rate": 0.002556,
+      "loss": 4.5892,
+      "step": 852
+    },
+    {
+      "epoch": 0.00853,
+      "grad_norm": 0.6835383176803589,
+      "learning_rate": 0.002559,
+      "loss": 4.5743,
+      "step": 853
+    },
+    {
+      "epoch": 0.00854,
+      "grad_norm": 0.6499077081680298,
+      "learning_rate": 0.002562,
+      "loss": 4.5629,
+      "step": 854
+    },
+    {
+      "epoch": 0.00855,
+      "grad_norm": 0.5777466297149658,
+      "learning_rate": 0.002565,
+      "loss": 4.5388,
+      "step": 855
+    },
+    {
+      "epoch": 0.00856,
+      "grad_norm": 0.6060221195220947,
+      "learning_rate": 0.002568,
+      "loss": 4.5724,
+      "step": 856
+    },
+    {
+      "epoch": 0.00857,
+      "grad_norm": 0.6906097531318665,
+      "learning_rate": 0.002571,
+      "loss": 4.5615,
+      "step": 857
+    },
+    {
+      "epoch": 0.00858,
+      "grad_norm": 0.7318242788314819,
+      "learning_rate": 0.002574,
+      "loss": 4.5575,
+      "step": 858
+    },
+    {
+      "epoch": 0.00859,
+      "grad_norm": 0.6517086029052734,
+      "learning_rate": 0.002577,
+      "loss": 4.5498,
+      "step": 859
+    },
+    {
+      "epoch": 0.0086,
+      "grad_norm": 0.6514325737953186,
+      "learning_rate": 0.00258,
+      "loss": 4.5756,
+      "step": 860
+    },
+    {
+      "epoch": 0.00861,
+      "grad_norm": 0.7899180054664612,
+      "learning_rate": 0.0025830000000000002,
+      "loss": 4.594,
+      "step": 861
+    },
+    {
+      "epoch": 0.00862,
+      "grad_norm": 0.874433696269989,
+      "learning_rate": 0.002586,
+      "loss": 4.5519,
+      "step": 862
+    },
+    {
+      "epoch": 0.00863,
+      "grad_norm": 0.8605120182037354,
+      "learning_rate": 0.002589,
+      "loss": 4.5772,
+      "step": 863
+    },
+    {
+      "epoch": 0.00864,
+      "grad_norm": 0.9631415009498596,
+      "learning_rate": 0.002592,
+      "loss": 4.5403,
+      "step": 864
+    },
+    {
+      "epoch": 0.00865,
+      "grad_norm": 1.1293022632598877,
+      "learning_rate": 0.002595,
+      "loss": 4.5658,
+      "step": 865
+    },
+    {
+      "epoch": 0.00866,
+      "grad_norm": 1.0691903829574585,
+      "learning_rate": 0.002598,
+      "loss": 4.5962,
+      "step": 866
+    },
+    {
+      "epoch": 0.00867,
+      "grad_norm": 0.9885998368263245,
+      "learning_rate": 0.002601,
+      "loss": 4.5818,
+      "step": 867
+    },
+    {
+      "epoch": 0.00868,
+      "grad_norm": 1.3688061237335205,
+      "learning_rate": 0.002604,
+      "loss": 4.5938,
+      "step": 868
+    },
+    {
+      "epoch": 0.00869,
+      "grad_norm": 0.8156008124351501,
+      "learning_rate": 0.002607,
+      "loss": 4.5787,
+      "step": 869
+    },
+    {
+      "epoch": 0.0087,
+      "grad_norm": 0.8631764054298401,
+      "learning_rate": 0.00261,
+      "loss": 4.5676,
+      "step": 870
+    },
+    {
+      "epoch": 0.00871,
+      "grad_norm": 0.9996145963668823,
+      "learning_rate": 0.002613,
+      "loss": 4.5932,
+      "step": 871
+    },
+    {
+      "epoch": 0.00872,
+      "grad_norm": 0.972501814365387,
+      "learning_rate": 0.002616,
+      "loss": 4.5764,
+      "step": 872
+    },
+    {
+      "epoch": 0.00873,
+      "grad_norm": 1.0140340328216553,
+      "learning_rate": 0.0026190000000000002,
+      "loss": 4.5878,
+      "step": 873
+    },
+    {
+      "epoch": 0.00874,
+      "grad_norm": 0.9380632042884827,
+      "learning_rate": 0.002622,
+      "loss": 4.6105,
+      "step": 874
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 0.9407688975334167,
+      "learning_rate": 0.002625,
+      "loss": 4.5588,
+      "step": 875
+    },
+    {
+      "epoch": 0.00876,
+      "grad_norm": 1.0034210681915283,
+      "learning_rate": 0.002628,
+      "loss": 4.5949,
+      "step": 876
+    },
+    {
+      "epoch": 0.00877,
+      "grad_norm": 1.2963709831237793,
+      "learning_rate": 0.002631,
+      "loss": 4.6268,
+      "step": 877
+    },
+    {
+      "epoch": 0.00878,
+      "grad_norm": 0.7438150644302368,
+      "learning_rate": 0.002634,
+      "loss": 4.537,
+      "step": 878
+    },
+    {
+      "epoch": 0.00879,
+      "grad_norm": 0.8438600301742554,
+      "learning_rate": 0.002637,
+      "loss": 4.5729,
+      "step": 879
+    },
+    {
+      "epoch": 0.0088,
+      "grad_norm": 0.7277801036834717,
+      "learning_rate": 0.00264,
+      "loss": 4.5704,
+      "step": 880
+    },
+    {
+      "epoch": 0.00881,
+      "grad_norm": 0.9043455123901367,
+      "learning_rate": 0.002643,
+      "loss": 4.5794,
+      "step": 881
+    },
+    {
+      "epoch": 0.00882,
+      "grad_norm": 0.9441054463386536,
+      "learning_rate": 0.002646,
+      "loss": 4.5807,
+      "step": 882
+    },
+    {
+      "epoch": 0.00883,
+      "grad_norm": 0.8490539193153381,
+      "learning_rate": 0.002649,
+      "loss": 4.5704,
+      "step": 883
+    },
+    {
+      "epoch": 0.00884,
+      "grad_norm": 0.8639208078384399,
+      "learning_rate": 0.0026520000000000003,
+      "loss": 4.555,
+      "step": 884
+    },
+    {
+      "epoch": 0.00885,
+      "grad_norm": 0.8000319600105286,
+      "learning_rate": 0.0026550000000000002,
+      "loss": 4.5531,
+      "step": 885
+    },
+    {
+      "epoch": 0.00886,
+      "grad_norm": 0.8000409007072449,
+      "learning_rate": 0.002658,
+      "loss": 4.5702,
+      "step": 886
+    },
+    {
+      "epoch": 0.00887,
+      "grad_norm": 0.5944011807441711,
+      "learning_rate": 0.002661,
+      "loss": 4.5499,
+      "step": 887
+    },
+    {
+      "epoch": 0.00888,
+      "grad_norm": 0.5328640341758728,
+      "learning_rate": 0.002664,
+      "loss": 4.5494,
+      "step": 888
+    },
+    {
+      "epoch": 0.00889,
+      "grad_norm": 0.5338924527168274,
+      "learning_rate": 0.002667,
+      "loss": 4.5018,
+      "step": 889
+    },
+    {
+      "epoch": 0.0089,
+      "grad_norm": 0.4938536286354065,
+      "learning_rate": 0.00267,
+      "loss": 4.545,
+      "step": 890
+    },
+    {
+      "epoch": 0.00891,
+      "grad_norm": 0.4882456064224243,
+      "learning_rate": 0.002673,
+      "loss": 4.5301,
+      "step": 891
+    },
+    {
+      "epoch": 0.00892,
+      "grad_norm": 0.4257451891899109,
+      "learning_rate": 0.002676,
+      "loss": 4.5393,
+      "step": 892
+    },
+    {
+      "epoch": 0.00893,
+      "grad_norm": 0.5505130290985107,
+      "learning_rate": 0.002679,
+      "loss": 4.5171,
+      "step": 893
+    },
+    {
+      "epoch": 0.00894,
+      "grad_norm": 0.6718395352363586,
+      "learning_rate": 0.002682,
+      "loss": 4.5132,
+      "step": 894
+    },
+    {
+      "epoch": 0.00895,
+      "grad_norm": 0.7553327679634094,
+      "learning_rate": 0.0026850000000000003,
+      "loss": 4.5223,
+      "step": 895
+    },
+    {
+      "epoch": 0.00896,
+      "grad_norm": 0.7149863243103027,
+      "learning_rate": 0.0026880000000000003,
+      "loss": 4.5119,
+      "step": 896
+    },
+    {
+      "epoch": 0.00897,
+      "grad_norm": 0.8281179070472717,
+      "learning_rate": 0.0026910000000000002,
+      "loss": 4.5328,
+      "step": 897
+    },
+    {
+      "epoch": 0.00898,
+      "grad_norm": 0.9519450068473816,
+      "learning_rate": 0.002694,
+      "loss": 4.5342,
+      "step": 898
+    },
+    {
+      "epoch": 0.00899,
+      "grad_norm": 0.9227859377861023,
+      "learning_rate": 0.002697,
+      "loss": 4.5445,
+      "step": 899
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 1.040459156036377,
+      "learning_rate": 0.0027,
+      "loss": 4.5756,
+      "step": 900
+    },
+    {
+      "epoch": 0.00901,
+      "grad_norm": 0.8912736773490906,
+      "learning_rate": 0.002703,
+      "loss": 4.5234,
+      "step": 901
+    },
+    {
+      "epoch": 0.00902,
+      "grad_norm": 0.768046498298645,
+      "learning_rate": 0.002706,
+      "loss": 4.5317,
+      "step": 902
+    },
+    {
+      "epoch": 0.00903,
+      "grad_norm": 0.698384702205658,
+      "learning_rate": 0.002709,
+      "loss": 4.5227,
+      "step": 903
+    },
+    {
+      "epoch": 0.00904,
+      "grad_norm": 0.7553470134735107,
+      "learning_rate": 0.002712,
+      "loss": 4.5046,
+      "step": 904
+    },
+    {
+      "epoch": 0.00905,
+      "grad_norm": 0.9005333185195923,
+      "learning_rate": 0.002715,
+      "loss": 4.5273,
+      "step": 905
+    },
+    {
+      "epoch": 0.00906,
+      "grad_norm": 0.9150082468986511,
+      "learning_rate": 0.002718,
+      "loss": 4.54,
+      "step": 906
+    },
+    {
+      "epoch": 0.00907,
+      "grad_norm": 0.8349012732505798,
+      "learning_rate": 0.0027210000000000003,
+      "loss": 4.5543,
+      "step": 907
+    },
+    {
+      "epoch": 0.00908,
+      "grad_norm": 0.7856695055961609,
+      "learning_rate": 0.0027240000000000003,
+      "loss": 4.5219,
+      "step": 908
+    },
+    {
+      "epoch": 0.00909,
+      "grad_norm": 0.8433918952941895,
+      "learning_rate": 0.0027270000000000003,
+      "loss": 4.5423,
+      "step": 909
+    },
+    {
+      "epoch": 0.0091,
+      "grad_norm": 0.9650413393974304,
+      "learning_rate": 0.0027300000000000002,
+      "loss": 4.5577,
+      "step": 910
+    },
+    {
+      "epoch": 0.00911,
+      "grad_norm": 0.9894043803215027,
+      "learning_rate": 0.002733,
+      "loss": 4.514,
+      "step": 911
+    },
+    {
+      "epoch": 0.00912,
+      "grad_norm": 0.9117268919944763,
+      "learning_rate": 0.002736,
+      "loss": 4.5245,
+      "step": 912
+    },
+    {
+      "epoch": 0.00913,
+      "grad_norm": 1.020936369895935,
+      "learning_rate": 0.002739,
+      "loss": 4.5244,
+      "step": 913
+    },
+    {
+      "epoch": 0.00914,
+      "grad_norm": 1.1810083389282227,
+      "learning_rate": 0.002742,
+      "loss": 4.5723,
+      "step": 914
+    },
+    {
+      "epoch": 0.00915,
+      "grad_norm": 0.8899980187416077,
+      "learning_rate": 0.002745,
+      "loss": 4.5252,
+      "step": 915
+    },
+    {
+      "epoch": 0.00916,
+      "grad_norm": 0.9896888136863708,
+      "learning_rate": 0.002748,
+      "loss": 4.5028,
+      "step": 916
+    },
+    {
+      "epoch": 0.00917,
+      "grad_norm": 1.3935941457748413,
+      "learning_rate": 0.002751,
+      "loss": 4.5471,
+      "step": 917
+    },
+    {
+      "epoch": 0.00918,
+      "grad_norm": 1.1309375762939453,
+      "learning_rate": 0.0027540000000000004,
+      "loss": 4.562,
+      "step": 918
+    },
+    {
+      "epoch": 0.00919,
+      "grad_norm": 0.8358169198036194,
+      "learning_rate": 0.0027570000000000003,
+      "loss": 4.5255,
+      "step": 919
+    },
+    {
+      "epoch": 0.0092,
+      "grad_norm": 0.8860310316085815,
+      "learning_rate": 0.0027600000000000003,
+      "loss": 4.5423,
+      "step": 920
+    },
+    {
+      "epoch": 0.00921,
+      "grad_norm": 1.0510568618774414,
+      "learning_rate": 0.0027630000000000003,
+      "loss": 4.5336,
+      "step": 921
+    },
+    {
+      "epoch": 0.00922,
+      "grad_norm": 1.1311016082763672,
+      "learning_rate": 0.0027660000000000002,
+      "loss": 4.5483,
+      "step": 922
+    },
+    {
+      "epoch": 0.00923,
+      "grad_norm": 1.1111584901809692,
+      "learning_rate": 0.002769,
+      "loss": 4.5339,
+      "step": 923
+    },
+    {
+      "epoch": 0.00924,
+      "grad_norm": 0.9117261171340942,
+      "learning_rate": 0.002772,
+      "loss": 4.5412,
+      "step": 924
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 0.9757253527641296,
+      "learning_rate": 0.002775,
+      "loss": 4.5263,
+      "step": 925
+    },
+    {
+      "epoch": 0.00926,
+      "grad_norm": 0.907768189907074,
+      "learning_rate": 0.002778,
+      "loss": 4.5422,
+      "step": 926
+    },
+    {
+      "epoch": 0.00927,
+      "grad_norm": 0.8181371092796326,
+      "learning_rate": 0.002781,
+      "loss": 4.4999,
+      "step": 927
+    },
+    {
+      "epoch": 0.00928,
+      "grad_norm": 0.7728373408317566,
+      "learning_rate": 0.002784,
+      "loss": 4.5461,
+      "step": 928
+    },
+    {
+      "epoch": 0.00929,
+      "grad_norm": 0.624686598777771,
+      "learning_rate": 0.0027870000000000004,
+      "loss": 4.5192,
+      "step": 929
+    },
+    {
+      "epoch": 0.0093,
+      "grad_norm": 0.6143611073493958,
+      "learning_rate": 0.0027900000000000004,
+      "loss": 4.4968,
+      "step": 930
+    },
+    {
+      "epoch": 0.00931,
+      "grad_norm": 0.5353983044624329,
+      "learning_rate": 0.0027930000000000003,
+      "loss": 4.5041,
+      "step": 931
+    },
+    {
+      "epoch": 0.00932,
+      "grad_norm": 0.5034843683242798,
+      "learning_rate": 0.0027960000000000003,
+      "loss": 4.5081,
+      "step": 932
+    },
+    {
+      "epoch": 0.00933,
+      "grad_norm": 0.43557336926460266,
+      "learning_rate": 0.0027990000000000003,
+      "loss": 4.47,
+      "step": 933
+    },
+    {
+      "epoch": 0.00934,
+      "grad_norm": 0.42429375648498535,
+      "learning_rate": 0.0028020000000000002,
+      "loss": 4.4927,
+      "step": 934
+    },
+    {
+      "epoch": 0.00935,
+      "grad_norm": 0.4439206123352051,
+      "learning_rate": 0.002805,
+      "loss": 4.5015,
+      "step": 935
+    },
+    {
+      "epoch": 0.00936,
+      "grad_norm": 0.4363570511341095,
+      "learning_rate": 0.002808,
+      "loss": 4.4989,
+      "step": 936
+    },
+    {
+      "epoch": 0.00937,
+      "grad_norm": 0.4996969401836395,
+      "learning_rate": 0.002811,
+      "loss": 4.5097,
+      "step": 937
+    },
+    {
+      "epoch": 0.00938,
+      "grad_norm": 0.5577415227890015,
+      "learning_rate": 0.002814,
+      "loss": 4.4882,
+      "step": 938
+    },
+    {
+      "epoch": 0.00939,
+      "grad_norm": 0.6116971373558044,
+      "learning_rate": 0.002817,
+      "loss": 4.4823,
+      "step": 939
+    },
+    {
+      "epoch": 0.0094,
+      "grad_norm": 0.675699770450592,
+      "learning_rate": 0.00282,
+      "loss": 4.4838,
+      "step": 940
+    },
+    {
+      "epoch": 0.00941,
+      "grad_norm": 0.8027246594429016,
+      "learning_rate": 0.002823,
+      "loss": 4.4671,
+      "step": 941
+    },
+    {
+      "epoch": 0.00942,
+      "grad_norm": 0.9442744851112366,
+      "learning_rate": 0.002826,
+      "loss": 4.5129,
+      "step": 942
+    },
+    {
+      "epoch": 0.00943,
+      "grad_norm": 0.963239312171936,
+      "learning_rate": 0.002829,
+      "loss": 4.4975,
+      "step": 943
+    },
+    {
+      "epoch": 0.00944,
+      "grad_norm": 0.8721352815628052,
+      "learning_rate": 0.002832,
+      "loss": 4.493,
+      "step": 944
+    },
+    {
+      "epoch": 0.00945,
+      "grad_norm": 1.0316184759140015,
+      "learning_rate": 0.002835,
+      "loss": 4.5084,
+      "step": 945
+    },
+    {
+      "epoch": 0.00946,
+      "grad_norm": 0.7907041907310486,
+      "learning_rate": 0.002838,
+      "loss": 4.5093,
+      "step": 946
+    },
+    {
+      "epoch": 0.00947,
+      "grad_norm": 0.8508433699607849,
+      "learning_rate": 0.0028409999999999998,
+      "loss": 4.4859,
+      "step": 947
+    },
+    {
+      "epoch": 0.00948,
+      "grad_norm": 0.8942288756370544,
+      "learning_rate": 0.0028439999999999997,
+      "loss": 4.4805,
+      "step": 948
+    },
+    {
+      "epoch": 0.00949,
+      "grad_norm": 0.8804354071617126,
+      "learning_rate": 0.002847,
+      "loss": 4.5178,
+      "step": 949
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 0.9102524518966675,
+      "learning_rate": 0.00285,
+      "loss": 4.5104,
+      "step": 950
+    },
+    {
+      "epoch": 0.00951,
+      "grad_norm": 0.9140040278434753,
+      "learning_rate": 0.002853,
+      "loss": 4.4913,
+      "step": 951
+    },
+    {
+      "epoch": 0.00952,
+      "grad_norm": 0.996699869632721,
+      "learning_rate": 0.002856,
+      "loss": 4.5192,
+      "step": 952
+    },
+    {
+      "epoch": 0.00953,
+      "grad_norm": 1.0743249654769897,
+      "learning_rate": 0.002859,
+      "loss": 4.4742,
+      "step": 953
+    },
+    {
+      "epoch": 0.00954,
+      "grad_norm": 0.9619385004043579,
+      "learning_rate": 0.002862,
+      "loss": 4.532,
+      "step": 954
+    },
+    {
+      "epoch": 0.00955,
+      "grad_norm": 0.9820901155471802,
+      "learning_rate": 0.002865,
+      "loss": 4.5178,
+      "step": 955
+    },
+    {
+      "epoch": 0.00956,
+      "grad_norm": 1.0036413669586182,
+      "learning_rate": 0.002868,
+      "loss": 4.501,
+      "step": 956
+    },
+    {
+      "epoch": 0.00957,
+      "grad_norm": 0.8994410634040833,
+      "learning_rate": 0.002871,
+      "loss": 4.5266,
+      "step": 957
+    },
+    {
+      "epoch": 0.00958,
+      "grad_norm": 0.910679817199707,
+      "learning_rate": 0.002874,
+      "loss": 4.5103,
+      "step": 958
+    },
+    {
+      "epoch": 0.00959,
+      "grad_norm": 0.7567980885505676,
+      "learning_rate": 0.002877,
+      "loss": 4.4958,
+      "step": 959
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 0.7841255068778992,
+      "learning_rate": 0.0028799999999999997,
+      "loss": 4.4967,
+      "step": 960
+    },
+    {
+      "epoch": 0.00961,
+      "grad_norm": 0.6947323083877563,
+      "learning_rate": 0.002883,
+      "loss": 4.4665,
+      "step": 961
+    },
+    {
+      "epoch": 0.00962,
+      "grad_norm": 0.5831562280654907,
+      "learning_rate": 0.002886,
+      "loss": 4.4887,
+      "step": 962
+    },
+    {
+      "epoch": 0.00963,
+      "grad_norm": 0.5601492524147034,
+      "learning_rate": 0.002889,
+      "loss": 4.4846,
+      "step": 963
+    },
+    {
+      "epoch": 0.00964,
+      "grad_norm": 0.474755197763443,
+      "learning_rate": 0.002892,
+      "loss": 4.47,
+      "step": 964
+    },
+    {
+      "epoch": 0.00965,
+      "grad_norm": 0.5203716158866882,
+      "learning_rate": 0.002895,
+      "loss": 4.4451,
+      "step": 965
+    },
+    {
+      "epoch": 0.00966,
+      "grad_norm": 0.4714201092720032,
+      "learning_rate": 0.002898,
+      "loss": 4.4572,
+      "step": 966
+    },
+    {
+      "epoch": 0.00967,
+      "grad_norm": 0.4852457344532013,
+      "learning_rate": 0.002901,
+      "loss": 4.4924,
+      "step": 967
+    },
+    {
+      "epoch": 0.00968,
+      "grad_norm": 0.5026896595954895,
+      "learning_rate": 0.002904,
+      "loss": 4.4559,
+      "step": 968
+    },
+    {
+      "epoch": 0.00969,
+      "grad_norm": 0.5138400793075562,
+      "learning_rate": 0.002907,
+      "loss": 4.4843,
+      "step": 969
+    },
+    {
+      "epoch": 0.0097,
+      "grad_norm": 0.46154963970184326,
+      "learning_rate": 0.00291,
+      "loss": 4.4402,
+      "step": 970
+    },
+    {
+      "epoch": 0.00971,
+      "grad_norm": 0.4250973165035248,
+      "learning_rate": 0.002913,
+      "loss": 4.4437,
+      "step": 971
+    },
+    {
+      "epoch": 0.00972,
+      "grad_norm": 0.46983814239501953,
+      "learning_rate": 0.002916,
+      "loss": 4.45,
+      "step": 972
+    },
+    {
+      "epoch": 0.00973,
+      "grad_norm": 0.578143835067749,
+      "learning_rate": 0.002919,
+      "loss": 4.4478,
+      "step": 973
+    },
+    {
+      "epoch": 0.00974,
+      "grad_norm": 0.7184126377105713,
+      "learning_rate": 0.002922,
+      "loss": 4.4366,
+      "step": 974
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 0.8322773575782776,
+      "learning_rate": 0.002925,
+      "loss": 4.4683,
+      "step": 975
+    },
+    {
+      "epoch": 0.00976,
+      "grad_norm": 0.7910879850387573,
+      "learning_rate": 0.002928,
+      "loss": 4.4497,
+      "step": 976
+    },
+    {
+      "epoch": 0.00977,
+      "grad_norm": 0.7662490606307983,
+      "learning_rate": 0.002931,
+      "loss": 4.4868,
+      "step": 977
+    },
+    {
+      "epoch": 0.00978,
+      "grad_norm": 0.6879754066467285,
+      "learning_rate": 0.002934,
+      "loss": 4.4568,
+      "step": 978
+    },
+    {
+      "epoch": 0.00979,
+      "grad_norm": 0.7417387366294861,
+      "learning_rate": 0.002937,
+      "loss": 4.4491,
+      "step": 979
+    },
+    {
+      "epoch": 0.0098,
+      "grad_norm": 0.6087706089019775,
+      "learning_rate": 0.00294,
+      "loss": 4.4405,
+      "step": 980
+    },
+    {
+      "epoch": 0.00981,
+      "grad_norm": 0.5661059617996216,
+      "learning_rate": 0.002943,
+      "loss": 4.4516,
+      "step": 981
+    },
+    {
+      "epoch": 0.00982,
+      "grad_norm": 0.5970472097396851,
+      "learning_rate": 0.002946,
+      "loss": 4.4435,
+      "step": 982
+    },
+    {
+      "epoch": 0.00983,
+      "grad_norm": 0.6391454935073853,
+      "learning_rate": 0.0029490000000000002,
+      "loss": 4.4444,
+      "step": 983
+    },
+    {
+      "epoch": 0.00984,
+      "grad_norm": 0.5948253870010376,
+      "learning_rate": 0.002952,
+      "loss": 4.4303,
+      "step": 984
+    },
+    {
+      "epoch": 0.00985,
+      "grad_norm": 0.5908463597297668,
+      "learning_rate": 0.002955,
+      "loss": 4.4509,
+      "step": 985
+    },
+    {
+      "epoch": 0.00986,
+      "grad_norm": 0.6963019371032715,
+      "learning_rate": 0.002958,
+      "loss": 4.4177,
+      "step": 986
+    },
+    {
+      "epoch": 0.00987,
+      "grad_norm": 0.7601437568664551,
+      "learning_rate": 0.002961,
+      "loss": 4.4332,
+      "step": 987
+    },
+    {
+      "epoch": 0.00988,
+      "grad_norm": 0.8287732005119324,
+      "learning_rate": 0.002964,
+      "loss": 4.4707,
+      "step": 988
+    },
+    {
+      "epoch": 0.00989,
+      "grad_norm": 0.9172109961509705,
+      "learning_rate": 0.002967,
+      "loss": 4.4168,
+      "step": 989
+    },
+    {
+      "epoch": 0.0099,
+      "grad_norm": 0.9710732102394104,
+      "learning_rate": 0.00297,
+      "loss": 4.4601,
+      "step": 990
+    },
+    {
+      "epoch": 0.00991,
+      "grad_norm": 0.9308454990386963,
+      "learning_rate": 0.002973,
+      "loss": 4.4596,
+      "step": 991
+    },
+    {
+      "epoch": 0.00992,
+      "grad_norm": 0.8575865030288696,
+      "learning_rate": 0.002976,
+      "loss": 4.4567,
+      "step": 992
+    },
+    {
+      "epoch": 0.00993,
+      "grad_norm": 0.8846513032913208,
+      "learning_rate": 0.002979,
+      "loss": 4.4578,
+      "step": 993
+    },
+    {
+      "epoch": 0.00994,
+      "grad_norm": 1.0507467985153198,
+      "learning_rate": 0.002982,
+      "loss": 4.4695,
+      "step": 994
+    },
+    {
+      "epoch": 0.00995,
+      "grad_norm": 0.9225064516067505,
+      "learning_rate": 0.0029850000000000002,
+      "loss": 4.472,
+      "step": 995
+    },
+    {
+      "epoch": 0.00996,
+      "grad_norm": 1.0122137069702148,
+      "learning_rate": 0.002988,
+      "loss": 4.4787,
+      "step": 996
+    },
+    {
+      "epoch": 0.00997,
+      "grad_norm": 1.1055110692977905,
+      "learning_rate": 0.002991,
+      "loss": 4.4822,
+      "step": 997
+    },
+    {
+      "epoch": 0.00998,
+      "grad_norm": 1.0749611854553223,
+      "learning_rate": 0.002994,
+      "loss": 4.4866,
+      "step": 998
+    },
+    {
+      "epoch": 0.00999,
+      "grad_norm": 1.0997530221939087,
+      "learning_rate": 0.002997,
+      "loss": 4.469,
+      "step": 999
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.853782057762146,
+      "learning_rate": 0.003,
+      "loss": 4.4844,
+      "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": 2.5210740801536e+16,
+  "train_batch_size": 256,
+  "trial_name": null,
+  "trial_params": null
+}