diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7040 @@
+{
+  "best_metric": 0.29665234684944153,
+  "best_model_checkpoint": "sentiment_classification/checkpoint-1000",
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.7555551528930664,
+      "learning_rate": 9.99e-05,
+      "loss": 0.0054,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 7.883056640625,
+      "learning_rate": 9.98e-05,
+      "loss": 0.0587,
+      "step": 2
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 12.585994720458984,
+      "learning_rate": 9.970000000000001e-05,
+      "loss": 0.1564,
+      "step": 3
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 36.28175735473633,
+      "learning_rate": 9.960000000000001e-05,
+      "loss": 1.0358,
+      "step": 4
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 35.03889083862305,
+      "learning_rate": 9.95e-05,
+      "loss": 1.0603,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 46.80266189575195,
+      "learning_rate": 9.94e-05,
+      "loss": 0.8565,
+      "step": 6
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 55.776710510253906,
+      "learning_rate": 9.93e-05,
+      "loss": 1.0901,
+      "step": 7
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 30.31947898864746,
+      "learning_rate": 9.92e-05,
+      "loss": 0.3881,
+      "step": 8
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 37.395389556884766,
+      "learning_rate": 9.910000000000001e-05,
+      "loss": 0.8026,
+      "step": 9
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 46.29269790649414,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 1.1394,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 44.72748565673828,
+      "learning_rate": 9.89e-05,
+      "loss": 1.8889,
+      "step": 11
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 50.34558868408203,
+      "learning_rate": 9.88e-05,
+      "loss": 1.9163,
+      "step": 12
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 50.587432861328125,
+      "learning_rate": 9.87e-05,
+      "loss": 1.101,
+      "step": 13
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 40.331085205078125,
+      "learning_rate": 9.86e-05,
+      "loss": 1.4982,
+      "step": 14
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 42.99494552612305,
+      "learning_rate": 9.850000000000001e-05,
+      "loss": 1.2145,
+      "step": 15
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 65.54414367675781,
+      "learning_rate": 9.84e-05,
+      "loss": 1.7914,
+      "step": 16
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 23.524738311767578,
+      "learning_rate": 9.83e-05,
+      "loss": 0.4952,
+      "step": 17
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 54.68332290649414,
+      "learning_rate": 9.82e-05,
+      "loss": 0.8135,
+      "step": 18
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 25.295198440551758,
+      "learning_rate": 9.81e-05,
+      "loss": 0.42,
+      "step": 19
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 64.28269958496094,
+      "learning_rate": 9.8e-05,
+      "loss": 1.8965,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 59.879634857177734,
+      "learning_rate": 9.790000000000001e-05,
+      "loss": 1.9614,
+      "step": 21
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 28.60529136657715,
+      "learning_rate": 9.78e-05,
+      "loss": 1.1231,
+      "step": 22
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 52.29844665527344,
+      "learning_rate": 9.77e-05,
+      "loss": 1.0977,
+      "step": 23
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 50.24287033081055,
+      "learning_rate": 9.76e-05,
+      "loss": 2.2355,
+      "step": 24
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 17.639894485473633,
+      "learning_rate": 9.75e-05,
+      "loss": 0.1371,
+      "step": 25
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 40.8584098815918,
+      "learning_rate": 9.74e-05,
+      "loss": 1.0995,
+      "step": 26
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 55.463260650634766,
+      "learning_rate": 9.730000000000001e-05,
+      "loss": 1.0458,
+      "step": 27
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 42.90127944946289,
+      "learning_rate": 9.72e-05,
+      "loss": 1.0041,
+      "step": 28
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 22.913488388061523,
+      "learning_rate": 9.71e-05,
+      "loss": 0.4278,
+      "step": 29
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 38.93981170654297,
+      "learning_rate": 9.7e-05,
+      "loss": 0.5715,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 21.701683044433594,
+      "learning_rate": 9.69e-05,
+      "loss": 0.3351,
+      "step": 31
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 47.243526458740234,
+      "learning_rate": 9.680000000000001e-05,
+      "loss": 0.9803,
+      "step": 32
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.447603225708008,
+      "learning_rate": 9.67e-05,
+      "loss": 0.0597,
+      "step": 33
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 51.16413879394531,
+      "learning_rate": 9.66e-05,
+      "loss": 1.8308,
+      "step": 34
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 21.44324493408203,
+      "learning_rate": 9.65e-05,
+      "loss": 0.206,
+      "step": 35
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 31.796432495117188,
+      "learning_rate": 9.64e-05,
+      "loss": 0.8214,
+      "step": 36
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 37.89706802368164,
+      "learning_rate": 9.63e-05,
+      "loss": 0.7969,
+      "step": 37
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 34.30683898925781,
+      "learning_rate": 9.620000000000001e-05,
+      "loss": 0.7503,
+      "step": 38
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 44.750328063964844,
+      "learning_rate": 9.61e-05,
+      "loss": 0.9113,
+      "step": 39
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 29.14406967163086,
+      "learning_rate": 9.6e-05,
+      "loss": 0.399,
+      "step": 40
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 28.217470169067383,
+      "learning_rate": 9.59e-05,
+      "loss": 0.4654,
+      "step": 41
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 17.639678955078125,
+      "learning_rate": 9.58e-05,
+      "loss": 0.3132,
+      "step": 42
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 21.909076690673828,
+      "learning_rate": 9.57e-05,
+      "loss": 0.448,
+      "step": 43
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.7724447846412659,
+      "learning_rate": 9.56e-05,
+      "loss": 0.0066,
+      "step": 44
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.8613312244415283,
+      "learning_rate": 9.55e-05,
+      "loss": 0.0286,
+      "step": 45
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 34.55177307128906,
+      "learning_rate": 9.54e-05,
+      "loss": 0.7055,
+      "step": 46
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 31.46555519104004,
+      "learning_rate": 9.53e-05,
+      "loss": 0.5292,
+      "step": 47
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 28.770160675048828,
+      "learning_rate": 9.52e-05,
+      "loss": 1.0285,
+      "step": 48
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 21.542871475219727,
+      "learning_rate": 9.51e-05,
+      "loss": 0.2557,
+      "step": 49
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 44.05634307861328,
+      "learning_rate": 9.5e-05,
+      "loss": 0.6056,
+      "step": 50
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 28.97549819946289,
+      "learning_rate": 9.49e-05,
+      "loss": 0.6857,
+      "step": 51
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 9.51518440246582,
+      "learning_rate": 9.48e-05,
+      "loss": 0.09,
+      "step": 52
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.6027193665504456,
+      "learning_rate": 9.47e-05,
+      "loss": 0.0071,
+      "step": 53
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.3073718547821045,
+      "learning_rate": 9.46e-05,
+      "loss": 0.0252,
+      "step": 54
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 25.46698570251465,
+      "learning_rate": 9.449999999999999e-05,
+      "loss": 0.5108,
+      "step": 55
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 14.047112464904785,
+      "learning_rate": 9.44e-05,
+      "loss": 0.1018,
+      "step": 56
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 9.882685661315918,
+      "learning_rate": 9.43e-05,
+      "loss": 0.0764,
+      "step": 57
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 42.48798751831055,
+      "learning_rate": 9.42e-05,
+      "loss": 1.0458,
+      "step": 58
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 31.454246520996094,
+      "learning_rate": 9.41e-05,
+      "loss": 0.5511,
+      "step": 59
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 16.312801361083984,
+      "learning_rate": 9.4e-05,
+      "loss": 0.1688,
+      "step": 60
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 31.64558219909668,
+      "learning_rate": 9.39e-05,
+      "loss": 0.4796,
+      "step": 61
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 16.89689826965332,
+      "learning_rate": 9.38e-05,
+      "loss": 0.1242,
+      "step": 62
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 38.343963623046875,
+      "learning_rate": 9.370000000000001e-05,
+      "loss": 0.4748,
+      "step": 63
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 30.578535079956055,
+      "learning_rate": 9.360000000000001e-05,
+      "loss": 0.8823,
+      "step": 64
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 26.075117111206055,
+      "learning_rate": 9.350000000000001e-05,
+      "loss": 0.3553,
+      "step": 65
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 30.658485412597656,
+      "learning_rate": 9.340000000000001e-05,
+      "loss": 0.5925,
+      "step": 66
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 16.315160751342773,
+      "learning_rate": 9.33e-05,
+      "loss": 0.1272,
+      "step": 67
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 29.514232635498047,
+      "learning_rate": 9.320000000000002e-05,
+      "loss": 0.5304,
+      "step": 68
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.8858259320259094,
+      "learning_rate": 9.310000000000001e-05,
+      "loss": 0.0042,
+      "step": 69
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 34.349971771240234,
+      "learning_rate": 9.300000000000001e-05,
+      "loss": 1.4723,
+      "step": 70
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 12.096094131469727,
+      "learning_rate": 9.290000000000001e-05,
+      "loss": 0.0912,
+      "step": 71
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 25.67418098449707,
+      "learning_rate": 9.28e-05,
+      "loss": 0.2763,
+      "step": 72
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 9.287317276000977,
+      "learning_rate": 9.27e-05,
+      "loss": 0.0993,
+      "step": 73
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 13.056856155395508,
+      "learning_rate": 9.260000000000001e-05,
+      "loss": 0.1464,
+      "step": 74
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 19.23573875427246,
+      "learning_rate": 9.250000000000001e-05,
+      "loss": 0.1995,
+      "step": 75
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 30.34964942932129,
+      "learning_rate": 9.240000000000001e-05,
+      "loss": 0.331,
+      "step": 76
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 34.08774948120117,
+      "learning_rate": 9.230000000000001e-05,
+      "loss": 0.5528,
+      "step": 77
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 25.965503692626953,
+      "learning_rate": 9.22e-05,
+      "loss": 0.4419,
+      "step": 78
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 27.7506160736084,
+      "learning_rate": 9.21e-05,
+      "loss": 0.3613,
+      "step": 79
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 64.53148651123047,
+      "learning_rate": 9.200000000000001e-05,
+      "loss": 1.8911,
+      "step": 80
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 14.74899959564209,
+      "learning_rate": 9.190000000000001e-05,
+      "loss": 0.2018,
+      "step": 81
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 29.797199249267578,
+      "learning_rate": 9.180000000000001e-05,
+      "loss": 0.7492,
+      "step": 82
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 24.35639762878418,
+      "learning_rate": 9.17e-05,
+      "loss": 0.2577,
+      "step": 83
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 39.93398666381836,
+      "learning_rate": 9.16e-05,
+      "loss": 0.7685,
+      "step": 84
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 17.78274154663086,
+      "learning_rate": 9.15e-05,
+      "loss": 0.2235,
+      "step": 85
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 21.428129196166992,
+      "learning_rate": 9.140000000000001e-05,
+      "loss": 0.236,
+      "step": 86
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.9476423263549805,
+      "learning_rate": 9.130000000000001e-05,
+      "loss": 0.0323,
+      "step": 87
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.092775344848633,
+      "learning_rate": 9.120000000000001e-05,
+      "loss": 0.0189,
+      "step": 88
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 38.60344314575195,
+      "learning_rate": 9.11e-05,
+      "loss": 0.9759,
+      "step": 89
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 10.748154640197754,
+      "learning_rate": 9.1e-05,
+      "loss": 0.088,
+      "step": 90
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 31.464191436767578,
+      "learning_rate": 9.090000000000001e-05,
+      "loss": 0.5222,
+      "step": 91
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 27.113365173339844,
+      "learning_rate": 9.080000000000001e-05,
+      "loss": 0.2638,
+      "step": 92
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.8783392906188965,
+      "learning_rate": 9.070000000000001e-05,
+      "loss": 0.0234,
+      "step": 93
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 51.70059585571289,
+      "learning_rate": 9.06e-05,
+      "loss": 1.9132,
+      "step": 94
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 35.822940826416016,
+      "learning_rate": 9.05e-05,
+      "loss": 0.7936,
+      "step": 95
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 41.59641647338867,
+      "learning_rate": 9.04e-05,
+      "loss": 1.2705,
+      "step": 96
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.7700830698013306,
+      "learning_rate": 9.030000000000001e-05,
+      "loss": 0.0097,
+      "step": 97
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 78.31722259521484,
+      "learning_rate": 9.020000000000001e-05,
+      "loss": 2.2656,
+      "step": 98
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 35.84953689575195,
+      "learning_rate": 9.010000000000001e-05,
+      "loss": 0.4499,
+      "step": 99
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 32.15023422241211,
+      "learning_rate": 9e-05,
+      "loss": 1.3309,
+      "step": 100
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.091931343078613,
+      "learning_rate": 8.99e-05,
+      "loss": 0.0577,
+      "step": 101
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.4728281497955322,
+      "learning_rate": 8.98e-05,
+      "loss": 0.0151,
+      "step": 102
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 22.939411163330078,
+      "learning_rate": 8.970000000000001e-05,
+      "loss": 0.2908,
+      "step": 103
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 33.68719482421875,
+      "learning_rate": 8.960000000000001e-05,
+      "loss": 1.0016,
+      "step": 104
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 15.805360794067383,
+      "learning_rate": 8.950000000000001e-05,
+      "loss": 0.1251,
+      "step": 105
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 28.76931381225586,
+      "learning_rate": 8.94e-05,
+      "loss": 0.486,
+      "step": 106
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 63.06796646118164,
+      "learning_rate": 8.93e-05,
+      "loss": 1.705,
+      "step": 107
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 35.362327575683594,
+      "learning_rate": 8.92e-05,
+      "loss": 2.3002,
+      "step": 108
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 30.09642791748047,
+      "learning_rate": 8.910000000000001e-05,
+      "loss": 0.7912,
+      "step": 109
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 31.83744239807129,
+      "learning_rate": 8.900000000000001e-05,
+      "loss": 0.652,
+      "step": 110
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 26.239160537719727,
+      "learning_rate": 8.89e-05,
+      "loss": 0.4104,
+      "step": 111
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 17.760883331298828,
+      "learning_rate": 8.88e-05,
+      "loss": 0.2015,
+      "step": 112
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 27.477123260498047,
+      "learning_rate": 8.87e-05,
+      "loss": 0.4919,
+      "step": 113
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 51.267391204833984,
+      "learning_rate": 8.86e-05,
+      "loss": 0.6284,
+      "step": 114
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 29.60396957397461,
+      "learning_rate": 8.850000000000001e-05,
+      "loss": 0.2356,
+      "step": 115
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.065886497497559,
+      "learning_rate": 8.840000000000001e-05,
+      "loss": 0.034,
+      "step": 116
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 28.858583450317383,
+      "learning_rate": 8.83e-05,
+      "loss": 0.2402,
+      "step": 117
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.959737777709961,
+      "learning_rate": 8.82e-05,
+      "loss": 0.0418,
+      "step": 118
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 7.274974346160889,
+      "learning_rate": 8.81e-05,
+      "loss": 0.0469,
+      "step": 119
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 49.039154052734375,
+      "learning_rate": 8.800000000000001e-05,
+      "loss": 1.8188,
+      "step": 120
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 26.21894645690918,
+      "learning_rate": 8.790000000000001e-05,
+      "loss": 0.4633,
+      "step": 121
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.200336933135986,
+      "learning_rate": 8.78e-05,
+      "loss": 0.0348,
+      "step": 122
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.5993304252624512,
+      "learning_rate": 8.77e-05,
+      "loss": 0.0085,
+      "step": 123
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 9.848234176635742,
+      "learning_rate": 8.76e-05,
+      "loss": 0.0598,
+      "step": 124
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 31.79520034790039,
+      "learning_rate": 8.75e-05,
+      "loss": 0.9076,
+      "step": 125
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 28.447763442993164,
+      "learning_rate": 8.740000000000001e-05,
+      "loss": 0.8934,
+      "step": 126
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 27.019044876098633,
+      "learning_rate": 8.730000000000001e-05,
+      "loss": 0.6513,
+      "step": 127
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 48.12791061401367,
+      "learning_rate": 8.72e-05,
+      "loss": 2.4184,
+      "step": 128
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 39.648468017578125,
+      "learning_rate": 8.71e-05,
+      "loss": 0.9709,
+      "step": 129
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 19.506731033325195,
+      "learning_rate": 8.7e-05,
+      "loss": 0.1482,
+      "step": 130
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.624091148376465,
+      "learning_rate": 8.69e-05,
+      "loss": 0.0289,
+      "step": 131
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 33.03536605834961,
+      "learning_rate": 8.680000000000001e-05,
+      "loss": 0.523,
+      "step": 132
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 26.916067123413086,
+      "learning_rate": 8.67e-05,
+      "loss": 0.4418,
+      "step": 133
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.7219924926757812,
+      "learning_rate": 8.66e-05,
+      "loss": 0.0268,
+      "step": 134
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.1921474933624268,
+      "learning_rate": 8.65e-05,
+      "loss": 0.0063,
+      "step": 135
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 32.356571197509766,
+      "learning_rate": 8.64e-05,
+      "loss": 0.4848,
+      "step": 136
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 8.539983749389648,
+      "learning_rate": 8.63e-05,
+      "loss": 0.0491,
+      "step": 137
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 28.862144470214844,
+      "learning_rate": 8.620000000000001e-05,
+      "loss": 0.7974,
+      "step": 138
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 15.244535446166992,
+      "learning_rate": 8.61e-05,
+      "loss": 0.1273,
+      "step": 139
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 27.476613998413086,
+      "learning_rate": 8.6e-05,
+      "loss": 0.6099,
+      "step": 140
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 25.30959129333496,
+      "learning_rate": 8.59e-05,
+      "loss": 0.5524,
+      "step": 141
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 14.604026794433594,
+      "learning_rate": 8.58e-05,
+      "loss": 0.1036,
+      "step": 142
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 25.0202579498291,
+      "learning_rate": 8.57e-05,
+      "loss": 0.4411,
+      "step": 143
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 35.9989128112793,
+      "learning_rate": 8.560000000000001e-05,
+      "loss": 0.7138,
+      "step": 144
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.4863531291484833,
+      "learning_rate": 8.55e-05,
+      "loss": 0.0038,
+      "step": 145
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.6537394523620605,
+      "learning_rate": 8.54e-05,
+      "loss": 0.0223,
+      "step": 146
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 21.37197494506836,
+      "learning_rate": 8.53e-05,
+      "loss": 0.3206,
+      "step": 147
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8339968919754028,
+      "learning_rate": 8.52e-05,
+      "loss": 0.0056,
+      "step": 148
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 33.149024963378906,
+      "learning_rate": 8.510000000000001e-05,
+      "loss": 0.3708,
+      "step": 149
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 18.96430778503418,
+      "learning_rate": 8.5e-05,
+      "loss": 0.2126,
+      "step": 150
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.3087029457092285,
+      "learning_rate": 8.49e-05,
+      "loss": 0.0343,
+      "step": 151
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.4322396516799927,
+      "learning_rate": 8.48e-05,
+      "loss": 0.0099,
+      "step": 152
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 27.87354850769043,
+      "learning_rate": 8.47e-05,
+      "loss": 0.4536,
+      "step": 153
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 24.29855728149414,
+      "learning_rate": 8.46e-05,
+      "loss": 0.2568,
+      "step": 154
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 32.701412200927734,
+      "learning_rate": 8.450000000000001e-05,
+      "loss": 0.6025,
+      "step": 155
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 29.71053123474121,
+      "learning_rate": 8.44e-05,
+      "loss": 0.3161,
+      "step": 156
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 24.117475509643555,
+      "learning_rate": 8.43e-05,
+      "loss": 0.3355,
+      "step": 157
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.675106406211853,
+      "learning_rate": 8.42e-05,
+      "loss": 0.0131,
+      "step": 158
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.6950457096099854,
+      "learning_rate": 8.41e-05,
+      "loss": 0.0139,
+      "step": 159
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.965531349182129,
+      "learning_rate": 8.4e-05,
+      "loss": 0.0381,
+      "step": 160
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.490444660186768,
+      "learning_rate": 8.39e-05,
+      "loss": 0.0311,
+      "step": 161
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 48.338592529296875,
+      "learning_rate": 8.38e-05,
+      "loss": 2.2775,
+      "step": 162
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.9811841249465942,
+      "learning_rate": 8.37e-05,
+      "loss": 0.0054,
+      "step": 163
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 28.258628845214844,
+      "learning_rate": 8.36e-05,
+      "loss": 1.0993,
+      "step": 164
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 31.349571228027344,
+      "learning_rate": 8.35e-05,
+      "loss": 0.4425,
+      "step": 165
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.468545913696289,
+      "learning_rate": 8.34e-05,
+      "loss": 0.0119,
+      "step": 166
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.3878319263458252,
+      "learning_rate": 8.33e-05,
+      "loss": 0.0027,
+      "step": 167
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 17.001144409179688,
+      "learning_rate": 8.32e-05,
+      "loss": 0.1208,
+      "step": 168
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 11.259538650512695,
+      "learning_rate": 8.31e-05,
+      "loss": 0.0934,
+      "step": 169
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 28.391857147216797,
+      "learning_rate": 8.3e-05,
+      "loss": 0.6119,
+      "step": 170
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 12.239534378051758,
+      "learning_rate": 8.29e-05,
+      "loss": 0.1011,
+      "step": 171
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 19.20558738708496,
+      "learning_rate": 8.28e-05,
+      "loss": 0.1845,
+      "step": 172
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 28.89300537109375,
+      "learning_rate": 8.27e-05,
+      "loss": 0.2756,
+      "step": 173
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 34.91083908081055,
+      "learning_rate": 8.26e-05,
+      "loss": 0.7058,
+      "step": 174
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 7.6595845222473145,
+      "learning_rate": 8.25e-05,
+      "loss": 0.0658,
+      "step": 175
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 35.96283721923828,
+      "learning_rate": 8.24e-05,
+      "loss": 0.5961,
+      "step": 176
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.20796895027160645,
+      "learning_rate": 8.23e-05,
+      "loss": 0.0018,
+      "step": 177
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 11.105677604675293,
+      "learning_rate": 8.22e-05,
+      "loss": 0.0699,
+      "step": 178
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 45.597267150878906,
+      "learning_rate": 8.21e-05,
+      "loss": 0.7625,
+      "step": 179
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 14.93590259552002,
+      "learning_rate": 8.2e-05,
+      "loss": 0.1476,
+      "step": 180
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.5446606874465942,
+      "learning_rate": 8.19e-05,
+      "loss": 0.0101,
+      "step": 181
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.4029202461242676,
+      "learning_rate": 8.18e-05,
+      "loss": 0.0188,
+      "step": 182
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 26.945552825927734,
+      "learning_rate": 8.17e-05,
+      "loss": 1.0054,
+      "step": 183
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 17.689714431762695,
+      "learning_rate": 8.16e-05,
+      "loss": 0.1296,
+      "step": 184
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.4410698413848877,
+      "learning_rate": 8.15e-05,
+      "loss": 0.0168,
+      "step": 185
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 54.830902099609375,
+      "learning_rate": 8.14e-05,
+      "loss": 3.1109,
+      "step": 186
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 38.1295166015625,
+      "learning_rate": 8.13e-05,
+      "loss": 0.9705,
+      "step": 187
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.8395466804504395,
+      "learning_rate": 8.120000000000001e-05,
+      "loss": 0.0493,
+      "step": 188
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.1371989250183105,
+      "learning_rate": 8.11e-05,
+      "loss": 0.0165,
+      "step": 189
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 23.55173683166504,
+      "learning_rate": 8.1e-05,
+      "loss": 0.2548,
+      "step": 190
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 8.50500202178955,
+      "learning_rate": 8.090000000000001e-05,
+      "loss": 0.0517,
+      "step": 191
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 32.87669372558594,
+      "learning_rate": 8.080000000000001e-05,
+      "loss": 0.3359,
+      "step": 192
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.529865741729736,
+      "learning_rate": 8.070000000000001e-05,
+      "loss": 0.0391,
+      "step": 193
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 24.900270462036133,
+      "learning_rate": 8.060000000000001e-05,
+      "loss": 0.2896,
+      "step": 194
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 17.59227752685547,
+      "learning_rate": 8.05e-05,
+      "loss": 0.1772,
+      "step": 195
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.2958277463912964,
+      "learning_rate": 8.04e-05,
+      "loss": 0.0067,
+      "step": 196
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 16.34239959716797,
+      "learning_rate": 8.030000000000001e-05,
+      "loss": 0.1426,
+      "step": 197
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 27.314735412597656,
+      "learning_rate": 8.020000000000001e-05,
+      "loss": 1.4348,
+      "step": 198
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 19.829275131225586,
+      "learning_rate": 8.010000000000001e-05,
+      "loss": 0.5608,
+      "step": 199
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 28.39422607421875,
+      "learning_rate": 8e-05,
+      "loss": 1.4606,
+      "step": 200
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 31.092449188232422,
+      "learning_rate": 7.99e-05,
+      "loss": 0.2932,
+      "step": 201
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 51.73237991333008,
+      "learning_rate": 7.98e-05,
+      "loss": 1.8982,
+      "step": 202
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 8.397103309631348,
+      "learning_rate": 7.970000000000001e-05,
+      "loss": 0.0508,
+      "step": 203
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 22.450349807739258,
+      "learning_rate": 7.960000000000001e-05,
+      "loss": 0.1964,
+      "step": 204
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 38.33378982543945,
+      "learning_rate": 7.950000000000001e-05,
+      "loss": 0.9211,
+      "step": 205
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 53.80275344848633,
+      "learning_rate": 7.94e-05,
+      "loss": 1.2505,
+      "step": 206
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.610823154449463,
+      "learning_rate": 7.93e-05,
+      "loss": 0.0086,
+      "step": 207
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.0060709137469530106,
+      "learning_rate": 7.920000000000001e-05,
+      "loss": 0.0,
+      "step": 208
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0124675035476685,
+      "learning_rate": 7.910000000000001e-05,
+      "loss": 0.0066,
+      "step": 209
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 27.3523006439209,
+      "learning_rate": 7.900000000000001e-05,
+      "loss": 0.3906,
+      "step": 210
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 23.502004623413086,
+      "learning_rate": 7.890000000000001e-05,
+      "loss": 0.3856,
+      "step": 211
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.276196241378784,
+      "learning_rate": 7.88e-05,
+      "loss": 0.0203,
+      "step": 212
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 20.402345657348633,
+      "learning_rate": 7.87e-05,
+      "loss": 0.1852,
+      "step": 213
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.2247250080108643,
+      "learning_rate": 7.860000000000001e-05,
+      "loss": 0.0105,
+      "step": 214
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.4587538540363312,
+      "learning_rate": 7.850000000000001e-05,
+      "loss": 0.0025,
+      "step": 215
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 25.925575256347656,
+      "learning_rate": 7.840000000000001e-05,
+      "loss": 0.3174,
+      "step": 216
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 23.675764083862305,
+      "learning_rate": 7.83e-05,
+      "loss": 0.433,
+      "step": 217
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 20.47338104248047,
+      "learning_rate": 7.82e-05,
+      "loss": 0.1766,
+      "step": 218
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 27.251806259155273,
+      "learning_rate": 7.81e-05,
+      "loss": 0.9316,
+      "step": 219
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 40.24387741088867,
+      "learning_rate": 7.800000000000001e-05,
+      "loss": 0.6914,
+      "step": 220
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 34.86661911010742,
+      "learning_rate": 7.790000000000001e-05,
+      "loss": 0.705,
+      "step": 221
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.656433343887329,
+      "learning_rate": 7.780000000000001e-05,
+      "loss": 0.015,
+      "step": 222
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.383270740509033,
+      "learning_rate": 7.77e-05,
+      "loss": 0.0246,
+      "step": 223
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 25.344125747680664,
+      "learning_rate": 7.76e-05,
+      "loss": 0.6604,
+      "step": 224
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.3705878257751465,
+      "learning_rate": 7.75e-05,
+      "loss": 0.0108,
+      "step": 225
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.18879830837249756,
+      "learning_rate": 7.740000000000001e-05,
+      "loss": 0.0013,
+      "step": 226
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 38.107086181640625,
+      "learning_rate": 7.730000000000001e-05,
+      "loss": 1.5827,
+      "step": 227
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 8.14284610748291,
+      "learning_rate": 7.72e-05,
+      "loss": 0.0449,
+      "step": 228
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 10.337870597839355,
+      "learning_rate": 7.71e-05,
+      "loss": 0.06,
+      "step": 229
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 18.004863739013672,
+      "learning_rate": 7.7e-05,
+      "loss": 0.1319,
+      "step": 230
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 34.32221984863281,
+      "learning_rate": 7.69e-05,
+      "loss": 0.2753,
+      "step": 231
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 43.25252914428711,
+      "learning_rate": 7.680000000000001e-05,
+      "loss": 1.961,
+      "step": 232
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 18.302125930786133,
+      "learning_rate": 7.670000000000001e-05,
+      "loss": 0.1674,
+      "step": 233
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 7.26860237121582,
+      "learning_rate": 7.66e-05,
+      "loss": 0.0528,
+      "step": 234
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 17.053834915161133,
+      "learning_rate": 7.65e-05,
+      "loss": 0.1233,
+      "step": 235
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 27.578536987304688,
+      "learning_rate": 7.64e-05,
+      "loss": 0.4018,
+      "step": 236
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 31.73558807373047,
+      "learning_rate": 7.630000000000001e-05,
+      "loss": 1.2531,
+      "step": 237
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 27.72031593322754,
+      "learning_rate": 7.620000000000001e-05,
+      "loss": 1.6853,
+      "step": 238
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 9.15406608581543,
+      "learning_rate": 7.61e-05,
+      "loss": 0.0539,
+      "step": 239
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 12.243758201599121,
+      "learning_rate": 7.6e-05,
+      "loss": 0.0753,
+      "step": 240
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.238646030426025,
+      "learning_rate": 7.59e-05,
+      "loss": 0.0304,
+      "step": 241
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 12.011713981628418,
+      "learning_rate": 7.58e-05,
+      "loss": 0.0777,
+      "step": 242
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 25.589540481567383,
+      "learning_rate": 7.570000000000001e-05,
+      "loss": 0.7472,
+      "step": 243
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.752915859222412,
+      "learning_rate": 7.560000000000001e-05,
+      "loss": 0.0184,
+      "step": 244
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3850865960121155,
+      "learning_rate": 7.55e-05,
+      "loss": 0.0023,
+      "step": 245
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.0657126903533936,
+      "learning_rate": 7.54e-05,
+      "loss": 0.0139,
+      "step": 246
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.209764003753662,
+      "learning_rate": 7.53e-05,
+      "loss": 0.0278,
+      "step": 247
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 26.664770126342773,
+      "learning_rate": 7.52e-05,
+      "loss": 0.4915,
+      "step": 248
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 22.27918815612793,
+      "learning_rate": 7.510000000000001e-05,
+      "loss": 0.2126,
+      "step": 249
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 13.991991996765137,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 0.0893,
+      "step": 250
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 28.546430587768555,
+      "learning_rate": 7.49e-05,
+      "loss": 0.7366,
+      "step": 251
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 40.69116973876953,
+      "learning_rate": 7.48e-05,
+      "loss": 0.825,
+      "step": 252
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 10.466350555419922,
+      "learning_rate": 7.47e-05,
+      "loss": 0.0609,
+      "step": 253
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 41.81674575805664,
+      "learning_rate": 7.46e-05,
+      "loss": 0.7904,
+      "step": 254
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 24.998729705810547,
+      "learning_rate": 7.450000000000001e-05,
+      "loss": 0.3079,
+      "step": 255
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 29.997278213500977,
+      "learning_rate": 7.44e-05,
+      "loss": 0.2313,
+      "step": 256
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 24.29697608947754,
+      "learning_rate": 7.43e-05,
+      "loss": 0.2985,
+      "step": 257
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.0109050273895264,
+      "learning_rate": 7.42e-05,
+      "loss": 0.008,
+      "step": 258
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.4110793173313141,
+      "learning_rate": 7.41e-05,
+      "loss": 0.0022,
+      "step": 259
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 22.185529708862305,
+      "learning_rate": 7.4e-05,
+      "loss": 0.2034,
+      "step": 260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 38.921875,
+      "learning_rate": 7.390000000000001e-05,
+      "loss": 0.7375,
+      "step": 261
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 21.488807678222656,
+      "learning_rate": 7.38e-05,
+      "loss": 0.194,
+      "step": 262
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 10.119009017944336,
+      "learning_rate": 7.37e-05,
+      "loss": 0.0688,
+      "step": 263
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 23.405567169189453,
+      "learning_rate": 7.36e-05,
+      "loss": 0.3999,
+      "step": 264
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.5199978351593018,
+      "learning_rate": 7.35e-05,
+      "loss": 0.0152,
+      "step": 265
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.482628583908081,
+      "learning_rate": 7.340000000000001e-05,
+      "loss": 0.0127,
+      "step": 266
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.1299862265586853,
+      "learning_rate": 7.33e-05,
+      "loss": 0.0012,
+      "step": 267
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.690964460372925,
+      "learning_rate": 7.32e-05,
+      "loss": 0.0133,
+      "step": 268
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.540533065795898,
+      "learning_rate": 7.31e-05,
+      "loss": 0.0407,
+      "step": 269
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.2752537727355957,
+      "learning_rate": 7.3e-05,
+      "loss": 0.0121,
+      "step": 270
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 13.641573905944824,
+      "learning_rate": 7.29e-05,
+      "loss": 0.0854,
+      "step": 271
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 28.05483055114746,
+      "learning_rate": 7.280000000000001e-05,
+      "loss": 0.5136,
+      "step": 272
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 23.613855361938477,
+      "learning_rate": 7.27e-05,
+      "loss": 0.3377,
+      "step": 273
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 18.591848373413086,
+      "learning_rate": 7.26e-05,
+      "loss": 0.2001,
+      "step": 274
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.20727111399173737,
+      "learning_rate": 7.25e-05,
+      "loss": 0.0017,
+      "step": 275
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 10.257818222045898,
+      "learning_rate": 7.24e-05,
+      "loss": 0.0567,
+      "step": 276
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.06034255027771,
+      "learning_rate": 7.23e-05,
+      "loss": 0.017,
+      "step": 277
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 26.740501403808594,
+      "learning_rate": 7.22e-05,
+      "loss": 1.564,
+      "step": 278
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.23033320903778076,
+      "learning_rate": 7.21e-05,
+      "loss": 0.0013,
+      "step": 279
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 18.267688751220703,
+      "learning_rate": 7.2e-05,
+      "loss": 0.1606,
+      "step": 280
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.9552924633026123,
+      "learning_rate": 7.19e-05,
+      "loss": 0.0119,
+      "step": 281
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3651779294013977,
+      "learning_rate": 7.18e-05,
+      "loss": 0.0032,
+      "step": 282
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 9.202610969543457,
+      "learning_rate": 7.17e-05,
+      "loss": 0.0506,
+      "step": 283
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 36.84799575805664,
+      "learning_rate": 7.16e-05,
+      "loss": 1.1391,
+      "step": 284
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 36.70115661621094,
+      "learning_rate": 7.15e-05,
+      "loss": 1.0974,
+      "step": 285
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 24.444910049438477,
+      "learning_rate": 7.14e-05,
+      "loss": 0.2421,
+      "step": 286
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 20.180049896240234,
+      "learning_rate": 7.13e-05,
+      "loss": 0.2202,
+      "step": 287
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 23.31342124938965,
+      "learning_rate": 7.12e-05,
+      "loss": 0.1958,
+      "step": 288
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 15.620962142944336,
+      "learning_rate": 7.11e-05,
+      "loss": 0.1034,
+      "step": 289
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 27.025894165039062,
+      "learning_rate": 7.1e-05,
+      "loss": 0.5174,
+      "step": 290
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 25.132476806640625,
+      "learning_rate": 7.09e-05,
+      "loss": 0.423,
+      "step": 291
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.431723117828369,
+      "learning_rate": 7.08e-05,
+      "loss": 0.0509,
+      "step": 292
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 16.48543357849121,
+      "learning_rate": 7.07e-05,
+      "loss": 0.1131,
+      "step": 293
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 29.093488693237305,
+      "learning_rate": 7.06e-05,
+      "loss": 0.5578,
+      "step": 294
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 11.249940872192383,
+      "learning_rate": 7.05e-05,
+      "loss": 0.0791,
+      "step": 295
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 9.518302917480469,
+      "learning_rate": 7.04e-05,
+      "loss": 0.0549,
+      "step": 296
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.03421126306056976,
+      "learning_rate": 7.03e-05,
+      "loss": 0.0003,
+      "step": 297
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 30.86328125,
+      "learning_rate": 7.02e-05,
+      "loss": 0.8866,
+      "step": 298
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 25.39554214477539,
+      "learning_rate": 7.01e-05,
+      "loss": 0.2173,
+      "step": 299
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 27.889158248901367,
+      "learning_rate": 7e-05,
+      "loss": 0.9736,
+      "step": 300
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 11.48615837097168,
+      "learning_rate": 6.99e-05,
+      "loss": 0.0671,
+      "step": 301
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 17.868688583374023,
+      "learning_rate": 6.98e-05,
+      "loss": 0.1292,
+      "step": 302
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 11.254850387573242,
+      "learning_rate": 6.97e-05,
+      "loss": 0.0739,
+      "step": 303
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.120415687561035,
+      "learning_rate": 6.96e-05,
+      "loss": 0.0103,
+      "step": 304
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.4435783922672272,
+      "learning_rate": 6.95e-05,
+      "loss": 0.0029,
+      "step": 305
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 7.28007173538208,
+      "learning_rate": 6.939999999999999e-05,
+      "loss": 0.0388,
+      "step": 306
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 10.124960899353027,
+      "learning_rate": 6.93e-05,
+      "loss": 0.0583,
+      "step": 307
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 25.847856521606445,
+      "learning_rate": 6.92e-05,
+      "loss": 0.7839,
+      "step": 308
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 24.378753662109375,
+      "learning_rate": 6.91e-05,
+      "loss": 0.4693,
+      "step": 309
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 23.50893783569336,
+      "learning_rate": 6.9e-05,
+      "loss": 0.3675,
+      "step": 310
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 23.423568725585938,
+      "learning_rate": 6.89e-05,
+      "loss": 0.4385,
+      "step": 311
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 22.86809539794922,
+      "learning_rate": 6.879999999999999e-05,
+      "loss": 0.2332,
+      "step": 312
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 23.574260711669922,
+      "learning_rate": 6.87e-05,
+      "loss": 0.2794,
+      "step": 313
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 11.08251667022705,
+      "learning_rate": 6.860000000000001e-05,
+      "loss": 0.0687,
+      "step": 314
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 22.305559158325195,
+      "learning_rate": 6.850000000000001e-05,
+      "loss": 0.2469,
+      "step": 315
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 27.41745376586914,
+      "learning_rate": 6.840000000000001e-05,
+      "loss": 0.3026,
+      "step": 316
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 11.78591251373291,
+      "learning_rate": 6.83e-05,
+      "loss": 0.0773,
+      "step": 317
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 28.363525390625,
+      "learning_rate": 6.82e-05,
+      "loss": 0.3241,
+      "step": 318
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.026371236890554428,
+      "learning_rate": 6.81e-05,
+      "loss": 0.0002,
+      "step": 319
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 26.836273193359375,
+      "learning_rate": 6.800000000000001e-05,
+      "loss": 0.6225,
+      "step": 320
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 20.1362361907959,
+      "learning_rate": 6.790000000000001e-05,
+      "loss": 0.1332,
+      "step": 321
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.151141881942749,
+      "learning_rate": 6.780000000000001e-05,
+      "loss": 0.0161,
+      "step": 322
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.3739681243896484,
+      "learning_rate": 6.77e-05,
+      "loss": 0.02,
+      "step": 323
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 11.643929481506348,
+      "learning_rate": 6.76e-05,
+      "loss": 0.0814,
+      "step": 324
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.02318856306374073,
+      "learning_rate": 6.750000000000001e-05,
+      "loss": 0.0002,
+      "step": 325
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.5685155987739563,
+      "learning_rate": 6.740000000000001e-05,
+      "loss": 0.0027,
+      "step": 326
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.9630302786827087,
+      "learning_rate": 6.730000000000001e-05,
+      "loss": 0.0129,
+      "step": 327
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 27.39959144592285,
+      "learning_rate": 6.720000000000001e-05,
+      "loss": 0.9036,
+      "step": 328
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 27.959325790405273,
+      "learning_rate": 6.71e-05,
+      "loss": 0.3344,
+      "step": 329
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 25.573692321777344,
+      "learning_rate": 6.7e-05,
+      "loss": 0.5308,
+      "step": 330
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 8.346728324890137,
+      "learning_rate": 6.690000000000001e-05,
+      "loss": 0.1063,
+      "step": 331
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 36.26382064819336,
+      "learning_rate": 6.680000000000001e-05,
+      "loss": 0.8096,
+      "step": 332
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 20.98823356628418,
+      "learning_rate": 6.670000000000001e-05,
+      "loss": 0.1938,
+      "step": 333
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 11.60432243347168,
+      "learning_rate": 6.66e-05,
+      "loss": 0.0702,
+      "step": 334
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.3954720497131348,
+      "learning_rate": 6.65e-05,
+      "loss": 0.0085,
+      "step": 335
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 25.829086303710938,
+      "learning_rate": 6.64e-05,
+      "loss": 0.5444,
+      "step": 336
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 25.731420516967773,
+      "learning_rate": 6.630000000000001e-05,
+      "loss": 0.7691,
+      "step": 337
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 17.625537872314453,
+      "learning_rate": 6.620000000000001e-05,
+      "loss": 0.1764,
+      "step": 338
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.233189344406128,
+      "learning_rate": 6.610000000000001e-05,
+      "loss": 0.0057,
+      "step": 339
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 38.987022399902344,
+      "learning_rate": 6.6e-05,
+      "loss": 1.1134,
+      "step": 340
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 28.655746459960938,
+      "learning_rate": 6.59e-05,
+      "loss": 0.7567,
+      "step": 341
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 24.218215942382812,
+      "learning_rate": 6.58e-05,
+      "loss": 0.2627,
+      "step": 342
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 27.85608673095703,
+      "learning_rate": 6.570000000000001e-05,
+      "loss": 1.3316,
+      "step": 343
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 16.30738067626953,
+      "learning_rate": 6.560000000000001e-05,
+      "loss": 0.1128,
+      "step": 344
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 27.84078598022461,
+      "learning_rate": 6.55e-05,
+      "loss": 1.3145,
+      "step": 345
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.4055051803588867,
+      "learning_rate": 6.54e-05,
+      "loss": 0.0163,
+      "step": 346
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 36.960227966308594,
+      "learning_rate": 6.53e-05,
+      "loss": 1.1153,
+      "step": 347
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 27.966150283813477,
+      "learning_rate": 6.52e-05,
+      "loss": 0.8653,
+      "step": 348
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 16.647764205932617,
+      "learning_rate": 6.510000000000001e-05,
+      "loss": 0.1341,
+      "step": 349
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 27.271018981933594,
+      "learning_rate": 6.500000000000001e-05,
+      "loss": 0.8342,
+      "step": 350
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.560828685760498,
+      "learning_rate": 6.49e-05,
+      "loss": 0.0271,
+      "step": 351
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 27.721872329711914,
+      "learning_rate": 6.48e-05,
+      "loss": 0.5398,
+      "step": 352
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 27.782337188720703,
+      "learning_rate": 6.47e-05,
+      "loss": 0.8772,
+      "step": 353
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 17.337244033813477,
+      "learning_rate": 6.460000000000001e-05,
+      "loss": 0.1268,
+      "step": 354
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 48.405452728271484,
+      "learning_rate": 6.450000000000001e-05,
+      "loss": 0.8516,
+      "step": 355
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.5789971351623535,
+      "learning_rate": 6.440000000000001e-05,
+      "loss": 0.0035,
+      "step": 356
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 43.46677017211914,
+      "learning_rate": 6.43e-05,
+      "loss": 1.6386,
+      "step": 357
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 20.956693649291992,
+      "learning_rate": 6.42e-05,
+      "loss": 0.2639,
+      "step": 358
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.5366334319114685,
+      "learning_rate": 6.41e-05,
+      "loss": 0.003,
+      "step": 359
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 14.680846214294434,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 0.1709,
+      "step": 360
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.22275477647781372,
+      "learning_rate": 6.390000000000001e-05,
+      "loss": 0.0019,
+      "step": 361
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 20.443256378173828,
+      "learning_rate": 6.38e-05,
+      "loss": 0.1746,
+      "step": 362
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 23.81313705444336,
+      "learning_rate": 6.37e-05,
+      "loss": 0.402,
+      "step": 363
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 15.194694519042969,
+      "learning_rate": 6.36e-05,
+      "loss": 0.1142,
+      "step": 364
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 17.877086639404297,
+      "learning_rate": 6.35e-05,
+      "loss": 0.1906,
+      "step": 365
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 13.516918182373047,
+      "learning_rate": 6.340000000000001e-05,
+      "loss": 0.1061,
+      "step": 366
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 42.48619079589844,
+      "learning_rate": 6.330000000000001e-05,
+      "loss": 2.2565,
+      "step": 367
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 26.311281204223633,
+      "learning_rate": 6.32e-05,
+      "loss": 0.7296,
+      "step": 368
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 34.43406677246094,
+      "learning_rate": 6.31e-05,
+      "loss": 1.9544,
+      "step": 369
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6160056591033936,
+      "learning_rate": 6.3e-05,
+      "loss": 0.003,
+      "step": 370
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 8.770819664001465,
+      "learning_rate": 6.29e-05,
+      "loss": 0.0579,
+      "step": 371
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 45.1691780090332,
+      "learning_rate": 6.280000000000001e-05,
+      "loss": 0.924,
+      "step": 372
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 33.87689208984375,
+      "learning_rate": 6.27e-05,
+      "loss": 1.5384,
+      "step": 373
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.7516889572143555,
+      "learning_rate": 6.26e-05,
+      "loss": 0.0374,
+      "step": 374
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6744739413261414,
+      "learning_rate": 6.25e-05,
+      "loss": 0.0035,
+      "step": 375
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 29.63498306274414,
+      "learning_rate": 6.24e-05,
+      "loss": 0.398,
+      "step": 376
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 23.468154907226562,
+      "learning_rate": 6.23e-05,
+      "loss": 0.4509,
+      "step": 377
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 46.63499069213867,
+      "learning_rate": 6.220000000000001e-05,
+      "loss": 1.2205,
+      "step": 378
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.985834121704102,
+      "learning_rate": 6.21e-05,
+      "loss": 0.0551,
+      "step": 379
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 20.038787841796875,
+      "learning_rate": 6.2e-05,
+      "loss": 1.2499,
+      "step": 380
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.5343722701072693,
+      "learning_rate": 6.19e-05,
+      "loss": 0.0043,
+      "step": 381
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.8306647539138794,
+      "learning_rate": 6.18e-05,
+      "loss": 0.0041,
+      "step": 382
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.906933784484863,
+      "learning_rate": 6.170000000000001e-05,
+      "loss": 0.0328,
+      "step": 383
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 26.180572509765625,
+      "learning_rate": 6.16e-05,
+      "loss": 0.4335,
+      "step": 384
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.458011656999588,
+      "learning_rate": 6.15e-05,
+      "loss": 0.0032,
+      "step": 385
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 24.316070556640625,
+      "learning_rate": 6.14e-05,
+      "loss": 0.2748,
+      "step": 386
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 23.521617889404297,
+      "learning_rate": 6.13e-05,
+      "loss": 0.2207,
+      "step": 387
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 32.9349365234375,
+      "learning_rate": 6.12e-05,
+      "loss": 0.4958,
+      "step": 388
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.459490776062012,
+      "learning_rate": 6.110000000000001e-05,
+      "loss": 0.03,
+      "step": 389
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.2316316366195679,
+      "learning_rate": 6.1e-05,
+      "loss": 0.0072,
+      "step": 390
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 29.919815063476562,
+      "learning_rate": 6.09e-05,
+      "loss": 0.3748,
+      "step": 391
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.5657413005828857,
+      "learning_rate": 6.08e-05,
+      "loss": 0.0287,
+      "step": 392
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.026643753051758,
+      "learning_rate": 6.07e-05,
+      "loss": 0.0368,
+      "step": 393
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 27.373517990112305,
+      "learning_rate": 6.06e-05,
+      "loss": 1.2679,
+      "step": 394
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 15.123701095581055,
+      "learning_rate": 6.05e-05,
+      "loss": 0.1167,
+      "step": 395
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.5672614574432373,
+      "learning_rate": 6.04e-05,
+      "loss": 0.0104,
+      "step": 396
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 27.24859619140625,
+      "learning_rate": 6.03e-05,
+      "loss": 0.7966,
+      "step": 397
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.6505637168884277,
+      "learning_rate": 6.02e-05,
+      "loss": 0.0084,
+      "step": 398
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 38.031185150146484,
+      "learning_rate": 6.0100000000000004e-05,
+      "loss": 1.3056,
+      "step": 399
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.0308609008789062,
+      "learning_rate": 6e-05,
+      "loss": 0.0268,
+      "step": 400
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9318943023681641,
+      "learning_rate": 5.99e-05,
+      "loss": 0.0068,
+      "step": 401
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 30.836875915527344,
+      "learning_rate": 5.9800000000000003e-05,
+      "loss": 0.2937,
+      "step": 402
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.094520092010498,
+      "learning_rate": 5.97e-05,
+      "loss": 0.037,
+      "step": 403
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 49.60438537597656,
+      "learning_rate": 5.96e-05,
+      "loss": 0.7892,
+      "step": 404
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 26.148279190063477,
+      "learning_rate": 5.95e-05,
+      "loss": 0.8323,
+      "step": 405
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 13.96594524383545,
+      "learning_rate": 5.94e-05,
+      "loss": 0.0942,
+      "step": 406
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.893899917602539,
+      "learning_rate": 5.93e-05,
+      "loss": 0.0486,
+      "step": 407
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.9445253610610962,
+      "learning_rate": 5.92e-05,
+      "loss": 0.0047,
+      "step": 408
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 19.779830932617188,
+      "learning_rate": 5.91e-05,
+      "loss": 0.1503,
+      "step": 409
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 44.186283111572266,
+      "learning_rate": 5.9e-05,
+      "loss": 1.0128,
+      "step": 410
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 26.87023162841797,
+      "learning_rate": 5.89e-05,
+      "loss": 0.6065,
+      "step": 411
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 28.162328720092773,
+      "learning_rate": 5.88e-05,
+      "loss": 1.2739,
+      "step": 412
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 38.76342010498047,
+      "learning_rate": 5.87e-05,
+      "loss": 1.3519,
+      "step": 413
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.24123622477054596,
+      "learning_rate": 5.86e-05,
+      "loss": 0.0017,
+      "step": 414
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 26.10549545288086,
+      "learning_rate": 5.85e-05,
+      "loss": 0.6323,
+      "step": 415
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 24.87130355834961,
+      "learning_rate": 5.8399999999999997e-05,
+      "loss": 0.3113,
+      "step": 416
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 26.145565032958984,
+      "learning_rate": 5.83e-05,
+      "loss": 0.4853,
+      "step": 417
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 23.986276626586914,
+      "learning_rate": 5.82e-05,
+      "loss": 0.4204,
+      "step": 418
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 25.39900779724121,
+      "learning_rate": 5.8099999999999996e-05,
+      "loss": 0.3669,
+      "step": 419
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 24.91488265991211,
+      "learning_rate": 5.8e-05,
+      "loss": 0.4622,
+      "step": 420
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.05866813659668,
+      "learning_rate": 5.79e-05,
+      "loss": 0.0574,
+      "step": 421
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 22.513336181640625,
+      "learning_rate": 5.7799999999999995e-05,
+      "loss": 0.7316,
+      "step": 422
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 20.641204833984375,
+      "learning_rate": 5.77e-05,
+      "loss": 0.8384,
+      "step": 423
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.3587974309921265,
+      "learning_rate": 5.76e-05,
+      "loss": 0.0065,
+      "step": 424
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 21.501081466674805,
+      "learning_rate": 5.7499999999999995e-05,
+      "loss": 0.2061,
+      "step": 425
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 42.36915588378906,
+      "learning_rate": 5.74e-05,
+      "loss": 1.3581,
+      "step": 426
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.361469745635986,
+      "learning_rate": 5.73e-05,
+      "loss": 0.052,
+      "step": 427
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 30.654287338256836,
+      "learning_rate": 5.72e-05,
+      "loss": 0.3091,
+      "step": 428
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.8391109108924866,
+      "learning_rate": 5.71e-05,
+      "loss": 0.0061,
+      "step": 429
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.208446979522705,
+      "learning_rate": 5.6999999999999996e-05,
+      "loss": 0.0193,
+      "step": 430
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.1669785976409912,
+      "learning_rate": 5.69e-05,
+      "loss": 0.0075,
+      "step": 431
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 15.361763000488281,
+      "learning_rate": 5.68e-05,
+      "loss": 0.1838,
+      "step": 432
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.7279540300369263,
+      "learning_rate": 5.6699999999999996e-05,
+      "loss": 0.0069,
+      "step": 433
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 35.00238800048828,
+      "learning_rate": 5.66e-05,
+      "loss": 0.7085,
+      "step": 434
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 27.048877716064453,
+      "learning_rate": 5.65e-05,
+      "loss": 0.3643,
+      "step": 435
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5507223606109619,
+      "learning_rate": 5.6399999999999995e-05,
+      "loss": 0.0036,
+      "step": 436
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.34695854783058167,
+      "learning_rate": 5.63e-05,
+      "loss": 0.0021,
+      "step": 437
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 27.41823959350586,
+      "learning_rate": 5.620000000000001e-05,
+      "loss": 0.6107,
+      "step": 438
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 27.145366668701172,
+      "learning_rate": 5.610000000000001e-05,
+      "loss": 0.6633,
+      "step": 439
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.264007091522217,
+      "learning_rate": 5.6000000000000006e-05,
+      "loss": 0.0222,
+      "step": 440
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 26.830528259277344,
+      "learning_rate": 5.590000000000001e-05,
+      "loss": 0.5476,
+      "step": 441
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 40.81327438354492,
+      "learning_rate": 5.580000000000001e-05,
+      "loss": 0.7979,
+      "step": 442
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.6571967005729675,
+      "learning_rate": 5.5700000000000005e-05,
+      "loss": 0.0043,
+      "step": 443
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 18.67711067199707,
+      "learning_rate": 5.560000000000001e-05,
+      "loss": 0.1879,
+      "step": 444
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 21.878129959106445,
+      "learning_rate": 5.550000000000001e-05,
+      "loss": 0.2265,
+      "step": 445
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 9.461925506591797,
+      "learning_rate": 5.5400000000000005e-05,
+      "loss": 0.056,
+      "step": 446
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.5579652786254883,
+      "learning_rate": 5.530000000000001e-05,
+      "loss": 0.0132,
+      "step": 447
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.1434270143508911,
+      "learning_rate": 5.520000000000001e-05,
+      "loss": 0.0083,
+      "step": 448
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.296957015991211,
+      "learning_rate": 5.5100000000000004e-05,
+      "loss": 0.0537,
+      "step": 449
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.7849231958389282,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 0.0041,
+      "step": 450
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.7314896583557129,
+      "learning_rate": 5.4900000000000006e-05,
+      "loss": 0.0057,
+      "step": 451
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.517040252685547,
+      "learning_rate": 5.4800000000000004e-05,
+      "loss": 0.0533,
+      "step": 452
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.680517673492432,
+      "learning_rate": 5.470000000000001e-05,
+      "loss": 0.0452,
+      "step": 453
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 50.83726119995117,
+      "learning_rate": 5.4600000000000006e-05,
+      "loss": 2.289,
+      "step": 454
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 29.895687103271484,
+      "learning_rate": 5.45e-05,
+      "loss": 0.6449,
+      "step": 455
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 30.133241653442383,
+      "learning_rate": 5.440000000000001e-05,
+      "loss": 0.4874,
+      "step": 456
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.48309120535850525,
+      "learning_rate": 5.4300000000000005e-05,
+      "loss": 0.004,
+      "step": 457
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 25.740129470825195,
+      "learning_rate": 5.420000000000001e-05,
+      "loss": 0.3231,
+      "step": 458
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 13.617050170898438,
+      "learning_rate": 5.410000000000001e-05,
+      "loss": 0.1009,
+      "step": 459
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 31.53325080871582,
+      "learning_rate": 5.4000000000000005e-05,
+      "loss": 0.7426,
+      "step": 460
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 36.63671875,
+      "learning_rate": 5.390000000000001e-05,
+      "loss": 1.9978,
+      "step": 461
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 10.66323184967041,
+      "learning_rate": 5.380000000000001e-05,
+      "loss": 0.0627,
+      "step": 462
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 8.019021034240723,
+      "learning_rate": 5.3700000000000004e-05,
+      "loss": 0.0588,
+      "step": 463
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 26.658863067626953,
+      "learning_rate": 5.360000000000001e-05,
+      "loss": 0.4552,
+      "step": 464
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.3222432136535645,
+      "learning_rate": 5.3500000000000006e-05,
+      "loss": 0.0576,
+      "step": 465
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 25.450960159301758,
+      "learning_rate": 5.3400000000000004e-05,
+      "loss": 0.8938,
+      "step": 466
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 21.743581771850586,
+      "learning_rate": 5.330000000000001e-05,
+      "loss": 0.1594,
+      "step": 467
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 9.093583106994629,
+      "learning_rate": 5.3200000000000006e-05,
+      "loss": 0.0525,
+      "step": 468
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 12.705286979675293,
+      "learning_rate": 5.31e-05,
+      "loss": 0.1962,
+      "step": 469
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.836313247680664,
+      "learning_rate": 5.300000000000001e-05,
+      "loss": 0.0176,
+      "step": 470
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 30.454456329345703,
+      "learning_rate": 5.2900000000000005e-05,
+      "loss": 1.425,
+      "step": 471
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 27.88629150390625,
+      "learning_rate": 5.28e-05,
+      "loss": 1.7417,
+      "step": 472
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 28.916534423828125,
+      "learning_rate": 5.270000000000001e-05,
+      "loss": 0.759,
+      "step": 473
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.10350897908210754,
+      "learning_rate": 5.2600000000000005e-05,
+      "loss": 0.0007,
+      "step": 474
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.06146182492375374,
+      "learning_rate": 5.25e-05,
+      "loss": 0.0005,
+      "step": 475
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 23.231124877929688,
+      "learning_rate": 5.2400000000000007e-05,
+      "loss": 0.318,
+      "step": 476
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 39.07556915283203,
+      "learning_rate": 5.2300000000000004e-05,
+      "loss": 1.574,
+      "step": 477
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.361273765563965,
+      "learning_rate": 5.22e-05,
+      "loss": 0.0221,
+      "step": 478
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 66.95952606201172,
+      "learning_rate": 5.2100000000000006e-05,
+      "loss": 2.6285,
+      "step": 479
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 17.135887145996094,
+      "learning_rate": 5.2000000000000004e-05,
+      "loss": 0.9921,
+      "step": 480
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 8.985684394836426,
+      "learning_rate": 5.19e-05,
+      "loss": 0.0569,
+      "step": 481
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 9.756717681884766,
+      "learning_rate": 5.1800000000000005e-05,
+      "loss": 0.0604,
+      "step": 482
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 13.56096076965332,
+      "learning_rate": 5.17e-05,
+      "loss": 0.1157,
+      "step": 483
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.52184534072876,
+      "learning_rate": 5.16e-05,
+      "loss": 0.0449,
+      "step": 484
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 30.49310874938965,
+      "learning_rate": 5.1500000000000005e-05,
+      "loss": 0.781,
+      "step": 485
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.6025214195251465,
+      "learning_rate": 5.14e-05,
+      "loss": 0.029,
+      "step": 486
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.667913436889648,
+      "learning_rate": 5.130000000000001e-05,
+      "loss": 0.0281,
+      "step": 487
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 21.860044479370117,
+      "learning_rate": 5.1200000000000004e-05,
+      "loss": 0.24,
+      "step": 488
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.007333993911743,
+      "learning_rate": 5.11e-05,
+      "loss": 0.0184,
+      "step": 489
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.179823398590088,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 0.0204,
+      "step": 490
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.7667524814605713,
+      "learning_rate": 5.0900000000000004e-05,
+      "loss": 0.0062,
+      "step": 491
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.3707985877990723,
+      "learning_rate": 5.08e-05,
+      "loss": 0.0147,
+      "step": 492
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 22.363510131835938,
+      "learning_rate": 5.0700000000000006e-05,
+      "loss": 0.3099,
+      "step": 493
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 17.151803970336914,
+      "learning_rate": 5.0600000000000003e-05,
+      "loss": 0.1448,
+      "step": 494
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 26.439014434814453,
+      "learning_rate": 5.05e-05,
+      "loss": 0.2519,
+      "step": 495
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 11.708067893981934,
+      "learning_rate": 5.0400000000000005e-05,
+      "loss": 0.0841,
+      "step": 496
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.6435901522636414,
+      "learning_rate": 5.03e-05,
+      "loss": 0.0045,
+      "step": 497
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 22.941650390625,
+      "learning_rate": 5.02e-05,
+      "loss": 0.2113,
+      "step": 498
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.059579696506261826,
+      "learning_rate": 5.0100000000000005e-05,
+      "loss": 0.0004,
+      "step": 499
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.3625144958496094,
+      "learning_rate": 5e-05,
+      "loss": 0.025,
+      "step": 500
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4082565307617188,
+      "learning_rate": 4.99e-05,
+      "loss": 0.013,
+      "step": 501
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 11.377418518066406,
+      "learning_rate": 4.9800000000000004e-05,
+      "loss": 0.0745,
+      "step": 502
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 32.57530212402344,
+      "learning_rate": 4.97e-05,
+      "loss": 0.6956,
+      "step": 503
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.3960559666156769,
+      "learning_rate": 4.96e-05,
+      "loss": 0.0025,
+      "step": 504
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9463507533073425,
+      "learning_rate": 4.9500000000000004e-05,
+      "loss": 0.0047,
+      "step": 505
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 26.72979164123535,
+      "learning_rate": 4.94e-05,
+      "loss": 0.5934,
+      "step": 506
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.7482037544250488,
+      "learning_rate": 4.93e-05,
+      "loss": 0.0105,
+      "step": 507
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 31.476030349731445,
+      "learning_rate": 4.92e-05,
+      "loss": 0.4363,
+      "step": 508
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 17.86579132080078,
+      "learning_rate": 4.91e-05,
+      "loss": 0.1511,
+      "step": 509
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 3.154860496520996,
+      "learning_rate": 4.9e-05,
+      "loss": 0.0167,
+      "step": 510
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.15494592487812042,
+      "learning_rate": 4.89e-05,
+      "loss": 0.001,
+      "step": 511
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 22.509258270263672,
+      "learning_rate": 4.88e-05,
+      "loss": 0.2224,
+      "step": 512
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 3.79130220413208,
+      "learning_rate": 4.87e-05,
+      "loss": 0.0198,
+      "step": 513
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.28567245602607727,
+      "learning_rate": 4.86e-05,
+      "loss": 0.0023,
+      "step": 514
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.115846157073975,
+      "learning_rate": 4.85e-05,
+      "loss": 0.0433,
+      "step": 515
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 21.192081451416016,
+      "learning_rate": 4.8400000000000004e-05,
+      "loss": 0.1827,
+      "step": 516
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 32.70637512207031,
+      "learning_rate": 4.83e-05,
+      "loss": 2.368,
+      "step": 517
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 26.6806697845459,
+      "learning_rate": 4.82e-05,
+      "loss": 0.8146,
+      "step": 518
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 25.684167861938477,
+      "learning_rate": 4.8100000000000004e-05,
+      "loss": 0.7391,
+      "step": 519
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 13.313119888305664,
+      "learning_rate": 4.8e-05,
+      "loss": 0.08,
+      "step": 520
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 39.951602935791016,
+      "learning_rate": 4.79e-05,
+      "loss": 1.68,
+      "step": 521
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 29.313465118408203,
+      "learning_rate": 4.78e-05,
+      "loss": 0.8004,
+      "step": 522
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 3.168497085571289,
+      "learning_rate": 4.77e-05,
+      "loss": 0.0208,
+      "step": 523
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 11.69185733795166,
+      "learning_rate": 4.76e-05,
+      "loss": 0.0701,
+      "step": 524
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 15.70853328704834,
+      "learning_rate": 4.75e-05,
+      "loss": 0.141,
+      "step": 525
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.7469470500946045,
+      "learning_rate": 4.74e-05,
+      "loss": 0.0057,
+      "step": 526
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03925801441073418,
+      "learning_rate": 4.73e-05,
+      "loss": 0.0002,
+      "step": 527
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 28.931766510009766,
+      "learning_rate": 4.72e-05,
+      "loss": 0.4199,
+      "step": 528
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.8265246748924255,
+      "learning_rate": 4.71e-05,
+      "loss": 0.0044,
+      "step": 529
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 16.333219528198242,
+      "learning_rate": 4.7e-05,
+      "loss": 0.1149,
+      "step": 530
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.08890499919652939,
+      "learning_rate": 4.69e-05,
+      "loss": 0.0007,
+      "step": 531
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.24474568665027618,
+      "learning_rate": 4.6800000000000006e-05,
+      "loss": 0.0013,
+      "step": 532
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 42.735389709472656,
+      "learning_rate": 4.6700000000000003e-05,
+      "loss": 1.2246,
+      "step": 533
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 15.926461219787598,
+      "learning_rate": 4.660000000000001e-05,
+      "loss": 0.1089,
+      "step": 534
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 26.88334846496582,
+      "learning_rate": 4.6500000000000005e-05,
+      "loss": 0.2097,
+      "step": 535
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 25.04203987121582,
+      "learning_rate": 4.64e-05,
+      "loss": 0.5348,
+      "step": 536
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 38.08080291748047,
+      "learning_rate": 4.630000000000001e-05,
+      "loss": 2.7062,
+      "step": 537
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.8540865778923035,
+      "learning_rate": 4.6200000000000005e-05,
+      "loss": 0.0062,
+      "step": 538
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 26.498910903930664,
+      "learning_rate": 4.61e-05,
+      "loss": 0.7067,
+      "step": 539
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 7.919168472290039,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.0499,
+      "step": 540
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 3.056781530380249,
+      "learning_rate": 4.5900000000000004e-05,
+      "loss": 0.0155,
+      "step": 541
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 26.98896026611328,
+      "learning_rate": 4.58e-05,
+      "loss": 1.8109,
+      "step": 542
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.0453416109085083,
+      "learning_rate": 4.5700000000000006e-05,
+      "loss": 0.0051,
+      "step": 543
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 26.941593170166016,
+      "learning_rate": 4.5600000000000004e-05,
+      "loss": 1.3416,
+      "step": 544
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 7.245401382446289,
+      "learning_rate": 4.55e-05,
+      "loss": 0.0806,
+      "step": 545
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 28.137914657592773,
+      "learning_rate": 4.5400000000000006e-05,
+      "loss": 1.3048,
+      "step": 546
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 27.121395111083984,
+      "learning_rate": 4.53e-05,
+      "loss": 0.9772,
+      "step": 547
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 25.434829711914062,
+      "learning_rate": 4.52e-05,
+      "loss": 0.481,
+      "step": 548
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 29.52955436706543,
+      "learning_rate": 4.5100000000000005e-05,
+      "loss": 0.7622,
+      "step": 549
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.06173492223024368,
+      "learning_rate": 4.5e-05,
+      "loss": 0.0008,
+      "step": 550
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3727266192436218,
+      "learning_rate": 4.49e-05,
+      "loss": 0.0025,
+      "step": 551
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.017284419387578964,
+      "learning_rate": 4.4800000000000005e-05,
+      "loss": 0.0001,
+      "step": 552
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.07270216196775436,
+      "learning_rate": 4.47e-05,
+      "loss": 0.0009,
+      "step": 553
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 7.1794257164001465,
+      "learning_rate": 4.46e-05,
+      "loss": 0.0394,
+      "step": 554
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5796927213668823,
+      "learning_rate": 4.4500000000000004e-05,
+      "loss": 0.0035,
+      "step": 555
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.966355085372925,
+      "learning_rate": 4.44e-05,
+      "loss": 0.015,
+      "step": 556
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.8540501594543457,
+      "learning_rate": 4.43e-05,
+      "loss": 0.0045,
+      "step": 557
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.8207895755767822,
+      "learning_rate": 4.4200000000000004e-05,
+      "loss": 0.0069,
+      "step": 558
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 19.900699615478516,
+      "learning_rate": 4.41e-05,
+      "loss": 0.2333,
+      "step": 559
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.041948080062866,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 0.0157,
+      "step": 560
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 40.66261672973633,
+      "learning_rate": 4.39e-05,
+      "loss": 0.7749,
+      "step": 561
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 23.92762565612793,
+      "learning_rate": 4.38e-05,
+      "loss": 0.2517,
+      "step": 562
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 26.802112579345703,
+      "learning_rate": 4.3700000000000005e-05,
+      "loss": 1.1189,
+      "step": 563
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 25.978343963623047,
+      "learning_rate": 4.36e-05,
+      "loss": 0.3535,
+      "step": 564
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.6375006437301636,
+      "learning_rate": 4.35e-05,
+      "loss": 0.003,
+      "step": 565
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 4.264609336853027,
+      "learning_rate": 4.3400000000000005e-05,
+      "loss": 0.0295,
+      "step": 566
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2793530523777008,
+      "learning_rate": 4.33e-05,
+      "loss": 0.0029,
+      "step": 567
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 10.96884822845459,
+      "learning_rate": 4.32e-05,
+      "loss": 0.0667,
+      "step": 568
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.39111682772636414,
+      "learning_rate": 4.3100000000000004e-05,
+      "loss": 0.0027,
+      "step": 569
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.712225079536438,
+      "learning_rate": 4.3e-05,
+      "loss": 0.0106,
+      "step": 570
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 13.446296691894531,
+      "learning_rate": 4.29e-05,
+      "loss": 0.0828,
+      "step": 571
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 28.18335723876953,
+      "learning_rate": 4.2800000000000004e-05,
+      "loss": 1.3561,
+      "step": 572
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.2135007530450821,
+      "learning_rate": 4.27e-05,
+      "loss": 0.0019,
+      "step": 573
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 43.12370300292969,
+      "learning_rate": 4.26e-05,
+      "loss": 2.1706,
+      "step": 574
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.703223466873169,
+      "learning_rate": 4.25e-05,
+      "loss": 0.0057,
+      "step": 575
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.19304363429546356,
+      "learning_rate": 4.24e-05,
+      "loss": 0.0013,
+      "step": 576
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9636431336402893,
+      "learning_rate": 4.23e-05,
+      "loss": 0.0065,
+      "step": 577
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.248726487159729,
+      "learning_rate": 4.22e-05,
+      "loss": 0.0092,
+      "step": 578
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 26.916940689086914,
+      "learning_rate": 4.21e-05,
+      "loss": 0.4018,
+      "step": 579
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 20.4180965423584,
+      "learning_rate": 4.2e-05,
+      "loss": 0.2221,
+      "step": 580
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 24.25115966796875,
+      "learning_rate": 4.19e-05,
+      "loss": 0.3603,
+      "step": 581
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 14.176848411560059,
+      "learning_rate": 4.18e-05,
+      "loss": 0.096,
+      "step": 582
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 18.621923446655273,
+      "learning_rate": 4.17e-05,
+      "loss": 1.9419,
+      "step": 583
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.3514004051685333,
+      "learning_rate": 4.16e-05,
+      "loss": 0.002,
+      "step": 584
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 22.820341110229492,
+      "learning_rate": 4.15e-05,
+      "loss": 0.1718,
+      "step": 585
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.11256919801235199,
+      "learning_rate": 4.14e-05,
+      "loss": 0.0014,
+      "step": 586
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 48.12772750854492,
+      "learning_rate": 4.13e-05,
+      "loss": 1.0516,
+      "step": 587
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.6842830181121826,
+      "learning_rate": 4.12e-05,
+      "loss": 0.0058,
+      "step": 588
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 27.722042083740234,
+      "learning_rate": 4.11e-05,
+      "loss": 0.3808,
+      "step": 589
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 28.378963470458984,
+      "learning_rate": 4.1e-05,
+      "loss": 1.2508,
+      "step": 590
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 21.205421447753906,
+      "learning_rate": 4.09e-05,
+      "loss": 0.3059,
+      "step": 591
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 11.505128860473633,
+      "learning_rate": 4.08e-05,
+      "loss": 0.0675,
+      "step": 592
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.10159604996442795,
+      "learning_rate": 4.07e-05,
+      "loss": 0.0006,
+      "step": 593
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.013346759602427483,
+      "learning_rate": 4.0600000000000004e-05,
+      "loss": 0.0001,
+      "step": 594
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 28.66274642944336,
+      "learning_rate": 4.05e-05,
+      "loss": 0.7989,
+      "step": 595
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.537345290184021,
+      "learning_rate": 4.0400000000000006e-05,
+      "loss": 0.0073,
+      "step": 596
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.15267232060432434,
+      "learning_rate": 4.0300000000000004e-05,
+      "loss": 0.0011,
+      "step": 597
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.3610873222351074,
+      "learning_rate": 4.02e-05,
+      "loss": 0.0154,
+      "step": 598
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.7249945402145386,
+      "learning_rate": 4.0100000000000006e-05,
+      "loss": 0.006,
+      "step": 599
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 9.683416366577148,
+      "learning_rate": 4e-05,
+      "loss": 0.0602,
+      "step": 600
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 22.996389389038086,
+      "learning_rate": 3.99e-05,
+      "loss": 0.4727,
+      "step": 601
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 25.594470977783203,
+      "learning_rate": 3.9800000000000005e-05,
+      "loss": 0.4286,
+      "step": 602
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 27.750869750976562,
+      "learning_rate": 3.97e-05,
+      "loss": 0.7911,
+      "step": 603
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 26.540616989135742,
+      "learning_rate": 3.960000000000001e-05,
+      "loss": 0.6582,
+      "step": 604
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 26.802492141723633,
+      "learning_rate": 3.9500000000000005e-05,
+      "loss": 0.6329,
+      "step": 605
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 27.42936134338379,
+      "learning_rate": 3.94e-05,
+      "loss": 1.0816,
+      "step": 606
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 3.6503002643585205,
+      "learning_rate": 3.9300000000000007e-05,
+      "loss": 0.023,
+      "step": 607
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 31.658214569091797,
+      "learning_rate": 3.9200000000000004e-05,
+      "loss": 0.4239,
+      "step": 608
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.08632346242666245,
+      "learning_rate": 3.91e-05,
+      "loss": 0.0007,
+      "step": 609
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.26137351989746094,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 0.0048,
+      "step": 610
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 22.23630142211914,
+      "learning_rate": 3.8900000000000004e-05,
+      "loss": 0.3182,
+      "step": 611
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 4.530112266540527,
+      "learning_rate": 3.88e-05,
+      "loss": 0.0271,
+      "step": 612
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 16.424596786499023,
+      "learning_rate": 3.8700000000000006e-05,
+      "loss": 0.1404,
+      "step": 613
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 2.2123844623565674,
+      "learning_rate": 3.86e-05,
+      "loss": 0.0127,
+      "step": 614
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 26.423566818237305,
+      "learning_rate": 3.85e-05,
+      "loss": 0.4926,
+      "step": 615
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 24.903274536132812,
+      "learning_rate": 3.8400000000000005e-05,
+      "loss": 0.5274,
+      "step": 616
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 28.026695251464844,
+      "learning_rate": 3.83e-05,
+      "loss": 0.7997,
+      "step": 617
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4405864477157593,
+      "learning_rate": 3.82e-05,
+      "loss": 0.0245,
+      "step": 618
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.11772224307060242,
+      "learning_rate": 3.8100000000000005e-05,
+      "loss": 0.0011,
+      "step": 619
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.7985239028930664,
+      "learning_rate": 3.8e-05,
+      "loss": 0.0294,
+      "step": 620
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 22.811920166015625,
+      "learning_rate": 3.79e-05,
+      "loss": 0.3031,
+      "step": 621
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 16.438217163085938,
+      "learning_rate": 3.7800000000000004e-05,
+      "loss": 0.283,
+      "step": 622
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 16.60034942626953,
+      "learning_rate": 3.77e-05,
+      "loss": 0.1261,
+      "step": 623
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 12.6220121383667,
+      "learning_rate": 3.76e-05,
+      "loss": 0.0818,
+      "step": 624
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 26.638370513916016,
+      "learning_rate": 3.7500000000000003e-05,
+      "loss": 0.8201,
+      "step": 625
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 28.03763771057129,
+      "learning_rate": 3.74e-05,
+      "loss": 1.0075,
+      "step": 626
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 45.880584716796875,
+      "learning_rate": 3.73e-05,
+      "loss": 0.6867,
+      "step": 627
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 6.816001892089844,
+      "learning_rate": 3.72e-05,
+      "loss": 0.0382,
+      "step": 628
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 10.210158348083496,
+      "learning_rate": 3.71e-05,
+      "loss": 0.0685,
+      "step": 629
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.01786009781062603,
+      "learning_rate": 3.7e-05,
+      "loss": 0.0002,
+      "step": 630
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.49422261118888855,
+      "learning_rate": 3.69e-05,
+      "loss": 0.0045,
+      "step": 631
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 19.256742477416992,
+      "learning_rate": 3.68e-05,
+      "loss": 0.1697,
+      "step": 632
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 18.655818939208984,
+      "learning_rate": 3.6700000000000004e-05,
+      "loss": 0.2467,
+      "step": 633
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 30.253002166748047,
+      "learning_rate": 3.66e-05,
+      "loss": 1.181,
+      "step": 634
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 25.4757137298584,
+      "learning_rate": 3.65e-05,
+      "loss": 1.4477,
+      "step": 635
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 50.04316329956055,
+      "learning_rate": 3.6400000000000004e-05,
+      "loss": 0.7256,
+      "step": 636
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 5.345543384552002,
+      "learning_rate": 3.63e-05,
+      "loss": 0.0279,
+      "step": 637
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.8220503926277161,
+      "learning_rate": 3.62e-05,
+      "loss": 0.0059,
+      "step": 638
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.3052619993686676,
+      "learning_rate": 3.61e-05,
+      "loss": 0.0032,
+      "step": 639
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11896578967571259,
+      "learning_rate": 3.6e-05,
+      "loss": 0.0007,
+      "step": 640
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.0606658011674881,
+      "learning_rate": 3.59e-05,
+      "loss": 0.0006,
+      "step": 641
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.5427333116531372,
+      "learning_rate": 3.58e-05,
+      "loss": 0.0036,
+      "step": 642
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.271037757396698,
+      "learning_rate": 3.57e-05,
+      "loss": 0.0016,
+      "step": 643
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.1393972635269165,
+      "learning_rate": 3.56e-05,
+      "loss": 0.007,
+      "step": 644
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.5412964820861816,
+      "learning_rate": 3.55e-05,
+      "loss": 0.0131,
+      "step": 645
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.731440782546997,
+      "learning_rate": 3.54e-05,
+      "loss": 0.0091,
+      "step": 646
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.20476283133029938,
+      "learning_rate": 3.53e-05,
+      "loss": 0.0015,
+      "step": 647
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 7.767834186553955,
+      "learning_rate": 3.52e-05,
+      "loss": 0.0518,
+      "step": 648
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 39.933876037597656,
+      "learning_rate": 3.51e-05,
+      "loss": 0.7671,
+      "step": 649
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 39.605648040771484,
+      "learning_rate": 3.5e-05,
+      "loss": 1.9882,
+      "step": 650
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.08953946828842163,
+      "learning_rate": 3.49e-05,
+      "loss": 0.0007,
+      "step": 651
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 11.983209609985352,
+      "learning_rate": 3.48e-05,
+      "loss": 0.0733,
+      "step": 652
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.12467926740646362,
+      "learning_rate": 3.4699999999999996e-05,
+      "loss": 0.0009,
+      "step": 653
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 28.761707305908203,
+      "learning_rate": 3.46e-05,
+      "loss": 1.0135,
+      "step": 654
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9357807040214539,
+      "learning_rate": 3.45e-05,
+      "loss": 0.0098,
+      "step": 655
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 33.30131149291992,
+      "learning_rate": 3.4399999999999996e-05,
+      "loss": 0.3859,
+      "step": 656
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.06523477286100388,
+      "learning_rate": 3.430000000000001e-05,
+      "loss": 0.0004,
+      "step": 657
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 26.81741714477539,
+      "learning_rate": 3.4200000000000005e-05,
+      "loss": 1.4735,
+      "step": 658
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.06154371052980423,
+      "learning_rate": 3.41e-05,
+      "loss": 0.0004,
+      "step": 659
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 24.583940505981445,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.4161,
+      "step": 660
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 36.68991470336914,
+      "learning_rate": 3.3900000000000004e-05,
+      "loss": 1.0415,
+      "step": 661
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 45.04407501220703,
+      "learning_rate": 3.38e-05,
+      "loss": 0.9511,
+      "step": 662
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 9.760159492492676,
+      "learning_rate": 3.3700000000000006e-05,
+      "loss": 0.0636,
+      "step": 663
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 37.501834869384766,
+      "learning_rate": 3.3600000000000004e-05,
+      "loss": 0.5429,
+      "step": 664
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 7.125507354736328,
+      "learning_rate": 3.35e-05,
+      "loss": 0.0742,
+      "step": 665
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.6868292093276978,
+      "learning_rate": 3.3400000000000005e-05,
+      "loss": 0.0104,
+      "step": 666
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 44.928775787353516,
+      "learning_rate": 3.33e-05,
+      "loss": 0.5577,
+      "step": 667
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 23.767250061035156,
+      "learning_rate": 3.32e-05,
+      "loss": 0.2924,
+      "step": 668
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 27.04254722595215,
+      "learning_rate": 3.3100000000000005e-05,
+      "loss": 0.4817,
+      "step": 669
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.3137115240097046,
+      "learning_rate": 3.3e-05,
+      "loss": 0.0021,
+      "step": 670
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 10.24022388458252,
+      "learning_rate": 3.29e-05,
+      "loss": 0.0798,
+      "step": 671
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.1032443046569824,
+      "learning_rate": 3.2800000000000004e-05,
+      "loss": 0.0103,
+      "step": 672
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.888711452484131,
+      "learning_rate": 3.27e-05,
+      "loss": 0.0211,
+      "step": 673
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.06245700269937515,
+      "learning_rate": 3.26e-05,
+      "loss": 0.0004,
+      "step": 674
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 17.818830490112305,
+      "learning_rate": 3.2500000000000004e-05,
+      "loss": 0.1672,
+      "step": 675
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.842295527458191,
+      "learning_rate": 3.24e-05,
+      "loss": 0.0102,
+      "step": 676
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.6570789813995361,
+      "learning_rate": 3.2300000000000006e-05,
+      "loss": 0.016,
+      "step": 677
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5554737448692322,
+      "learning_rate": 3.2200000000000003e-05,
+      "loss": 0.0032,
+      "step": 678
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 26.97112464904785,
+      "learning_rate": 3.21e-05,
+      "loss": 0.8228,
+      "step": 679
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.09880402684211731,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 0.0007,
+      "step": 680
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 26.534805297851562,
+      "learning_rate": 3.19e-05,
+      "loss": 1.1114,
+      "step": 681
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 13.699009895324707,
+      "learning_rate": 3.18e-05,
+      "loss": 0.104,
+      "step": 682
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 27.22563362121582,
+      "learning_rate": 3.1700000000000005e-05,
+      "loss": 1.0485,
+      "step": 683
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 13.79177474975586,
+      "learning_rate": 3.16e-05,
+      "loss": 0.1139,
+      "step": 684
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.6369801163673401,
+      "learning_rate": 3.15e-05,
+      "loss": 0.004,
+      "step": 685
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.0804632157087326,
+      "learning_rate": 3.1400000000000004e-05,
+      "loss": 0.0006,
+      "step": 686
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.0806119441986084,
+      "learning_rate": 3.13e-05,
+      "loss": 0.0084,
+      "step": 687
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 19.096689224243164,
+      "learning_rate": 3.12e-05,
+      "loss": 0.1812,
+      "step": 688
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 11.853796005249023,
+      "learning_rate": 3.1100000000000004e-05,
+      "loss": 0.0738,
+      "step": 689
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 13.981377601623535,
+      "learning_rate": 3.1e-05,
+      "loss": 0.1197,
+      "step": 690
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 19.2493896484375,
+      "learning_rate": 3.09e-05,
+      "loss": 0.1576,
+      "step": 691
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 26.30409812927246,
+      "learning_rate": 3.08e-05,
+      "loss": 0.497,
+      "step": 692
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 24.468488693237305,
+      "learning_rate": 3.07e-05,
+      "loss": 0.3814,
+      "step": 693
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.004858358763158321,
+      "learning_rate": 3.06e-05,
+      "loss": 0.0,
+      "step": 694
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 20.56285858154297,
+      "learning_rate": 3.05e-05,
+      "loss": 0.2559,
+      "step": 695
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 10.914053916931152,
+      "learning_rate": 3.04e-05,
+      "loss": 0.0581,
+      "step": 696
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 53.04142761230469,
+      "learning_rate": 3.03e-05,
+      "loss": 1.5222,
+      "step": 697
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.3830277919769287,
+      "learning_rate": 3.02e-05,
+      "loss": 0.016,
+      "step": 698
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 24.649484634399414,
+      "learning_rate": 3.01e-05,
+      "loss": 0.8276,
+      "step": 699
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 26.6253604888916,
+      "learning_rate": 3e-05,
+      "loss": 0.5528,
+      "step": 700
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.891787052154541,
+      "learning_rate": 2.9900000000000002e-05,
+      "loss": 0.005,
+      "step": 701
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 3.0499017238616943,
+      "learning_rate": 2.98e-05,
+      "loss": 0.0156,
+      "step": 702
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.6534267663955688,
+      "learning_rate": 2.97e-05,
+      "loss": 0.0084,
+      "step": 703
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.562388896942139,
+      "learning_rate": 2.96e-05,
+      "loss": 0.0455,
+      "step": 704
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.963227272033691,
+      "learning_rate": 2.95e-05,
+      "loss": 0.0556,
+      "step": 705
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 2.26806902885437,
+      "learning_rate": 2.94e-05,
+      "loss": 0.0181,
+      "step": 706
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.621293544769287,
+      "learning_rate": 2.93e-05,
+      "loss": 0.0342,
+      "step": 707
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.3657660484313965,
+      "learning_rate": 2.9199999999999998e-05,
+      "loss": 0.0171,
+      "step": 708
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 8.781088829040527,
+      "learning_rate": 2.91e-05,
+      "loss": 0.0495,
+      "step": 709
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 24.759685516357422,
+      "learning_rate": 2.9e-05,
+      "loss": 0.3487,
+      "step": 710
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 23.32072639465332,
+      "learning_rate": 2.8899999999999998e-05,
+      "loss": 0.2415,
+      "step": 711
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 14.044452667236328,
+      "learning_rate": 2.88e-05,
+      "loss": 0.1113,
+      "step": 712
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.096346855163574,
+      "learning_rate": 2.87e-05,
+      "loss": 0.0154,
+      "step": 713
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9582540392875671,
+      "learning_rate": 2.86e-05,
+      "loss": 0.0084,
+      "step": 714
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.22299572825431824,
+      "learning_rate": 2.8499999999999998e-05,
+      "loss": 0.0018,
+      "step": 715
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 23.300783157348633,
+      "learning_rate": 2.84e-05,
+      "loss": 0.2524,
+      "step": 716
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 26.694652557373047,
+      "learning_rate": 2.83e-05,
+      "loss": 0.3081,
+      "step": 717
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 26.942161560058594,
+      "learning_rate": 2.8199999999999998e-05,
+      "loss": 0.2603,
+      "step": 718
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.22826315462589264,
+      "learning_rate": 2.8100000000000005e-05,
+      "loss": 0.0013,
+      "step": 719
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.042904481291770935,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 0.0003,
+      "step": 720
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 5.835173606872559,
+      "learning_rate": 2.7900000000000004e-05,
+      "loss": 0.0336,
+      "step": 721
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2858074903488159,
+      "learning_rate": 2.7800000000000005e-05,
+      "loss": 0.0027,
+      "step": 722
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 4.612087249755859,
+      "learning_rate": 2.7700000000000002e-05,
+      "loss": 0.0243,
+      "step": 723
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 8.472134590148926,
+      "learning_rate": 2.7600000000000003e-05,
+      "loss": 0.0506,
+      "step": 724
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 13.461259841918945,
+      "learning_rate": 2.7500000000000004e-05,
+      "loss": 0.1264,
+      "step": 725
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.7797164916992188,
+      "learning_rate": 2.7400000000000002e-05,
+      "loss": 0.0093,
+      "step": 726
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.120004653930664,
+      "learning_rate": 2.7300000000000003e-05,
+      "loss": 0.052,
+      "step": 727
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 27.213024139404297,
+      "learning_rate": 2.7200000000000004e-05,
+      "loss": 0.6143,
+      "step": 728
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 23.018705368041992,
+      "learning_rate": 2.7100000000000005e-05,
+      "loss": 0.355,
+      "step": 729
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 5.7414655685424805,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 0.0318,
+      "step": 730
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 23.56236457824707,
+      "learning_rate": 2.6900000000000003e-05,
+      "loss": 0.3694,
+      "step": 731
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.875092029571533,
+      "learning_rate": 2.6800000000000004e-05,
+      "loss": 0.0448,
+      "step": 732
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 26.399948120117188,
+      "learning_rate": 2.6700000000000002e-05,
+      "loss": 0.7737,
+      "step": 733
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 28.907119750976562,
+      "learning_rate": 2.6600000000000003e-05,
+      "loss": 0.2974,
+      "step": 734
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 2.2717533111572266,
+      "learning_rate": 2.6500000000000004e-05,
+      "loss": 0.0148,
+      "step": 735
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.15304788947105408,
+      "learning_rate": 2.64e-05,
+      "loss": 0.0009,
+      "step": 736
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 28.949296951293945,
+      "learning_rate": 2.6300000000000002e-05,
+      "loss": 0.5319,
+      "step": 737
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.7885856628417969,
+      "learning_rate": 2.6200000000000003e-05,
+      "loss": 0.0058,
+      "step": 738
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 26.116670608520508,
+      "learning_rate": 2.61e-05,
+      "loss": 0.3608,
+      "step": 739
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.3367154598236084,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.0029,
+      "step": 740
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 23.798566818237305,
+      "learning_rate": 2.5900000000000003e-05,
+      "loss": 0.2741,
+      "step": 741
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 32.780723571777344,
+      "learning_rate": 2.58e-05,
+      "loss": 0.4461,
+      "step": 742
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.2866144180297852,
+      "learning_rate": 2.57e-05,
+      "loss": 0.0072,
+      "step": 743
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.4138367176055908,
+      "learning_rate": 2.5600000000000002e-05,
+      "loss": 0.0022,
+      "step": 744
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.8757226467132568,
+      "learning_rate": 2.5500000000000003e-05,
+      "loss": 0.0049,
+      "step": 745
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 27.533611297607422,
+      "learning_rate": 2.54e-05,
+      "loss": 0.4811,
+      "step": 746
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 12.152800559997559,
+      "learning_rate": 2.5300000000000002e-05,
+      "loss": 0.1209,
+      "step": 747
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 25.30514144897461,
+      "learning_rate": 2.5200000000000003e-05,
+      "loss": 0.3028,
+      "step": 748
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.702767252922058,
+      "learning_rate": 2.51e-05,
+      "loss": 0.0094,
+      "step": 749
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 12.766653060913086,
+      "learning_rate": 2.5e-05,
+      "loss": 0.0826,
+      "step": 750
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 23.281902313232422,
+      "learning_rate": 2.4900000000000002e-05,
+      "loss": 0.8941,
+      "step": 751
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.5401146411895752,
+      "learning_rate": 2.48e-05,
+      "loss": 0.008,
+      "step": 752
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 16.982078552246094,
+      "learning_rate": 2.47e-05,
+      "loss": 0.1315,
+      "step": 753
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.11115346103906631,
+      "learning_rate": 2.46e-05,
+      "loss": 0.0007,
+      "step": 754
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 31.15369987487793,
+      "learning_rate": 2.45e-05,
+      "loss": 0.759,
+      "step": 755
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 36.960872650146484,
+      "learning_rate": 2.44e-05,
+      "loss": 0.4665,
+      "step": 756
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.038327042013406754,
+      "learning_rate": 2.43e-05,
+      "loss": 0.0003,
+      "step": 757
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.14821617305278778,
+      "learning_rate": 2.4200000000000002e-05,
+      "loss": 0.0013,
+      "step": 758
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 20.849529266357422,
+      "learning_rate": 2.41e-05,
+      "loss": 0.1943,
+      "step": 759
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 21.639812469482422,
+      "learning_rate": 2.4e-05,
+      "loss": 0.1947,
+      "step": 760
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 26.269208908081055,
+      "learning_rate": 2.39e-05,
+      "loss": 0.8363,
+      "step": 761
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 16.222475051879883,
+      "learning_rate": 2.38e-05,
+      "loss": 0.1137,
+      "step": 762
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 31.54730796813965,
+      "learning_rate": 2.37e-05,
+      "loss": 1.8999,
+      "step": 763
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 17.911474227905273,
+      "learning_rate": 2.36e-05,
+      "loss": 0.1302,
+      "step": 764
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 47.63165283203125,
+      "learning_rate": 2.35e-05,
+      "loss": 1.0344,
+      "step": 765
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 26.189237594604492,
+      "learning_rate": 2.3400000000000003e-05,
+      "loss": 0.4308,
+      "step": 766
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.35479736328125,
+      "learning_rate": 2.3300000000000004e-05,
+      "loss": 0.0129,
+      "step": 767
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 9.399853706359863,
+      "learning_rate": 2.32e-05,
+      "loss": 0.066,
+      "step": 768
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.1681160926818848,
+      "learning_rate": 2.3100000000000002e-05,
+      "loss": 0.0062,
+      "step": 769
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 29.163421630859375,
+      "learning_rate": 2.3000000000000003e-05,
+      "loss": 0.7343,
+      "step": 770
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 4.385603904724121,
+      "learning_rate": 2.29e-05,
+      "loss": 0.0237,
+      "step": 771
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.0795365571975708,
+      "learning_rate": 2.2800000000000002e-05,
+      "loss": 0.0062,
+      "step": 772
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 22.450462341308594,
+      "learning_rate": 2.2700000000000003e-05,
+      "loss": 0.2405,
+      "step": 773
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 4.557497978210449,
+      "learning_rate": 2.26e-05,
+      "loss": 0.0423,
+      "step": 774
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 7.581943511962891,
+      "learning_rate": 2.25e-05,
+      "loss": 0.0433,
+      "step": 775
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 13.19905948638916,
+      "learning_rate": 2.2400000000000002e-05,
+      "loss": 0.1416,
+      "step": 776
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 24.024709701538086,
+      "learning_rate": 2.23e-05,
+      "loss": 0.4463,
+      "step": 777
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 8.119006156921387,
+      "learning_rate": 2.22e-05,
+      "loss": 0.0608,
+      "step": 778
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 3.916139841079712,
+      "learning_rate": 2.2100000000000002e-05,
+      "loss": 0.0216,
+      "step": 779
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.002933024661615491,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.0,
+      "step": 780
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.1204041242599487,
+      "learning_rate": 2.19e-05,
+      "loss": 0.0081,
+      "step": 781
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 31.576692581176758,
+      "learning_rate": 2.18e-05,
+      "loss": 1.0689,
+      "step": 782
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 31.140485763549805,
+      "learning_rate": 2.1700000000000002e-05,
+      "loss": 0.335,
+      "step": 783
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.42327681183815,
+      "learning_rate": 2.16e-05,
+      "loss": 0.0046,
+      "step": 784
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.32425838708877563,
+      "learning_rate": 2.15e-05,
+      "loss": 0.0017,
+      "step": 785
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 29.009767532348633,
+      "learning_rate": 2.1400000000000002e-05,
+      "loss": 1.1913,
+      "step": 786
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 43.424034118652344,
+      "learning_rate": 2.13e-05,
+      "loss": 1.1988,
+      "step": 787
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.1210248470306396,
+      "learning_rate": 2.12e-05,
+      "loss": 0.0065,
+      "step": 788
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.054579153656959534,
+      "learning_rate": 2.11e-05,
+      "loss": 0.0003,
+      "step": 789
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.4357629716396332,
+      "learning_rate": 2.1e-05,
+      "loss": 0.0034,
+      "step": 790
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 34.34247589111328,
+      "learning_rate": 2.09e-05,
+      "loss": 0.5637,
+      "step": 791
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 36.91853713989258,
+      "learning_rate": 2.08e-05,
+      "loss": 0.7341,
+      "step": 792
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 8.309200286865234,
+      "learning_rate": 2.07e-05,
+      "loss": 0.0432,
+      "step": 793
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 12.616776466369629,
+      "learning_rate": 2.06e-05,
+      "loss": 0.0952,
+      "step": 794
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.600636899471283,
+      "learning_rate": 2.05e-05,
+      "loss": 0.0058,
+      "step": 795
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 44.27837371826172,
+      "learning_rate": 2.04e-05,
+      "loss": 1.0362,
+      "step": 796
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 26.939998626708984,
+      "learning_rate": 2.0300000000000002e-05,
+      "loss": 0.9799,
+      "step": 797
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 45.84077835083008,
+      "learning_rate": 2.0200000000000003e-05,
+      "loss": 0.5936,
+      "step": 798
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 5.504016399383545,
+      "learning_rate": 2.01e-05,
+      "loss": 0.0332,
+      "step": 799
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 3.4088833332061768,
+      "learning_rate": 2e-05,
+      "loss": 0.0165,
+      "step": 800
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.32817769050598145,
+      "learning_rate": 1.9900000000000003e-05,
+      "loss": 0.0019,
+      "step": 801
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 19.7551212310791,
+      "learning_rate": 1.9800000000000004e-05,
+      "loss": 0.2079,
+      "step": 802
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 27.117473602294922,
+      "learning_rate": 1.97e-05,
+      "loss": 1.0161,
+      "step": 803
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.30872708559036255,
+      "learning_rate": 1.9600000000000002e-05,
+      "loss": 0.0023,
+      "step": 804
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 32.718196868896484,
+      "learning_rate": 1.9500000000000003e-05,
+      "loss": 0.8794,
+      "step": 805
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 26.284860610961914,
+      "learning_rate": 1.94e-05,
+      "loss": 0.6486,
+      "step": 806
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.6852843761444092,
+      "learning_rate": 1.93e-05,
+      "loss": 0.0036,
+      "step": 807
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 24.214462280273438,
+      "learning_rate": 1.9200000000000003e-05,
+      "loss": 0.2703,
+      "step": 808
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 16.324054718017578,
+      "learning_rate": 1.91e-05,
+      "loss": 0.1177,
+      "step": 809
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 24.803068161010742,
+      "learning_rate": 1.9e-05,
+      "loss": 1.05,
+      "step": 810
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 48.46371841430664,
+      "learning_rate": 1.8900000000000002e-05,
+      "loss": 0.7774,
+      "step": 811
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 19.621374130249023,
+      "learning_rate": 1.88e-05,
+      "loss": 0.1642,
+      "step": 812
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 32.60996627807617,
+      "learning_rate": 1.87e-05,
+      "loss": 0.9919,
+      "step": 813
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 14.111787796020508,
+      "learning_rate": 1.86e-05,
+      "loss": 0.094,
+      "step": 814
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.027604341506958,
+      "learning_rate": 1.85e-05,
+      "loss": 0.007,
+      "step": 815
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5485715866088867,
+      "learning_rate": 1.84e-05,
+      "loss": 0.0026,
+      "step": 816
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.3733130395412445,
+      "learning_rate": 1.83e-05,
+      "loss": 0.0024,
+      "step": 817
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 26.762765884399414,
+      "learning_rate": 1.8200000000000002e-05,
+      "loss": 0.5703,
+      "step": 818
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 11.364485740661621,
+      "learning_rate": 1.81e-05,
+      "loss": 0.0749,
+      "step": 819
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 16.991313934326172,
+      "learning_rate": 1.8e-05,
+      "loss": 0.1318,
+      "step": 820
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.18820422887802124,
+      "learning_rate": 1.79e-05,
+      "loss": 0.001,
+      "step": 821
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 27.79010009765625,
+      "learning_rate": 1.78e-05,
+      "loss": 1.011,
+      "step": 822
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 37.26203155517578,
+      "learning_rate": 1.77e-05,
+      "loss": 0.8081,
+      "step": 823
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.4303054213523865,
+      "learning_rate": 1.76e-05,
+      "loss": 0.0037,
+      "step": 824
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.8750838041305542,
+      "learning_rate": 1.75e-05,
+      "loss": 0.0047,
+      "step": 825
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 5.441610336303711,
+      "learning_rate": 1.74e-05,
+      "loss": 0.0306,
+      "step": 826
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.0719880685210228,
+      "learning_rate": 1.73e-05,
+      "loss": 0.0005,
+      "step": 827
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 26.52375030517578,
+      "learning_rate": 1.7199999999999998e-05,
+      "loss": 0.8269,
+      "step": 828
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 3.149235963821411,
+      "learning_rate": 1.7100000000000002e-05,
+      "loss": 0.0166,
+      "step": 829
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.7330241799354553,
+      "learning_rate": 1.7000000000000003e-05,
+      "loss": 0.0038,
+      "step": 830
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 29.252592086791992,
+      "learning_rate": 1.69e-05,
+      "loss": 0.5713,
+      "step": 831
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 27.370386123657227,
+      "learning_rate": 1.6800000000000002e-05,
+      "loss": 0.8151,
+      "step": 832
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.770989179611206,
+      "learning_rate": 1.6700000000000003e-05,
+      "loss": 0.0283,
+      "step": 833
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.7080297470092773,
+      "learning_rate": 1.66e-05,
+      "loss": 0.0085,
+      "step": 834
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 11.255769729614258,
+      "learning_rate": 1.65e-05,
+      "loss": 0.1052,
+      "step": 835
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 26.47229766845703,
+      "learning_rate": 1.6400000000000002e-05,
+      "loss": 1.0482,
+      "step": 836
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 26.88467025756836,
+      "learning_rate": 1.63e-05,
+      "loss": 0.5596,
+      "step": 837
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 27.23533058166504,
+      "learning_rate": 1.62e-05,
+      "loss": 0.7561,
+      "step": 838
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 30.387807846069336,
+      "learning_rate": 1.6100000000000002e-05,
+      "loss": 1.4771,
+      "step": 839
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.03761901333928108,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.0006,
+      "step": 840
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 2.5605359077453613,
+      "learning_rate": 1.59e-05,
+      "loss": 0.015,
+      "step": 841
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.10990981757640839,
+      "learning_rate": 1.58e-05,
+      "loss": 0.0008,
+      "step": 842
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 11.03862476348877,
+      "learning_rate": 1.5700000000000002e-05,
+      "loss": 0.0686,
+      "step": 843
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 20.627206802368164,
+      "learning_rate": 1.56e-05,
+      "loss": 0.3718,
+      "step": 844
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.19297750294208527,
+      "learning_rate": 1.55e-05,
+      "loss": 0.0013,
+      "step": 845
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.0797531604766846,
+      "learning_rate": 1.54e-05,
+      "loss": 0.0102,
+      "step": 846
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.0616090297698975,
+      "learning_rate": 1.53e-05,
+      "loss": 0.0124,
+      "step": 847
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 5.628384590148926,
+      "learning_rate": 1.52e-05,
+      "loss": 0.0354,
+      "step": 848
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 25.70692253112793,
+      "learning_rate": 1.51e-05,
+      "loss": 0.4203,
+      "step": 849
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 21.292953491210938,
+      "learning_rate": 1.5e-05,
+      "loss": 1.182,
+      "step": 850
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 8.425320625305176,
+      "learning_rate": 1.49e-05,
+      "loss": 0.0462,
+      "step": 851
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 26.563316345214844,
+      "learning_rate": 1.48e-05,
+      "loss": 0.4144,
+      "step": 852
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.6242803931236267,
+      "learning_rate": 1.47e-05,
+      "loss": 0.0052,
+      "step": 853
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.20703597366809845,
+      "learning_rate": 1.4599999999999999e-05,
+      "loss": 0.0013,
+      "step": 854
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 19.229211807250977,
+      "learning_rate": 1.45e-05,
+      "loss": 0.1673,
+      "step": 855
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.8394806385040283,
+      "learning_rate": 1.44e-05,
+      "loss": 0.0141,
+      "step": 856
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.4133518636226654,
+      "learning_rate": 1.43e-05,
+      "loss": 0.0021,
+      "step": 857
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.384779453277588,
+      "learning_rate": 1.42e-05,
+      "loss": 0.0126,
+      "step": 858
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 12.650036811828613,
+      "learning_rate": 1.4099999999999999e-05,
+      "loss": 0.1045,
+      "step": 859
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.14024153351783752,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 0.0015,
+      "step": 860
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 25.9268741607666,
+      "learning_rate": 1.3900000000000002e-05,
+      "loss": 0.2269,
+      "step": 861
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 22.659929275512695,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 0.5073,
+      "step": 862
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.20133636891841888,
+      "learning_rate": 1.3700000000000001e-05,
+      "loss": 0.0011,
+      "step": 863
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 16.163724899291992,
+      "learning_rate": 1.3600000000000002e-05,
+      "loss": 0.1342,
+      "step": 864
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 35.01076126098633,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 1.3155,
+      "step": 865
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 24.18101692199707,
+      "learning_rate": 1.3400000000000002e-05,
+      "loss": 0.2992,
+      "step": 866
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.3129206895828247,
+      "learning_rate": 1.3300000000000001e-05,
+      "loss": 0.0019,
+      "step": 867
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 3.156733751296997,
+      "learning_rate": 1.32e-05,
+      "loss": 0.0197,
+      "step": 868
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 20.24107551574707,
+      "learning_rate": 1.3100000000000002e-05,
+      "loss": 0.2469,
+      "step": 869
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.023805420845746994,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.0002,
+      "step": 870
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 25.843679428100586,
+      "learning_rate": 1.29e-05,
+      "loss": 0.4148,
+      "step": 871
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 46.338687896728516,
+      "learning_rate": 1.2800000000000001e-05,
+      "loss": 0.7211,
+      "step": 872
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 28.163171768188477,
+      "learning_rate": 1.27e-05,
+      "loss": 1.2761,
+      "step": 873
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 5.789135456085205,
+      "learning_rate": 1.2600000000000001e-05,
+      "loss": 0.0272,
+      "step": 874
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 8.08776569366455,
+      "learning_rate": 1.25e-05,
+      "loss": 0.0441,
+      "step": 875
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.10339970886707306,
+      "learning_rate": 1.24e-05,
+      "loss": 0.0014,
+      "step": 876
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 26.45261573791504,
+      "learning_rate": 1.23e-05,
+      "loss": 0.8787,
+      "step": 877
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 7.62965202331543,
+      "learning_rate": 1.22e-05,
+      "loss": 0.0466,
+      "step": 878
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 10.6347017288208,
+      "learning_rate": 1.2100000000000001e-05,
+      "loss": 0.0693,
+      "step": 879
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 26.655921936035156,
+      "learning_rate": 1.2e-05,
+      "loss": 0.528,
+      "step": 880
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.786623477935791,
+      "learning_rate": 1.19e-05,
+      "loss": 0.0181,
+      "step": 881
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.6593170166015625,
+      "learning_rate": 1.18e-05,
+      "loss": 0.005,
+      "step": 882
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 42.40369415283203,
+      "learning_rate": 1.1700000000000001e-05,
+      "loss": 0.889,
+      "step": 883
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 27.659780502319336,
+      "learning_rate": 1.16e-05,
+      "loss": 1.0498,
+      "step": 884
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 15.126568794250488,
+      "learning_rate": 1.1500000000000002e-05,
+      "loss": 0.1185,
+      "step": 885
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 29.32221794128418,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 0.895,
+      "step": 886
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 5.868095397949219,
+      "learning_rate": 1.13e-05,
+      "loss": 0.0429,
+      "step": 887
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 25.551660537719727,
+      "learning_rate": 1.1200000000000001e-05,
+      "loss": 0.2916,
+      "step": 888
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 26.83506202697754,
+      "learning_rate": 1.11e-05,
+      "loss": 0.5324,
+      "step": 889
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.322347640991211,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.0221,
+      "step": 890
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 28.00304412841797,
+      "learning_rate": 1.09e-05,
+      "loss": 0.6972,
+      "step": 891
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 2.1361875534057617,
+      "learning_rate": 1.08e-05,
+      "loss": 0.0102,
+      "step": 892
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.4008387327194214,
+      "learning_rate": 1.0700000000000001e-05,
+      "loss": 0.0104,
+      "step": 893
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 21.043048858642578,
+      "learning_rate": 1.06e-05,
+      "loss": 0.1851,
+      "step": 894
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.06448391079902649,
+      "learning_rate": 1.05e-05,
+      "loss": 0.0004,
+      "step": 895
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.15726327896118164,
+      "learning_rate": 1.04e-05,
+      "loss": 0.0012,
+      "step": 896
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 44.16018295288086,
+      "learning_rate": 1.03e-05,
+      "loss": 1.0449,
+      "step": 897
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.02686498314142227,
+      "learning_rate": 1.02e-05,
+      "loss": 0.0002,
+      "step": 898
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.17026451230049133,
+      "learning_rate": 1.0100000000000002e-05,
+      "loss": 0.0012,
+      "step": 899
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 44.215797424316406,
+      "learning_rate": 1e-05,
+      "loss": 1.2545,
+      "step": 900
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9567099809646606,
+      "learning_rate": 9.900000000000002e-06,
+      "loss": 0.0068,
+      "step": 901
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 4.834523677825928,
+      "learning_rate": 9.800000000000001e-06,
+      "loss": 0.027,
+      "step": 902
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.023485347628593445,
+      "learning_rate": 9.7e-06,
+      "loss": 0.0001,
+      "step": 903
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 14.029927253723145,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 0.0971,
+      "step": 904
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 16.840566635131836,
+      "learning_rate": 9.5e-06,
+      "loss": 0.1671,
+      "step": 905
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9835864901542664,
+      "learning_rate": 9.4e-06,
+      "loss": 0.005,
+      "step": 906
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 3.175522804260254,
+      "learning_rate": 9.3e-06,
+      "loss": 0.0173,
+      "step": 907
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 27.09811019897461,
+      "learning_rate": 9.2e-06,
+      "loss": 1.2902,
+      "step": 908
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 22.86342430114746,
+      "learning_rate": 9.100000000000001e-06,
+      "loss": 0.2469,
+      "step": 909
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.795168399810791,
+      "learning_rate": 9e-06,
+      "loss": 0.0045,
+      "step": 910
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.3939276933670044,
+      "learning_rate": 8.9e-06,
+      "loss": 0.0021,
+      "step": 911
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 27.596954345703125,
+      "learning_rate": 8.8e-06,
+      "loss": 0.9408,
+      "step": 912
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 15.893352508544922,
+      "learning_rate": 8.7e-06,
+      "loss": 0.1598,
+      "step": 913
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.5591702461242676,
+      "learning_rate": 8.599999999999999e-06,
+      "loss": 0.0103,
+      "step": 914
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 26.889225006103516,
+      "learning_rate": 8.500000000000002e-06,
+      "loss": 0.5883,
+      "step": 915
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 47.645057678222656,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 2.2247,
+      "step": 916
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 4.39687442779541,
+      "learning_rate": 8.3e-06,
+      "loss": 0.0234,
+      "step": 917
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.21854829788208,
+      "learning_rate": 8.200000000000001e-06,
+      "loss": 0.0102,
+      "step": 918
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.2136417627334595,
+      "learning_rate": 8.1e-06,
+      "loss": 0.0094,
+      "step": 919
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.374004602432251,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.0069,
+      "step": 920
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.16840314865112305,
+      "learning_rate": 7.9e-06,
+      "loss": 0.0014,
+      "step": 921
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.11105885356664658,
+      "learning_rate": 7.8e-06,
+      "loss": 0.0011,
+      "step": 922
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 19.66978645324707,
+      "learning_rate": 7.7e-06,
+      "loss": 0.3866,
+      "step": 923
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.2701959609985352,
+      "learning_rate": 7.6e-06,
+      "loss": 0.0061,
+      "step": 924
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 27.793781280517578,
+      "learning_rate": 7.5e-06,
+      "loss": 0.8543,
+      "step": 925
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 30.1798095703125,
+      "learning_rate": 7.4e-06,
+      "loss": 0.5122,
+      "step": 926
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.5201961398124695,
+      "learning_rate": 7.2999999999999996e-06,
+      "loss": 0.0027,
+      "step": 927
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 10.465607643127441,
+      "learning_rate": 7.2e-06,
+      "loss": 0.0573,
+      "step": 928
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.1310155391693115,
+      "learning_rate": 7.1e-06,
+      "loss": 0.0131,
+      "step": 929
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 5.451208114624023,
+      "learning_rate": 7.000000000000001e-06,
+      "loss": 0.0284,
+      "step": 930
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 24.128276824951172,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 0.3309,
+      "step": 931
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 25.905508041381836,
+      "learning_rate": 6.800000000000001e-06,
+      "loss": 0.389,
+      "step": 932
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 31.932828903198242,
+      "learning_rate": 6.700000000000001e-06,
+      "loss": 1.2094,
+      "step": 933
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.16106216609477997,
+      "learning_rate": 6.6e-06,
+      "loss": 0.0013,
+      "step": 934
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 28.226865768432617,
+      "learning_rate": 6.5000000000000004e-06,
+      "loss": 0.5602,
+      "step": 935
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.7520970702171326,
+      "learning_rate": 6.4000000000000006e-06,
+      "loss": 0.005,
+      "step": 936
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.36321502923965454,
+      "learning_rate": 6.300000000000001e-06,
+      "loss": 0.003,
+      "step": 937
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.2656748294830322,
+      "learning_rate": 6.2e-06,
+      "loss": 0.0064,
+      "step": 938
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 25.317880630493164,
+      "learning_rate": 6.1e-06,
+      "loss": 0.5348,
+      "step": 939
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.06496942788362503,
+      "learning_rate": 6e-06,
+      "loss": 0.0004,
+      "step": 940
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.31098636984825134,
+      "learning_rate": 5.9e-06,
+      "loss": 0.0026,
+      "step": 941
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.3573431968688965,
+      "learning_rate": 5.8e-06,
+      "loss": 0.0117,
+      "step": 942
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.8721911907196045,
+      "learning_rate": 5.7000000000000005e-06,
+      "loss": 0.0127,
+      "step": 943
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 27.062021255493164,
+      "learning_rate": 5.600000000000001e-06,
+      "loss": 0.6997,
+      "step": 944
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.5216832756996155,
+      "learning_rate": 5.500000000000001e-06,
+      "loss": 0.003,
+      "step": 945
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 26.380741119384766,
+      "learning_rate": 5.4e-06,
+      "loss": 0.273,
+      "step": 946
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 25.798185348510742,
+      "learning_rate": 5.3e-06,
+      "loss": 1.0317,
+      "step": 947
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 11.128933906555176,
+      "learning_rate": 5.2e-06,
+      "loss": 0.0781,
+      "step": 948
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.875777244567871,
+      "learning_rate": 5.1e-06,
+      "loss": 0.0096,
+      "step": 949
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.302217036485672,
+      "learning_rate": 5e-06,
+      "loss": 0.0016,
+      "step": 950
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.35395440459251404,
+      "learning_rate": 4.9000000000000005e-06,
+      "loss": 0.0021,
+      "step": 951
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 31.090530395507812,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.5176,
+      "step": 952
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.993788242340088,
+      "learning_rate": 4.7e-06,
+      "loss": 0.0169,
+      "step": 953
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 3.8373305797576904,
+      "learning_rate": 4.6e-06,
+      "loss": 0.0227,
+      "step": 954
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.49911898374557495,
+      "learning_rate": 4.5e-06,
+      "loss": 0.0037,
+      "step": 955
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 26.027299880981445,
+      "learning_rate": 4.4e-06,
+      "loss": 0.9036,
+      "step": 956
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 11.176959037780762,
+      "learning_rate": 4.2999999999999995e-06,
+      "loss": 0.0678,
+      "step": 957
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 2.6982274055480957,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.0144,
+      "step": 958
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 37.264869689941406,
+      "learning_rate": 4.1000000000000006e-06,
+      "loss": 1.4107,
+      "step": 959
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 40.02341842651367,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.3686,
+      "step": 960
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 38.77351760864258,
+      "learning_rate": 3.9e-06,
+      "loss": 0.4521,
+      "step": 961
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 4.801632881164551,
+      "learning_rate": 3.8e-06,
+      "loss": 0.0299,
+      "step": 962
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 8.612468719482422,
+      "learning_rate": 3.7e-06,
+      "loss": 0.0573,
+      "step": 963
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.5201643705368042,
+      "learning_rate": 3.6e-06,
+      "loss": 0.0041,
+      "step": 964
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.17031624913215637,
+      "learning_rate": 3.5000000000000004e-06,
+      "loss": 0.0025,
+      "step": 965
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8196523785591125,
+      "learning_rate": 3.4000000000000005e-06,
+      "loss": 0.0041,
+      "step": 966
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.013019957579672337,
+      "learning_rate": 3.3e-06,
+      "loss": 0.0001,
+      "step": 967
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 27.062700271606445,
+      "learning_rate": 3.2000000000000003e-06,
+      "loss": 0.658,
+      "step": 968
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 15.487645149230957,
+      "learning_rate": 3.1e-06,
+      "loss": 0.118,
+      "step": 969
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 5.086794853210449,
+      "learning_rate": 3e-06,
+      "loss": 0.0259,
+      "step": 970
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 27.120859146118164,
+      "learning_rate": 2.9e-06,
+      "loss": 1.1966,
+      "step": 971
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 16.72373390197754,
+      "learning_rate": 2.8000000000000003e-06,
+      "loss": 0.151,
+      "step": 972
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.03614487126469612,
+      "learning_rate": 2.7e-06,
+      "loss": 0.0004,
+      "step": 973
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.07811087369918823,
+      "learning_rate": 2.6e-06,
+      "loss": 0.0004,
+      "step": 974
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.030874168500304222,
+      "learning_rate": 2.5e-06,
+      "loss": 0.0002,
+      "step": 975
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.020734872668981552,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 0.0001,
+      "step": 976
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 32.341888427734375,
+      "learning_rate": 2.3e-06,
+      "loss": 0.3712,
+      "step": 977
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.11538070440292358,
+      "learning_rate": 2.2e-06,
+      "loss": 0.0008,
+      "step": 978
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 24.41085433959961,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 0.2808,
+      "step": 979
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 3.5011379718780518,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.0195,
+      "step": 980
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.7570369839668274,
+      "learning_rate": 1.9e-06,
+      "loss": 0.0047,
+      "step": 981
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 14.30388355255127,
+      "learning_rate": 1.8e-06,
+      "loss": 0.0972,
+      "step": 982
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.019772866740822792,
+      "learning_rate": 1.7000000000000002e-06,
+      "loss": 0.0002,
+      "step": 983
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 25.79269790649414,
+      "learning_rate": 1.6000000000000001e-06,
+      "loss": 0.3179,
+      "step": 984
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 4.481821060180664,
+      "learning_rate": 1.5e-06,
+      "loss": 0.0273,
+      "step": 985
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.6944172382354736,
+      "learning_rate": 1.4000000000000001e-06,
+      "loss": 0.0168,
+      "step": 986
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.0176031589508057,
+      "learning_rate": 1.3e-06,
+      "loss": 0.005,
+      "step": 987
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 8.57127857208252,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 0.05,
+      "step": 988
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 27.0198917388916,
+      "learning_rate": 1.1e-06,
+      "loss": 1.6885,
+      "step": 989
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 16.693689346313477,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.1505,
+      "step": 990
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 26.509174346923828,
+      "learning_rate": 9e-07,
+      "loss": 0.7264,
+      "step": 991
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 7.112478256225586,
+      "learning_rate": 8.000000000000001e-07,
+      "loss": 0.0374,
+      "step": 992
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 31.123777389526367,
+      "learning_rate": 7.000000000000001e-07,
+      "loss": 0.9894,
+      "step": 993
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 8.979686737060547,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 0.0554,
+      "step": 994
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.1432526707649231,
+      "learning_rate": 5.000000000000001e-07,
+      "loss": 0.0007,
+      "step": 995
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 27.64889907836914,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 1.3425,
+      "step": 996
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.3497440218925476,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 0.0017,
+      "step": 997
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.7565171122550964,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 0.0036,
+      "step": 998
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 18.668054580688477,
+      "learning_rate": 1.0000000000000001e-07,
+      "loss": 1.0386,
+      "step": 999
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 33.273231506347656,
+      "learning_rate": 0.0,
+      "loss": 1.1228,
+      "step": 1000
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.9305,
+      "eval_balanced_accuracy": 0.9307943873632947,
+      "eval_loss": 0.29665234684944153,
+      "eval_runtime": 737.3091,
+      "eval_samples_per_second": 2.713,
+      "eval_steps_per_second": 0.339,
+      "step": 1000
+    },
+    {
+      "epoch": 1.0,
+      "step": 1000,
+      "total_flos": 2.636836569631949e+16,
+      "train_loss": 0.3728995402829132,
+      "train_runtime": 9397.7049,
+      "train_samples_per_second": 0.851,
+      "train_steps_per_second": 0.106
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "total_flos": 2.636836569631949e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}