diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7066 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 250,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.001,
+      "grad_norm": 55.0,
+      "learning_rate": 0.0,
+      "loss": 15.8026,
+      "step": 1
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 53.5,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 15.7515,
+      "step": 2
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 54.0,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 15.6805,
+      "step": 3
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 53.75,
+      "learning_rate": 6e-06,
+      "loss": 15.6899,
+      "step": 4
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 51.75,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 15.4635,
+      "step": 5
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 52.0,
+      "learning_rate": 1e-05,
+      "loss": 15.2861,
+      "step": 6
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 50.25,
+      "learning_rate": 1.2e-05,
+      "loss": 15.0195,
+      "step": 7
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 51.75,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 14.5896,
+      "step": 8
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 48.5,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 14.428,
+      "step": 9
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 43.0,
+      "learning_rate": 1.8e-05,
+      "loss": 13.8742,
+      "step": 10
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 35.75,
+      "learning_rate": 2e-05,
+      "loss": 13.1404,
+      "step": 11
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 25.75,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 12.7393,
+      "step": 12
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 28.25,
+      "learning_rate": 2.4e-05,
+      "loss": 12.4467,
+      "step": 13
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 39.75,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 12.3137,
+      "step": 14
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 40.75,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 11.8957,
+      "step": 15
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 39.75,
+      "learning_rate": 3e-05,
+      "loss": 11.3899,
+      "step": 16
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 32.0,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 10.4023,
+      "step": 17
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 25.0,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 9.4691,
+      "step": 18
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 23.5,
+      "learning_rate": 3.6e-05,
+      "loss": 8.8468,
+      "step": 19
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 20.5,
+      "learning_rate": 3.8e-05,
+      "loss": 8.201,
+      "step": 20
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 17.5,
+      "learning_rate": 4e-05,
+      "loss": 7.7326,
+      "step": 21
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 17.375,
+      "learning_rate": 4.2e-05,
+      "loss": 7.3021,
+      "step": 22
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 14.8125,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 7.0761,
+      "step": 23
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 12.0625,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 7.0736,
+      "step": 24
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 9.625,
+      "learning_rate": 4.8e-05,
+      "loss": 6.6617,
+      "step": 25
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 7.125,
+      "learning_rate": 5e-05,
+      "loss": 6.2627,
+      "step": 26
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 6.0,
+      "learning_rate": 5.2000000000000004e-05,
+      "loss": 6.108,
+      "step": 27
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 4.65625,
+      "learning_rate": 5.4000000000000005e-05,
+      "loss": 5.7079,
+      "step": 28
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 4.5625,
+      "learning_rate": 5.6000000000000006e-05,
+      "loss": 6.0111,
+      "step": 29
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.78125,
+      "learning_rate": 5.8e-05,
+      "loss": 6.0342,
+      "step": 30
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 3.4375,
+      "learning_rate": 6e-05,
+      "loss": 5.8238,
+      "step": 31
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 3.4375,
+      "learning_rate": 6.2e-05,
+      "loss": 5.732,
+      "step": 32
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 3.171875,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 5.3643,
+      "step": 33
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 3.546875,
+      "learning_rate": 6.6e-05,
+      "loss": 5.2917,
+      "step": 34
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 4.21875,
+      "learning_rate": 6.800000000000001e-05,
+      "loss": 5.2339,
+      "step": 35
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 2.859375,
+      "learning_rate": 7e-05,
+      "loss": 5.2663,
+      "step": 36
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 2.125,
+      "learning_rate": 7.2e-05,
+      "loss": 5.0012,
+      "step": 37
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 2.84375,
+      "learning_rate": 7.4e-05,
+      "loss": 4.8893,
+      "step": 38
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 4.09375,
+      "learning_rate": 7.6e-05,
+      "loss": 4.9856,
+      "step": 39
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.046875,
+      "learning_rate": 7.800000000000001e-05,
+      "loss": 5.0183,
+      "step": 40
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 8.875,
+      "learning_rate": 8e-05,
+      "loss": 4.9491,
+      "step": 41
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 4.65625,
+      "learning_rate": 8.2e-05,
+      "loss": 4.7829,
+      "step": 42
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 4.40625,
+      "learning_rate": 8.4e-05,
+      "loss": 4.9717,
+      "step": 43
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.6e-05,
+      "loss": 4.817,
+      "step": 44
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 2.1875,
+      "learning_rate": 8.800000000000001e-05,
+      "loss": 4.8171,
+      "step": 45
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 5.1875,
+      "learning_rate": 9e-05,
+      "loss": 4.8199,
+      "step": 46
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 3.40625,
+      "learning_rate": 9.200000000000001e-05,
+      "loss": 4.7965,
+      "step": 47
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 3.609375,
+      "learning_rate": 9.4e-05,
+      "loss": 4.649,
+      "step": 48
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 3.59375,
+      "learning_rate": 9.6e-05,
+      "loss": 4.6714,
+      "step": 49
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.25,
+      "learning_rate": 9.8e-05,
+      "loss": 4.5177,
+      "step": 50
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 3.03125,
+      "learning_rate": 0.0001,
+      "loss": 4.5112,
+      "step": 51
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.00010200000000000001,
+      "loss": 4.5739,
+      "step": 52
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 2.1875,
+      "learning_rate": 0.00010400000000000001,
+      "loss": 4.4601,
+      "step": 53
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 2.46875,
+      "learning_rate": 0.00010600000000000002,
+      "loss": 4.5269,
+      "step": 54
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 1.8828125,
+      "learning_rate": 0.00010800000000000001,
+      "loss": 4.2753,
+      "step": 55
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.00011000000000000002,
+      "loss": 4.3292,
+      "step": 56
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 2.390625,
+      "learning_rate": 0.00011200000000000001,
+      "loss": 4.3489,
+      "step": 57
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 4.125,
+      "learning_rate": 0.00011399999999999999,
+      "loss": 4.4857,
+      "step": 58
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 2.96875,
+      "learning_rate": 0.000116,
+      "loss": 4.3047,
+      "step": 59
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.000118,
+      "loss": 4.4213,
+      "step": 60
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 11.4375,
+      "learning_rate": 0.00012,
+      "loss": 4.3308,
+      "step": 61
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 2.59375,
+      "learning_rate": 0.000122,
+      "loss": 4.3524,
+      "step": 62
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 3.375,
+      "learning_rate": 0.000124,
+      "loss": 4.2236,
+      "step": 63
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 2.234375,
+      "learning_rate": 0.000126,
+      "loss": 4.3674,
+      "step": 64
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 3.0,
+      "learning_rate": 0.00012800000000000002,
+      "loss": 4.1765,
+      "step": 65
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 2.234375,
+      "learning_rate": 0.00013000000000000002,
+      "loss": 4.1043,
+      "step": 66
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 3.8125,
+      "learning_rate": 0.000132,
+      "loss": 4.0828,
+      "step": 67
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 2.75,
+      "learning_rate": 0.000134,
+      "loss": 4.1654,
+      "step": 68
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00013600000000000003,
+      "loss": 4.0984,
+      "step": 69
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.000138,
+      "loss": 4.0978,
+      "step": 70
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 3.84375,
+      "learning_rate": 0.00014,
+      "loss": 4.1044,
+      "step": 71
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 2.5625,
+      "learning_rate": 0.000142,
+      "loss": 3.967,
+      "step": 72
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 2.0625,
+      "learning_rate": 0.000144,
+      "loss": 4.0976,
+      "step": 73
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 2.328125,
+      "learning_rate": 0.000146,
+      "loss": 3.8734,
+      "step": 74
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 2.015625,
+      "learning_rate": 0.000148,
+      "loss": 3.9381,
+      "step": 75
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 3.97,
+      "step": 76
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 2.046875,
+      "learning_rate": 0.000152,
+      "loss": 3.9806,
+      "step": 77
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 3.25,
+      "learning_rate": 0.000154,
+      "loss": 3.7804,
+      "step": 78
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 2.578125,
+      "learning_rate": 0.00015600000000000002,
+      "loss": 3.9961,
+      "step": 79
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.9765625,
+      "learning_rate": 0.00015800000000000002,
+      "loss": 3.9899,
+      "step": 80
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.00016,
+      "loss": 3.8394,
+      "step": 81
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.000162,
+      "loss": 3.8614,
+      "step": 82
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.000164,
+      "loss": 3.7687,
+      "step": 83
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.000166,
+      "loss": 3.8413,
+      "step": 84
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 2.375,
+      "learning_rate": 0.000168,
+      "loss": 3.7533,
+      "step": 85
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 1.890625,
+      "learning_rate": 0.00017,
+      "loss": 3.8058,
+      "step": 86
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.000172,
+      "loss": 3.8243,
+      "step": 87
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 2.25,
+      "learning_rate": 0.000174,
+      "loss": 3.7795,
+      "step": 88
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 1.8359375,
+      "learning_rate": 0.00017600000000000002,
+      "loss": 3.7421,
+      "step": 89
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.7421875,
+      "learning_rate": 0.00017800000000000002,
+      "loss": 3.7222,
+      "step": 90
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 1.5078125,
+      "learning_rate": 0.00018,
+      "loss": 3.727,
+      "step": 91
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 2.328125,
+      "learning_rate": 0.000182,
+      "loss": 3.8749,
+      "step": 92
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.00018400000000000003,
+      "loss": 3.6911,
+      "step": 93
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.00018600000000000002,
+      "loss": 3.6221,
+      "step": 94
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 2.265625,
+      "learning_rate": 0.000188,
+      "loss": 3.6225,
+      "step": 95
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.00019,
+      "loss": 3.5456,
+      "step": 96
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 3.0625,
+      "learning_rate": 0.000192,
+      "loss": 3.5316,
+      "step": 97
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 1.75,
+      "learning_rate": 0.000194,
+      "loss": 3.5942,
+      "step": 98
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 1.5078125,
+      "learning_rate": 0.000196,
+      "loss": 3.5513,
+      "step": 99
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.8125,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 3.673,
+      "step": 100
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 2.328125,
+      "learning_rate": 0.0002,
+      "loss": 3.6088,
+      "step": 101
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.00019999939076577905,
+      "loss": 3.5971,
+      "step": 102
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.00019999756307053948,
+      "loss": 3.5318,
+      "step": 103
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 2.046875,
+      "learning_rate": 0.00019999451693655123,
+      "loss": 3.5199,
+      "step": 104
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.00019999025240093044,
+      "loss": 3.5049,
+      "step": 105
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 2.234375,
+      "learning_rate": 0.00019998476951563915,
+      "loss": 3.5481,
+      "step": 106
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 1.9921875,
+      "learning_rate": 0.00019997806834748456,
+      "loss": 3.4391,
+      "step": 107
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.00019997014897811833,
+      "loss": 3.5096,
+      "step": 108
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.00019996101150403543,
+      "loss": 3.424,
+      "step": 109
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.00019995065603657316,
+      "loss": 3.4862,
+      "step": 110
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 1.5859375,
+      "learning_rate": 0.0001999390827019096,
+      "loss": 3.5117,
+      "step": 111
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.0001999262916410621,
+      "loss": 3.452,
+      "step": 112
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.00019991228300988585,
+      "loss": 3.4834,
+      "step": 113
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 1.8984375,
+      "learning_rate": 0.00019989705697907149,
+      "loss": 3.4673,
+      "step": 114
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 1.625,
+      "learning_rate": 0.0001998806137341434,
+      "loss": 3.2031,
+      "step": 115
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.0001998629534754574,
+      "loss": 3.3329,
+      "step": 116
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.00019984407641819812,
+      "loss": 3.4215,
+      "step": 117
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 1.65625,
+      "learning_rate": 0.00019982398279237655,
+      "loss": 3.3886,
+      "step": 118
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.00019980267284282717,
+      "loss": 3.2904,
+      "step": 119
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.000199780146829205,
+      "loss": 3.3135,
+      "step": 120
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.00019975640502598244,
+      "loss": 3.4523,
+      "step": 121
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 1.796875,
+      "learning_rate": 0.00019973144772244582,
+      "loss": 3.248,
+      "step": 122
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00019970527522269205,
+      "loss": 3.3419,
+      "step": 123
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.00019967788784562473,
+      "loss": 3.296,
+      "step": 124
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.00019964928592495045,
+      "loss": 3.2151,
+      "step": 125
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 1.3828125,
+      "learning_rate": 0.00019961946980917456,
+      "loss": 3.3786,
+      "step": 126
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 1.859375,
+      "learning_rate": 0.00019958843986159704,
+      "loss": 3.4188,
+      "step": 127
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.00019955619646030802,
+      "loss": 3.4489,
+      "step": 128
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.0001995227399981831,
+      "loss": 3.3746,
+      "step": 129
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.921875,
+      "learning_rate": 0.00019948807088287883,
+      "loss": 3.29,
+      "step": 130
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.00019945218953682734,
+      "loss": 3.2717,
+      "step": 131
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00019941509639723155,
+      "loss": 3.2601,
+      "step": 132
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.00019937679191605963,
+      "loss": 3.3615,
+      "step": 133
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 2.390625,
+      "learning_rate": 0.00019933727656003963,
+      "loss": 3.4326,
+      "step": 134
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 1.9296875,
+      "learning_rate": 0.0001992965508106537,
+      "loss": 3.3733,
+      "step": 135
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 1.6953125,
+      "learning_rate": 0.00019925461516413223,
+      "loss": 3.3488,
+      "step": 136
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.0001992114701314478,
+      "loss": 3.2991,
+      "step": 137
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 1.375,
+      "learning_rate": 0.00019916711623830903,
+      "loss": 3.2037,
+      "step": 138
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 1.5234375,
+      "learning_rate": 0.00019912155402515417,
+      "loss": 3.2748,
+      "step": 139
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.00019907478404714436,
+      "loss": 3.2753,
+      "step": 140
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 2.0,
+      "learning_rate": 0.00019902680687415705,
+      "loss": 3.3406,
+      "step": 141
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 1.4296875,
+      "learning_rate": 0.0001989776230907789,
+      "loss": 3.3422,
+      "step": 142
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 1.703125,
+      "learning_rate": 0.00019892723329629887,
+      "loss": 3.4146,
+      "step": 143
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.0001988756381047006,
+      "loss": 3.3321,
+      "step": 144
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.0001988228381446553,
+      "loss": 3.3225,
+      "step": 145
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.00019876883405951377,
+      "loss": 3.2079,
+      "step": 146
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.0001987136265072988,
+      "loss": 3.1757,
+      "step": 147
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.00019865721616069696,
+      "loss": 3.1882,
+      "step": 148
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.0001985996037070505,
+      "loss": 3.2314,
+      "step": 149
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00019854078984834903,
+      "loss": 3.1942,
+      "step": 150
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.00019848077530122083,
+      "loss": 3.2477,
+      "step": 151
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.0001984195607969242,
+      "loss": 3.1375,
+      "step": 152
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.00019835714708133862,
+      "loss": 3.145,
+      "step": 153
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 1.2265625,
+      "learning_rate": 0.00019829353491495545,
+      "loss": 3.1747,
+      "step": 154
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.0001982287250728689,
+      "loss": 3.2027,
+      "step": 155
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.00019816271834476642,
+      "loss": 3.09,
+      "step": 156
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 2.578125,
+      "learning_rate": 0.00019809551553491916,
+      "loss": 3.2477,
+      "step": 157
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 1.7421875,
+      "learning_rate": 0.00019802711746217218,
+      "loss": 3.1619,
+      "step": 158
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.0001979575249599344,
+      "loss": 3.2682,
+      "step": 159
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.0001978867388761685,
+      "loss": 3.0885,
+      "step": 160
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.00019781476007338058,
+      "loss": 3.2491,
+      "step": 161
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 1.5,
+      "learning_rate": 0.0001977415894286096,
+      "loss": 3.21,
+      "step": 162
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.0001976672278334168,
+      "loss": 3.1559,
+      "step": 163
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.00019759167619387476,
+      "loss": 3.0837,
+      "step": 164
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.00019751493543055632,
+      "loss": 3.1893,
+      "step": 165
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.00019743700647852354,
+      "loss": 3.1023,
+      "step": 166
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00019735789028731604,
+      "loss": 3.0936,
+      "step": 167
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.00019727758782093967,
+      "loss": 3.2097,
+      "step": 168
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 1.6171875,
+      "learning_rate": 0.00019719610005785465,
+      "loss": 3.1198,
+      "step": 169
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.00019711342799096361,
+      "loss": 3.1823,
+      "step": 170
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00019702957262759965,
+      "loss": 3.1136,
+      "step": 171
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.0001969445349895139,
+      "loss": 3.1076,
+      "step": 172
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 1.625,
+      "learning_rate": 0.0001968583161128631,
+      "loss": 3.113,
+      "step": 173
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 1.6640625,
+      "learning_rate": 0.00019677091704819715,
+      "loss": 3.1752,
+      "step": 174
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 1.5390625,
+      "learning_rate": 0.00019668233886044597,
+      "loss": 3.1309,
+      "step": 175
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 1.34375,
+      "learning_rate": 0.00019659258262890683,
+      "loss": 3.1505,
+      "step": 176
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.00019650164944723115,
+      "loss": 3.1151,
+      "step": 177
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.00019640954042341103,
+      "loss": 3.085,
+      "step": 178
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.00019631625667976583,
+      "loss": 3.0857,
+      "step": 179
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00019622179935292855,
+      "loss": 3.2549,
+      "step": 180
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0001961261695938319,
+      "loss": 3.2138,
+      "step": 181
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0001960293685676943,
+      "loss": 3.0507,
+      "step": 182
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.00019593139745400576,
+      "loss": 3.2578,
+      "step": 183
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00019583225744651333,
+      "loss": 3.0951,
+      "step": 184
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 1.6875,
+      "learning_rate": 0.00019573194975320673,
+      "loss": 3.1273,
+      "step": 185
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.00019563047559630357,
+      "loss": 3.0946,
+      "step": 186
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00019552783621223436,
+      "loss": 3.1003,
+      "step": 187
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.0001954240328516277,
+      "loss": 3.2026,
+      "step": 188
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.0001953190667792947,
+      "loss": 3.1465,
+      "step": 189
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5390625,
+      "learning_rate": 0.00019521293927421388,
+      "loss": 3.1967,
+      "step": 190
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00019510565162951537,
+      "loss": 3.1578,
+      "step": 191
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.00019499720515246525,
+      "loss": 3.1388,
+      "step": 192
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.00019488760116444966,
+      "loss": 3.1873,
+      "step": 193
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 1.5703125,
+      "learning_rate": 0.0001947768410009586,
+      "loss": 3.1005,
+      "step": 194
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.00019466492601156966,
+      "loss": 3.0942,
+      "step": 195
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 1.25,
+      "learning_rate": 0.0001945518575599317,
+      "loss": 3.0119,
+      "step": 196
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00019443763702374812,
+      "loss": 3.0152,
+      "step": 197
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.0001943222657947601,
+      "loss": 2.9887,
+      "step": 198
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.00019420574527872968,
+      "loss": 3.0333,
+      "step": 199
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00019408807689542257,
+      "loss": 3.1974,
+      "step": 200
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.00019396926207859084,
+      "loss": 3.1598,
+      "step": 201
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0001938493022759556,
+      "loss": 3.1167,
+      "step": 202
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.00019372819894918915,
+      "loss": 3.0386,
+      "step": 203
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 1.890625,
+      "learning_rate": 0.00019360595357389735,
+      "loss": 3.0719,
+      "step": 204
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.00019348256763960145,
+      "loss": 3.1225,
+      "step": 205
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 1.578125,
+      "learning_rate": 0.00019335804264972018,
+      "loss": 3.1224,
+      "step": 206
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00019323238012155123,
+      "loss": 3.0044,
+      "step": 207
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 1.25,
+      "learning_rate": 0.00019310558158625285,
+      "loss": 3.0157,
+      "step": 208
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 1.375,
+      "learning_rate": 0.00019297764858882514,
+      "loss": 2.9538,
+      "step": 209
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.375,
+      "learning_rate": 0.00019284858268809137,
+      "loss": 3.0122,
+      "step": 210
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.00019271838545667876,
+      "loss": 3.0153,
+      "step": 211
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.0001925870584809995,
+      "loss": 3.0157,
+      "step": 212
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.00019245460336123134,
+      "loss": 3.1123,
+      "step": 213
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00019232102171129811,
+      "loss": 3.0119,
+      "step": 214
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.00019218631515885006,
+      "loss": 3.0785,
+      "step": 215
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00019205048534524406,
+      "loss": 2.9962,
+      "step": 216
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.00019191353392552344,
+      "loss": 2.98,
+      "step": 217
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.00019177546256839812,
+      "loss": 3.1094,
+      "step": 218
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 1.84375,
+      "learning_rate": 0.00019163627295622397,
+      "loss": 3.0632,
+      "step": 219
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.0001914959667849825,
+      "loss": 3.0286,
+      "step": 220
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.0001913545457642601,
+      "loss": 3.2119,
+      "step": 221
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.0001912120116172273,
+      "loss": 3.1813,
+      "step": 222
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.00019106836608061772,
+      "loss": 3.144,
+      "step": 223
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.00019092361090470688,
+      "loss": 3.012,
+      "step": 224
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00019077774785329087,
+      "loss": 3.0367,
+      "step": 225
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.000190630778703665,
+      "loss": 3.0175,
+      "step": 226
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.00019048270524660196,
+      "loss": 2.9683,
+      "step": 227
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.0001903335292863301,
+      "loss": 3.0164,
+      "step": 228
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.0001901832526405114,
+      "loss": 2.9548,
+      "step": 229
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.6875,
+      "learning_rate": 0.00019003187714021938,
+      "loss": 2.4954,
+      "step": 230
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 5.59375,
+      "learning_rate": 0.0001898794046299167,
+      "loss": 2.6911,
+      "step": 231
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 132.0,
+      "learning_rate": 0.00018972583696743285,
+      "loss": 3.6355,
+      "step": 232
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 2.875,
+      "learning_rate": 0.0001895711760239413,
+      "loss": 3.0587,
+      "step": 233
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.0001894154236839368,
+      "loss": 3.0121,
+      "step": 234
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 52.0,
+      "learning_rate": 0.00018925858184521256,
+      "loss": 3.0583,
+      "step": 235
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 2.484375,
+      "learning_rate": 0.0001891006524188368,
+      "loss": 3.1517,
+      "step": 236
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.00018894163732912977,
+      "loss": 3.1375,
+      "step": 237
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.00018878153851364013,
+      "loss": 3.1155,
+      "step": 238
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.00018862035792312147,
+      "loss": 3.0236,
+      "step": 239
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.0001884580975215084,
+      "loss": 3.2067,
+      "step": 240
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00018829475928589271,
+      "loss": 3.1324,
+      "step": 241
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.0001881303452064992,
+      "loss": 3.0433,
+      "step": 242
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.00018796485728666165,
+      "loss": 2.9619,
+      "step": 243
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.00018779829754279805,
+      "loss": 3.052,
+      "step": 244
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00018763066800438636,
+      "loss": 3.0409,
+      "step": 245
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.00018746197071393958,
+      "loss": 2.9298,
+      "step": 246
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.00018729220772698097,
+      "loss": 3.0326,
+      "step": 247
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.00018712138111201895,
+      "loss": 3.0219,
+      "step": 248
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0001869494929505219,
+      "loss": 3.0315,
+      "step": 249
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 11.5,
+      "learning_rate": 0.00018677654533689287,
+      "loss": 2.9556,
+      "step": 250
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 3.0629920959472656,
+      "eval_runtime": 31.7573,
+      "eval_samples_per_second": 15.744,
+      "eval_steps_per_second": 2.645,
+      "step": 250
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 1.953125,
+      "learning_rate": 0.00018660254037844388,
+      "loss": 3.0847,
+      "step": 251
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.0001864274801953705,
+      "loss": 3.0586,
+      "step": 252
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 1.5390625,
+      "learning_rate": 0.00018625136692072575,
+      "loss": 3.1137,
+      "step": 253
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.0001860742027003944,
+      "loss": 3.021,
+      "step": 254
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.00018589598969306645,
+      "loss": 3.0077,
+      "step": 255
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.00018571673007021123,
+      "loss": 3.0533,
+      "step": 256
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.00018553642601605068,
+      "loss": 3.038,
+      "step": 257
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 2.125,
+      "learning_rate": 0.00018535507972753274,
+      "loss": 2.9965,
+      "step": 258
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 1.34375,
+      "learning_rate": 0.00018517269341430476,
+      "loss": 3.0632,
+      "step": 259
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.00018498926929868642,
+      "loss": 3.1137,
+      "step": 260
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 23.75,
+      "learning_rate": 0.0001848048096156426,
+      "loss": 2.9885,
+      "step": 261
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 3.75,
+      "learning_rate": 0.00018461931661275643,
+      "loss": 3.12,
+      "step": 262
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 2.0625,
+      "learning_rate": 0.00018443279255020152,
+      "loss": 3.1726,
+      "step": 263
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 1.921875,
+      "learning_rate": 0.00018424523970071477,
+      "loss": 3.2129,
+      "step": 264
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 1.71875,
+      "learning_rate": 0.00018405666034956844,
+      "loss": 2.9784,
+      "step": 265
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 71.0,
+      "learning_rate": 0.00018386705679454242,
+      "loss": 3.0557,
+      "step": 266
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.00018367643134589617,
+      "loss": 3.1084,
+      "step": 267
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 1.8984375,
+      "learning_rate": 0.00018348478632634066,
+      "loss": 3.0229,
+      "step": 268
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.00018329212407100994,
+      "loss": 2.9938,
+      "step": 269
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.00018309844692743283,
+      "loss": 3.029,
+      "step": 270
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 2.015625,
+      "learning_rate": 0.00018290375725550417,
+      "loss": 2.9886,
+      "step": 271
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 2.015625,
+      "learning_rate": 0.00018270805742745617,
+      "loss": 3.0823,
+      "step": 272
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 3.28125,
+      "learning_rate": 0.00018251134982782952,
+      "loss": 3.0021,
+      "step": 273
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 2.203125,
+      "learning_rate": 0.0001823136368534442,
+      "loss": 2.9724,
+      "step": 274
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 1.390625,
+      "learning_rate": 0.00018211492091337042,
+      "loss": 2.8975,
+      "step": 275
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.0001819152044288992,
+      "loss": 2.9522,
+      "step": 276
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.00018171448983351284,
+      "loss": 2.9637,
+      "step": 277
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 1.984375,
+      "learning_rate": 0.00018151277957285543,
+      "loss": 3.0239,
+      "step": 278
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.00018131007610470276,
+      "loss": 2.9014,
+      "step": 279
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.8828125,
+      "learning_rate": 0.00018110638189893267,
+      "loss": 3.0008,
+      "step": 280
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 1.4765625,
+      "learning_rate": 0.00018090169943749476,
+      "loss": 3.1132,
+      "step": 281
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 5.09375,
+      "learning_rate": 0.00018069603121438022,
+      "loss": 3.0337,
+      "step": 282
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 2.765625,
+      "learning_rate": 0.0001804893797355914,
+      "loss": 3.1402,
+      "step": 283
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.00018028174751911146,
+      "loss": 2.9044,
+      "step": 284
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.00018007313709487334,
+      "loss": 3.0565,
+      "step": 285
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.00017986355100472928,
+      "loss": 3.1032,
+      "step": 286
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.00017965299180241963,
+      "loss": 2.9643,
+      "step": 287
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.00017944146205354182,
+      "loss": 3.0511,
+      "step": 288
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.00017922896433551907,
+      "loss": 2.8223,
+      "step": 289
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.00017901550123756906,
+      "loss": 2.9322,
+      "step": 290
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 13.0,
+      "learning_rate": 0.00017880107536067218,
+      "loss": 3.054,
+      "step": 291
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.0001785856893175402,
+      "loss": 3.0824,
+      "step": 292
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.000178369345732584,
+      "loss": 3.0968,
+      "step": 293
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 9.25,
+      "learning_rate": 0.00017815204724188187,
+      "loss": 3.138,
+      "step": 294
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 1.703125,
+      "learning_rate": 0.00017793379649314744,
+      "loss": 3.0392,
+      "step": 295
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 20.625,
+      "learning_rate": 0.0001777145961456971,
+      "loss": 2.9701,
+      "step": 296
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 12.0,
+      "learning_rate": 0.00017749444887041799,
+      "loss": 3.0403,
+      "step": 297
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 2.125,
+      "learning_rate": 0.00017727335734973512,
+      "loss": 2.8949,
+      "step": 298
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 25.875,
+      "learning_rate": 0.00017705132427757895,
+      "loss": 3.0195,
+      "step": 299
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 18.25,
+      "learning_rate": 0.00017682835235935236,
+      "loss": 2.9629,
+      "step": 300
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 3.0,
+      "learning_rate": 0.0001766044443118978,
+      "loss": 3.0801,
+      "step": 301
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.00017637960286346425,
+      "loss": 2.9368,
+      "step": 302
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 2.203125,
+      "learning_rate": 0.0001761538307536737,
+      "loss": 2.9964,
+      "step": 303
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 0.984375,
+      "learning_rate": 0.00017592713073348807,
+      "loss": 2.9558,
+      "step": 304
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.00017569950556517566,
+      "loss": 2.9006,
+      "step": 305
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00017547095802227723,
+      "loss": 3.0037,
+      "step": 306
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 1.0,
+      "learning_rate": 0.00017524149088957245,
+      "loss": 2.8517,
+      "step": 307
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00017501110696304596,
+      "loss": 2.9163,
+      "step": 308
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.0001747798090498532,
+      "loss": 2.9563,
+      "step": 309
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.00017454759996828623,
+      "loss": 2.8861,
+      "step": 310
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00017431448254773944,
+      "loss": 3.0871,
+      "step": 311
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.000174080459628675,
+      "loss": 2.8852,
+      "step": 312
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 26.0,
+      "learning_rate": 0.00017384553406258842,
+      "loss": 3.0938,
+      "step": 313
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.00017360970871197346,
+      "loss": 3.0905,
+      "step": 314
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.00017337298645028764,
+      "loss": 3.0793,
+      "step": 315
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 1.25,
+      "learning_rate": 0.00017313537016191706,
+      "loss": 3.0879,
+      "step": 316
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.00017289686274214118,
+      "loss": 3.1008,
+      "step": 317
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.0001726574670970976,
+      "loss": 3.0737,
+      "step": 318
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.00017241718614374678,
+      "loss": 2.9513,
+      "step": 319
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.00017217602280983623,
+      "loss": 3.1167,
+      "step": 320
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 1.0,
+      "learning_rate": 0.0001719339800338651,
+      "loss": 3.0823,
+      "step": 321
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0001716910607650483,
+      "loss": 2.9707,
+      "step": 322
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.00017144726796328034,
+      "loss": 3.0313,
+      "step": 323
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 0.93359375,
+      "learning_rate": 0.00017120260459909967,
+      "loss": 2.9786,
+      "step": 324
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0001709570736536521,
+      "loss": 2.9752,
+      "step": 325
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00017071067811865476,
+      "loss": 3.0147,
+      "step": 326
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00017046342099635948,
+      "loss": 3.0414,
+      "step": 327
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.00017021530529951625,
+      "loss": 3.0018,
+      "step": 328
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.00016996633405133655,
+      "loss": 3.0249,
+      "step": 329
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00016971651028545648,
+      "loss": 2.9958,
+      "step": 330
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.00016946583704589973,
+      "loss": 2.9505,
+      "step": 331
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0001692143173870407,
+      "loss": 2.9806,
+      "step": 332
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 248.0,
+      "learning_rate": 0.000168961954373567,
+      "loss": 2.9853,
+      "step": 333
+    },
+    {
+      "epoch": 0.334,
+      "grad_norm": 1.75,
+      "learning_rate": 0.0001687087510804423,
+      "loss": 2.9122,
+      "step": 334
+    },
+    {
+      "epoch": 0.335,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00016845471059286887,
+      "loss": 2.9686,
+      "step": 335
+    },
+    {
+      "epoch": 0.336,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.00016819983600624986,
+      "loss": 3.0128,
+      "step": 336
+    },
+    {
+      "epoch": 0.337,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.00016794413042615168,
+      "loss": 2.9692,
+      "step": 337
+    },
+    {
+      "epoch": 0.338,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00016768759696826608,
+      "loss": 2.9639,
+      "step": 338
+    },
+    {
+      "epoch": 0.339,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.00016743023875837233,
+      "loss": 2.8997,
+      "step": 339
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.00016717205893229903,
+      "loss": 2.9825,
+      "step": 340
+    },
+    {
+      "epoch": 0.341,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00016691306063588583,
+      "loss": 2.9442,
+      "step": 341
+    },
+    {
+      "epoch": 0.342,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00016665324702494524,
+      "loss": 2.9611,
+      "step": 342
+    },
+    {
+      "epoch": 0.343,
+      "grad_norm": 1.0,
+      "learning_rate": 0.00016639262126522418,
+      "loss": 3.0123,
+      "step": 343
+    },
+    {
+      "epoch": 0.344,
+      "grad_norm": 1.6953125,
+      "learning_rate": 0.00016613118653236518,
+      "loss": 2.9498,
+      "step": 344
+    },
+    {
+      "epoch": 0.345,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00016586894601186805,
+      "loss": 2.8588,
+      "step": 345
+    },
+    {
+      "epoch": 0.346,
+      "grad_norm": 0.9609375,
+      "learning_rate": 0.00016560590289905073,
+      "loss": 3.0459,
+      "step": 346
+    },
+    {
+      "epoch": 0.347,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00016534206039901057,
+      "loss": 2.8838,
+      "step": 347
+    },
+    {
+      "epoch": 0.348,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.0001650774217265851,
+      "loss": 2.9694,
+      "step": 348
+    },
+    {
+      "epoch": 0.349,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0001648119901063131,
+      "loss": 2.9102,
+      "step": 349
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00016454576877239507,
+      "loss": 2.8904,
+      "step": 350
+    },
+    {
+      "epoch": 0.351,
+      "grad_norm": 0.85546875,
+      "learning_rate": 0.00016427876096865394,
+      "loss": 2.8798,
+      "step": 351
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 0.93359375,
+      "learning_rate": 0.00016401096994849557,
+      "loss": 2.9685,
+      "step": 352
+    },
+    {
+      "epoch": 0.353,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.000163742398974869,
+      "loss": 2.9498,
+      "step": 353
+    },
+    {
+      "epoch": 0.354,
+      "grad_norm": 0.93359375,
+      "learning_rate": 0.00016347305132022677,
+      "loss": 2.9719,
+      "step": 354
+    },
+    {
+      "epoch": 0.355,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.0001632029302664851,
+      "loss": 2.9964,
+      "step": 355
+    },
+    {
+      "epoch": 0.356,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00016293203910498376,
+      "loss": 2.9672,
+      "step": 356
+    },
+    {
+      "epoch": 0.357,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00016266038113644607,
+      "loss": 2.944,
+      "step": 357
+    },
+    {
+      "epoch": 0.358,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.00016238795967093864,
+      "loss": 2.928,
+      "step": 358
+    },
+    {
+      "epoch": 0.359,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00016211477802783103,
+      "loss": 2.8984,
+      "step": 359
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.6640625,
+      "learning_rate": 0.0001618408395357554,
+      "loss": 2.807,
+      "step": 360
+    },
+    {
+      "epoch": 0.361,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.0001615661475325658,
+      "loss": 2.9199,
+      "step": 361
+    },
+    {
+      "epoch": 0.362,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00016129070536529766,
+      "loss": 2.9442,
+      "step": 362
+    },
+    {
+      "epoch": 0.363,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.0001610145163901268,
+      "loss": 2.8548,
+      "step": 363
+    },
+    {
+      "epoch": 0.364,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.00016073758397232868,
+      "loss": 2.9272,
+      "step": 364
+    },
+    {
+      "epoch": 0.365,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.0001604599114862375,
+      "loss": 2.9774,
+      "step": 365
+    },
+    {
+      "epoch": 0.366,
+      "grad_norm": 0.9609375,
+      "learning_rate": 0.00016018150231520486,
+      "loss": 2.8988,
+      "step": 366
+    },
+    {
+      "epoch": 0.367,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0001599023598515586,
+      "loss": 2.8573,
+      "step": 367
+    },
+    {
+      "epoch": 0.368,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.0001596224874965616,
+      "loss": 2.86,
+      "step": 368
+    },
+    {
+      "epoch": 0.369,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00015934188866037016,
+      "loss": 2.9432,
+      "step": 369
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00015906056676199255,
+      "loss": 2.8969,
+      "step": 370
+    },
+    {
+      "epoch": 0.371,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00015877852522924732,
+      "loss": 2.968,
+      "step": 371
+    },
+    {
+      "epoch": 0.372,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.00015849576749872157,
+      "loss": 2.885,
+      "step": 372
+    },
+    {
+      "epoch": 0.373,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.00015821229701572896,
+      "loss": 2.9352,
+      "step": 373
+    },
+    {
+      "epoch": 0.374,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.0001579281172342679,
+      "loss": 2.7847,
+      "step": 374
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00015764323161697935,
+      "loss": 2.9347,
+      "step": 375
+    },
+    {
+      "epoch": 0.376,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.0001573576436351046,
+      "loss": 2.8265,
+      "step": 376
+    },
+    {
+      "epoch": 0.377,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0001570713567684432,
+      "loss": 2.8173,
+      "step": 377
+    },
+    {
+      "epoch": 0.378,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.00015678437450531013,
+      "loss": 2.8425,
+      "step": 378
+    },
+    {
+      "epoch": 0.379,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.0001564967003424938,
+      "loss": 2.7723,
+      "step": 379
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.00015620833778521307,
+      "loss": 2.9031,
+      "step": 380
+    },
+    {
+      "epoch": 0.381,
+      "grad_norm": 0.85546875,
+      "learning_rate": 0.0001559192903470747,
+      "loss": 2.9154,
+      "step": 381
+    },
+    {
+      "epoch": 0.382,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.0001556295615500305,
+      "loss": 2.8226,
+      "step": 382
+    },
+    {
+      "epoch": 0.383,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00015533915492433443,
+      "loss": 2.8911,
+      "step": 383
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.00015504807400849958,
+      "loss": 2.8114,
+      "step": 384
+    },
+    {
+      "epoch": 0.385,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00015475632234925504,
+      "loss": 2.8762,
+      "step": 385
+    },
+    {
+      "epoch": 0.386,
+      "grad_norm": 0.828125,
+      "learning_rate": 0.00015446390350150273,
+      "loss": 2.8918,
+      "step": 386
+    },
+    {
+      "epoch": 0.387,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.000154170821028274,
+      "loss": 2.9433,
+      "step": 387
+    },
+    {
+      "epoch": 0.388,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.0001538770785006863,
+      "loss": 2.871,
+      "step": 388
+    },
+    {
+      "epoch": 0.389,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.00015358267949789966,
+      "loss": 2.8932,
+      "step": 389
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.000153287627607073,
+      "loss": 2.8251,
+      "step": 390
+    },
+    {
+      "epoch": 0.391,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0001529919264233205,
+      "loss": 2.9062,
+      "step": 391
+    },
+    {
+      "epoch": 0.392,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00015269557954966778,
+      "loss": 2.8995,
+      "step": 392
+    },
+    {
+      "epoch": 0.393,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.00015239859059700794,
+      "loss": 2.7726,
+      "step": 393
+    },
+    {
+      "epoch": 0.394,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00015210096318405767,
+      "loss": 2.7649,
+      "step": 394
+    },
+    {
+      "epoch": 0.395,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00015180270093731303,
+      "loss": 2.778,
+      "step": 395
+    },
+    {
+      "epoch": 0.396,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.00015150380749100545,
+      "loss": 2.8322,
+      "step": 396
+    },
+    {
+      "epoch": 0.397,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00015120428648705717,
+      "loss": 2.8831,
+      "step": 397
+    },
+    {
+      "epoch": 0.398,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.00015090414157503714,
+      "loss": 2.8862,
+      "step": 398
+    },
+    {
+      "epoch": 0.399,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00015060337641211637,
+      "loss": 2.8219,
+      "step": 399
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00015030199466302353,
+      "loss": 2.8894,
+      "step": 400
+    },
+    {
+      "epoch": 0.401,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 2.8594,
+      "step": 401
+    },
+    {
+      "epoch": 0.402,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00014969739610275556,
+      "loss": 2.7943,
+      "step": 402
+    },
+    {
+      "epoch": 0.403,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.0001493941866584231,
+      "loss": 2.9105,
+      "step": 403
+    },
+    {
+      "epoch": 0.404,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.00014909037536151409,
+      "loss": 2.9341,
+      "step": 404
+    },
+    {
+      "epoch": 0.405,
+      "grad_norm": 0.984375,
+      "learning_rate": 0.0001487859659138733,
+      "loss": 2.8726,
+      "step": 405
+    },
+    {
+      "epoch": 0.406,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.00014848096202463372,
+      "loss": 2.9266,
+      "step": 406
+    },
+    {
+      "epoch": 0.407,
+      "grad_norm": 12.5,
+      "learning_rate": 0.00014817536741017152,
+      "loss": 2.8098,
+      "step": 407
+    },
+    {
+      "epoch": 0.408,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0001478691857940607,
+      "loss": 2.8836,
+      "step": 408
+    },
+    {
+      "epoch": 0.409,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00014756242090702756,
+      "loss": 2.8176,
+      "step": 409
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.00014725507648690543,
+      "loss": 2.8858,
+      "step": 410
+    },
+    {
+      "epoch": 0.411,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00014694715627858908,
+      "loss": 2.8898,
+      "step": 411
+    },
+    {
+      "epoch": 0.412,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.00014663866403398913,
+      "loss": 2.8529,
+      "step": 412
+    },
+    {
+      "epoch": 0.413,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00014632960351198618,
+      "loss": 2.8151,
+      "step": 413
+    },
+    {
+      "epoch": 0.414,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00014601997847838518,
+      "loss": 2.9201,
+      "step": 414
+    },
+    {
+      "epoch": 0.415,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00014570979270586945,
+      "loss": 2.7901,
+      "step": 415
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.00014539904997395468,
+      "loss": 2.8675,
+      "step": 416
+    },
+    {
+      "epoch": 0.417,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.00014508775406894307,
+      "loss": 2.7743,
+      "step": 417
+    },
+    {
+      "epoch": 0.418,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.00014477590878387696,
+      "loss": 2.8796,
+      "step": 418
+    },
+    {
+      "epoch": 0.419,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.00014446351791849276,
+      "loss": 2.9855,
+      "step": 419
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00014415058527917452,
+      "loss": 2.8645,
+      "step": 420
+    },
+    {
+      "epoch": 0.421,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00014383711467890774,
+      "loss": 2.7821,
+      "step": 421
+    },
+    {
+      "epoch": 0.422,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.00014352310993723277,
+      "loss": 2.9062,
+      "step": 422
+    },
+    {
+      "epoch": 0.423,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.00014320857488019824,
+      "loss": 2.8564,
+      "step": 423
+    },
+    {
+      "epoch": 0.424,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.0001428935133403146,
+      "loss": 2.855,
+      "step": 424
+    },
+    {
+      "epoch": 0.425,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00014257792915650728,
+      "loss": 2.907,
+      "step": 425
+    },
+    {
+      "epoch": 0.426,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.00014226182617406996,
+      "loss": 2.8238,
+      "step": 426
+    },
+    {
+      "epoch": 0.427,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.00014194520824461771,
+      "loss": 2.8523,
+      "step": 427
+    },
+    {
+      "epoch": 0.428,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.00014162807922604012,
+      "loss": 2.91,
+      "step": 428
+    },
+    {
+      "epoch": 0.429,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.0001413104429824542,
+      "loss": 2.9161,
+      "step": 429
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00014099230338415728,
+      "loss": 2.9304,
+      "step": 430
+    },
+    {
+      "epoch": 0.431,
+      "grad_norm": 0.75,
+      "learning_rate": 0.00014067366430758004,
+      "loss": 2.8266,
+      "step": 431
+    },
+    {
+      "epoch": 0.432,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00014035452963523902,
+      "loss": 2.8303,
+      "step": 432
+    },
+    {
+      "epoch": 0.433,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00014003490325568954,
+      "loss": 2.8311,
+      "step": 433
+    },
+    {
+      "epoch": 0.434,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.00013971478906347806,
+      "loss": 2.826,
+      "step": 434
+    },
+    {
+      "epoch": 0.435,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.00013939419095909512,
+      "loss": 2.6267,
+      "step": 435
+    },
+    {
+      "epoch": 0.436,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.00013907311284892736,
+      "loss": 2.7203,
+      "step": 436
+    },
+    {
+      "epoch": 0.437,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.0001387515586452103,
+      "loss": 2.8801,
+      "step": 437
+    },
+    {
+      "epoch": 0.438,
+      "grad_norm": 2.421875,
+      "learning_rate": 0.00013842953226598037,
+      "loss": 2.9568,
+      "step": 438
+    },
+    {
+      "epoch": 0.439,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.00013810703763502744,
+      "loss": 2.758,
+      "step": 439
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.00013778407868184672,
+      "loss": 2.8374,
+      "step": 440
+    },
+    {
+      "epoch": 0.441,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00013746065934159123,
+      "loss": 2.7894,
+      "step": 441
+    },
+    {
+      "epoch": 0.442,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00013713678355502351,
+      "loss": 2.855,
+      "step": 442
+    },
+    {
+      "epoch": 0.443,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.00013681245526846783,
+      "loss": 2.777,
+      "step": 443
+    },
+    {
+      "epoch": 0.444,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00013648767843376196,
+      "loss": 2.8408,
+      "step": 444
+    },
+    {
+      "epoch": 0.445,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.00013616245700820922,
+      "loss": 2.8602,
+      "step": 445
+    },
+    {
+      "epoch": 0.446,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.00013583679495453,
+      "loss": 2.9452,
+      "step": 446
+    },
+    {
+      "epoch": 0.447,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.0001355106962408137,
+      "loss": 2.799,
+      "step": 447
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.00013518416484047018,
+      "loss": 2.8164,
+      "step": 448
+    },
+    {
+      "epoch": 0.449,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00013485720473218154,
+      "loss": 2.9159,
+      "step": 449
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.00013452981989985348,
+      "loss": 2.8874,
+      "step": 450
+    },
+    {
+      "epoch": 0.451,
+      "grad_norm": 0.984375,
+      "learning_rate": 0.00013420201433256689,
+      "loss": 2.8846,
+      "step": 451
+    },
+    {
+      "epoch": 0.452,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00013387379202452917,
+      "loss": 2.8474,
+      "step": 452
+    },
+    {
+      "epoch": 0.453,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.00013354515697502553,
+      "loss": 2.8697,
+      "step": 453
+    },
+    {
+      "epoch": 0.454,
+      "grad_norm": 2.875,
+      "learning_rate": 0.00013321611318837032,
+      "loss": 2.8103,
+      "step": 454
+    },
+    {
+      "epoch": 0.455,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.00013288666467385833,
+      "loss": 2.7893,
+      "step": 455
+    },
+    {
+      "epoch": 0.456,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.00013255681544571568,
+      "loss": 2.8302,
+      "step": 456
+    },
+    {
+      "epoch": 0.457,
+      "grad_norm": 0.76953125,
+      "learning_rate": 0.00013222656952305113,
+      "loss": 2.8056,
+      "step": 457
+    },
+    {
+      "epoch": 0.458,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.00013189593092980702,
+      "loss": 2.8463,
+      "step": 458
+    },
+    {
+      "epoch": 0.459,
+      "grad_norm": 0.75,
+      "learning_rate": 0.00013156490369471027,
+      "loss": 2.7264,
+      "step": 459
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.00013123349185122327,
+      "loss": 2.7938,
+      "step": 460
+    },
+    {
+      "epoch": 0.461,
+      "grad_norm": 0.76953125,
+      "learning_rate": 0.00013090169943749476,
+      "loss": 2.7834,
+      "step": 461
+    },
+    {
+      "epoch": 0.462,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.00013056953049631057,
+      "loss": 2.8079,
+      "step": 462
+    },
+    {
+      "epoch": 0.463,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.00013023698907504446,
+      "loss": 2.7598,
+      "step": 463
+    },
+    {
+      "epoch": 0.464,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.00012990407922560868,
+      "loss": 2.8624,
+      "step": 464
+    },
+    {
+      "epoch": 0.465,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00012957080500440468,
+      "loss": 2.8337,
+      "step": 465
+    },
+    {
+      "epoch": 0.466,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.00012923717047227368,
+      "loss": 2.8641,
+      "step": 466
+    },
+    {
+      "epoch": 0.467,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00012890317969444716,
+      "loss": 2.832,
+      "step": 467
+    },
+    {
+      "epoch": 0.468,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.00012856883674049736,
+      "loss": 2.9126,
+      "step": 468
+    },
+    {
+      "epoch": 0.469,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.00012823414568428768,
+      "loss": 2.9386,
+      "step": 469
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.00012789911060392294,
+      "loss": 2.7886,
+      "step": 470
+    },
+    {
+      "epoch": 0.471,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.0001275637355816999,
+      "loss": 2.9174,
+      "step": 471
+    },
+    {
+      "epoch": 0.472,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00012722802470405744,
+      "loss": 2.7964,
+      "step": 472
+    },
+    {
+      "epoch": 0.473,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00012689198206152657,
+      "loss": 2.779,
+      "step": 473
+    },
+    {
+      "epoch": 0.474,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00012655561174868088,
+      "loss": 2.8909,
+      "step": 474
+    },
+    {
+      "epoch": 0.475,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00012621891786408648,
+      "loss": 2.8705,
+      "step": 475
+    },
+    {
+      "epoch": 0.476,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00012588190451025207,
+      "loss": 2.9862,
+      "step": 476
+    },
+    {
+      "epoch": 0.477,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.00012554457579357905,
+      "loss": 2.8599,
+      "step": 477
+    },
+    {
+      "epoch": 0.478,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.0001252069358243114,
+      "loss": 2.9299,
+      "step": 478
+    },
+    {
+      "epoch": 0.479,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.0001248689887164855,
+      "loss": 2.8572,
+      "step": 479
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00012453073858788026,
+      "loss": 2.8889,
+      "step": 480
+    },
+    {
+      "epoch": 0.481,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.00012419218955996676,
+      "loss": 2.8363,
+      "step": 481
+    },
+    {
+      "epoch": 0.482,
+      "grad_norm": 0.7578125,
+      "learning_rate": 0.0001238533457578581,
+      "loss": 2.8624,
+      "step": 482
+    },
+    {
+      "epoch": 0.483,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.000123514211310259,
+      "loss": 2.8517,
+      "step": 483
+    },
+    {
+      "epoch": 0.484,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00012317479034941573,
+      "loss": 2.8247,
+      "step": 484
+    },
+    {
+      "epoch": 0.485,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.00012283508701106557,
+      "loss": 2.8114,
+      "step": 485
+    },
+    {
+      "epoch": 0.486,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0001224951054343865,
+      "loss": 2.7308,
+      "step": 486
+    },
+    {
+      "epoch": 0.487,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.00012215484976194676,
+      "loss": 2.8445,
+      "step": 487
+    },
+    {
+      "epoch": 0.488,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.00012181432413965428,
+      "loss": 2.933,
+      "step": 488
+    },
+    {
+      "epoch": 0.489,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.00012147353271670634,
+      "loss": 2.8725,
+      "step": 489
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.00012113247964553888,
+      "loss": 2.7981,
+      "step": 490
+    },
+    {
+      "epoch": 0.491,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.00012079116908177593,
+      "loss": 2.8803,
+      "step": 491
+    },
+    {
+      "epoch": 0.492,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.00012044960518417903,
+      "loss": 2.7637,
+      "step": 492
+    },
+    {
+      "epoch": 0.493,
+      "grad_norm": 0.75,
+      "learning_rate": 0.00012010779211459648,
+      "loss": 2.888,
+      "step": 493
+    },
+    {
+      "epoch": 0.494,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.00011976573403791262,
+      "loss": 2.815,
+      "step": 494
+    },
+    {
+      "epoch": 0.495,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.0001194234351219972,
+      "loss": 2.7684,
+      "step": 495
+    },
+    {
+      "epoch": 0.496,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00011908089953765449,
+      "loss": 2.7292,
+      "step": 496
+    },
+    {
+      "epoch": 0.497,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.00011873813145857249,
+      "loss": 2.7949,
+      "step": 497
+    },
+    {
+      "epoch": 0.498,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00011839513506127203,
+      "loss": 2.7681,
+      "step": 498
+    },
+    {
+      "epoch": 0.499,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.00011805191452505602,
+      "loss": 2.6686,
+      "step": 499
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.00011770847403195834,
+      "loss": 2.7635,
+      "step": 500
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 2.7894091606140137,
+      "eval_runtime": 31.8024,
+      "eval_samples_per_second": 15.722,
+      "eval_steps_per_second": 2.641,
+      "step": 500
+    },
+    {
+      "epoch": 0.501,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00011736481776669306,
+      "loss": 2.7591,
+      "step": 501
+    },
+    {
+      "epoch": 0.502,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.00011702094991660326,
+      "loss": 2.8107,
+      "step": 502
+    },
+    {
+      "epoch": 0.503,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.00011667687467161024,
+      "loss": 2.8147,
+      "step": 503
+    },
+    {
+      "epoch": 0.504,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.00011633259622416224,
+      "loss": 2.8342,
+      "step": 504
+    },
+    {
+      "epoch": 0.505,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.0001159881187691835,
+      "loss": 2.8037,
+      "step": 505
+    },
+    {
+      "epoch": 0.506,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.0001156434465040231,
+      "loss": 2.8258,
+      "step": 506
+    },
+    {
+      "epoch": 0.507,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00011529858362840382,
+      "loss": 2.8476,
+      "step": 507
+    },
+    {
+      "epoch": 0.508,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.00011495353434437098,
+      "loss": 2.7726,
+      "step": 508
+    },
+    {
+      "epoch": 0.509,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00011460830285624118,
+      "loss": 2.7936,
+      "step": 509
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.00011426289337055119,
+      "loss": 2.7585,
+      "step": 510
+    },
+    {
+      "epoch": 0.511,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.00011391731009600654,
+      "loss": 2.7722,
+      "step": 511
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.00011357155724343045,
+      "loss": 2.8518,
+      "step": 512
+    },
+    {
+      "epoch": 0.513,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00011322563902571226,
+      "loss": 2.8035,
+      "step": 513
+    },
+    {
+      "epoch": 0.514,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.0001128795596577563,
+      "loss": 2.8544,
+      "step": 514
+    },
+    {
+      "epoch": 0.515,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.00011253332335643043,
+      "loss": 2.8212,
+      "step": 515
+    },
+    {
+      "epoch": 0.516,
+      "grad_norm": 2.21875,
+      "learning_rate": 0.00011218693434051475,
+      "loss": 2.7369,
+      "step": 516
+    },
+    {
+      "epoch": 0.517,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00011184039683065013,
+      "loss": 2.7366,
+      "step": 517
+    },
+    {
+      "epoch": 0.518,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.00011149371504928668,
+      "loss": 2.8155,
+      "step": 518
+    },
+    {
+      "epoch": 0.519,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.00011114689322063255,
+      "loss": 2.7464,
+      "step": 519
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.0001107999355706023,
+      "loss": 2.796,
+      "step": 520
+    },
+    {
+      "epoch": 0.521,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.00011045284632676536,
+      "loss": 2.7958,
+      "step": 521
+    },
+    {
+      "epoch": 0.522,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.00011010562971829463,
+      "loss": 2.7453,
+      "step": 522
+    },
+    {
+      "epoch": 0.523,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00010975828997591495,
+      "loss": 2.7748,
+      "step": 523
+    },
+    {
+      "epoch": 0.524,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.00010941083133185146,
+      "loss": 2.8185,
+      "step": 524
+    },
+    {
+      "epoch": 0.525,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00010906325801977804,
+      "loss": 2.7946,
+      "step": 525
+    },
+    {
+      "epoch": 0.526,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.00010871557427476583,
+      "loss": 2.8366,
+      "step": 526
+    },
+    {
+      "epoch": 0.527,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00010836778433323158,
+      "loss": 2.81,
+      "step": 527
+    },
+    {
+      "epoch": 0.528,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.00010801989243288589,
+      "loss": 2.7869,
+      "step": 528
+    },
+    {
+      "epoch": 0.529,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00010767190281268187,
+      "loss": 2.7461,
+      "step": 529
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.00010732381971276318,
+      "loss": 2.7465,
+      "step": 530
+    },
+    {
+      "epoch": 0.531,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00010697564737441252,
+      "loss": 2.7434,
+      "step": 531
+    },
+    {
+      "epoch": 0.532,
+      "grad_norm": 6.3125,
+      "learning_rate": 0.00010662739004000005,
+      "loss": 2.7999,
+      "step": 532
+    },
+    {
+      "epoch": 0.533,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00010627905195293135,
+      "loss": 2.7405,
+      "step": 533
+    },
+    {
+      "epoch": 0.534,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.00010593063735759618,
+      "loss": 2.8709,
+      "step": 534
+    },
+    {
+      "epoch": 0.535,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.00010558215049931638,
+      "loss": 2.9645,
+      "step": 535
+    },
+    {
+      "epoch": 0.536,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.0001052335956242944,
+      "loss": 2.7778,
+      "step": 536
+    },
+    {
+      "epoch": 0.537,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.00010488497697956135,
+      "loss": 2.7532,
+      "step": 537
+    },
+    {
+      "epoch": 0.538,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00010453629881292538,
+      "loss": 2.8057,
+      "step": 538
+    },
+    {
+      "epoch": 0.539,
+      "grad_norm": 0.70703125,
+      "learning_rate": 0.00010418756537291996,
+      "loss": 2.75,
+      "step": 539
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.00010383878090875201,
+      "loss": 2.8593,
+      "step": 540
+    },
+    {
+      "epoch": 0.541,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.00010348994967025012,
+      "loss": 2.7264,
+      "step": 541
+    },
+    {
+      "epoch": 0.542,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.00010314107590781284,
+      "loss": 2.8229,
+      "step": 542
+    },
+    {
+      "epoch": 0.543,
+      "grad_norm": 0.68359375,
+      "learning_rate": 0.0001027921638723569,
+      "loss": 2.7506,
+      "step": 543
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.00010244321781526533,
+      "loss": 2.7689,
+      "step": 544
+    },
+    {
+      "epoch": 0.545,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.0001020942419883357,
+      "loss": 2.766,
+      "step": 545
+    },
+    {
+      "epoch": 0.546,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.00010174524064372837,
+      "loss": 2.8915,
+      "step": 546
+    },
+    {
+      "epoch": 0.547,
+      "grad_norm": 2.625,
+      "learning_rate": 0.00010139621803391455,
+      "loss": 2.7575,
+      "step": 547
+    },
+    {
+      "epoch": 0.548,
+      "grad_norm": 0.7578125,
+      "learning_rate": 0.00010104717841162458,
+      "loss": 2.7613,
+      "step": 548
+    },
+    {
+      "epoch": 0.549,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.00010069812602979615,
+      "loss": 2.8354,
+      "step": 549
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.00010034906514152238,
+      "loss": 2.815,
+      "step": 550
+    },
+    {
+      "epoch": 0.551,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0001,
+      "loss": 2.9268,
+      "step": 551
+    },
+    {
+      "epoch": 0.552,
+      "grad_norm": 0.7109375,
+      "learning_rate": 9.965093485847767e-05,
+      "loss": 2.7682,
+      "step": 552
+    },
+    {
+      "epoch": 0.553,
+      "grad_norm": 0.6484375,
+      "learning_rate": 9.930187397020386e-05,
+      "loss": 2.8294,
+      "step": 553
+    },
+    {
+      "epoch": 0.554,
+      "grad_norm": 0.734375,
+      "learning_rate": 9.895282158837545e-05,
+      "loss": 2.7828,
+      "step": 554
+    },
+    {
+      "epoch": 0.555,
+      "grad_norm": 0.6953125,
+      "learning_rate": 9.860378196608549e-05,
+      "loss": 2.7539,
+      "step": 555
+    },
+    {
+      "epoch": 0.556,
+      "grad_norm": 0.6953125,
+      "learning_rate": 9.825475935627165e-05,
+      "loss": 2.7614,
+      "step": 556
+    },
+    {
+      "epoch": 0.557,
+      "grad_norm": 0.73828125,
+      "learning_rate": 9.790575801166432e-05,
+      "loss": 2.7728,
+      "step": 557
+    },
+    {
+      "epoch": 0.558,
+      "grad_norm": 0.7265625,
+      "learning_rate": 9.755678218473469e-05,
+      "loss": 2.7955,
+      "step": 558
+    },
+    {
+      "epoch": 0.559,
+      "grad_norm": 0.6796875,
+      "learning_rate": 9.720783612764314e-05,
+      "loss": 2.8389,
+      "step": 559
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.70703125,
+      "learning_rate": 9.685892409218717e-05,
+      "loss": 2.7221,
+      "step": 560
+    },
+    {
+      "epoch": 0.561,
+      "grad_norm": 0.69140625,
+      "learning_rate": 9.651005032974994e-05,
+      "loss": 2.7392,
+      "step": 561
+    },
+    {
+      "epoch": 0.562,
+      "grad_norm": 0.65625,
+      "learning_rate": 9.616121909124801e-05,
+      "loss": 2.7522,
+      "step": 562
+    },
+    {
+      "epoch": 0.563,
+      "grad_norm": 4.125,
+      "learning_rate": 9.581243462708006e-05,
+      "loss": 2.7441,
+      "step": 563
+    },
+    {
+      "epoch": 0.564,
+      "grad_norm": 0.81640625,
+      "learning_rate": 9.546370118707463e-05,
+      "loss": 2.7218,
+      "step": 564
+    },
+    {
+      "epoch": 0.565,
+      "grad_norm": 0.73046875,
+      "learning_rate": 9.511502302043868e-05,
+      "loss": 2.7805,
+      "step": 565
+    },
+    {
+      "epoch": 0.566,
+      "grad_norm": 0.67578125,
+      "learning_rate": 9.476640437570562e-05,
+      "loss": 2.7791,
+      "step": 566
+    },
+    {
+      "epoch": 0.567,
+      "grad_norm": 0.69921875,
+      "learning_rate": 9.441784950068362e-05,
+      "loss": 2.8229,
+      "step": 567
+    },
+    {
+      "epoch": 0.568,
+      "grad_norm": 0.71875,
+      "learning_rate": 9.406936264240386e-05,
+      "loss": 2.8902,
+      "step": 568
+    },
+    {
+      "epoch": 0.569,
+      "grad_norm": 0.71484375,
+      "learning_rate": 9.372094804706867e-05,
+      "loss": 2.8009,
+      "step": 569
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.62890625,
+      "learning_rate": 9.337260996000002e-05,
+      "loss": 2.7867,
+      "step": 570
+    },
+    {
+      "epoch": 0.571,
+      "grad_norm": 0.71875,
+      "learning_rate": 9.302435262558747e-05,
+      "loss": 2.7513,
+      "step": 571
+    },
+    {
+      "epoch": 0.572,
+      "grad_norm": 0.6875,
+      "learning_rate": 9.267618028723686e-05,
+      "loss": 2.8066,
+      "step": 572
+    },
+    {
+      "epoch": 0.573,
+      "grad_norm": 0.6484375,
+      "learning_rate": 9.232809718731814e-05,
+      "loss": 2.818,
+      "step": 573
+    },
+    {
+      "epoch": 0.574,
+      "grad_norm": 0.69921875,
+      "learning_rate": 9.198010756711412e-05,
+      "loss": 2.6698,
+      "step": 574
+    },
+    {
+      "epoch": 0.575,
+      "grad_norm": 0.7109375,
+      "learning_rate": 9.163221566676847e-05,
+      "loss": 2.7901,
+      "step": 575
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 2.0625,
+      "learning_rate": 9.128442572523417e-05,
+      "loss": 2.8648,
+      "step": 576
+    },
+    {
+      "epoch": 0.577,
+      "grad_norm": 0.78515625,
+      "learning_rate": 9.093674198022201e-05,
+      "loss": 2.8698,
+      "step": 577
+    },
+    {
+      "epoch": 0.578,
+      "grad_norm": 0.66796875,
+      "learning_rate": 9.058916866814858e-05,
+      "loss": 2.8594,
+      "step": 578
+    },
+    {
+      "epoch": 0.579,
+      "grad_norm": 3.65625,
+      "learning_rate": 9.024171002408506e-05,
+      "loss": 2.788,
+      "step": 579
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.8125,
+      "learning_rate": 8.989437028170537e-05,
+      "loss": 2.7126,
+      "step": 580
+    },
+    {
+      "epoch": 0.581,
+      "grad_norm": 0.734375,
+      "learning_rate": 8.954715367323468e-05,
+      "loss": 2.7795,
+      "step": 581
+    },
+    {
+      "epoch": 0.582,
+      "grad_norm": 0.72265625,
+      "learning_rate": 8.920006442939772e-05,
+      "loss": 2.7243,
+      "step": 582
+    },
+    {
+      "epoch": 0.583,
+      "grad_norm": 0.828125,
+      "learning_rate": 8.885310677936746e-05,
+      "loss": 2.8055,
+      "step": 583
+    },
+    {
+      "epoch": 0.584,
+      "grad_norm": 0.66796875,
+      "learning_rate": 8.850628495071336e-05,
+      "loss": 2.8449,
+      "step": 584
+    },
+    {
+      "epoch": 0.585,
+      "grad_norm": 0.83203125,
+      "learning_rate": 8.81596031693499e-05,
+      "loss": 2.7153,
+      "step": 585
+    },
+    {
+      "epoch": 0.586,
+      "grad_norm": 0.7890625,
+      "learning_rate": 8.781306565948528e-05,
+      "loss": 2.8445,
+      "step": 586
+    },
+    {
+      "epoch": 0.587,
+      "grad_norm": 0.60546875,
+      "learning_rate": 8.746667664356956e-05,
+      "loss": 2.7783,
+      "step": 587
+    },
+    {
+      "epoch": 0.588,
+      "grad_norm": 0.6953125,
+      "learning_rate": 8.712044034224374e-05,
+      "loss": 2.8369,
+      "step": 588
+    },
+    {
+      "epoch": 0.589,
+      "grad_norm": 0.92578125,
+      "learning_rate": 8.677436097428775e-05,
+      "loss": 2.7912,
+      "step": 589
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.6484375,
+      "learning_rate": 8.642844275656957e-05,
+      "loss": 2.8142,
+      "step": 590
+    },
+    {
+      "epoch": 0.591,
+      "grad_norm": 0.72265625,
+      "learning_rate": 8.608268990399349e-05,
+      "loss": 2.6924,
+      "step": 591
+    },
+    {
+      "epoch": 0.592,
+      "grad_norm": 0.734375,
+      "learning_rate": 8.573710662944885e-05,
+      "loss": 2.7975,
+      "step": 592
+    },
+    {
+      "epoch": 0.593,
+      "grad_norm": 0.69140625,
+      "learning_rate": 8.539169714375885e-05,
+      "loss": 2.8273,
+      "step": 593
+    },
+    {
+      "epoch": 0.594,
+      "grad_norm": 3.609375,
+      "learning_rate": 8.504646565562906e-05,
+      "loss": 2.7106,
+      "step": 594
+    },
+    {
+      "epoch": 0.595,
+      "grad_norm": 0.8359375,
+      "learning_rate": 8.47014163715962e-05,
+      "loss": 2.773,
+      "step": 595
+    },
+    {
+      "epoch": 0.596,
+      "grad_norm": 0.8203125,
+      "learning_rate": 8.435655349597689e-05,
+      "loss": 2.759,
+      "step": 596
+    },
+    {
+      "epoch": 0.597,
+      "grad_norm": 0.7265625,
+      "learning_rate": 8.401188123081653e-05,
+      "loss": 2.7223,
+      "step": 597
+    },
+    {
+      "epoch": 0.598,
+      "grad_norm": 0.73828125,
+      "learning_rate": 8.366740377583781e-05,
+      "loss": 2.7163,
+      "step": 598
+    },
+    {
+      "epoch": 0.599,
+      "grad_norm": 0.76953125,
+      "learning_rate": 8.332312532838978e-05,
+      "loss": 2.7544,
+      "step": 599
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.71484375,
+      "learning_rate": 8.297905008339677e-05,
+      "loss": 2.7424,
+      "step": 600
+    },
+    {
+      "epoch": 0.601,
+      "grad_norm": 0.70703125,
+      "learning_rate": 8.263518223330697e-05,
+      "loss": 2.8959,
+      "step": 601
+    },
+    {
+      "epoch": 0.602,
+      "grad_norm": 0.6640625,
+      "learning_rate": 8.229152596804168e-05,
+      "loss": 2.665,
+      "step": 602
+    },
+    {
+      "epoch": 0.603,
+      "grad_norm": 0.6640625,
+      "learning_rate": 8.194808547494401e-05,
+      "loss": 2.8718,
+      "step": 603
+    },
+    {
+      "epoch": 0.604,
+      "grad_norm": 0.66796875,
+      "learning_rate": 8.160486493872798e-05,
+      "loss": 2.7585,
+      "step": 604
+    },
+    {
+      "epoch": 0.605,
+      "grad_norm": 0.65625,
+      "learning_rate": 8.126186854142752e-05,
+      "loss": 2.7911,
+      "step": 605
+    },
+    {
+      "epoch": 0.606,
+      "grad_norm": 0.92578125,
+      "learning_rate": 8.091910046234552e-05,
+      "loss": 2.6548,
+      "step": 606
+    },
+    {
+      "epoch": 0.607,
+      "grad_norm": 0.625,
+      "learning_rate": 8.057656487800282e-05,
+      "loss": 2.7227,
+      "step": 607
+    },
+    {
+      "epoch": 0.608,
+      "grad_norm": 0.66796875,
+      "learning_rate": 8.023426596208739e-05,
+      "loss": 2.7485,
+      "step": 608
+    },
+    {
+      "epoch": 0.609,
+      "grad_norm": 0.65625,
+      "learning_rate": 7.989220788540355e-05,
+      "loss": 2.7348,
+      "step": 609
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.546875,
+      "learning_rate": 7.955039481582097e-05,
+      "loss": 2.7158,
+      "step": 610
+    },
+    {
+      "epoch": 0.611,
+      "grad_norm": 0.69921875,
+      "learning_rate": 7.920883091822408e-05,
+      "loss": 2.6516,
+      "step": 611
+    },
+    {
+      "epoch": 0.612,
+      "grad_norm": 0.6484375,
+      "learning_rate": 7.886752035446114e-05,
+      "loss": 2.8051,
+      "step": 612
+    },
+    {
+      "epoch": 0.613,
+      "grad_norm": 0.59765625,
+      "learning_rate": 7.852646728329368e-05,
+      "loss": 2.7428,
+      "step": 613
+    },
+    {
+      "epoch": 0.614,
+      "grad_norm": 0.70703125,
+      "learning_rate": 7.818567586034577e-05,
+      "loss": 2.8212,
+      "step": 614
+    },
+    {
+      "epoch": 0.615,
+      "grad_norm": 0.66015625,
+      "learning_rate": 7.784515023805328e-05,
+      "loss": 2.7348,
+      "step": 615
+    },
+    {
+      "epoch": 0.616,
+      "grad_norm": 0.69921875,
+      "learning_rate": 7.750489456561352e-05,
+      "loss": 2.7394,
+      "step": 616
+    },
+    {
+      "epoch": 0.617,
+      "grad_norm": 0.67578125,
+      "learning_rate": 7.716491298893442e-05,
+      "loss": 2.6866,
+      "step": 617
+    },
+    {
+      "epoch": 0.618,
+      "grad_norm": 0.67578125,
+      "learning_rate": 7.682520965058428e-05,
+      "loss": 2.6922,
+      "step": 618
+    },
+    {
+      "epoch": 0.619,
+      "grad_norm": 0.64453125,
+      "learning_rate": 7.6485788689741e-05,
+      "loss": 2.6569,
+      "step": 619
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.640625,
+      "learning_rate": 7.614665424214193e-05,
+      "loss": 2.752,
+      "step": 620
+    },
+    {
+      "epoch": 0.621,
+      "grad_norm": 0.69921875,
+      "learning_rate": 7.580781044003324e-05,
+      "loss": 2.7281,
+      "step": 621
+    },
+    {
+      "epoch": 0.622,
+      "grad_norm": 1.171875,
+      "learning_rate": 7.546926141211974e-05,
+      "loss": 2.8616,
+      "step": 622
+    },
+    {
+      "epoch": 0.623,
+      "grad_norm": 0.6796875,
+      "learning_rate": 7.513101128351454e-05,
+      "loss": 2.8221,
+      "step": 623
+    },
+    {
+      "epoch": 0.624,
+      "grad_norm": 0.640625,
+      "learning_rate": 7.479306417568864e-05,
+      "loss": 2.8145,
+      "step": 624
+    },
+    {
+      "epoch": 0.625,
+      "grad_norm": 1.8359375,
+      "learning_rate": 7.445542420642097e-05,
+      "loss": 2.6816,
+      "step": 625
+    },
+    {
+      "epoch": 0.626,
+      "grad_norm": 0.640625,
+      "learning_rate": 7.411809548974792e-05,
+      "loss": 2.6947,
+      "step": 626
+    },
+    {
+      "epoch": 0.627,
+      "grad_norm": 0.66796875,
+      "learning_rate": 7.378108213591355e-05,
+      "loss": 2.7485,
+      "step": 627
+    },
+    {
+      "epoch": 0.628,
+      "grad_norm": 0.66015625,
+      "learning_rate": 7.344438825131911e-05,
+      "loss": 2.7634,
+      "step": 628
+    },
+    {
+      "epoch": 0.629,
+      "grad_norm": 0.60546875,
+      "learning_rate": 7.310801793847344e-05,
+      "loss": 2.7374,
+      "step": 629
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.65625,
+      "learning_rate": 7.277197529594257e-05,
+      "loss": 2.6431,
+      "step": 630
+    },
+    {
+      "epoch": 0.631,
+      "grad_norm": 0.66796875,
+      "learning_rate": 7.243626441830009e-05,
+      "loss": 2.6069,
+      "step": 631
+    },
+    {
+      "epoch": 0.632,
+      "grad_norm": 0.6328125,
+      "learning_rate": 7.210088939607708e-05,
+      "loss": 2.7996,
+      "step": 632
+    },
+    {
+      "epoch": 0.633,
+      "grad_norm": 1.046875,
+      "learning_rate": 7.176585431571235e-05,
+      "loss": 2.7437,
+      "step": 633
+    },
+    {
+      "epoch": 0.634,
+      "grad_norm": 0.62890625,
+      "learning_rate": 7.143116325950265e-05,
+      "loss": 2.7624,
+      "step": 634
+    },
+    {
+      "epoch": 0.635,
+      "grad_norm": 0.59765625,
+      "learning_rate": 7.109682030555283e-05,
+      "loss": 2.715,
+      "step": 635
+    },
+    {
+      "epoch": 0.636,
+      "grad_norm": 0.70703125,
+      "learning_rate": 7.076282952772633e-05,
+      "loss": 2.7208,
+      "step": 636
+    },
+    {
+      "epoch": 0.637,
+      "grad_norm": 0.6328125,
+      "learning_rate": 7.042919499559537e-05,
+      "loss": 2.7576,
+      "step": 637
+    },
+    {
+      "epoch": 0.638,
+      "grad_norm": 0.58203125,
+      "learning_rate": 7.009592077439134e-05,
+      "loss": 2.8255,
+      "step": 638
+    },
+    {
+      "epoch": 0.639,
+      "grad_norm": 0.640625,
+      "learning_rate": 6.976301092495556e-05,
+      "loss": 2.7823,
+      "step": 639
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.64453125,
+      "learning_rate": 6.943046950368944e-05,
+      "loss": 2.6574,
+      "step": 640
+    },
+    {
+      "epoch": 0.641,
+      "grad_norm": 1.2578125,
+      "learning_rate": 6.909830056250527e-05,
+      "loss": 2.6497,
+      "step": 641
+    },
+    {
+      "epoch": 0.642,
+      "grad_norm": 0.6171875,
+      "learning_rate": 6.876650814877674e-05,
+      "loss": 2.7926,
+      "step": 642
+    },
+    {
+      "epoch": 0.643,
+      "grad_norm": 0.63671875,
+      "learning_rate": 6.843509630528977e-05,
+      "loss": 2.6997,
+      "step": 643
+    },
+    {
+      "epoch": 0.644,
+      "grad_norm": 0.61328125,
+      "learning_rate": 6.8104069070193e-05,
+      "loss": 2.7251,
+      "step": 644
+    },
+    {
+      "epoch": 0.645,
+      "grad_norm": 0.62109375,
+      "learning_rate": 6.77734304769489e-05,
+      "loss": 2.6186,
+      "step": 645
+    },
+    {
+      "epoch": 0.646,
+      "grad_norm": 0.6171875,
+      "learning_rate": 6.744318455428436e-05,
+      "loss": 2.7101,
+      "step": 646
+    },
+    {
+      "epoch": 0.647,
+      "grad_norm": 0.66015625,
+      "learning_rate": 6.711333532614168e-05,
+      "loss": 2.6802,
+      "step": 647
+    },
+    {
+      "epoch": 0.648,
+      "grad_norm": 0.671875,
+      "learning_rate": 6.67838868116297e-05,
+      "loss": 2.7727,
+      "step": 648
+    },
+    {
+      "epoch": 0.649,
+      "grad_norm": 0.640625,
+      "learning_rate": 6.64548430249745e-05,
+      "loss": 2.8216,
+      "step": 649
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.6640625,
+      "learning_rate": 6.612620797547087e-05,
+      "loss": 2.8565,
+      "step": 650
+    },
+    {
+      "epoch": 0.651,
+      "grad_norm": 0.671875,
+      "learning_rate": 6.579798566743314e-05,
+      "loss": 2.795,
+      "step": 651
+    },
+    {
+      "epoch": 0.652,
+      "grad_norm": 0.66015625,
+      "learning_rate": 6.547018010014654e-05,
+      "loss": 2.7028,
+      "step": 652
+    },
+    {
+      "epoch": 0.653,
+      "grad_norm": 0.625,
+      "learning_rate": 6.51427952678185e-05,
+      "loss": 2.7782,
+      "step": 653
+    },
+    {
+      "epoch": 0.654,
+      "grad_norm": 0.60546875,
+      "learning_rate": 6.481583515952983e-05,
+      "loss": 2.7242,
+      "step": 654
+    },
+    {
+      "epoch": 0.655,
+      "grad_norm": 0.6640625,
+      "learning_rate": 6.448930375918631e-05,
+      "loss": 2.7598,
+      "step": 655
+    },
+    {
+      "epoch": 0.656,
+      "grad_norm": 0.64453125,
+      "learning_rate": 6.416320504546997e-05,
+      "loss": 2.6908,
+      "step": 656
+    },
+    {
+      "epoch": 0.657,
+      "grad_norm": 1.265625,
+      "learning_rate": 6.383754299179079e-05,
+      "loss": 2.7308,
+      "step": 657
+    },
+    {
+      "epoch": 0.658,
+      "grad_norm": 0.66796875,
+      "learning_rate": 6.351232156623803e-05,
+      "loss": 2.7679,
+      "step": 658
+    },
+    {
+      "epoch": 0.659,
+      "grad_norm": 0.66015625,
+      "learning_rate": 6.318754473153221e-05,
+      "loss": 2.8747,
+      "step": 659
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.62890625,
+      "learning_rate": 6.286321644497655e-05,
+      "loss": 2.7716,
+      "step": 660
+    },
+    {
+      "epoch": 0.661,
+      "grad_norm": 0.62890625,
+      "learning_rate": 6.25393406584088e-05,
+      "loss": 2.6582,
+      "step": 661
+    },
+    {
+      "epoch": 0.662,
+      "grad_norm": 0.60546875,
+      "learning_rate": 6.22159213181533e-05,
+      "loss": 2.7907,
+      "step": 662
+    },
+    {
+      "epoch": 0.663,
+      "grad_norm": 0.67578125,
+      "learning_rate": 6.18929623649726e-05,
+      "loss": 2.6852,
+      "step": 663
+    },
+    {
+      "epoch": 0.664,
+      "grad_norm": 0.61328125,
+      "learning_rate": 6.157046773401964e-05,
+      "loss": 2.7763,
+      "step": 664
+    },
+    {
+      "epoch": 0.665,
+      "grad_norm": 0.640625,
+      "learning_rate": 6.12484413547897e-05,
+      "loss": 2.713,
+      "step": 665
+    },
+    {
+      "epoch": 0.666,
+      "grad_norm": 0.75390625,
+      "learning_rate": 6.092688715107264e-05,
+      "loss": 2.8589,
+      "step": 666
+    },
+    {
+      "epoch": 0.667,
+      "grad_norm": 0.67578125,
+      "learning_rate": 6.0605809040904894e-05,
+      "loss": 2.813,
+      "step": 667
+    },
+    {
+      "epoch": 0.668,
+      "grad_norm": 0.68359375,
+      "learning_rate": 6.0285210936521955e-05,
+      "loss": 2.7389,
+      "step": 668
+    },
+    {
+      "epoch": 0.669,
+      "grad_norm": 0.625,
+      "learning_rate": 5.9965096744310526e-05,
+      "loss": 2.7049,
+      "step": 669
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.6875,
+      "learning_rate": 5.964547036476099e-05,
+      "loss": 2.7361,
+      "step": 670
+    },
+    {
+      "epoch": 0.671,
+      "grad_norm": 0.61328125,
+      "learning_rate": 5.9326335692419995e-05,
+      "loss": 2.7189,
+      "step": 671
+    },
+    {
+      "epoch": 0.672,
+      "grad_norm": 1.5078125,
+      "learning_rate": 5.900769661584272e-05,
+      "loss": 2.7206,
+      "step": 672
+    },
+    {
+      "epoch": 0.673,
+      "grad_norm": 0.69921875,
+      "learning_rate": 5.868955701754584e-05,
+      "loss": 2.7581,
+      "step": 673
+    },
+    {
+      "epoch": 0.674,
+      "grad_norm": 0.63671875,
+      "learning_rate": 5.83719207739599e-05,
+      "loss": 2.5903,
+      "step": 674
+    },
+    {
+      "epoch": 0.675,
+      "grad_norm": 0.65234375,
+      "learning_rate": 5.805479175538229e-05,
+      "loss": 2.7786,
+      "step": 675
+    },
+    {
+      "epoch": 0.676,
+      "grad_norm": 0.59765625,
+      "learning_rate": 5.773817382593008e-05,
+      "loss": 2.7335,
+      "step": 676
+    },
+    {
+      "epoch": 0.677,
+      "grad_norm": 0.6171875,
+      "learning_rate": 5.7422070843492734e-05,
+      "loss": 2.6639,
+      "step": 677
+    },
+    {
+      "epoch": 0.678,
+      "grad_norm": 0.66015625,
+      "learning_rate": 5.710648665968543e-05,
+      "loss": 2.6919,
+      "step": 678
+    },
+    {
+      "epoch": 0.679,
+      "grad_norm": 0.6328125,
+      "learning_rate": 5.679142511980175e-05,
+      "loss": 2.7755,
+      "step": 679
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.6171875,
+      "learning_rate": 5.647689006276726e-05,
+      "loss": 2.8639,
+      "step": 680
+    },
+    {
+      "epoch": 0.681,
+      "grad_norm": 0.6953125,
+      "learning_rate": 5.616288532109225e-05,
+      "loss": 2.9127,
+      "step": 681
+    },
+    {
+      "epoch": 0.682,
+      "grad_norm": 0.73046875,
+      "learning_rate": 5.584941472082549e-05,
+      "loss": 2.8527,
+      "step": 682
+    },
+    {
+      "epoch": 0.683,
+      "grad_norm": 0.6875,
+      "learning_rate": 5.553648208150728e-05,
+      "loss": 2.7896,
+      "step": 683
+    },
+    {
+      "epoch": 0.684,
+      "grad_norm": 0.6171875,
+      "learning_rate": 5.522409121612304e-05,
+      "loss": 2.7857,
+      "step": 684
+    },
+    {
+      "epoch": 0.685,
+      "grad_norm": 0.65625,
+      "learning_rate": 5.491224593105695e-05,
+      "loss": 2.7679,
+      "step": 685
+    },
+    {
+      "epoch": 0.686,
+      "grad_norm": 0.6640625,
+      "learning_rate": 5.4600950026045326e-05,
+      "loss": 2.7092,
+      "step": 686
+    },
+    {
+      "epoch": 0.687,
+      "grad_norm": 0.64453125,
+      "learning_rate": 5.4290207294130615e-05,
+      "loss": 2.7738,
+      "step": 687
+    },
+    {
+      "epoch": 0.688,
+      "grad_norm": 0.99609375,
+      "learning_rate": 5.398002152161484e-05,
+      "loss": 2.7504,
+      "step": 688
+    },
+    {
+      "epoch": 0.689,
+      "grad_norm": 0.609375,
+      "learning_rate": 5.3670396488013854e-05,
+      "loss": 2.783,
+      "step": 689
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.640625,
+      "learning_rate": 5.33613359660109e-05,
+      "loss": 2.7702,
+      "step": 690
+    },
+    {
+      "epoch": 0.691,
+      "grad_norm": 0.66796875,
+      "learning_rate": 5.305284372141095e-05,
+      "loss": 2.7703,
+      "step": 691
+    },
+    {
+      "epoch": 0.692,
+      "grad_norm": 0.67578125,
+      "learning_rate": 5.274492351309461e-05,
+      "loss": 2.8481,
+      "step": 692
+    },
+    {
+      "epoch": 0.693,
+      "grad_norm": 0.640625,
+      "learning_rate": 5.243757909297247e-05,
+      "loss": 2.8131,
+      "step": 693
+    },
+    {
+      "epoch": 0.694,
+      "grad_norm": 0.62109375,
+      "learning_rate": 5.213081420593933e-05,
+      "loss": 2.8698,
+      "step": 694
+    },
+    {
+      "epoch": 0.695,
+      "grad_norm": 0.68359375,
+      "learning_rate": 5.182463258982846e-05,
+      "loss": 2.8107,
+      "step": 695
+    },
+    {
+      "epoch": 0.696,
+      "grad_norm": 0.6796875,
+      "learning_rate": 5.15190379753663e-05,
+      "loss": 2.8036,
+      "step": 696
+    },
+    {
+      "epoch": 0.697,
+      "grad_norm": 0.64453125,
+      "learning_rate": 5.121403408612672e-05,
+      "loss": 2.8471,
+      "step": 697
+    },
+    {
+      "epoch": 0.698,
+      "grad_norm": 0.609375,
+      "learning_rate": 5.090962463848592e-05,
+      "loss": 2.8396,
+      "step": 698
+    },
+    {
+      "epoch": 0.699,
+      "grad_norm": 0.62890625,
+      "learning_rate": 5.0605813341576924e-05,
+      "loss": 2.8636,
+      "step": 699
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.70703125,
+      "learning_rate": 5.0302603897244474e-05,
+      "loss": 2.9405,
+      "step": 700
+    },
+    {
+      "epoch": 0.701,
+      "grad_norm": 0.62890625,
+      "learning_rate": 5.000000000000002e-05,
+      "loss": 2.7732,
+      "step": 701
+    },
+    {
+      "epoch": 0.702,
+      "grad_norm": 0.671875,
+      "learning_rate": 4.969800533697649e-05,
+      "loss": 2.9202,
+      "step": 702
+    },
+    {
+      "epoch": 0.703,
+      "grad_norm": 0.65234375,
+      "learning_rate": 4.939662358788364e-05,
+      "loss": 2.8386,
+      "step": 703
+    },
+    {
+      "epoch": 0.704,
+      "grad_norm": 0.82421875,
+      "learning_rate": 4.909585842496287e-05,
+      "loss": 2.8088,
+      "step": 704
+    },
+    {
+      "epoch": 0.705,
+      "grad_norm": 0.66796875,
+      "learning_rate": 4.8795713512942865e-05,
+      "loss": 2.6467,
+      "step": 705
+    },
+    {
+      "epoch": 0.706,
+      "grad_norm": 0.65234375,
+      "learning_rate": 4.8496192508994576e-05,
+      "loss": 2.7861,
+      "step": 706
+    },
+    {
+      "epoch": 0.707,
+      "grad_norm": 0.52734375,
+      "learning_rate": 4.8197299062686995e-05,
+      "loss": 2.7869,
+      "step": 707
+    },
+    {
+      "epoch": 0.708,
+      "grad_norm": 0.6015625,
+      "learning_rate": 4.78990368159424e-05,
+      "loss": 2.6457,
+      "step": 708
+    },
+    {
+      "epoch": 0.709,
+      "grad_norm": 0.57421875,
+      "learning_rate": 4.7601409402992106e-05,
+      "loss": 2.6031,
+      "step": 709
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.58203125,
+      "learning_rate": 4.7304420450332244e-05,
+      "loss": 2.8342,
+      "step": 710
+    },
+    {
+      "epoch": 0.711,
+      "grad_norm": 0.56640625,
+      "learning_rate": 4.700807357667952e-05,
+      "loss": 2.7622,
+      "step": 711
+    },
+    {
+      "epoch": 0.712,
+      "grad_norm": 0.61328125,
+      "learning_rate": 4.6712372392927e-05,
+      "loss": 2.7514,
+      "step": 712
+    },
+    {
+      "epoch": 0.713,
+      "grad_norm": 0.6640625,
+      "learning_rate": 4.6417320502100316e-05,
+      "loss": 2.8732,
+      "step": 713
+    },
+    {
+      "epoch": 0.714,
+      "grad_norm": 0.6015625,
+      "learning_rate": 4.612292149931369e-05,
+      "loss": 2.8161,
+      "step": 714
+    },
+    {
+      "epoch": 0.715,
+      "grad_norm": 0.578125,
+      "learning_rate": 4.582917897172603e-05,
+      "loss": 2.7898,
+      "step": 715
+    },
+    {
+      "epoch": 0.716,
+      "grad_norm": 0.59765625,
+      "learning_rate": 4.5536096498497295e-05,
+      "loss": 2.7848,
+      "step": 716
+    },
+    {
+      "epoch": 0.717,
+      "grad_norm": 0.59765625,
+      "learning_rate": 4.524367765074499e-05,
+      "loss": 2.703,
+      "step": 717
+    },
+    {
+      "epoch": 0.718,
+      "grad_norm": 0.59375,
+      "learning_rate": 4.495192599150044e-05,
+      "loss": 2.7208,
+      "step": 718
+    },
+    {
+      "epoch": 0.719,
+      "grad_norm": 1.3671875,
+      "learning_rate": 4.46608450756656e-05,
+      "loss": 2.767,
+      "step": 719
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.57421875,
+      "learning_rate": 4.437043844996952e-05,
+      "loss": 2.7353,
+      "step": 720
+    },
+    {
+      "epoch": 0.721,
+      "grad_norm": 0.640625,
+      "learning_rate": 4.4080709652925336e-05,
+      "loss": 2.7186,
+      "step": 721
+    },
+    {
+      "epoch": 0.722,
+      "grad_norm": 18.625,
+      "learning_rate": 4.379166221478697e-05,
+      "loss": 2.6783,
+      "step": 722
+    },
+    {
+      "epoch": 0.723,
+      "grad_norm": 0.6015625,
+      "learning_rate": 4.350329965750621e-05,
+      "loss": 2.718,
+      "step": 723
+    },
+    {
+      "epoch": 0.724,
+      "grad_norm": 0.64453125,
+      "learning_rate": 4.32156254946899e-05,
+      "loss": 2.6782,
+      "step": 724
+    },
+    {
+      "epoch": 0.725,
+      "grad_norm": 0.6484375,
+      "learning_rate": 4.2928643231556844e-05,
+      "loss": 2.7787,
+      "step": 725
+    },
+    {
+      "epoch": 0.726,
+      "grad_norm": 0.59765625,
+      "learning_rate": 4.264235636489542e-05,
+      "loss": 2.7355,
+      "step": 726
+    },
+    {
+      "epoch": 0.727,
+      "grad_norm": 0.59765625,
+      "learning_rate": 4.235676838302068e-05,
+      "loss": 2.8365,
+      "step": 727
+    },
+    {
+      "epoch": 0.728,
+      "grad_norm": 0.59765625,
+      "learning_rate": 4.207188276573214e-05,
+      "loss": 2.7514,
+      "step": 728
+    },
+    {
+      "epoch": 0.729,
+      "grad_norm": 0.60546875,
+      "learning_rate": 4.1787702984271074e-05,
+      "loss": 2.7733,
+      "step": 729
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.8984375,
+      "learning_rate": 4.150423250127845e-05,
+      "loss": 2.809,
+      "step": 730
+    },
+    {
+      "epoch": 0.731,
+      "grad_norm": 0.5859375,
+      "learning_rate": 4.12214747707527e-05,
+      "loss": 2.7394,
+      "step": 731
+    },
+    {
+      "epoch": 0.732,
+      "grad_norm": 0.578125,
+      "learning_rate": 4.093943323800745e-05,
+      "loss": 2.6793,
+      "step": 732
+    },
+    {
+      "epoch": 0.733,
+      "grad_norm": 0.5625,
+      "learning_rate": 4.065811133962987e-05,
+      "loss": 2.7069,
+      "step": 733
+    },
+    {
+      "epoch": 0.734,
+      "grad_norm": 0.57421875,
+      "learning_rate": 4.037751250343841e-05,
+      "loss": 2.6702,
+      "step": 734
+    },
+    {
+      "epoch": 0.735,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.009764014844143e-05,
+      "loss": 2.8871,
+      "step": 735
+    },
+    {
+      "epoch": 0.736,
+      "grad_norm": 0.62890625,
+      "learning_rate": 3.981849768479517e-05,
+      "loss": 2.7682,
+      "step": 736
+    },
+    {
+      "epoch": 0.737,
+      "grad_norm": 0.59765625,
+      "learning_rate": 3.954008851376252e-05,
+      "loss": 2.7024,
+      "step": 737
+    },
+    {
+      "epoch": 0.738,
+      "grad_norm": 0.59765625,
+      "learning_rate": 3.9262416027671356e-05,
+      "loss": 2.6956,
+      "step": 738
+    },
+    {
+      "epoch": 0.739,
+      "grad_norm": 0.6640625,
+      "learning_rate": 3.8985483609873244e-05,
+      "loss": 2.8372,
+      "step": 739
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5703125,
+      "learning_rate": 3.8709294634702376e-05,
+      "loss": 2.7524,
+      "step": 740
+    },
+    {
+      "epoch": 0.741,
+      "grad_norm": 0.59375,
+      "learning_rate": 3.843385246743417e-05,
+      "loss": 2.8114,
+      "step": 741
+    },
+    {
+      "epoch": 0.742,
+      "grad_norm": 0.60546875,
+      "learning_rate": 3.8159160464244606e-05,
+      "loss": 2.8749,
+      "step": 742
+    },
+    {
+      "epoch": 0.743,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.788522197216897e-05,
+      "loss": 2.7037,
+      "step": 743
+    },
+    {
+      "epoch": 0.744,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.7612040329061405e-05,
+      "loss": 2.7504,
+      "step": 744
+    },
+    {
+      "epoch": 0.745,
+      "grad_norm": 0.59765625,
+      "learning_rate": 3.733961886355398e-05,
+      "loss": 2.7219,
+      "step": 745
+    },
+    {
+      "epoch": 0.746,
+      "grad_norm": 0.60546875,
+      "learning_rate": 3.7067960895016275e-05,
+      "loss": 2.741,
+      "step": 746
+    },
+    {
+      "epoch": 0.747,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.679706973351491e-05,
+      "loss": 2.7085,
+      "step": 747
+    },
+    {
+      "epoch": 0.748,
+      "grad_norm": 0.59375,
+      "learning_rate": 3.6526948679773257e-05,
+      "loss": 2.7705,
+      "step": 748
+    },
+    {
+      "epoch": 0.749,
+      "grad_norm": 0.58984375,
+      "learning_rate": 3.6257601025131026e-05,
+      "loss": 2.7403,
+      "step": 749
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9375,
+      "learning_rate": 3.5989030051504434e-05,
+      "loss": 2.7612,
+      "step": 750
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 2.7113237380981445,
+      "eval_runtime": 31.7342,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 2.647,
+      "step": 750
+    },
+    {
+      "epoch": 0.751,
+      "grad_norm": 0.59375,
+      "learning_rate": 3.5721239031346066e-05,
+      "loss": 2.8165,
+      "step": 751
+    },
+    {
+      "epoch": 0.752,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.545423122760493e-05,
+      "loss": 2.6717,
+      "step": 752
+    },
+    {
+      "epoch": 0.753,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.518800989368691e-05,
+      "loss": 2.5895,
+      "step": 753
+    },
+    {
+      "epoch": 0.754,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.492257827341492e-05,
+      "loss": 2.7452,
+      "step": 754
+    },
+    {
+      "epoch": 0.755,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.465793960098945e-05,
+      "loss": 2.7445,
+      "step": 755
+    },
+    {
+      "epoch": 0.756,
+      "grad_norm": 0.59765625,
+      "learning_rate": 3.439409710094929e-05,
+      "loss": 2.805,
+      "step": 756
+    },
+    {
+      "epoch": 0.757,
+      "grad_norm": 0.640625,
+      "learning_rate": 3.413105398813195e-05,
+      "loss": 2.8297,
+      "step": 757
+    },
+    {
+      "epoch": 0.758,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.386881346763483e-05,
+      "loss": 2.6497,
+      "step": 758
+    },
+    {
+      "epoch": 0.759,
+      "grad_norm": 0.56640625,
+      "learning_rate": 3.360737873477584e-05,
+      "loss": 2.7854,
+      "step": 759
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.58203125,
+      "learning_rate": 3.334675297505476e-05,
+      "loss": 2.7295,
+      "step": 760
+    },
+    {
+      "epoch": 0.761,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.308693936411421e-05,
+      "loss": 2.7563,
+      "step": 761
+    },
+    {
+      "epoch": 0.762,
+      "grad_norm": 0.5859375,
+      "learning_rate": 3.2827941067700996e-05,
+      "loss": 2.8956,
+      "step": 762
+    },
+    {
+      "epoch": 0.763,
+      "grad_norm": 0.76953125,
+      "learning_rate": 3.2569761241627696e-05,
+      "loss": 2.7614,
+      "step": 763
+    },
+    {
+      "epoch": 0.764,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.231240303173394e-05,
+      "loss": 2.7303,
+      "step": 764
+    },
+    {
+      "epoch": 0.765,
+      "grad_norm": 0.58984375,
+      "learning_rate": 3.205586957384838e-05,
+      "loss": 2.7826,
+      "step": 765
+    },
+    {
+      "epoch": 0.766,
+      "grad_norm": 1.3359375,
+      "learning_rate": 3.1800163993750166e-05,
+      "loss": 2.7078,
+      "step": 766
+    },
+    {
+      "epoch": 0.767,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.154528940713113e-05,
+      "loss": 2.6754,
+      "step": 767
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.59765625,
+      "learning_rate": 3.129124891955771e-05,
+      "loss": 2.7744,
+      "step": 768
+    },
+    {
+      "epoch": 0.769,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.103804562643302e-05,
+      "loss": 2.653,
+      "step": 769
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.078568261295933e-05,
+      "loss": 2.7278,
+      "step": 770
+    },
+    {
+      "epoch": 0.771,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.053416295410026e-05,
+      "loss": 2.7533,
+      "step": 771
+    },
+    {
+      "epoch": 0.772,
+      "grad_norm": 0.60546875,
+      "learning_rate": 3.0283489714543556e-05,
+      "loss": 2.7639,
+      "step": 772
+    },
+    {
+      "epoch": 0.773,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.0033665948663448e-05,
+      "loss": 2.7951,
+      "step": 773
+    },
+    {
+      "epoch": 0.774,
+      "grad_norm": 0.6015625,
+      "learning_rate": 2.9784694700483762e-05,
+      "loss": 2.7375,
+      "step": 774
+    },
+    {
+      "epoch": 0.775,
+      "grad_norm": 0.55859375,
+      "learning_rate": 2.953657900364053e-05,
+      "loss": 2.7481,
+      "step": 775
+    },
+    {
+      "epoch": 0.776,
+      "grad_norm": 0.59765625,
+      "learning_rate": 2.9289321881345254e-05,
+      "loss": 2.7437,
+      "step": 776
+    },
+    {
+      "epoch": 0.777,
+      "grad_norm": 0.58984375,
+      "learning_rate": 2.904292634634793e-05,
+      "loss": 2.7637,
+      "step": 777
+    },
+    {
+      "epoch": 0.778,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.879739540090036e-05,
+      "loss": 2.7252,
+      "step": 778
+    },
+    {
+      "epoch": 0.779,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.8552732036719687e-05,
+      "loss": 2.5999,
+      "step": 779
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.8308939234951726e-05,
+      "loss": 2.8094,
+      "step": 780
+    },
+    {
+      "epoch": 0.781,
+      "grad_norm": 0.58984375,
+      "learning_rate": 2.8066019966134904e-05,
+      "loss": 2.7139,
+      "step": 781
+    },
+    {
+      "epoch": 0.782,
+      "grad_norm": 1.46875,
+      "learning_rate": 2.7823977190163786e-05,
+      "loss": 2.8623,
+      "step": 782
+    },
+    {
+      "epoch": 0.783,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.7582813856253275e-05,
+      "loss": 2.7122,
+      "step": 783
+    },
+    {
+      "epoch": 0.784,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.734253290290242e-05,
+      "loss": 2.7772,
+      "step": 784
+    },
+    {
+      "epoch": 0.785,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.7103137257858868e-05,
+      "loss": 2.7612,
+      "step": 785
+    },
+    {
+      "epoch": 0.786,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.6864629838082956e-05,
+      "loss": 2.7303,
+      "step": 786
+    },
+    {
+      "epoch": 0.787,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.6627013549712355e-05,
+      "loss": 2.6967,
+      "step": 787
+    },
+    {
+      "epoch": 0.788,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.639029128802657e-05,
+      "loss": 2.675,
+      "step": 788
+    },
+    {
+      "epoch": 0.789,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.615446593741161e-05,
+      "loss": 2.771,
+      "step": 789
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.59765625,
+      "learning_rate": 2.5919540371325e-05,
+      "loss": 2.7169,
+      "step": 790
+    },
+    {
+      "epoch": 0.791,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.5685517452260567e-05,
+      "loss": 2.6604,
+      "step": 791
+    },
+    {
+      "epoch": 0.792,
+      "grad_norm": 0.59765625,
+      "learning_rate": 2.5452400031713785e-05,
+      "loss": 2.6749,
+      "step": 792
+    },
+    {
+      "epoch": 0.793,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.5220190950146827e-05,
+      "loss": 2.6759,
+      "step": 793
+    },
+    {
+      "epoch": 0.794,
+      "grad_norm": 0.57421875,
+      "learning_rate": 2.4988893036954043e-05,
+      "loss": 2.5863,
+      "step": 794
+    },
+    {
+      "epoch": 0.795,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.4758509110427575e-05,
+      "loss": 2.7512,
+      "step": 795
+    },
+    {
+      "epoch": 0.796,
+      "grad_norm": 0.59765625,
+      "learning_rate": 2.45290419777228e-05,
+      "loss": 2.7619,
+      "step": 796
+    },
+    {
+      "epoch": 0.797,
+      "grad_norm": 0.74609375,
+      "learning_rate": 2.4300494434824373e-05,
+      "loss": 2.6975,
+      "step": 797
+    },
+    {
+      "epoch": 0.798,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.407286926651192e-05,
+      "loss": 2.6837,
+      "step": 798
+    },
+    {
+      "epoch": 0.799,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.3846169246326343e-05,
+      "loss": 2.7131,
+      "step": 799
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.362039713653581e-05,
+      "loss": 2.7429,
+      "step": 800
+    },
+    {
+      "epoch": 0.801,
+      "grad_norm": 0.578125,
+      "learning_rate": 2.339555568810221e-05,
+      "loss": 2.6673,
+      "step": 801
+    },
+    {
+      "epoch": 0.802,
+      "grad_norm": 0.57421875,
+      "learning_rate": 2.3171647640647687e-05,
+      "loss": 2.6941,
+      "step": 802
+    },
+    {
+      "epoch": 0.803,
+      "grad_norm": 0.6484375,
+      "learning_rate": 2.2948675722421086e-05,
+      "loss": 2.735,
+      "step": 803
+    },
+    {
+      "epoch": 0.804,
+      "grad_norm": 0.79296875,
+      "learning_rate": 2.2726642650264895e-05,
+      "loss": 2.7529,
+      "step": 804
+    },
+    {
+      "epoch": 0.805,
+      "grad_norm": 0.55859375,
+      "learning_rate": 2.2505551129582047e-05,
+      "loss": 2.6125,
+      "step": 805
+    },
+    {
+      "epoch": 0.806,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.2285403854302912e-05,
+      "loss": 2.8059,
+      "step": 806
+    },
+    {
+      "epoch": 0.807,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.2066203506852566e-05,
+      "loss": 2.6916,
+      "step": 807
+    },
+    {
+      "epoch": 0.808,
+      "grad_norm": 0.7265625,
+      "learning_rate": 2.1847952758118117e-05,
+      "loss": 2.7839,
+      "step": 808
+    },
+    {
+      "epoch": 0.809,
+      "grad_norm": 0.58984375,
+      "learning_rate": 2.163065426741603e-05,
+      "loss": 2.7451,
+      "step": 809
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.53515625,
+      "learning_rate": 2.1414310682459802e-05,
+      "loss": 2.7282,
+      "step": 810
+    },
+    {
+      "epoch": 0.811,
+      "grad_norm": 0.57421875,
+      "learning_rate": 2.119892463932781e-05,
+      "loss": 2.6989,
+      "step": 811
+    },
+    {
+      "epoch": 0.812,
+      "grad_norm": 0.5859375,
+      "learning_rate": 2.098449876243096e-05,
+      "loss": 2.6681,
+      "step": 812
+    },
+    {
+      "epoch": 0.813,
+      "grad_norm": 0.70703125,
+      "learning_rate": 2.0771035664480942e-05,
+      "loss": 2.7579,
+      "step": 813
+    },
+    {
+      "epoch": 0.814,
+      "grad_norm": 0.57421875,
+      "learning_rate": 2.0558537946458177e-05,
+      "loss": 2.6096,
+      "step": 814
+    },
+    {
+      "epoch": 0.815,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.0347008197580374e-05,
+      "loss": 2.5568,
+      "step": 815
+    },
+    {
+      "epoch": 0.816,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.013644899527074e-05,
+      "loss": 2.7679,
+      "step": 816
+    },
+    {
+      "epoch": 0.817,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.9926862905126665e-05,
+      "loss": 2.751,
+      "step": 817
+    },
+    {
+      "epoch": 0.818,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.9718252480888566e-05,
+      "loss": 2.7757,
+      "step": 818
+    },
+    {
+      "epoch": 0.819,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.9510620264408596e-05,
+      "loss": 2.6745,
+      "step": 819
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.930396878561983e-05,
+      "loss": 2.6728,
+      "step": 820
+    },
+    {
+      "epoch": 0.821,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.9098300562505266e-05,
+      "loss": 2.6893,
+      "step": 821
+    },
+    {
+      "epoch": 0.822,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.8893618101067355e-05,
+      "loss": 2.8697,
+      "step": 822
+    },
+    {
+      "epoch": 0.823,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.8689923895297245e-05,
+      "loss": 2.7268,
+      "step": 823
+    },
+    {
+      "epoch": 0.824,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.848722042714457e-05,
+      "loss": 2.7974,
+      "step": 824
+    },
+    {
+      "epoch": 0.825,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.8285510166487152e-05,
+      "loss": 2.691,
+      "step": 825
+    },
+    {
+      "epoch": 0.826,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.808479557110081e-05,
+      "loss": 2.6877,
+      "step": 826
+    },
+    {
+      "epoch": 0.827,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.78850790866296e-05,
+      "loss": 2.6971,
+      "step": 827
+    },
+    {
+      "epoch": 0.828,
+      "grad_norm": 0.5625,
+      "learning_rate": 1.7686363146555805e-05,
+      "loss": 2.8624,
+      "step": 828
+    },
+    {
+      "epoch": 0.829,
+      "grad_norm": 2.421875,
+      "learning_rate": 1.7488650172170496e-05,
+      "loss": 2.6911,
+      "step": 829
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.7291942572543807e-05,
+      "loss": 2.6905,
+      "step": 830
+    },
+    {
+      "epoch": 0.831,
+      "grad_norm": 0.58984375,
+      "learning_rate": 1.7096242744495837e-05,
+      "loss": 2.7459,
+      "step": 831
+    },
+    {
+      "epoch": 0.832,
+      "grad_norm": 0.6875,
+      "learning_rate": 1.690155307256719e-05,
+      "loss": 2.738,
+      "step": 832
+    },
+    {
+      "epoch": 0.833,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.6707875928990058e-05,
+      "loss": 2.7459,
+      "step": 833
+    },
+    {
+      "epoch": 0.834,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.6515213673659357e-05,
+      "loss": 2.7432,
+      "step": 834
+    },
+    {
+      "epoch": 0.835,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.632356865410384e-05,
+      "loss": 2.6985,
+      "step": 835
+    },
+    {
+      "epoch": 0.836,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.6132943205457606e-05,
+      "loss": 2.8408,
+      "step": 836
+    },
+    {
+      "epoch": 0.837,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.5943339650431576e-05,
+      "loss": 2.662,
+      "step": 837
+    },
+    {
+      "epoch": 0.838,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.5754760299285252e-05,
+      "loss": 2.85,
+      "step": 838
+    },
+    {
+      "epoch": 0.839,
+      "grad_norm": 0.5390625,
+      "learning_rate": 1.5567207449798515e-05,
+      "loss": 2.6535,
+      "step": 839
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.538068338724361e-05,
+      "loss": 2.7144,
+      "step": 840
+    },
+    {
+      "epoch": 0.841,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.5195190384357404e-05,
+      "loss": 2.8185,
+      "step": 841
+    },
+    {
+      "epoch": 0.842,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.5010730701313625e-05,
+      "loss": 2.7978,
+      "step": 842
+    },
+    {
+      "epoch": 0.843,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.4827306585695234e-05,
+      "loss": 2.7138,
+      "step": 843
+    },
+    {
+      "epoch": 0.844,
+      "grad_norm": 0.890625,
+      "learning_rate": 1.4644920272467244e-05,
+      "loss": 2.7772,
+      "step": 844
+    },
+    {
+      "epoch": 0.845,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.4463573983949341e-05,
+      "loss": 2.7598,
+      "step": 845
+    },
+    {
+      "epoch": 0.846,
+      "grad_norm": 0.6953125,
+      "learning_rate": 1.4283269929788779e-05,
+      "loss": 2.7201,
+      "step": 846
+    },
+    {
+      "epoch": 0.847,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.4104010306933557e-05,
+      "loss": 2.8091,
+      "step": 847
+    },
+    {
+      "epoch": 0.848,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.3925797299605647e-05,
+      "loss": 2.7008,
+      "step": 848
+    },
+    {
+      "epoch": 0.849,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.3748633079274253e-05,
+      "loss": 2.655,
+      "step": 849
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.3572519804629536e-05,
+      "loss": 2.6479,
+      "step": 850
+    },
+    {
+      "epoch": 0.851,
+      "grad_norm": 0.5625,
+      "learning_rate": 1.339745962155613e-05,
+      "loss": 2.8183,
+      "step": 851
+    },
+    {
+      "epoch": 0.852,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.3223454663107172e-05,
+      "loss": 2.7465,
+      "step": 852
+    },
+    {
+      "epoch": 0.853,
+      "grad_norm": 0.6015625,
+      "learning_rate": 1.30505070494781e-05,
+      "loss": 2.7232,
+      "step": 853
+    },
+    {
+      "epoch": 0.854,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.2878618887981064e-05,
+      "loss": 2.7987,
+      "step": 854
+    },
+    {
+      "epoch": 0.855,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.2707792273019048e-05,
+      "loss": 2.7005,
+      "step": 855
+    },
+    {
+      "epoch": 0.856,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.2538029286060426e-05,
+      "loss": 2.7156,
+      "step": 856
+    },
+    {
+      "epoch": 0.857,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.2369331995613665e-05,
+      "loss": 2.757,
+      "step": 857
+    },
+    {
+      "epoch": 0.858,
+      "grad_norm": 0.91015625,
+      "learning_rate": 1.2201702457201947e-05,
+      "loss": 2.6486,
+      "step": 858
+    },
+    {
+      "epoch": 0.859,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.2035142713338366e-05,
+      "loss": 2.7237,
+      "step": 859
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.125,
+      "learning_rate": 1.1869654793500784e-05,
+      "loss": 2.6579,
+      "step": 860
+    },
+    {
+      "epoch": 0.861,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.1705240714107302e-05,
+      "loss": 2.719,
+      "step": 861
+    },
+    {
+      "epoch": 0.862,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.1541902478491606e-05,
+      "loss": 2.6551,
+      "step": 862
+    },
+    {
+      "epoch": 0.863,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.1379642076878527e-05,
+      "loss": 2.6851,
+      "step": 863
+    },
+    {
+      "epoch": 0.864,
+      "grad_norm": 0.5390625,
+      "learning_rate": 1.1218461486359877e-05,
+      "loss": 2.7538,
+      "step": 864
+    },
+    {
+      "epoch": 0.865,
+      "grad_norm": 0.58984375,
+      "learning_rate": 1.1058362670870249e-05,
+      "loss": 2.693,
+      "step": 865
+    },
+    {
+      "epoch": 0.866,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.0899347581163221e-05,
+      "loss": 2.7243,
+      "step": 866
+    },
+    {
+      "epoch": 0.867,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.0741418154787442e-05,
+      "loss": 2.6959,
+      "step": 867
+    },
+    {
+      "epoch": 0.868,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.0584576316063188e-05,
+      "loss": 2.7305,
+      "step": 868
+    },
+    {
+      "epoch": 0.869,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.042882397605871e-05,
+      "loss": 2.7807,
+      "step": 869
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.0274163032567163e-05,
+      "loss": 2.7001,
+      "step": 870
+    },
+    {
+      "epoch": 0.871,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.0120595370083318e-05,
+      "loss": 2.7309,
+      "step": 871
+    },
+    {
+      "epoch": 0.872,
+      "grad_norm": 0.54296875,
+      "learning_rate": 9.968122859780648e-06,
+      "loss": 2.6748,
+      "step": 872
+    },
+    {
+      "epoch": 0.873,
+      "grad_norm": 0.58203125,
+      "learning_rate": 9.816747359488632e-06,
+      "loss": 2.7391,
+      "step": 873
+    },
+    {
+      "epoch": 0.874,
+      "grad_norm": 0.5625,
+      "learning_rate": 9.666470713669918e-06,
+      "loss": 2.8011,
+      "step": 874
+    },
+    {
+      "epoch": 0.875,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.517294753398064e-06,
+      "loss": 2.6658,
+      "step": 875
+    },
+    {
+      "epoch": 0.876,
+      "grad_norm": 0.546875,
+      "learning_rate": 9.369221296335006e-06,
+      "loss": 2.7542,
+      "step": 876
+    },
+    {
+      "epoch": 0.877,
+      "grad_norm": 0.5390625,
+      "learning_rate": 9.222252146709142e-06,
+      "loss": 2.6932,
+      "step": 877
+    },
+    {
+      "epoch": 0.878,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.076389095293148e-06,
+      "loss": 2.7678,
+      "step": 878
+    },
+    {
+      "epoch": 0.879,
+      "grad_norm": 0.53515625,
+      "learning_rate": 8.931633919382298e-06,
+      "loss": 2.6873,
+      "step": 879
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.52734375,
+      "learning_rate": 8.787988382772705e-06,
+      "loss": 2.812,
+      "step": 880
+    },
+    {
+      "epoch": 0.881,
+      "grad_norm": 0.57421875,
+      "learning_rate": 8.645454235739903e-06,
+      "loss": 2.6715,
+      "step": 881
+    },
+    {
+      "epoch": 0.882,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.504033215017527e-06,
+      "loss": 2.7504,
+      "step": 882
+    },
+    {
+      "epoch": 0.883,
+      "grad_norm": 0.515625,
+      "learning_rate": 8.363727043776038e-06,
+      "loss": 2.7513,
+      "step": 883
+    },
+    {
+      "epoch": 0.884,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.224537431601886e-06,
+      "loss": 2.8052,
+      "step": 884
+    },
+    {
+      "epoch": 0.885,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.086466074476563e-06,
+      "loss": 2.7368,
+      "step": 885
+    },
+    {
+      "epoch": 0.886,
+      "grad_norm": 0.55859375,
+      "learning_rate": 7.949514654755962e-06,
+      "loss": 2.6757,
+      "step": 886
+    },
+    {
+      "epoch": 0.887,
+      "grad_norm": 0.5546875,
+      "learning_rate": 7.81368484114996e-06,
+      "loss": 2.703,
+      "step": 887
+    },
+    {
+      "epoch": 0.888,
+      "grad_norm": 0.54296875,
+      "learning_rate": 7.67897828870191e-06,
+      "loss": 2.7158,
+      "step": 888
+    },
+    {
+      "epoch": 0.889,
+      "grad_norm": 0.55078125,
+      "learning_rate": 7.545396638768698e-06,
+      "loss": 2.7778,
+      "step": 889
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.5078125,
+      "learning_rate": 7.412941519000527e-06,
+      "loss": 2.6752,
+      "step": 890
+    },
+    {
+      "epoch": 0.891,
+      "grad_norm": 1.59375,
+      "learning_rate": 7.281614543321269e-06,
+      "loss": 2.7462,
+      "step": 891
+    },
+    {
+      "epoch": 0.892,
+      "grad_norm": 0.515625,
+      "learning_rate": 7.151417311908648e-06,
+      "loss": 2.6856,
+      "step": 892
+    },
+    {
+      "epoch": 0.893,
+      "grad_norm": 0.546875,
+      "learning_rate": 7.022351411174866e-06,
+      "loss": 2.7136,
+      "step": 893
+    },
+    {
+      "epoch": 0.894,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.894418413747183e-06,
+      "loss": 2.7652,
+      "step": 894
+    },
+    {
+      "epoch": 0.895,
+      "grad_norm": 0.5390625,
+      "learning_rate": 6.767619878448783e-06,
+      "loss": 2.687,
+      "step": 895
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.609375,
+      "learning_rate": 6.6419573502798374e-06,
+      "loss": 2.8021,
+      "step": 896
+    },
+    {
+      "epoch": 0.897,
+      "grad_norm": 0.53125,
+      "learning_rate": 6.517432360398556e-06,
+      "loss": 2.7689,
+      "step": 897
+    },
+    {
+      "epoch": 0.898,
+      "grad_norm": 0.5859375,
+      "learning_rate": 6.394046426102674e-06,
+      "loss": 2.7352,
+      "step": 898
+    },
+    {
+      "epoch": 0.899,
+      "grad_norm": 0.5546875,
+      "learning_rate": 6.2718010508108545e-06,
+      "loss": 2.7,
+      "step": 899
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.1506977240444074e-06,
+      "loss": 2.6995,
+      "step": 900
+    },
+    {
+      "epoch": 0.901,
+      "grad_norm": 0.5390625,
+      "learning_rate": 6.030737921409169e-06,
+      "loss": 2.6935,
+      "step": 901
+    },
+    {
+      "epoch": 0.902,
+      "grad_norm": 0.5234375,
+      "learning_rate": 5.911923104577455e-06,
+      "loss": 2.711,
+      "step": 902
+    },
+    {
+      "epoch": 0.903,
+      "grad_norm": 0.5078125,
+      "learning_rate": 5.7942547212703315e-06,
+      "loss": 2.7778,
+      "step": 903
+    },
+    {
+      "epoch": 0.904,
+      "grad_norm": 0.53125,
+      "learning_rate": 5.6777342052399045e-06,
+      "loss": 2.7373,
+      "step": 904
+    },
+    {
+      "epoch": 0.905,
+      "grad_norm": 0.515625,
+      "learning_rate": 5.562362976251901e-06,
+      "loss": 2.7644,
+      "step": 905
+    },
+    {
+      "epoch": 0.906,
+      "grad_norm": 0.54296875,
+      "learning_rate": 5.448142440068316e-06,
+      "loss": 2.6753,
+      "step": 906
+    },
+    {
+      "epoch": 0.907,
+      "grad_norm": 0.8515625,
+      "learning_rate": 5.335073988430372e-06,
+      "loss": 2.7881,
+      "step": 907
+    },
+    {
+      "epoch": 0.908,
+      "grad_norm": 0.5234375,
+      "learning_rate": 5.223158999041444e-06,
+      "loss": 2.8192,
+      "step": 908
+    },
+    {
+      "epoch": 0.909,
+      "grad_norm": 0.578125,
+      "learning_rate": 5.1123988355503475e-06,
+      "loss": 2.7979,
+      "step": 909
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.578125,
+      "learning_rate": 5.002794847534764e-06,
+      "loss": 2.7629,
+      "step": 910
+    },
+    {
+      "epoch": 0.911,
+      "grad_norm": 0.53515625,
+      "learning_rate": 4.8943483704846475e-06,
+      "loss": 2.7704,
+      "step": 911
+    },
+    {
+      "epoch": 0.912,
+      "grad_norm": 0.55859375,
+      "learning_rate": 4.7870607257861415e-06,
+      "loss": 2.6811,
+      "step": 912
+    },
+    {
+      "epoch": 0.913,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.680933220705308e-06,
+      "loss": 2.7518,
+      "step": 913
+    },
+    {
+      "epoch": 0.914,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.575967148372317e-06,
+      "loss": 2.7595,
+      "step": 914
+    },
+    {
+      "epoch": 0.915,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.4721637877656375e-06,
+      "loss": 2.7555,
+      "step": 915
+    },
+    {
+      "epoch": 0.916,
+      "grad_norm": 0.546875,
+      "learning_rate": 4.369524403696457e-06,
+      "loss": 2.728,
+      "step": 916
+    },
+    {
+      "epoch": 0.917,
+      "grad_norm": 0.546875,
+      "learning_rate": 4.268050246793276e-06,
+      "loss": 2.7264,
+      "step": 917
+    },
+    {
+      "epoch": 0.918,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.167742553486675e-06,
+      "loss": 2.7391,
+      "step": 918
+    },
+    {
+      "epoch": 0.919,
+      "grad_norm": 0.53515625,
+      "learning_rate": 4.068602545994249e-06,
+      "loss": 2.7269,
+      "step": 919
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.51953125,
+      "learning_rate": 3.970631432305694e-06,
+      "loss": 2.7273,
+      "step": 920
+    },
+    {
+      "epoch": 0.921,
+      "grad_norm": 0.53515625,
+      "learning_rate": 3.873830406168111e-06,
+      "loss": 2.8126,
+      "step": 921
+    },
+    {
+      "epoch": 0.922,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.7782006470714616e-06,
+      "loss": 2.7532,
+      "step": 922
+    },
+    {
+      "epoch": 0.923,
+      "grad_norm": 0.53125,
+      "learning_rate": 3.68374332023419e-06,
+      "loss": 2.7923,
+      "step": 923
+    },
+    {
+      "epoch": 0.924,
+      "grad_norm": 0.52734375,
+      "learning_rate": 3.5904595765890005e-06,
+      "loss": 2.7534,
+      "step": 924
+    },
+    {
+      "epoch": 0.925,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.4983505527688586e-06,
+      "loss": 2.7706,
+      "step": 925
+    },
+    {
+      "epoch": 0.926,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.40741737109318e-06,
+      "loss": 2.6821,
+      "step": 926
+    },
+    {
+      "epoch": 0.927,
+      "grad_norm": 0.56640625,
+      "learning_rate": 3.3176611395540626e-06,
+      "loss": 2.6992,
+      "step": 927
+    },
+    {
+      "epoch": 0.928,
+      "grad_norm": 0.5546875,
+      "learning_rate": 3.2290829518028862e-06,
+      "loss": 2.7818,
+      "step": 928
+    },
+    {
+      "epoch": 0.929,
+      "grad_norm": 0.56640625,
+      "learning_rate": 3.1416838871368924e-06,
+      "loss": 2.7538,
+      "step": 929
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.5546875,
+      "learning_rate": 3.0554650104861136e-06,
+      "loss": 2.7118,
+      "step": 930
+    },
+    {
+      "epoch": 0.931,
+      "grad_norm": 0.55859375,
+      "learning_rate": 2.970427372400353e-06,
+      "loss": 2.7252,
+      "step": 931
+    },
+    {
+      "epoch": 0.932,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.8865720090364034e-06,
+      "loss": 2.8308,
+      "step": 932
+    },
+    {
+      "epoch": 0.933,
+      "grad_norm": 0.53125,
+      "learning_rate": 2.8038999421453826e-06,
+      "loss": 2.6437,
+      "step": 933
+    },
+    {
+      "epoch": 0.934,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.7224121790603517e-06,
+      "loss": 2.764,
+      "step": 934
+    },
+    {
+      "epoch": 0.935,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.6421097126839712e-06,
+      "loss": 2.6876,
+      "step": 935
+    },
+    {
+      "epoch": 0.936,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.5629935214764865e-06,
+      "loss": 2.8505,
+      "step": 936
+    },
+    {
+      "epoch": 0.937,
+      "grad_norm": 0.53515625,
+      "learning_rate": 2.4850645694436736e-06,
+      "loss": 2.7782,
+      "step": 937
+    },
+    {
+      "epoch": 0.938,
+      "grad_norm": 1.0625,
+      "learning_rate": 2.4083238061252567e-06,
+      "loss": 2.6892,
+      "step": 938
+    },
+    {
+      "epoch": 0.939,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.332772166583208e-06,
+      "loss": 2.6548,
+      "step": 939
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.2584105713904125e-06,
+      "loss": 2.7622,
+      "step": 940
+    },
+    {
+      "epoch": 0.941,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.1852399266194314e-06,
+      "loss": 2.7183,
+      "step": 941
+    },
+    {
+      "epoch": 0.942,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.1132611238315003e-06,
+      "loss": 2.7956,
+      "step": 942
+    },
+    {
+      "epoch": 0.943,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.0424750400655947e-06,
+      "loss": 2.6357,
+      "step": 943
+    },
+    {
+      "epoch": 0.944,
+      "grad_norm": 0.55859375,
+      "learning_rate": 1.9728825378278246e-06,
+      "loss": 2.6503,
+      "step": 944
+    },
+    {
+      "epoch": 0.945,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.904484465080847e-06,
+      "loss": 2.6973,
+      "step": 945
+    },
+    {
+      "epoch": 0.946,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.8372816552336026e-06,
+      "loss": 2.7222,
+      "step": 946
+    },
+    {
+      "epoch": 0.947,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.771274927131139e-06,
+      "loss": 2.6586,
+      "step": 947
+    },
+    {
+      "epoch": 0.948,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.706465085044584e-06,
+      "loss": 2.657,
+      "step": 948
+    },
+    {
+      "epoch": 0.949,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.6428529186614195e-06,
+      "loss": 2.7474,
+      "step": 949
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.580439203075812e-06,
+      "loss": 2.6946,
+      "step": 950
+    },
+    {
+      "epoch": 0.951,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.5192246987791981e-06,
+      "loss": 2.6957,
+      "step": 951
+    },
+    {
+      "epoch": 0.952,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.4592101516509914e-06,
+      "loss": 2.7191,
+      "step": 952
+    },
+    {
+      "epoch": 0.953,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.400396292949513e-06,
+      "loss": 2.6774,
+      "step": 953
+    },
+    {
+      "epoch": 0.954,
+      "grad_norm": 0.640625,
+      "learning_rate": 1.3427838393030633e-06,
+      "loss": 2.785,
+      "step": 954
+    },
+    {
+      "epoch": 0.955,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.2863734927012095e-06,
+      "loss": 2.7072,
+      "step": 955
+    },
+    {
+      "epoch": 0.956,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.231165940486234e-06,
+      "loss": 2.7511,
+      "step": 956
+    },
+    {
+      "epoch": 0.957,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.1771618553447216e-06,
+      "loss": 2.7826,
+      "step": 957
+    },
+    {
+      "epoch": 0.958,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.1243618952994195e-06,
+      "loss": 2.7722,
+      "step": 958
+    },
+    {
+      "epoch": 0.959,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.0727667037011668e-06,
+      "loss": 2.6715,
+      "step": 959
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.0223769092211012e-06,
+      "loss": 2.7709,
+      "step": 960
+    },
+    {
+      "epoch": 0.961,
+      "grad_norm": 0.5390625,
+      "learning_rate": 9.731931258429638e-07,
+      "loss": 2.7166,
+      "step": 961
+    },
+    {
+      "epoch": 0.962,
+      "grad_norm": 0.53515625,
+      "learning_rate": 9.252159528556403e-07,
+      "loss": 2.7073,
+      "step": 962
+    },
+    {
+      "epoch": 0.963,
+      "grad_norm": 0.5546875,
+      "learning_rate": 8.784459748458318e-07,
+      "loss": 2.6861,
+      "step": 963
+    },
+    {
+      "epoch": 0.964,
+      "grad_norm": 0.515625,
+      "learning_rate": 8.328837616909613e-07,
+      "loss": 2.7969,
+      "step": 964
+    },
+    {
+      "epoch": 0.965,
+      "grad_norm": 0.53125,
+      "learning_rate": 7.885298685522235e-07,
+      "loss": 2.6999,
+      "step": 965
+    },
+    {
+      "epoch": 0.966,
+      "grad_norm": 0.56640625,
+      "learning_rate": 7.453848358678017e-07,
+      "loss": 2.7762,
+      "step": 966
+    },
+    {
+      "epoch": 0.967,
+      "grad_norm": 0.55078125,
+      "learning_rate": 7.034491893463058e-07,
+      "loss": 2.7231,
+      "step": 967
+    },
+    {
+      "epoch": 0.968,
+      "grad_norm": 0.515625,
+      "learning_rate": 6.627234399603555e-07,
+      "loss": 2.6857,
+      "step": 968
+    },
+    {
+      "epoch": 0.969,
+      "grad_norm": 2.28125,
+      "learning_rate": 6.232080839403631e-07,
+      "loss": 2.707,
+      "step": 969
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.56640625,
+      "learning_rate": 5.849036027684606e-07,
+      "loss": 2.7359,
+      "step": 970
+    },
+    {
+      "epoch": 0.971,
+      "grad_norm": 0.55078125,
+      "learning_rate": 5.478104631726711e-07,
+      "loss": 2.6725,
+      "step": 971
+    },
+    {
+      "epoch": 0.972,
+      "grad_norm": 0.53125,
+      "learning_rate": 5.119291171211793e-07,
+      "loss": 2.762,
+      "step": 972
+    },
+    {
+      "epoch": 0.973,
+      "grad_norm": 0.546875,
+      "learning_rate": 4.772600018168816e-07,
+      "loss": 2.7416,
+      "step": 973
+    },
+    {
+      "epoch": 0.974,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.438035396920004e-07,
+      "loss": 2.7811,
+      "step": 974
+    },
+    {
+      "epoch": 0.975,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.115601384029666e-07,
+      "loss": 2.7719,
+      "step": 975
+    },
+    {
+      "epoch": 0.976,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.805301908254455e-07,
+      "loss": 2.7054,
+      "step": 976
+    },
+    {
+      "epoch": 0.977,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.50714075049563e-07,
+      "loss": 2.8006,
+      "step": 977
+    },
+    {
+      "epoch": 0.978,
+      "grad_norm": 0.58203125,
+      "learning_rate": 3.2211215437528694e-07,
+      "loss": 2.7788,
+      "step": 978
+    },
+    {
+      "epoch": 0.979,
+      "grad_norm": 0.53125,
+      "learning_rate": 2.947247773079753e-07,
+      "loss": 2.6917,
+      "step": 979
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.685522775541904e-07,
+      "loss": 2.7921,
+      "step": 980
+    },
+    {
+      "epoch": 0.981,
+      "grad_norm": 0.58984375,
+      "learning_rate": 2.4359497401758024e-07,
+      "loss": 2.6928,
+      "step": 981
+    },
+    {
+      "epoch": 0.982,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.1985317079500356e-07,
+      "loss": 2.8196,
+      "step": 982
+    },
+    {
+      "epoch": 0.983,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.973271571728441e-07,
+      "loss": 2.6882,
+      "step": 983
+    },
+    {
+      "epoch": 0.984,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.7601720762346897e-07,
+      "loss": 2.7932,
+      "step": 984
+    },
+    {
+      "epoch": 0.985,
+      "grad_norm": 0.79296875,
+      "learning_rate": 1.5592358180189782e-07,
+      "loss": 2.7674,
+      "step": 985
+    },
+    {
+      "epoch": 0.986,
+      "grad_norm": 0.5,
+      "learning_rate": 1.3704652454261668e-07,
+      "loss": 2.7227,
+      "step": 986
+    },
+    {
+      "epoch": 0.987,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.193862658566025e-07,
+      "loss": 2.6739,
+      "step": 987
+    },
+    {
+      "epoch": 0.988,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.0294302092853647e-07,
+      "loss": 2.731,
+      "step": 988
+    },
+    {
+      "epoch": 0.989,
+      "grad_norm": 0.578125,
+      "learning_rate": 8.771699011416168e-08,
+      "loss": 2.8035,
+      "step": 989
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.53515625,
+      "learning_rate": 7.370835893788508e-08,
+      "loss": 2.7294,
+      "step": 990
+    },
+    {
+      "epoch": 0.991,
+      "grad_norm": 0.55078125,
+      "learning_rate": 6.09172980904238e-08,
+      "loss": 2.8871,
+      "step": 991
+    },
+    {
+      "epoch": 0.992,
+      "grad_norm": 0.5390625,
+      "learning_rate": 4.934396342684e-08,
+      "loss": 2.7629,
+      "step": 992
+    },
+    {
+      "epoch": 0.993,
+      "grad_norm": 0.5859375,
+      "learning_rate": 3.898849596456478e-08,
+      "loss": 2.7404,
+      "step": 993
+    },
+    {
+      "epoch": 0.994,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.985102188168831e-08,
+      "loss": 2.7835,
+      "step": 994
+    },
+    {
+      "epoch": 0.995,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.193165251545004e-08,
+      "loss": 2.7149,
+      "step": 995
+    },
+    {
+      "epoch": 0.996,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.5230484360873044e-08,
+      "loss": 2.7078,
+      "step": 996
+    },
+    {
+      "epoch": 0.997,
+      "grad_norm": 0.53515625,
+      "learning_rate": 9.747599069576119e-09,
+      "loss": 2.6312,
+      "step": 997
+    },
+    {
+      "epoch": 0.998,
+      "grad_norm": 0.5078125,
+      "learning_rate": 5.483063448785686e-09,
+      "loss": 2.7117,
+      "step": 998
+    },
+    {
+      "epoch": 0.999,
+      "grad_norm": 0.53515625,
+      "learning_rate": 2.4369294605253166e-09,
+      "loss": 2.7865,
+      "step": 999
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.5859375,
+      "learning_rate": 6.092342209607083e-10,
+      "loss": 2.7628,
+      "step": 1000
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 2.704561948776245,
+      "eval_runtime": 31.803,
+      "eval_samples_per_second": 15.722,
+      "eval_steps_per_second": 2.641,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.818479221014528e+18,
+  "train_batch_size": 6,
+  "trial_name": null,
+  "trial_params": null
+}