diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7041 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.08776958290248525,
+  "eval_steps": 1000,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 8.776958290248525e-05,
+      "grad_norm": 121.5,
+      "learning_rate": 3e-06,
+      "loss": 7.9287,
+      "step": 1
+    },
+    {
+      "epoch": 0.0001755391658049705,
+      "grad_norm": 112.0,
+      "learning_rate": 6e-06,
+      "loss": 7.9365,
+      "step": 2
+    },
+    {
+      "epoch": 0.00026330874870745575,
+      "grad_norm": 73.0,
+      "learning_rate": 9e-06,
+      "loss": 7.9922,
+      "step": 3
+    },
+    {
+      "epoch": 0.000351078331609941,
+      "grad_norm": 67.0,
+      "learning_rate": 1.2e-05,
+      "loss": 8.0557,
+      "step": 4
+    },
+    {
+      "epoch": 0.00043884791451242626,
+      "grad_norm": 82.0,
+      "learning_rate": 1.5e-05,
+      "loss": 7.8389,
+      "step": 5
+    },
+    {
+      "epoch": 0.0005266174974149115,
+      "grad_norm": 66.5,
+      "learning_rate": 1.8e-05,
+      "loss": 7.8574,
+      "step": 6
+    },
+    {
+      "epoch": 0.0006143870803173967,
+      "grad_norm": 158.0,
+      "learning_rate": 2.1000000000000002e-05,
+      "loss": 7.708,
+      "step": 7
+    },
+    {
+      "epoch": 0.000702156663219882,
+      "grad_norm": 76.5,
+      "learning_rate": 2.4e-05,
+      "loss": 7.4805,
+      "step": 8
+    },
+    {
+      "epoch": 0.0007899262461223672,
+      "grad_norm": 76.5,
+      "learning_rate": 2.7e-05,
+      "loss": 7.667,
+      "step": 9
+    },
+    {
+      "epoch": 0.0008776958290248525,
+      "grad_norm": 59.25,
+      "learning_rate": 3e-05,
+      "loss": 7.4551,
+      "step": 10
+    },
+    {
+      "epoch": 0.0009654654119273377,
+      "grad_norm": 60.0,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 7.4072,
+      "step": 11
+    },
+    {
+      "epoch": 0.001053234994829823,
+      "grad_norm": 169.0,
+      "learning_rate": 3.6e-05,
+      "loss": 7.2051,
+      "step": 12
+    },
+    {
+      "epoch": 0.0011410045777323083,
+      "grad_norm": 198.0,
+      "learning_rate": 3.9e-05,
+      "loss": 6.9873,
+      "step": 13
+    },
+    {
+      "epoch": 0.0012287741606347934,
+      "grad_norm": 52.75,
+      "learning_rate": 4.2000000000000004e-05,
+      "loss": 6.8613,
+      "step": 14
+    },
+    {
+      "epoch": 0.0013165437435372787,
+      "grad_norm": 84.5,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 6.665,
+      "step": 15
+    },
+    {
+      "epoch": 0.001404313326439764,
+      "grad_norm": 42.75,
+      "learning_rate": 4.8e-05,
+      "loss": 6.6465,
+      "step": 16
+    },
+    {
+      "epoch": 0.0014920829093422493,
+      "grad_norm": 139.0,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 6.5166,
+      "step": 17
+    },
+    {
+      "epoch": 0.0015798524922447344,
+      "grad_norm": 41.75,
+      "learning_rate": 5.4e-05,
+      "loss": 6.4814,
+      "step": 18
+    },
+    {
+      "epoch": 0.0016676220751472197,
+      "grad_norm": 39.5,
+      "learning_rate": 5.7e-05,
+      "loss": 6.2256,
+      "step": 19
+    },
+    {
+      "epoch": 0.001755391658049705,
+      "grad_norm": 17.125,
+      "learning_rate": 6e-05,
+      "loss": 6.0605,
+      "step": 20
+    },
+    {
+      "epoch": 0.0018431612409521903,
+      "grad_norm": 16.75,
+      "learning_rate": 6.3e-05,
+      "loss": 6.0781,
+      "step": 21
+    },
+    {
+      "epoch": 0.0019309308238546754,
+      "grad_norm": 83.5,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 6.0078,
+      "step": 22
+    },
+    {
+      "epoch": 0.002018700406757161,
+      "grad_norm": 44.75,
+      "learning_rate": 6.9e-05,
+      "loss": 5.9941,
+      "step": 23
+    },
+    {
+      "epoch": 0.002106469989659646,
+      "grad_norm": 25.625,
+      "learning_rate": 7.2e-05,
+      "loss": 5.917,
+      "step": 24
+    },
+    {
+      "epoch": 0.002194239572562131,
+      "grad_norm": 65.5,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 5.7607,
+      "step": 25
+    },
+    {
+      "epoch": 0.0022820091554646166,
+      "grad_norm": 16.875,
+      "learning_rate": 7.8e-05,
+      "loss": 5.7617,
+      "step": 26
+    },
+    {
+      "epoch": 0.0023697787383671017,
+      "grad_norm": 22.625,
+      "learning_rate": 8.1e-05,
+      "loss": 5.8018,
+      "step": 27
+    },
+    {
+      "epoch": 0.002457548321269587,
+      "grad_norm": 27.375,
+      "learning_rate": 8.400000000000001e-05,
+      "loss": 5.6592,
+      "step": 28
+    },
+    {
+      "epoch": 0.0025453179041720723,
+      "grad_norm": 28.875,
+      "learning_rate": 8.7e-05,
+      "loss": 5.5781,
+      "step": 29
+    },
+    {
+      "epoch": 0.0026330874870745574,
+      "grad_norm": 25.875,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 5.4229,
+      "step": 30
+    },
+    {
+      "epoch": 0.002720857069977043,
+      "grad_norm": 256.0,
+      "learning_rate": 9.3e-05,
+      "loss": 5.373,
+      "step": 31
+    },
+    {
+      "epoch": 0.002808626652879528,
+      "grad_norm": 21.5,
+      "learning_rate": 9.6e-05,
+      "loss": 5.374,
+      "step": 32
+    },
+    {
+      "epoch": 0.002896396235782013,
+      "grad_norm": 15.625,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 5.3916,
+      "step": 33
+    },
+    {
+      "epoch": 0.0029841658186844987,
+      "grad_norm": 17.625,
+      "learning_rate": 0.00010200000000000001,
+      "loss": 5.2695,
+      "step": 34
+    },
+    {
+      "epoch": 0.0030719354015869837,
+      "grad_norm": 26.75,
+      "learning_rate": 0.00010500000000000002,
+      "loss": 5.1943,
+      "step": 35
+    },
+    {
+      "epoch": 0.003159704984489469,
+      "grad_norm": 18.375,
+      "learning_rate": 0.000108,
+      "loss": 5.2988,
+      "step": 36
+    },
+    {
+      "epoch": 0.0032474745673919544,
+      "grad_norm": 15.375,
+      "learning_rate": 0.000111,
+      "loss": 5.1738,
+      "step": 37
+    },
+    {
+      "epoch": 0.0033352441502944394,
+      "grad_norm": 3.1875,
+      "learning_rate": 0.000114,
+      "loss": 5.0938,
+      "step": 38
+    },
+    {
+      "epoch": 0.003423013733196925,
+      "grad_norm": 13.875,
+      "learning_rate": 0.000117,
+      "loss": 5.0488,
+      "step": 39
+    },
+    {
+      "epoch": 0.00351078331609941,
+      "grad_norm": 18.375,
+      "learning_rate": 0.00012,
+      "loss": 5.0342,
+      "step": 40
+    },
+    {
+      "epoch": 0.003598552899001895,
+      "grad_norm": 13.25,
+      "learning_rate": 0.000123,
+      "loss": 4.9072,
+      "step": 41
+    },
+    {
+      "epoch": 0.0036863224819043807,
+      "grad_norm": 14.375,
+      "learning_rate": 0.000126,
+      "loss": 5.0068,
+      "step": 42
+    },
+    {
+      "epoch": 0.0037740920648068658,
+      "grad_norm": 8.5625,
+      "learning_rate": 0.000129,
+      "loss": 4.9346,
+      "step": 43
+    },
+    {
+      "epoch": 0.003861861647709351,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 4.9004,
+      "step": 44
+    },
+    {
+      "epoch": 0.003949631230611836,
+      "grad_norm": 44.5,
+      "learning_rate": 0.000135,
+      "loss": 4.96,
+      "step": 45
+    },
+    {
+      "epoch": 0.004037400813514322,
+      "grad_norm": 2.09375,
+      "learning_rate": 0.000138,
+      "loss": 4.8945,
+      "step": 46
+    },
+    {
+      "epoch": 0.0041251703964168066,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.000141,
+      "loss": 4.9229,
+      "step": 47
+    },
+    {
+      "epoch": 0.004212939979319292,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.000144,
+      "loss": 4.8545,
+      "step": 48
+    },
+    {
+      "epoch": 0.004300709562221778,
+      "grad_norm": 1.6484375,
+      "learning_rate": 0.000147,
+      "loss": 4.8369,
+      "step": 49
+    },
+    {
+      "epoch": 0.004388479145124262,
+      "grad_norm": 6.8125,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 4.9775,
+      "step": 50
+    },
+    {
+      "epoch": 0.004476248728026748,
+      "grad_norm": 5.0,
+      "learning_rate": 0.000153,
+      "loss": 4.9795,
+      "step": 51
+    },
+    {
+      "epoch": 0.004564018310929233,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.000156,
+      "loss": 4.7822,
+      "step": 52
+    },
+    {
+      "epoch": 0.004651787893831718,
+      "grad_norm": 7.40625,
+      "learning_rate": 0.000159,
+      "loss": 4.8193,
+      "step": 53
+    },
+    {
+      "epoch": 0.0047395574767342035,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.000162,
+      "loss": 4.8438,
+      "step": 54
+    },
+    {
+      "epoch": 0.004827327059636689,
+      "grad_norm": 6.0,
+      "learning_rate": 0.000165,
+      "loss": 4.8115,
+      "step": 55
+    },
+    {
+      "epoch": 0.004915096642539174,
+      "grad_norm": 6.75,
+      "learning_rate": 0.00016800000000000002,
+      "loss": 4.7568,
+      "step": 56
+    },
+    {
+      "epoch": 0.005002866225441659,
+      "grad_norm": 8.125,
+      "learning_rate": 0.000171,
+      "loss": 4.8145,
+      "step": 57
+    },
+    {
+      "epoch": 0.005090635808344145,
+      "grad_norm": 12.125,
+      "learning_rate": 0.000174,
+      "loss": 4.8076,
+      "step": 58
+    },
+    {
+      "epoch": 0.005178405391246629,
+      "grad_norm": 11.875,
+      "learning_rate": 0.000177,
+      "loss": 4.7861,
+      "step": 59
+    },
+    {
+      "epoch": 0.005266174974149115,
+      "grad_norm": 1.7109375,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 4.7412,
+      "step": 60
+    },
+    {
+      "epoch": 0.0053539445570516,
+      "grad_norm": 5.875,
+      "learning_rate": 0.000183,
+      "loss": 4.7051,
+      "step": 61
+    },
+    {
+      "epoch": 0.005441714139954086,
+      "grad_norm": 1.71875,
+      "learning_rate": 0.000186,
+      "loss": 4.7061,
+      "step": 62
+    },
+    {
+      "epoch": 0.005529483722856571,
+      "grad_norm": 5.90625,
+      "learning_rate": 0.000189,
+      "loss": 4.7012,
+      "step": 63
+    },
+    {
+      "epoch": 0.005617253305759056,
+      "grad_norm": 4.375,
+      "learning_rate": 0.000192,
+      "loss": 4.7412,
+      "step": 64
+    },
+    {
+      "epoch": 0.005705022888661542,
+      "grad_norm": 2.28125,
+      "learning_rate": 0.00019500000000000002,
+      "loss": 4.708,
+      "step": 65
+    },
+    {
+      "epoch": 0.005792792471564026,
+      "grad_norm": 5.5,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 4.7158,
+      "step": 66
+    },
+    {
+      "epoch": 0.005880562054466512,
+      "grad_norm": 3.375,
+      "learning_rate": 0.000201,
+      "loss": 4.6943,
+      "step": 67
+    },
+    {
+      "epoch": 0.005968331637368997,
+      "grad_norm": 2.1875,
+      "learning_rate": 0.00020400000000000003,
+      "loss": 4.6689,
+      "step": 68
+    },
+    {
+      "epoch": 0.006056101220271482,
+      "grad_norm": 5.625,
+      "learning_rate": 0.00020700000000000002,
+      "loss": 4.7139,
+      "step": 69
+    },
+    {
+      "epoch": 0.0061438708031739675,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.00021000000000000004,
+      "loss": 4.748,
+      "step": 70
+    },
+    {
+      "epoch": 0.006231640386076453,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 4.7041,
+      "step": 71
+    },
+    {
+      "epoch": 0.006319409968978938,
+      "grad_norm": 2.25,
+      "learning_rate": 0.000216,
+      "loss": 4.709,
+      "step": 72
+    },
+    {
+      "epoch": 0.006407179551881423,
+      "grad_norm": 13.25,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 4.8428,
+      "step": 73
+    },
+    {
+      "epoch": 0.006494949134783909,
+      "grad_norm": 14.5,
+      "learning_rate": 0.000222,
+      "loss": 4.7979,
+      "step": 74
+    },
+    {
+      "epoch": 0.006582718717686393,
+      "grad_norm": 7.4375,
+      "learning_rate": 0.000225,
+      "loss": 4.6299,
+      "step": 75
+    },
+    {
+      "epoch": 0.006670488300588879,
+      "grad_norm": 13.75,
+      "learning_rate": 0.000228,
+      "loss": 4.7627,
+      "step": 76
+    },
+    {
+      "epoch": 0.006758257883491364,
+      "grad_norm": 15.1875,
+      "learning_rate": 0.000231,
+      "loss": 4.8945,
+      "step": 77
+    },
+    {
+      "epoch": 0.00684602746639385,
+      "grad_norm": 12.5,
+      "learning_rate": 0.000234,
+      "loss": 4.7734,
+      "step": 78
+    },
+    {
+      "epoch": 0.006933797049296335,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.00023700000000000001,
+      "loss": 4.6152,
+      "step": 79
+    },
+    {
+      "epoch": 0.00702156663219882,
+      "grad_norm": 11.125,
+      "learning_rate": 0.00024,
+      "loss": 4.7637,
+      "step": 80
+    },
+    {
+      "epoch": 0.007109336215101306,
+      "grad_norm": 11.5,
+      "learning_rate": 0.00024300000000000002,
+      "loss": 4.7783,
+      "step": 81
+    },
+    {
+      "epoch": 0.00719710579800379,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.000246,
+      "loss": 4.6436,
+      "step": 82
+    },
+    {
+      "epoch": 0.007284875380906276,
+      "grad_norm": 12.6875,
+      "learning_rate": 0.00024900000000000004,
+      "loss": 4.8057,
+      "step": 83
+    },
+    {
+      "epoch": 0.007372644963808761,
+      "grad_norm": 14.0625,
+      "learning_rate": 0.000252,
+      "loss": 4.8232,
+      "step": 84
+    },
+    {
+      "epoch": 0.007460414546711246,
+      "grad_norm": 10.875,
+      "learning_rate": 0.000255,
+      "loss": 4.6621,
+      "step": 85
+    },
+    {
+      "epoch": 0.0075481841296137315,
+      "grad_norm": 1.5625,
+      "learning_rate": 0.000258,
+      "loss": 4.6309,
+      "step": 86
+    },
+    {
+      "epoch": 0.007635953712516217,
+      "grad_norm": 6.5,
+      "learning_rate": 0.000261,
+      "loss": 4.6016,
+      "step": 87
+    },
+    {
+      "epoch": 0.007723723295418702,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 4.6582,
+      "step": 88
+    },
+    {
+      "epoch": 0.007811492878321187,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.000267,
+      "loss": 4.6748,
+      "step": 89
+    },
+    {
+      "epoch": 0.007899262461223673,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00027,
+      "loss": 4.6484,
+      "step": 90
+    },
+    {
+      "epoch": 0.007987032044126157,
+      "grad_norm": 1.8828125,
+      "learning_rate": 0.000273,
+      "loss": 4.6318,
+      "step": 91
+    },
+    {
+      "epoch": 0.008074801627028644,
+      "grad_norm": 6.25,
+      "learning_rate": 0.000276,
+      "loss": 4.6748,
+      "step": 92
+    },
+    {
+      "epoch": 0.008162571209931128,
+      "grad_norm": 6.03125,
+      "learning_rate": 0.000279,
+      "loss": 4.6309,
+      "step": 93
+    },
+    {
+      "epoch": 0.008250340792833613,
+      "grad_norm": 2.09375,
+      "learning_rate": 0.000282,
+      "loss": 4.6621,
+      "step": 94
+    },
+    {
+      "epoch": 0.0083381103757361,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.000285,
+      "loss": 4.6953,
+      "step": 95
+    },
+    {
+      "epoch": 0.008425879958638584,
+      "grad_norm": 3.984375,
+      "learning_rate": 0.000288,
+      "loss": 4.6104,
+      "step": 96
+    },
+    {
+      "epoch": 0.008513649541541069,
+      "grad_norm": 2.546875,
+      "learning_rate": 0.000291,
+      "loss": 4.6055,
+      "step": 97
+    },
+    {
+      "epoch": 0.008601419124443555,
+      "grad_norm": 2.484375,
+      "learning_rate": 0.000294,
+      "loss": 4.6123,
+      "step": 98
+    },
+    {
+      "epoch": 0.00868918870734604,
+      "grad_norm": 2.328125,
+      "learning_rate": 0.000297,
+      "loss": 4.5986,
+      "step": 99
+    },
+    {
+      "epoch": 0.008776958290248525,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 4.5312,
+      "step": 100
+    },
+    {
+      "epoch": 0.008864727873151011,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.00030300000000000005,
+      "loss": 4.6201,
+      "step": 101
+    },
+    {
+      "epoch": 0.008952497456053496,
+      "grad_norm": 2.515625,
+      "learning_rate": 0.000306,
+      "loss": 4.626,
+      "step": 102
+    },
+    {
+      "epoch": 0.00904026703895598,
+      "grad_norm": 3.140625,
+      "learning_rate": 0.000309,
+      "loss": 4.624,
+      "step": 103
+    },
+    {
+      "epoch": 0.009128036621858467,
+      "grad_norm": 2.734375,
+      "learning_rate": 0.000312,
+      "loss": 4.5771,
+      "step": 104
+    },
+    {
+      "epoch": 0.009215806204760951,
+      "grad_norm": 2.9375,
+      "learning_rate": 0.000315,
+      "loss": 4.6221,
+      "step": 105
+    },
+    {
+      "epoch": 0.009303575787663436,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.000318,
+      "loss": 4.5703,
+      "step": 106
+    },
+    {
+      "epoch": 0.009391345370565922,
+      "grad_norm": 2.09375,
+      "learning_rate": 0.000321,
+      "loss": 4.5283,
+      "step": 107
+    },
+    {
+      "epoch": 0.009479114953468407,
+      "grad_norm": 1.9765625,
+      "learning_rate": 0.000324,
+      "loss": 4.6152,
+      "step": 108
+    },
+    {
+      "epoch": 0.009566884536370892,
+      "grad_norm": 2.765625,
+      "learning_rate": 0.000327,
+      "loss": 4.5557,
+      "step": 109
+    },
+    {
+      "epoch": 0.009654654119273378,
+      "grad_norm": 2.09375,
+      "learning_rate": 0.00033,
+      "loss": 4.6279,
+      "step": 110
+    },
+    {
+      "epoch": 0.009742423702175863,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.000333,
+      "loss": 4.542,
+      "step": 111
+    },
+    {
+      "epoch": 0.009830193285078347,
+      "grad_norm": 2.71875,
+      "learning_rate": 0.00033600000000000004,
+      "loss": 4.5586,
+      "step": 112
+    },
+    {
+      "epoch": 0.009917962867980834,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.000339,
+      "loss": 4.5127,
+      "step": 113
+    },
+    {
+      "epoch": 0.010005732450883318,
+      "grad_norm": 1.9609375,
+      "learning_rate": 0.000342,
+      "loss": 4.5742,
+      "step": 114
+    },
+    {
+      "epoch": 0.010093502033785803,
+      "grad_norm": 2.296875,
+      "learning_rate": 0.00034500000000000004,
+      "loss": 4.5088,
+      "step": 115
+    },
+    {
+      "epoch": 0.01018127161668829,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.000348,
+      "loss": 4.543,
+      "step": 116
+    },
+    {
+      "epoch": 0.010269041199590774,
+      "grad_norm": 3.140625,
+      "learning_rate": 0.000351,
+      "loss": 4.4609,
+      "step": 117
+    },
+    {
+      "epoch": 0.010356810782493259,
+      "grad_norm": 2.609375,
+      "learning_rate": 0.000354,
+      "loss": 4.5303,
+      "step": 118
+    },
+    {
+      "epoch": 0.010444580365395745,
+      "grad_norm": 2.765625,
+      "learning_rate": 0.000357,
+      "loss": 4.4971,
+      "step": 119
+    },
+    {
+      "epoch": 0.01053234994829823,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 4.5342,
+      "step": 120
+    },
+    {
+      "epoch": 0.010620119531200716,
+      "grad_norm": 2.640625,
+      "learning_rate": 0.000363,
+      "loss": 4.4941,
+      "step": 121
+    },
+    {
+      "epoch": 0.0107078891141032,
+      "grad_norm": 1.9453125,
+      "learning_rate": 0.000366,
+      "loss": 4.4658,
+      "step": 122
+    },
+    {
+      "epoch": 0.010795658697005685,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.000369,
+      "loss": 4.4941,
+      "step": 123
+    },
+    {
+      "epoch": 0.010883428279908172,
+      "grad_norm": 2.59375,
+      "learning_rate": 0.000372,
+      "loss": 4.4248,
+      "step": 124
+    },
+    {
+      "epoch": 0.010971197862810656,
+      "grad_norm": 1.84375,
+      "learning_rate": 0.000375,
+      "loss": 4.4961,
+      "step": 125
+    },
+    {
+      "epoch": 0.011058967445713141,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.000378,
+      "loss": 4.4292,
+      "step": 126
+    },
+    {
+      "epoch": 0.011146737028615628,
+      "grad_norm": 2.96875,
+      "learning_rate": 0.000381,
+      "loss": 4.5234,
+      "step": 127
+    },
+    {
+      "epoch": 0.011234506611518112,
+      "grad_norm": 2.640625,
+      "learning_rate": 0.000384,
+      "loss": 4.5049,
+      "step": 128
+    },
+    {
+      "epoch": 0.011322276194420597,
+      "grad_norm": 2.0625,
+      "learning_rate": 0.00038700000000000003,
+      "loss": 4.4512,
+      "step": 129
+    },
+    {
+      "epoch": 0.011410045777323083,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.00039000000000000005,
+      "loss": 4.4355,
+      "step": 130
+    },
+    {
+      "epoch": 0.011497815360225568,
+      "grad_norm": 1.4765625,
+      "learning_rate": 0.000393,
+      "loss": 4.4238,
+      "step": 131
+    },
+    {
+      "epoch": 0.011585584943128053,
+      "grad_norm": 2.046875,
+      "learning_rate": 0.00039600000000000003,
+      "loss": 4.5137,
+      "step": 132
+    },
+    {
+      "epoch": 0.011673354526030539,
+      "grad_norm": 1.734375,
+      "learning_rate": 0.00039900000000000005,
+      "loss": 4.4004,
+      "step": 133
+    },
+    {
+      "epoch": 0.011761124108933024,
+      "grad_norm": 2.21875,
+      "learning_rate": 0.000402,
+      "loss": 4.4141,
+      "step": 134
+    },
+    {
+      "epoch": 0.011848893691835508,
+      "grad_norm": 1.75,
+      "learning_rate": 0.00040500000000000003,
+      "loss": 4.4688,
+      "step": 135
+    },
+    {
+      "epoch": 0.011936663274737995,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00040800000000000005,
+      "loss": 4.4326,
+      "step": 136
+    },
+    {
+      "epoch": 0.01202443285764048,
+      "grad_norm": 2.90625,
+      "learning_rate": 0.000411,
+      "loss": 4.5049,
+      "step": 137
+    },
+    {
+      "epoch": 0.012112202440542964,
+      "grad_norm": 2.203125,
+      "learning_rate": 0.00041400000000000003,
+      "loss": 4.5088,
+      "step": 138
+    },
+    {
+      "epoch": 0.01219997202344545,
+      "grad_norm": 2.421875,
+      "learning_rate": 0.00041700000000000005,
+      "loss": 4.4717,
+      "step": 139
+    },
+    {
+      "epoch": 0.012287741606347935,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.00042000000000000007,
+      "loss": 4.4717,
+      "step": 140
+    },
+    {
+      "epoch": 0.01237551118925042,
+      "grad_norm": 1.8359375,
+      "learning_rate": 0.000423,
+      "loss": 4.3867,
+      "step": 141
+    },
+    {
+      "epoch": 0.012463280772152906,
+      "grad_norm": 1.390625,
+      "learning_rate": 0.00042599999999999995,
+      "loss": 4.3896,
+      "step": 142
+    },
+    {
+      "epoch": 0.01255105035505539,
+      "grad_norm": 1.9609375,
+      "learning_rate": 0.00042899999999999997,
+      "loss": 4.3428,
+      "step": 143
+    },
+    {
+      "epoch": 0.012638819937957875,
+      "grad_norm": 2.1875,
+      "learning_rate": 0.000432,
+      "loss": 4.4297,
+      "step": 144
+    },
+    {
+      "epoch": 0.012726589520860362,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.000435,
+      "loss": 4.418,
+      "step": 145
+    },
+    {
+      "epoch": 0.012814359103762846,
+      "grad_norm": 1.890625,
+      "learning_rate": 0.00043799999999999997,
+      "loss": 4.3877,
+      "step": 146
+    },
+    {
+      "epoch": 0.012902128686665331,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.000441,
+      "loss": 4.415,
+      "step": 147
+    },
+    {
+      "epoch": 0.012989898269567817,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.000444,
+      "loss": 4.4492,
+      "step": 148
+    },
+    {
+      "epoch": 0.013077667852470302,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.00044699999999999997,
+      "loss": 4.4111,
+      "step": 149
+    },
+    {
+      "epoch": 0.013165437435372787,
+      "grad_norm": 2.21875,
+      "learning_rate": 0.00045,
+      "loss": 4.3945,
+      "step": 150
+    },
+    {
+      "epoch": 0.013253207018275273,
+      "grad_norm": 2.234375,
+      "learning_rate": 0.000453,
+      "loss": 4.3818,
+      "step": 151
+    },
+    {
+      "epoch": 0.013340976601177758,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.000456,
+      "loss": 4.3496,
+      "step": 152
+    },
+    {
+      "epoch": 0.013428746184080244,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.000459,
+      "loss": 4.418,
+      "step": 153
+    },
+    {
+      "epoch": 0.013516515766982729,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.000462,
+      "loss": 4.4609,
+      "step": 154
+    },
+    {
+      "epoch": 0.013604285349885213,
+      "grad_norm": 6.5,
+      "learning_rate": 0.000465,
+      "loss": 4.4658,
+      "step": 155
+    },
+    {
+      "epoch": 0.0136920549327877,
+      "grad_norm": 5.8125,
+      "learning_rate": 0.000468,
+      "loss": 4.4727,
+      "step": 156
+    },
+    {
+      "epoch": 0.013779824515690185,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.000471,
+      "loss": 4.3848,
+      "step": 157
+    },
+    {
+      "epoch": 0.01386759409859267,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00047400000000000003,
+      "loss": 4.3906,
+      "step": 158
+    },
+    {
+      "epoch": 0.013955363681495156,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.000477,
+      "loss": 4.373,
+      "step": 159
+    },
+    {
+      "epoch": 0.01404313326439764,
+      "grad_norm": 3.671875,
+      "learning_rate": 0.00048,
+      "loss": 4.3984,
+      "step": 160
+    },
+    {
+      "epoch": 0.014130902847300125,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.00048300000000000003,
+      "loss": 4.4072,
+      "step": 161
+    },
+    {
+      "epoch": 0.014218672430202611,
+      "grad_norm": 3.703125,
+      "learning_rate": 0.00048600000000000005,
+      "loss": 4.418,
+      "step": 162
+    },
+    {
+      "epoch": 0.014306442013105096,
+      "grad_norm": 2.671875,
+      "learning_rate": 0.0004890000000000001,
+      "loss": 4.4209,
+      "step": 163
+    },
+    {
+      "epoch": 0.01439421159600758,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.000492,
+      "loss": 4.374,
+      "step": 164
+    },
+    {
+      "epoch": 0.014481981178910067,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.000495,
+      "loss": 4.4346,
+      "step": 165
+    },
+    {
+      "epoch": 0.014569750761812552,
+      "grad_norm": 1.828125,
+      "learning_rate": 0.0004980000000000001,
+      "loss": 4.3232,
+      "step": 166
+    },
+    {
+      "epoch": 0.014657520344715036,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.000501,
+      "loss": 4.332,
+      "step": 167
+    },
+    {
+      "epoch": 0.014745289927617523,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.000504,
+      "loss": 4.4102,
+      "step": 168
+    },
+    {
+      "epoch": 0.014833059510520007,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.0005070000000000001,
+      "loss": 4.3809,
+      "step": 169
+    },
+    {
+      "epoch": 0.014920829093422492,
+      "grad_norm": 2.828125,
+      "learning_rate": 0.00051,
+      "loss": 4.3984,
+      "step": 170
+    },
+    {
+      "epoch": 0.015008598676324978,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.000513,
+      "loss": 4.3887,
+      "step": 171
+    },
+    {
+      "epoch": 0.015096368259227463,
+      "grad_norm": 3.09375,
+      "learning_rate": 0.000516,
+      "loss": 4.3398,
+      "step": 172
+    },
+    {
+      "epoch": 0.015184137842129948,
+      "grad_norm": 2.625,
+      "learning_rate": 0.0005189999999999999,
+      "loss": 4.418,
+      "step": 173
+    },
+    {
+      "epoch": 0.015271907425032434,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.000522,
+      "loss": 4.3408,
+      "step": 174
+    },
+    {
+      "epoch": 0.015359677007934919,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.000525,
+      "loss": 4.4092,
+      "step": 175
+    },
+    {
+      "epoch": 0.015447446590837403,
+      "grad_norm": 1.5234375,
+      "learning_rate": 0.0005279999999999999,
+      "loss": 4.3594,
+      "step": 176
+    },
+    {
+      "epoch": 0.01553521617373989,
+      "grad_norm": 1.8828125,
+      "learning_rate": 0.000531,
+      "loss": 4.3291,
+      "step": 177
+    },
+    {
+      "epoch": 0.015622985756642374,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.000534,
+      "loss": 4.3228,
+      "step": 178
+    },
+    {
+      "epoch": 0.01571075533954486,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.000537,
+      "loss": 4.3418,
+      "step": 179
+    },
+    {
+      "epoch": 0.015798524922447345,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00054,
+      "loss": 4.3486,
+      "step": 180
+    },
+    {
+      "epoch": 0.01588629450534983,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.000543,
+      "loss": 4.3457,
+      "step": 181
+    },
+    {
+      "epoch": 0.015974064088252315,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.000546,
+      "loss": 4.3867,
+      "step": 182
+    },
+    {
+      "epoch": 0.0160618336711548,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.000549,
+      "loss": 4.3555,
+      "step": 183
+    },
+    {
+      "epoch": 0.016149603254057288,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.000552,
+      "loss": 4.334,
+      "step": 184
+    },
+    {
+      "epoch": 0.016237372836959772,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.000555,
+      "loss": 4.292,
+      "step": 185
+    },
+    {
+      "epoch": 0.016325142419862257,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.000558,
+      "loss": 4.3467,
+      "step": 186
+    },
+    {
+      "epoch": 0.01641291200276474,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.000561,
+      "loss": 4.3145,
+      "step": 187
+    },
+    {
+      "epoch": 0.016500681585667226,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.000564,
+      "loss": 4.2754,
+      "step": 188
+    },
+    {
+      "epoch": 0.01658845116856971,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.000567,
+      "loss": 4.2783,
+      "step": 189
+    },
+    {
+      "epoch": 0.0166762207514722,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00057,
+      "loss": 4.269,
+      "step": 190
+    },
+    {
+      "epoch": 0.016763990334374684,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.000573,
+      "loss": 4.332,
+      "step": 191
+    },
+    {
+      "epoch": 0.01685175991727717,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.000576,
+      "loss": 4.3018,
+      "step": 192
+    },
+    {
+      "epoch": 0.016939529500179653,
+      "grad_norm": 0.625,
+      "learning_rate": 0.000579,
+      "loss": 4.2607,
+      "step": 193
+    },
+    {
+      "epoch": 0.017027299083082138,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.000582,
+      "loss": 4.3398,
+      "step": 194
+    },
+    {
+      "epoch": 0.017115068665984622,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.000585,
+      "loss": 4.2788,
+      "step": 195
+    },
+    {
+      "epoch": 0.01720283824888711,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.000588,
+      "loss": 4.2266,
+      "step": 196
+    },
+    {
+      "epoch": 0.017290607831789595,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.000591,
+      "loss": 4.3252,
+      "step": 197
+    },
+    {
+      "epoch": 0.01737837741469208,
+      "grad_norm": 2.375,
+      "learning_rate": 0.000594,
+      "loss": 4.2686,
+      "step": 198
+    },
+    {
+      "epoch": 0.017466146997594564,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.0005970000000000001,
+      "loss": 4.2939,
+      "step": 199
+    },
+    {
+      "epoch": 0.01755391658049705,
+      "grad_norm": 1.84375,
+      "learning_rate": 0.0006000000000000001,
+      "loss": 4.2803,
+      "step": 200
+    },
+    {
+      "epoch": 0.017641686163399534,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.000603,
+      "loss": 4.2744,
+      "step": 201
+    },
+    {
+      "epoch": 0.017729455746302022,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0006060000000000001,
+      "loss": 4.292,
+      "step": 202
+    },
+    {
+      "epoch": 0.017817225329204506,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.0006090000000000001,
+      "loss": 4.2046,
+      "step": 203
+    },
+    {
+      "epoch": 0.01790499491210699,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.000612,
+      "loss": 4.2178,
+      "step": 204
+    },
+    {
+      "epoch": 0.017992764495009476,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.000615,
+      "loss": 4.2183,
+      "step": 205
+    },
+    {
+      "epoch": 0.01808053407791196,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.000618,
+      "loss": 4.2139,
+      "step": 206
+    },
+    {
+      "epoch": 0.01816830366081445,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.000621,
+      "loss": 4.2363,
+      "step": 207
+    },
+    {
+      "epoch": 0.018256073243716933,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.000624,
+      "loss": 4.2891,
+      "step": 208
+    },
+    {
+      "epoch": 0.018343842826619418,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.000627,
+      "loss": 4.2212,
+      "step": 209
+    },
+    {
+      "epoch": 0.018431612409521902,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00063,
+      "loss": 4.3203,
+      "step": 210
+    },
+    {
+      "epoch": 0.018519381992424387,
+      "grad_norm": 2.109375,
+      "learning_rate": 0.000633,
+      "loss": 4.3159,
+      "step": 211
+    },
+    {
+      "epoch": 0.018607151575326872,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.000636,
+      "loss": 4.272,
+      "step": 212
+    },
+    {
+      "epoch": 0.01869492115822936,
+      "grad_norm": 3.0625,
+      "learning_rate": 0.000639,
+      "loss": 4.2368,
+      "step": 213
+    },
+    {
+      "epoch": 0.018782690741131845,
+      "grad_norm": 2.875,
+      "learning_rate": 0.000642,
+      "loss": 4.2749,
+      "step": 214
+    },
+    {
+      "epoch": 0.01887046032403433,
+      "grad_norm": 1.8515625,
+      "learning_rate": 0.000645,
+      "loss": 4.1772,
+      "step": 215
+    },
+    {
+      "epoch": 0.018958229906936814,
+      "grad_norm": 2.125,
+      "learning_rate": 0.000648,
+      "loss": 4.3271,
+      "step": 216
+    },
+    {
+      "epoch": 0.0190459994898393,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.000651,
+      "loss": 4.2402,
+      "step": 217
+    },
+    {
+      "epoch": 0.019133769072741783,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.000654,
+      "loss": 4.2407,
+      "step": 218
+    },
+    {
+      "epoch": 0.01922153865564427,
+      "grad_norm": 1.84375,
+      "learning_rate": 0.000657,
+      "loss": 4.2031,
+      "step": 219
+    },
+    {
+      "epoch": 0.019309308238546756,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00066,
+      "loss": 4.2534,
+      "step": 220
+    },
+    {
+      "epoch": 0.01939707782144924,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.0006630000000000001,
+      "loss": 4.2451,
+      "step": 221
+    },
+    {
+      "epoch": 0.019484847404351725,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.000666,
+      "loss": 4.3438,
+      "step": 222
+    },
+    {
+      "epoch": 0.01957261698725421,
+      "grad_norm": 1.5625,
+      "learning_rate": 0.000669,
+      "loss": 4.2666,
+      "step": 223
+    },
+    {
+      "epoch": 0.019660386570156695,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.0006720000000000001,
+      "loss": 4.2119,
+      "step": 224
+    },
+    {
+      "epoch": 0.019748156153059183,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.000675,
+      "loss": 4.1357,
+      "step": 225
+    },
+    {
+      "epoch": 0.019835925735961667,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.000678,
+      "loss": 4.2373,
+      "step": 226
+    },
+    {
+      "epoch": 0.019923695318864152,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.0006810000000000001,
+      "loss": 4.2134,
+      "step": 227
+    },
+    {
+      "epoch": 0.020011464901766637,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.000684,
+      "loss": 4.1768,
+      "step": 228
+    },
+    {
+      "epoch": 0.02009923448466912,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.000687,
+      "loss": 4.165,
+      "step": 229
+    },
+    {
+      "epoch": 0.020187004067571606,
+      "grad_norm": 1.9609375,
+      "learning_rate": 0.0006900000000000001,
+      "loss": 4.1826,
+      "step": 230
+    },
+    {
+      "epoch": 0.020274773650474094,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.000693,
+      "loss": 4.1729,
+      "step": 231
+    },
+    {
+      "epoch": 0.02036254323337658,
+      "grad_norm": 2.265625,
+      "learning_rate": 0.000696,
+      "loss": 4.1919,
+      "step": 232
+    },
+    {
+      "epoch": 0.020450312816279063,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.0006990000000000001,
+      "loss": 4.2305,
+      "step": 233
+    },
+    {
+      "epoch": 0.020538082399181548,
+      "grad_norm": 2.71875,
+      "learning_rate": 0.000702,
+      "loss": 4.2744,
+      "step": 234
+    },
+    {
+      "epoch": 0.020625851982084033,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.000705,
+      "loss": 4.1924,
+      "step": 235
+    },
+    {
+      "epoch": 0.020713621564986517,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.000708,
+      "loss": 4.2466,
+      "step": 236
+    },
+    {
+      "epoch": 0.020801391147889006,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0007109999999999999,
+      "loss": 4.312,
+      "step": 237
+    },
+    {
+      "epoch": 0.02088916073079149,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.000714,
+      "loss": 4.1802,
+      "step": 238
+    },
+    {
+      "epoch": 0.020976930313693975,
+      "grad_norm": 2.75,
+      "learning_rate": 0.000717,
+      "loss": 4.3037,
+      "step": 239
+    },
+    {
+      "epoch": 0.02106469989659646,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 4.2271,
+      "step": 240
+    },
+    {
+      "epoch": 0.021152469479498944,
+      "grad_norm": 2.890625,
+      "learning_rate": 0.000723,
+      "loss": 4.2114,
+      "step": 241
+    },
+    {
+      "epoch": 0.021240239062401432,
+      "grad_norm": 1.65625,
+      "learning_rate": 0.000726,
+      "loss": 4.2104,
+      "step": 242
+    },
+    {
+      "epoch": 0.021328008645303917,
+      "grad_norm": 2.53125,
+      "learning_rate": 0.000729,
+      "loss": 4.1665,
+      "step": 243
+    },
+    {
+      "epoch": 0.0214157782282064,
+      "grad_norm": 1.53125,
+      "learning_rate": 0.000732,
+      "loss": 4.1475,
+      "step": 244
+    },
+    {
+      "epoch": 0.021503547811108886,
+      "grad_norm": 2.015625,
+      "learning_rate": 0.000735,
+      "loss": 4.2339,
+      "step": 245
+    },
+    {
+      "epoch": 0.02159131739401137,
+      "grad_norm": 1.125,
+      "learning_rate": 0.000738,
+      "loss": 4.2329,
+      "step": 246
+    },
+    {
+      "epoch": 0.021679086976913856,
+      "grad_norm": 2.609375,
+      "learning_rate": 0.000741,
+      "loss": 4.21,
+      "step": 247
+    },
+    {
+      "epoch": 0.021766856559816344,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.000744,
+      "loss": 4.2168,
+      "step": 248
+    },
+    {
+      "epoch": 0.02185462614271883,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.000747,
+      "loss": 4.1685,
+      "step": 249
+    },
+    {
+      "epoch": 0.021942395725621313,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.00075,
+      "loss": 4.1772,
+      "step": 250
+    },
+    {
+      "epoch": 0.022030165308523798,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.000753,
+      "loss": 4.1802,
+      "step": 251
+    },
+    {
+      "epoch": 0.022117934891426282,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.000756,
+      "loss": 4.1489,
+      "step": 252
+    },
+    {
+      "epoch": 0.022205704474328767,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.000759,
+      "loss": 4.1973,
+      "step": 253
+    },
+    {
+      "epoch": 0.022293474057231255,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.000762,
+      "loss": 4.1558,
+      "step": 254
+    },
+    {
+      "epoch": 0.02238124364013374,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.0007650000000000001,
+      "loss": 4.2026,
+      "step": 255
+    },
+    {
+      "epoch": 0.022469013223036224,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.000768,
+      "loss": 4.123,
+      "step": 256
+    },
+    {
+      "epoch": 0.02255678280593871,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.000771,
+      "loss": 4.1147,
+      "step": 257
+    },
+    {
+      "epoch": 0.022644552388841194,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.0007740000000000001,
+      "loss": 4.1553,
+      "step": 258
+    },
+    {
+      "epoch": 0.02273232197174368,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.000777,
+      "loss": 4.21,
+      "step": 259
+    },
+    {
+      "epoch": 0.022820091554646166,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 4.1685,
+      "step": 260
+    },
+    {
+      "epoch": 0.02290786113754865,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.0007830000000000001,
+      "loss": 4.0781,
+      "step": 261
+    },
+    {
+      "epoch": 0.022995630720451136,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.000786,
+      "loss": 4.1138,
+      "step": 262
+    },
+    {
+      "epoch": 0.02308340030335362,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0007890000000000001,
+      "loss": 4.1313,
+      "step": 263
+    },
+    {
+      "epoch": 0.023171169886256105,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.0007920000000000001,
+      "loss": 4.1675,
+      "step": 264
+    },
+    {
+      "epoch": 0.02325893946915859,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.000795,
+      "loss": 4.228,
+      "step": 265
+    },
+    {
+      "epoch": 0.023346709052061078,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.0007980000000000001,
+      "loss": 4.1475,
+      "step": 266
+    },
+    {
+      "epoch": 0.023434478634963563,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0008010000000000001,
+      "loss": 4.1196,
+      "step": 267
+    },
+    {
+      "epoch": 0.023522248217866047,
+      "grad_norm": 1.734375,
+      "learning_rate": 0.000804,
+      "loss": 4.1523,
+      "step": 268
+    },
+    {
+      "epoch": 0.023610017800768532,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.0008070000000000001,
+      "loss": 4.1338,
+      "step": 269
+    },
+    {
+      "epoch": 0.023697787383671017,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0008100000000000001,
+      "loss": 4.0962,
+      "step": 270
+    },
+    {
+      "epoch": 0.023785556966573505,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.000813,
+      "loss": 4.1567,
+      "step": 271
+    },
+    {
+      "epoch": 0.02387332654947599,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.0008160000000000001,
+      "loss": 4.1865,
+      "step": 272
+    },
+    {
+      "epoch": 0.023961096132378474,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.0008190000000000001,
+      "loss": 4.061,
+      "step": 273
+    },
+    {
+      "epoch": 0.02404886571528096,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.000822,
+      "loss": 4.0957,
+      "step": 274
+    },
+    {
+      "epoch": 0.024136635298183443,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.0008250000000000001,
+      "loss": 4.1304,
+      "step": 275
+    },
+    {
+      "epoch": 0.024224404881085928,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.0008280000000000001,
+      "loss": 4.1309,
+      "step": 276
+    },
+    {
+      "epoch": 0.024312174463988416,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0008310000000000001,
+      "loss": 4.1338,
+      "step": 277
+    },
+    {
+      "epoch": 0.0243999440468909,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0008340000000000001,
+      "loss": 4.0967,
+      "step": 278
+    },
+    {
+      "epoch": 0.024487713629793385,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.0008370000000000001,
+      "loss": 4.0981,
+      "step": 279
+    },
+    {
+      "epoch": 0.02457548321269587,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.0008400000000000001,
+      "loss": 4.0859,
+      "step": 280
+    },
+    {
+      "epoch": 0.024663252795598355,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.0008430000000000001,
+      "loss": 4.126,
+      "step": 281
+    },
+    {
+      "epoch": 0.02475102237850084,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.000846,
+      "loss": 4.2065,
+      "step": 282
+    },
+    {
+      "epoch": 0.024838791961403327,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0008489999999999999,
+      "loss": 4.0684,
+      "step": 283
+    },
+    {
+      "epoch": 0.024926561544305812,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.0008519999999999999,
+      "loss": 4.0898,
+      "step": 284
+    },
+    {
+      "epoch": 0.025014331127208297,
+      "grad_norm": 2.828125,
+      "learning_rate": 0.000855,
+      "loss": 4.1406,
+      "step": 285
+    },
+    {
+      "epoch": 0.02510210071011078,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.0008579999999999999,
+      "loss": 4.1655,
+      "step": 286
+    },
+    {
+      "epoch": 0.025189870293013266,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.000861,
+      "loss": 4.1816,
+      "step": 287
+    },
+    {
+      "epoch": 0.02527763987591575,
+      "grad_norm": 4.75,
+      "learning_rate": 0.000864,
+      "loss": 4.252,
+      "step": 288
+    },
+    {
+      "epoch": 0.02536540945881824,
+      "grad_norm": 1.5078125,
+      "learning_rate": 0.0008669999999999999,
+      "loss": 4.146,
+      "step": 289
+    },
+    {
+      "epoch": 0.025453179041720723,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.00087,
+      "loss": 4.1196,
+      "step": 290
+    },
+    {
+      "epoch": 0.025540948624623208,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.000873,
+      "loss": 4.123,
+      "step": 291
+    },
+    {
+      "epoch": 0.025628718207525693,
+      "grad_norm": 1.4765625,
+      "learning_rate": 0.0008759999999999999,
+      "loss": 4.1416,
+      "step": 292
+    },
+    {
+      "epoch": 0.025716487790428177,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.000879,
+      "loss": 4.1191,
+      "step": 293
+    },
+    {
+      "epoch": 0.025804257373330662,
+      "grad_norm": 0.9609375,
+      "learning_rate": 0.000882,
+      "loss": 4.1421,
+      "step": 294
+    },
+    {
+      "epoch": 0.02589202695623315,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0008849999999999999,
+      "loss": 4.123,
+      "step": 295
+    },
+    {
+      "epoch": 0.025979796539135635,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.000888,
+      "loss": 4.1802,
+      "step": 296
+    },
+    {
+      "epoch": 0.02606756612203812,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.000891,
+      "loss": 4.0566,
+      "step": 297
+    },
+    {
+      "epoch": 0.026155335704940604,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.0008939999999999999,
+      "loss": 4.1284,
+      "step": 298
+    },
+    {
+      "epoch": 0.02624310528784309,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.000897,
+      "loss": 4.0513,
+      "step": 299
+    },
+    {
+      "epoch": 0.026330874870745574,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0009,
+      "loss": 4.1318,
+      "step": 300
+    },
+    {
+      "epoch": 0.02641864445364806,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0009029999999999999,
+      "loss": 4.0923,
+      "step": 301
+    },
+    {
+      "epoch": 0.026506414036550546,
+      "grad_norm": 1.390625,
+      "learning_rate": 0.000906,
+      "loss": 4.063,
+      "step": 302
+    },
+    {
+      "epoch": 0.02659418361945303,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.000909,
+      "loss": 4.0669,
+      "step": 303
+    },
+    {
+      "epoch": 0.026681953202355516,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.000912,
+      "loss": 4.1074,
+      "step": 304
+    },
+    {
+      "epoch": 0.026769722785258,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.000915,
+      "loss": 4.1069,
+      "step": 305
+    },
+    {
+      "epoch": 0.02685749236816049,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.000918,
+      "loss": 4.0347,
+      "step": 306
+    },
+    {
+      "epoch": 0.026945261951062973,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.000921,
+      "loss": 4.0503,
+      "step": 307
+    },
+    {
+      "epoch": 0.027033031533965458,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.000924,
+      "loss": 4.0518,
+      "step": 308
+    },
+    {
+      "epoch": 0.027120801116867942,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.000927,
+      "loss": 4.0605,
+      "step": 309
+    },
+    {
+      "epoch": 0.027208570699770427,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00093,
+      "loss": 4.1006,
+      "step": 310
+    },
+    {
+      "epoch": 0.02729634028267291,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.000933,
+      "loss": 4.0596,
+      "step": 311
+    },
+    {
+      "epoch": 0.0273841098655754,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.000936,
+      "loss": 4.1206,
+      "step": 312
+    },
+    {
+      "epoch": 0.027471879448477884,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.0009390000000000001,
+      "loss": 4.0425,
+      "step": 313
+    },
+    {
+      "epoch": 0.02755964903138037,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.000942,
+      "loss": 4.1162,
+      "step": 314
+    },
+    {
+      "epoch": 0.027647418614282854,
+      "grad_norm": 3.515625,
+      "learning_rate": 0.000945,
+      "loss": 4.0737,
+      "step": 315
+    },
+    {
+      "epoch": 0.02773518819718534,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.0009480000000000001,
+      "loss": 4.0366,
+      "step": 316
+    },
+    {
+      "epoch": 0.027822957780087823,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.000951,
+      "loss": 4.021,
+      "step": 317
+    },
+    {
+      "epoch": 0.02791072736299031,
+      "grad_norm": 1.625,
+      "learning_rate": 0.000954,
+      "loss": 4.0576,
+      "step": 318
+    },
+    {
+      "epoch": 0.027998496945892796,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.0009570000000000001,
+      "loss": 4.0444,
+      "step": 319
+    },
+    {
+      "epoch": 0.02808626652879528,
+      "grad_norm": 1.5859375,
+      "learning_rate": 0.00096,
+      "loss": 4.0332,
+      "step": 320
+    },
+    {
+      "epoch": 0.028174036111697765,
+      "grad_norm": 1.671875,
+      "learning_rate": 0.000963,
+      "loss": 4.1543,
+      "step": 321
+    },
+    {
+      "epoch": 0.02826180569460025,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.0009660000000000001,
+      "loss": 4.0342,
+      "step": 322
+    },
+    {
+      "epoch": 0.028349575277502734,
+      "grad_norm": 3.53125,
+      "learning_rate": 0.000969,
+      "loss": 4.064,
+      "step": 323
+    },
+    {
+      "epoch": 0.028437344860405223,
+      "grad_norm": 1.9609375,
+      "learning_rate": 0.0009720000000000001,
+      "loss": 4.0474,
+      "step": 324
+    },
+    {
+      "epoch": 0.028525114443307707,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.0009750000000000001,
+      "loss": 4.1118,
+      "step": 325
+    },
+    {
+      "epoch": 0.028612884026210192,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.0009780000000000001,
+      "loss": 4.0039,
+      "step": 326
+    },
+    {
+      "epoch": 0.028700653609112677,
+      "grad_norm": 1.6953125,
+      "learning_rate": 0.000981,
+      "loss": 4.061,
+      "step": 327
+    },
+    {
+      "epoch": 0.02878842319201516,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.000984,
+      "loss": 4.0532,
+      "step": 328
+    },
+    {
+      "epoch": 0.028876192774917646,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.000987,
+      "loss": 4.0806,
+      "step": 329
+    },
+    {
+      "epoch": 0.028963962357820134,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00099,
+      "loss": 4.022,
+      "step": 330
+    },
+    {
+      "epoch": 0.02905173194072262,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.0009930000000000002,
+      "loss": 4.0552,
+      "step": 331
+    },
+    {
+      "epoch": 0.029139501523625103,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.0009960000000000001,
+      "loss": 3.9961,
+      "step": 332
+    },
+    {
+      "epoch": 0.029227271106527588,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.000999,
+      "loss": 4.0605,
+      "step": 333
+    },
+    {
+      "epoch": 0.029315040689430073,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.001002,
+      "loss": 3.9985,
+      "step": 334
+    },
+    {
+      "epoch": 0.029402810272332557,
+      "grad_norm": 2.0625,
+      "learning_rate": 0.001005,
+      "loss": 4.0396,
+      "step": 335
+    },
+    {
+      "epoch": 0.029490579855235045,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.001008,
+      "loss": 4.0532,
+      "step": 336
+    },
+    {
+      "epoch": 0.02957834943813753,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.0010110000000000002,
+      "loss": 4.0146,
+      "step": 337
+    },
+    {
+      "epoch": 0.029666119021040015,
+      "grad_norm": 2.0,
+      "learning_rate": 0.0010140000000000001,
+      "loss": 4.0566,
+      "step": 338
+    },
+    {
+      "epoch": 0.0297538886039425,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.0010170000000000001,
+      "loss": 4.0024,
+      "step": 339
+    },
+    {
+      "epoch": 0.029841658186844984,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.00102,
+      "loss": 4.0923,
+      "step": 340
+    },
+    {
+      "epoch": 0.029929427769747472,
+      "grad_norm": 2.109375,
+      "learning_rate": 0.001023,
+      "loss": 4.0166,
+      "step": 341
+    },
+    {
+      "epoch": 0.030017197352649957,
+      "grad_norm": 3.578125,
+      "learning_rate": 0.001026,
+      "loss": 4.1646,
+      "step": 342
+    },
+    {
+      "epoch": 0.03010496693555244,
+      "grad_norm": 2.296875,
+      "learning_rate": 0.0010290000000000002,
+      "loss": 4.0503,
+      "step": 343
+    },
+    {
+      "epoch": 0.030192736518454926,
+      "grad_norm": 4.125,
+      "learning_rate": 0.001032,
+      "loss": 4.1504,
+      "step": 344
+    },
+    {
+      "epoch": 0.03028050610135741,
+      "grad_norm": 2.734375,
+      "learning_rate": 0.001035,
+      "loss": 4.0747,
+      "step": 345
+    },
+    {
+      "epoch": 0.030368275684259895,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.0010379999999999999,
+      "loss": 4.1255,
+      "step": 346
+    },
+    {
+      "epoch": 0.030456045267162384,
+      "grad_norm": 7.4375,
+      "learning_rate": 0.001041,
+      "loss": 4.1182,
+      "step": 347
+    },
+    {
+      "epoch": 0.030543814850064868,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.001044,
+      "loss": 4.0532,
+      "step": 348
+    },
+    {
+      "epoch": 0.030631584432967353,
+      "grad_norm": 2.21875,
+      "learning_rate": 0.001047,
+      "loss": 4.0298,
+      "step": 349
+    },
+    {
+      "epoch": 0.030719354015869837,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00105,
+      "loss": 4.0298,
+      "step": 350
+    },
+    {
+      "epoch": 0.030807123598772322,
+      "grad_norm": 2.203125,
+      "learning_rate": 0.001053,
+      "loss": 4.0601,
+      "step": 351
+    },
+    {
+      "epoch": 0.030894893181674807,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0010559999999999999,
+      "loss": 3.9873,
+      "step": 352
+    },
+    {
+      "epoch": 0.030982662764577295,
+      "grad_norm": 2.375,
+      "learning_rate": 0.001059,
+      "loss": 4.0327,
+      "step": 353
+    },
+    {
+      "epoch": 0.03107043234747978,
+      "grad_norm": 1.390625,
+      "learning_rate": 0.001062,
+      "loss": 4.0474,
+      "step": 354
+    },
+    {
+      "epoch": 0.031158201930382264,
+      "grad_norm": 1.71875,
+      "learning_rate": 0.001065,
+      "loss": 4.085,
+      "step": 355
+    },
+    {
+      "epoch": 0.03124597151328475,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.001068,
+      "loss": 4.0176,
+      "step": 356
+    },
+    {
+      "epoch": 0.03133374109618724,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.001071,
+      "loss": 3.9917,
+      "step": 357
+    },
+    {
+      "epoch": 0.03142151067908972,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.001074,
+      "loss": 4.0342,
+      "step": 358
+    },
+    {
+      "epoch": 0.031509280261992206,
+      "grad_norm": 1.765625,
+      "learning_rate": 0.001077,
+      "loss": 4.0479,
+      "step": 359
+    },
+    {
+      "epoch": 0.03159704984489469,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00108,
+      "loss": 4.1104,
+      "step": 360
+    },
+    {
+      "epoch": 0.031684819427797176,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.001083,
+      "loss": 4.0,
+      "step": 361
+    },
+    {
+      "epoch": 0.03177258901069966,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.001086,
+      "loss": 4.1157,
+      "step": 362
+    },
+    {
+      "epoch": 0.031860358593602145,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.001089,
+      "loss": 4.0947,
+      "step": 363
+    },
+    {
+      "epoch": 0.03194812817650463,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.001092,
+      "loss": 3.9888,
+      "step": 364
+    },
+    {
+      "epoch": 0.032035897759407114,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.001095,
+      "loss": 4.0068,
+      "step": 365
+    },
+    {
+      "epoch": 0.0321236673423096,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.001098,
+      "loss": 4.0063,
+      "step": 366
+    },
+    {
+      "epoch": 0.032211436925212084,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.001101,
+      "loss": 4.0361,
+      "step": 367
+    },
+    {
+      "epoch": 0.032299206508114575,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.001104,
+      "loss": 4.042,
+      "step": 368
+    },
+    {
+      "epoch": 0.03238697609101706,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.001107,
+      "loss": 3.9937,
+      "step": 369
+    },
+    {
+      "epoch": 0.032474745673919544,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00111,
+      "loss": 4.0044,
+      "step": 370
+    },
+    {
+      "epoch": 0.03256251525682203,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.001113,
+      "loss": 3.9927,
+      "step": 371
+    },
+    {
+      "epoch": 0.032650284839724514,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.001116,
+      "loss": 3.9639,
+      "step": 372
+    },
+    {
+      "epoch": 0.032738054422627,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.001119,
+      "loss": 3.9951,
+      "step": 373
+    },
+    {
+      "epoch": 0.03282582400552948,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.001122,
+      "loss": 4.0352,
+      "step": 374
+    },
+    {
+      "epoch": 0.03291359358843197,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0011250000000000001,
+      "loss": 3.9482,
+      "step": 375
+    },
+    {
+      "epoch": 0.03300136317133445,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.001128,
+      "loss": 3.9829,
+      "step": 376
+    },
+    {
+      "epoch": 0.03308913275423694,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.001131,
+      "loss": 4.0322,
+      "step": 377
+    },
+    {
+      "epoch": 0.03317690233713942,
+      "grad_norm": 0.7578125,
+      "learning_rate": 0.001134,
+      "loss": 3.9609,
+      "step": 378
+    },
+    {
+      "epoch": 0.03326467192004191,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.001137,
+      "loss": 4.0239,
+      "step": 379
+    },
+    {
+      "epoch": 0.0333524415029444,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00114,
+      "loss": 3.9888,
+      "step": 380
+    },
+    {
+      "epoch": 0.03344021108584688,
+      "grad_norm": 0.49609375,
+      "learning_rate": 0.0011430000000000001,
+      "loss": 3.9292,
+      "step": 381
+    },
+    {
+      "epoch": 0.03352798066874937,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.001146,
+      "loss": 3.958,
+      "step": 382
+    },
+    {
+      "epoch": 0.03361575025165185,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.001149,
+      "loss": 3.9438,
+      "step": 383
+    },
+    {
+      "epoch": 0.03370351983455434,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.001152,
+      "loss": 3.9805,
+      "step": 384
+    },
+    {
+      "epoch": 0.03379128941745682,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.001155,
+      "loss": 4.0659,
+      "step": 385
+    },
+    {
+      "epoch": 0.033879059000359306,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.001158,
+      "loss": 3.9604,
+      "step": 386
+    },
+    {
+      "epoch": 0.03396682858326179,
+      "grad_norm": 1.5703125,
+      "learning_rate": 0.0011610000000000001,
+      "loss": 3.9927,
+      "step": 387
+    },
+    {
+      "epoch": 0.034054598166164275,
+      "grad_norm": 1.0,
+      "learning_rate": 0.001164,
+      "loss": 4.0088,
+      "step": 388
+    },
+    {
+      "epoch": 0.03414236774906676,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.001167,
+      "loss": 3.9712,
+      "step": 389
+    },
+    {
+      "epoch": 0.034230137331969245,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00117,
+      "loss": 3.9395,
+      "step": 390
+    },
+    {
+      "epoch": 0.034317906914871736,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.001173,
+      "loss": 3.9497,
+      "step": 391
+    },
+    {
+      "epoch": 0.03440567649777422,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.001176,
+      "loss": 3.998,
+      "step": 392
+    },
+    {
+      "epoch": 0.034493446080676705,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.0011790000000000001,
+      "loss": 3.9795,
+      "step": 393
+    },
+    {
+      "epoch": 0.03458121566357919,
+      "grad_norm": 0.875,
+      "learning_rate": 0.001182,
+      "loss": 3.9468,
+      "step": 394
+    },
+    {
+      "epoch": 0.034668985246481675,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.001185,
+      "loss": 3.9731,
+      "step": 395
+    },
+    {
+      "epoch": 0.03475675482938416,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.001188,
+      "loss": 3.9263,
+      "step": 396
+    },
+    {
+      "epoch": 0.034844524412286644,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.001191,
+      "loss": 3.9502,
+      "step": 397
+    },
+    {
+      "epoch": 0.03493229399518913,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0011940000000000002,
+      "loss": 3.9434,
+      "step": 398
+    },
+    {
+      "epoch": 0.03502006357809161,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.0011970000000000001,
+      "loss": 4.0049,
+      "step": 399
+    },
+    {
+      "epoch": 0.0351078331609941,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0012000000000000001,
+      "loss": 3.9722,
+      "step": 400
+    },
+    {
+      "epoch": 0.03519560274389658,
+      "grad_norm": 2.59375,
+      "learning_rate": 0.001203,
+      "loss": 3.9683,
+      "step": 401
+    },
+    {
+      "epoch": 0.03528337232679907,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.001206,
+      "loss": 3.9512,
+      "step": 402
+    },
+    {
+      "epoch": 0.03537114190970156,
+      "grad_norm": 3.140625,
+      "learning_rate": 0.001209,
+      "loss": 4.0049,
+      "step": 403
+    },
+    {
+      "epoch": 0.035458911492604044,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.0012120000000000002,
+      "loss": 4.0151,
+      "step": 404
+    },
+    {
+      "epoch": 0.03554668107550653,
+      "grad_norm": 3.0,
+      "learning_rate": 0.0012150000000000002,
+      "loss": 4.0083,
+      "step": 405
+    },
+    {
+      "epoch": 0.03563445065840901,
+      "grad_norm": 1.859375,
+      "learning_rate": 0.0012180000000000001,
+      "loss": 4.0024,
+      "step": 406
+    },
+    {
+      "epoch": 0.0357222202413115,
+      "grad_norm": 3.171875,
+      "learning_rate": 0.0012209999999999999,
+      "loss": 4.0562,
+      "step": 407
+    },
+    {
+      "epoch": 0.03580998982421398,
+      "grad_norm": 2.046875,
+      "learning_rate": 0.001224,
+      "loss": 3.9976,
+      "step": 408
+    },
+    {
+      "epoch": 0.03589775940711647,
+      "grad_norm": 2.234375,
+      "learning_rate": 0.001227,
+      "loss": 4.021,
+      "step": 409
+    },
+    {
+      "epoch": 0.03598552899001895,
+      "grad_norm": 1.6171875,
+      "learning_rate": 0.00123,
+      "loss": 4.002,
+      "step": 410
+    },
+    {
+      "epoch": 0.036073298572921436,
+      "grad_norm": 2.265625,
+      "learning_rate": 0.001233,
+      "loss": 4.0156,
+      "step": 411
+    },
+    {
+      "epoch": 0.03616106815582392,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.001236,
+      "loss": 4.0059,
+      "step": 412
+    },
+    {
+      "epoch": 0.036248837738726405,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.0012389999999999999,
+      "loss": 3.959,
+      "step": 413
+    },
+    {
+      "epoch": 0.0363366073216289,
+      "grad_norm": 1.78125,
+      "learning_rate": 0.001242,
+      "loss": 3.9736,
+      "step": 414
+    },
+    {
+      "epoch": 0.03642437690453138,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.001245,
+      "loss": 4.0049,
+      "step": 415
+    },
+    {
+      "epoch": 0.036512146487433866,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.001248,
+      "loss": 4.002,
+      "step": 416
+    },
+    {
+      "epoch": 0.03659991607033635,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.001251,
+      "loss": 3.9365,
+      "step": 417
+    },
+    {
+      "epoch": 0.036687685653238836,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.001254,
+      "loss": 3.9277,
+      "step": 418
+    },
+    {
+      "epoch": 0.03677545523614132,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.0012569999999999999,
+      "loss": 3.9312,
+      "step": 419
+    },
+    {
+      "epoch": 0.036863224819043805,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.00126,
+      "loss": 3.98,
+      "step": 420
+    },
+    {
+      "epoch": 0.03695099440194629,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.001263,
+      "loss": 3.9614,
+      "step": 421
+    },
+    {
+      "epoch": 0.037038763984848774,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.001266,
+      "loss": 3.9854,
+      "step": 422
+    },
+    {
+      "epoch": 0.03712653356775126,
+      "grad_norm": 0.70703125,
+      "learning_rate": 0.001269,
+      "loss": 3.9487,
+      "step": 423
+    },
+    {
+      "epoch": 0.037214303150653744,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.001272,
+      "loss": 3.9883,
+      "step": 424
+    },
+    {
+      "epoch": 0.03730207273355623,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.001275,
+      "loss": 3.9438,
+      "step": 425
+    },
+    {
+      "epoch": 0.03738984231645872,
+      "grad_norm": 0.5,
+      "learning_rate": 0.001278,
+      "loss": 3.9678,
+      "step": 426
+    },
+    {
+      "epoch": 0.037477611899361205,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.001281,
+      "loss": 3.9448,
+      "step": 427
+    },
+    {
+      "epoch": 0.03756538148226369,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.001284,
+      "loss": 3.9224,
+      "step": 428
+    },
+    {
+      "epoch": 0.037653151065166174,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.001287,
+      "loss": 3.874,
+      "step": 429
+    },
+    {
+      "epoch": 0.03774092064806866,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.00129,
+      "loss": 3.9141,
+      "step": 430
+    },
+    {
+      "epoch": 0.03782869023097114,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.001293,
+      "loss": 3.9487,
+      "step": 431
+    },
+    {
+      "epoch": 0.03791645981387363,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.001296,
+      "loss": 3.9058,
+      "step": 432
+    },
+    {
+      "epoch": 0.03800422939677611,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.001299,
+      "loss": 4.0063,
+      "step": 433
+    },
+    {
+      "epoch": 0.0380919989796786,
+      "grad_norm": 0.67578125,
+      "learning_rate": 0.001302,
+      "loss": 3.9141,
+      "step": 434
+    },
+    {
+      "epoch": 0.03817976856258108,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.001305,
+      "loss": 3.8501,
+      "step": 435
+    },
+    {
+      "epoch": 0.038267538145483566,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.001308,
+      "loss": 3.8936,
+      "step": 436
+    },
+    {
+      "epoch": 0.03835530772838605,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.001311,
+      "loss": 3.9009,
+      "step": 437
+    },
+    {
+      "epoch": 0.03844307731128854,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.001314,
+      "loss": 3.8857,
+      "step": 438
+    },
+    {
+      "epoch": 0.03853084689419103,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.001317,
+      "loss": 3.8975,
+      "step": 439
+    },
+    {
+      "epoch": 0.03861861647709351,
+      "grad_norm": 0.67578125,
+      "learning_rate": 0.00132,
+      "loss": 3.9072,
+      "step": 440
+    },
+    {
+      "epoch": 0.038706386059996,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.001323,
+      "loss": 3.9507,
+      "step": 441
+    },
+    {
+      "epoch": 0.03879415564289848,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.0013260000000000001,
+      "loss": 3.9082,
+      "step": 442
+    },
+    {
+      "epoch": 0.038881925225800966,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.001329,
+      "loss": 3.8965,
+      "step": 443
+    },
+    {
+      "epoch": 0.03896969480870345,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.001332,
+      "loss": 3.9629,
+      "step": 444
+    },
+    {
+      "epoch": 0.039057464391605935,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.001335,
+      "loss": 4.0127,
+      "step": 445
+    },
+    {
+      "epoch": 0.03914523397450842,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.001338,
+      "loss": 3.9023,
+      "step": 446
+    },
+    {
+      "epoch": 0.039233003557410905,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.001341,
+      "loss": 3.9448,
+      "step": 447
+    },
+    {
+      "epoch": 0.03932077314031339,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.0013440000000000001,
+      "loss": 3.9077,
+      "step": 448
+    },
+    {
+      "epoch": 0.03940854272321588,
+      "grad_norm": 2.890625,
+      "learning_rate": 0.001347,
+      "loss": 4.0171,
+      "step": 449
+    },
+    {
+      "epoch": 0.039496312306118365,
+      "grad_norm": 1.9765625,
+      "learning_rate": 0.00135,
+      "loss": 3.9956,
+      "step": 450
+    },
+    {
+      "epoch": 0.03958408188902085,
+      "grad_norm": 2.625,
+      "learning_rate": 0.001353,
+      "loss": 3.8921,
+      "step": 451
+    },
+    {
+      "epoch": 0.039671851471923335,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.001356,
+      "loss": 3.9917,
+      "step": 452
+    },
+    {
+      "epoch": 0.03975962105482582,
+      "grad_norm": 1.875,
+      "learning_rate": 0.001359,
+      "loss": 3.9468,
+      "step": 453
+    },
+    {
+      "epoch": 0.039847390637728304,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.0013620000000000001,
+      "loss": 3.8755,
+      "step": 454
+    },
+    {
+      "epoch": 0.03993516022063079,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.0013650000000000001,
+      "loss": 3.9326,
+      "step": 455
+    },
+    {
+      "epoch": 0.04002292980353327,
+      "grad_norm": 1.984375,
+      "learning_rate": 0.001368,
+      "loss": 3.918,
+      "step": 456
+    },
+    {
+      "epoch": 0.04011069938643576,
+      "grad_norm": 1.3828125,
+      "learning_rate": 0.001371,
+      "loss": 3.8735,
+      "step": 457
+    },
+    {
+      "epoch": 0.04019846896933824,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.001374,
+      "loss": 3.9316,
+      "step": 458
+    },
+    {
+      "epoch": 0.04028623855224073,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0013770000000000002,
+      "loss": 3.9751,
+      "step": 459
+    },
+    {
+      "epoch": 0.04037400813514321,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.0013800000000000002,
+      "loss": 3.9214,
+      "step": 460
+    },
+    {
+      "epoch": 0.040461777718045704,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0013830000000000001,
+      "loss": 3.9629,
+      "step": 461
+    },
+    {
+      "epoch": 0.04054954730094819,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.001386,
+      "loss": 3.9033,
+      "step": 462
+    },
+    {
+      "epoch": 0.04063731688385067,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.001389,
+      "loss": 3.9839,
+      "step": 463
+    },
+    {
+      "epoch": 0.04072508646675316,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.001392,
+      "loss": 3.937,
+      "step": 464
+    },
+    {
+      "epoch": 0.04081285604965564,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0013950000000000002,
+      "loss": 3.9165,
+      "step": 465
+    },
+    {
+      "epoch": 0.04090062563255813,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.0013980000000000002,
+      "loss": 3.9062,
+      "step": 466
+    },
+    {
+      "epoch": 0.04098839521546061,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.0014010000000000001,
+      "loss": 3.9053,
+      "step": 467
+    },
+    {
+      "epoch": 0.041076164798363096,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.001404,
+      "loss": 3.9014,
+      "step": 468
+    },
+    {
+      "epoch": 0.04116393438126558,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.001407,
+      "loss": 3.9634,
+      "step": 469
+    },
+    {
+      "epoch": 0.041251703964168066,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00141,
+      "loss": 3.9297,
+      "step": 470
+    },
+    {
+      "epoch": 0.04133947354707055,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.001413,
+      "loss": 3.9546,
+      "step": 471
+    },
+    {
+      "epoch": 0.041427243129973035,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.001416,
+      "loss": 3.9058,
+      "step": 472
+    },
+    {
+      "epoch": 0.041515012712875526,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.001419,
+      "loss": 3.8677,
+      "step": 473
+    },
+    {
+      "epoch": 0.04160278229577801,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.0014219999999999999,
+      "loss": 3.9424,
+      "step": 474
+    },
+    {
+      "epoch": 0.041690551878680496,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.001425,
+      "loss": 3.9219,
+      "step": 475
+    },
+    {
+      "epoch": 0.04177832146158298,
+      "grad_norm": 1.125,
+      "learning_rate": 0.001428,
+      "loss": 3.8311,
+      "step": 476
+    },
+    {
+      "epoch": 0.041866091044485465,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.001431,
+      "loss": 3.8345,
+      "step": 477
+    },
+    {
+      "epoch": 0.04195386062738795,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.001434,
+      "loss": 3.8394,
+      "step": 478
+    },
+    {
+      "epoch": 0.042041630210290434,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.001437,
+      "loss": 3.9087,
+      "step": 479
+    },
+    {
+      "epoch": 0.04212939979319292,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.0014399999999999999,
+      "loss": 3.8911,
+      "step": 480
+    },
+    {
+      "epoch": 0.042217169376095404,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.001443,
+      "loss": 3.9209,
+      "step": 481
+    },
+    {
+      "epoch": 0.04230493895899789,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.001446,
+      "loss": 3.9258,
+      "step": 482
+    },
+    {
+      "epoch": 0.04239270854190037,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.001449,
+      "loss": 3.9434,
+      "step": 483
+    },
+    {
+      "epoch": 0.042480478124802865,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.001452,
+      "loss": 3.9185,
+      "step": 484
+    },
+    {
+      "epoch": 0.04256824770770535,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.001455,
+      "loss": 3.957,
+      "step": 485
+    },
+    {
+      "epoch": 0.042656017290607834,
+      "grad_norm": 2.453125,
+      "learning_rate": 0.001458,
+      "loss": 3.9399,
+      "step": 486
+    },
+    {
+      "epoch": 0.04274378687351032,
+      "grad_norm": 1.765625,
+      "learning_rate": 0.001461,
+      "loss": 3.9497,
+      "step": 487
+    },
+    {
+      "epoch": 0.0428315564564128,
+      "grad_norm": 2.5625,
+      "learning_rate": 0.001464,
+      "loss": 3.8267,
+      "step": 488
+    },
+    {
+      "epoch": 0.04291932603931529,
+      "grad_norm": 2.25,
+      "learning_rate": 0.001467,
+      "loss": 3.918,
+      "step": 489
+    },
+    {
+      "epoch": 0.04300709562221777,
+      "grad_norm": 0.9609375,
+      "learning_rate": 0.00147,
+      "loss": 3.8647,
+      "step": 490
+    },
+    {
+      "epoch": 0.04309486520512026,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.001473,
+      "loss": 3.9663,
+      "step": 491
+    },
+    {
+      "epoch": 0.04318263478802274,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.001476,
+      "loss": 3.9067,
+      "step": 492
+    },
+    {
+      "epoch": 0.043270404370925226,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.001479,
+      "loss": 3.8442,
+      "step": 493
+    },
+    {
+      "epoch": 0.04335817395382771,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.001482,
+      "loss": 3.9346,
+      "step": 494
+    },
+    {
+      "epoch": 0.043445943536730196,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.001485,
+      "loss": 3.9224,
+      "step": 495
+    },
+    {
+      "epoch": 0.04353371311963269,
+      "grad_norm": 0.625,
+      "learning_rate": 0.001488,
+      "loss": 3.9116,
+      "step": 496
+    },
+    {
+      "epoch": 0.04362148270253517,
+      "grad_norm": 0.828125,
+      "learning_rate": 0.001491,
+      "loss": 3.8931,
+      "step": 497
+    },
+    {
+      "epoch": 0.04370925228543766,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.001494,
+      "loss": 3.9458,
+      "step": 498
+    },
+    {
+      "epoch": 0.04379702186834014,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.001497,
+      "loss": 3.8657,
+      "step": 499
+    },
+    {
+      "epoch": 0.043884791451242626,
+      "grad_norm": 1.25,
+      "learning_rate": 0.0015,
+      "loss": 3.937,
+      "step": 500
+    },
+    {
+      "epoch": 0.04397256103414511,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.001503,
+      "loss": 3.8296,
+      "step": 501
+    },
+    {
+      "epoch": 0.044060330617047595,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.001506,
+      "loss": 3.8706,
+      "step": 502
+    },
+    {
+      "epoch": 0.04414810019995008,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.0015090000000000001,
+      "loss": 3.8838,
+      "step": 503
+    },
+    {
+      "epoch": 0.044235869782852565,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.001512,
+      "loss": 3.8638,
+      "step": 504
+    },
+    {
+      "epoch": 0.04432363936575505,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.001515,
+      "loss": 3.936,
+      "step": 505
+    },
+    {
+      "epoch": 0.044411408948657534,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.001518,
+      "loss": 3.8638,
+      "step": 506
+    },
+    {
+      "epoch": 0.044499178531560026,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.001521,
+      "loss": 3.9365,
+      "step": 507
+    },
+    {
+      "epoch": 0.04458694811446251,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.001524,
+      "loss": 3.8721,
+      "step": 508
+    },
+    {
+      "epoch": 0.044674717697364995,
+      "grad_norm": 0.9609375,
+      "learning_rate": 0.0015270000000000001,
+      "loss": 3.8794,
+      "step": 509
+    },
+    {
+      "epoch": 0.04476248728026748,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0015300000000000001,
+      "loss": 3.8472,
+      "step": 510
+    },
+    {
+      "epoch": 0.044850256863169964,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.001533,
+      "loss": 3.8096,
+      "step": 511
+    },
+    {
+      "epoch": 0.04493802644607245,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.001536,
+      "loss": 3.877,
+      "step": 512
+    },
+    {
+      "epoch": 0.04502579602897493,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.001539,
+      "loss": 3.8677,
+      "step": 513
+    },
+    {
+      "epoch": 0.04511356561187742,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.001542,
+      "loss": 3.8823,
+      "step": 514
+    },
+    {
+      "epoch": 0.0452013351947799,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.0015450000000000001,
+      "loss": 3.8525,
+      "step": 515
+    },
+    {
+      "epoch": 0.04528910477768239,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.0015480000000000001,
+      "loss": 3.957,
+      "step": 516
+    },
+    {
+      "epoch": 0.04537687436058487,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.001551,
+      "loss": 3.8887,
+      "step": 517
+    },
+    {
+      "epoch": 0.04546464394348736,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.001554,
+      "loss": 3.8979,
+      "step": 518
+    },
+    {
+      "epoch": 0.04555241352638985,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.001557,
+      "loss": 3.8882,
+      "step": 519
+    },
+    {
+      "epoch": 0.04564018310929233,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.0015600000000000002,
+      "loss": 3.8604,
+      "step": 520
+    },
+    {
+      "epoch": 0.04572795269219482,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0015630000000000002,
+      "loss": 3.8828,
+      "step": 521
+    },
+    {
+      "epoch": 0.0458157222750973,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.0015660000000000001,
+      "loss": 3.9438,
+      "step": 522
+    },
+    {
+      "epoch": 0.04590349185799979,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.001569,
+      "loss": 3.873,
+      "step": 523
+    },
+    {
+      "epoch": 0.04599126144090227,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.001572,
+      "loss": 3.834,
+      "step": 524
+    },
+    {
+      "epoch": 0.046079031023804756,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.001575,
+      "loss": 3.9067,
+      "step": 525
+    },
+    {
+      "epoch": 0.04616680060670724,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.0015780000000000002,
+      "loss": 3.8608,
+      "step": 526
+    },
+    {
+      "epoch": 0.046254570189609726,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.0015810000000000002,
+      "loss": 3.8252,
+      "step": 527
+    },
+    {
+      "epoch": 0.04634233977251221,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.0015840000000000001,
+      "loss": 3.8516,
+      "step": 528
+    },
+    {
+      "epoch": 0.046430109355414695,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.001587,
+      "loss": 3.8423,
+      "step": 529
+    },
+    {
+      "epoch": 0.04651787893831718,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00159,
+      "loss": 3.9004,
+      "step": 530
+    },
+    {
+      "epoch": 0.04660564852121967,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.001593,
+      "loss": 3.9033,
+      "step": 531
+    },
+    {
+      "epoch": 0.046693418104122156,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.0015960000000000002,
+      "loss": 3.8877,
+      "step": 532
+    },
+    {
+      "epoch": 0.04678118768702464,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.0015990000000000002,
+      "loss": 3.833,
+      "step": 533
+    },
+    {
+      "epoch": 0.046868957269927125,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.0016020000000000001,
+      "loss": 3.8491,
+      "step": 534
+    },
+    {
+      "epoch": 0.04695672685282961,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.001605,
+      "loss": 3.7886,
+      "step": 535
+    },
+    {
+      "epoch": 0.047044496435732094,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.001608,
+      "loss": 3.8159,
+      "step": 536
+    },
+    {
+      "epoch": 0.04713226601863458,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.0016110000000000002,
+      "loss": 3.8486,
+      "step": 537
+    },
+    {
+      "epoch": 0.047220035601537064,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0016140000000000002,
+      "loss": 3.8882,
+      "step": 538
+    },
+    {
+      "epoch": 0.04730780518443955,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.0016170000000000002,
+      "loss": 3.8584,
+      "step": 539
+    },
+    {
+      "epoch": 0.04739557476734203,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0016200000000000001,
+      "loss": 3.8384,
+      "step": 540
+    },
+    {
+      "epoch": 0.04748334435024452,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.001623,
+      "loss": 3.9512,
+      "step": 541
+    },
+    {
+      "epoch": 0.04757111393314701,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.001626,
+      "loss": 3.9414,
+      "step": 542
+    },
+    {
+      "epoch": 0.047658883516049494,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.0016290000000000002,
+      "loss": 3.8887,
+      "step": 543
+    },
+    {
+      "epoch": 0.04774665309895198,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.0016320000000000002,
+      "loss": 3.9502,
+      "step": 544
+    },
+    {
+      "epoch": 0.04783442268185446,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.0016350000000000002,
+      "loss": 3.8711,
+      "step": 545
+    },
+    {
+      "epoch": 0.04792219226475695,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.0016380000000000001,
+      "loss": 3.9072,
+      "step": 546
+    },
+    {
+      "epoch": 0.04800996184765943,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.001641,
+      "loss": 3.9033,
+      "step": 547
+    },
+    {
+      "epoch": 0.04809773143056192,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.001644,
+      "loss": 3.8638,
+      "step": 548
+    },
+    {
+      "epoch": 0.0481855010134644,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.0016470000000000002,
+      "loss": 3.8281,
+      "step": 549
+    },
+    {
+      "epoch": 0.048273270596366886,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0016500000000000002,
+      "loss": 3.8472,
+      "step": 550
+    },
+    {
+      "epoch": 0.04836104017926937,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.0016530000000000002,
+      "loss": 3.8496,
+      "step": 551
+    },
+    {
+      "epoch": 0.048448809762171856,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.0016560000000000001,
+      "loss": 3.9116,
+      "step": 552
+    },
+    {
+      "epoch": 0.04853657934507434,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.001659,
+      "loss": 3.8252,
+      "step": 553
+    },
+    {
+      "epoch": 0.04862434892797683,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.0016620000000000003,
+      "loss": 3.79,
+      "step": 554
+    },
+    {
+      "epoch": 0.04871211851087932,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.0016650000000000002,
+      "loss": 3.8506,
+      "step": 555
+    },
+    {
+      "epoch": 0.0487998880937818,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.0016680000000000002,
+      "loss": 3.8159,
+      "step": 556
+    },
+    {
+      "epoch": 0.048887657676684286,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.0016710000000000002,
+      "loss": 3.811,
+      "step": 557
+    },
+    {
+      "epoch": 0.04897542725958677,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.0016740000000000001,
+      "loss": 3.877,
+      "step": 558
+    },
+    {
+      "epoch": 0.049063196842489255,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.001677,
+      "loss": 3.873,
+      "step": 559
+    },
+    {
+      "epoch": 0.04915096642539174,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0016800000000000003,
+      "loss": 3.8652,
+      "step": 560
+    },
+    {
+      "epoch": 0.049238736008294225,
+      "grad_norm": 0.482421875,
+      "learning_rate": 0.0016830000000000003,
+      "loss": 3.8296,
+      "step": 561
+    },
+    {
+      "epoch": 0.04932650559119671,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.0016860000000000002,
+      "loss": 3.9292,
+      "step": 562
+    },
+    {
+      "epoch": 0.049414275174099194,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.001689,
+      "loss": 3.7979,
+      "step": 563
+    },
+    {
+      "epoch": 0.04950204475700168,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.001692,
+      "loss": 3.8418,
+      "step": 564
+    },
+    {
+      "epoch": 0.04958981433990416,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.001695,
+      "loss": 3.8462,
+      "step": 565
+    },
+    {
+      "epoch": 0.049677583922806655,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.0016979999999999999,
+      "loss": 3.8521,
+      "step": 566
+    },
+    {
+      "epoch": 0.04976535350570914,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.0017009999999999998,
+      "loss": 3.8589,
+      "step": 567
+    },
+    {
+      "epoch": 0.049853123088611624,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0017039999999999998,
+      "loss": 3.7671,
+      "step": 568
+    },
+    {
+      "epoch": 0.04994089267151411,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.001707,
+      "loss": 3.8135,
+      "step": 569
+    },
+    {
+      "epoch": 0.050028662254416593,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.00171,
+      "loss": 3.8589,
+      "step": 570
+    },
+    {
+      "epoch": 0.05011643183731908,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.001713,
+      "loss": 3.8003,
+      "step": 571
+    },
+    {
+      "epoch": 0.05020420142022156,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.0017159999999999999,
+      "loss": 3.8389,
+      "step": 572
+    },
+    {
+      "epoch": 0.05029197100312405,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.0017189999999999998,
+      "loss": 3.7534,
+      "step": 573
+    },
+    {
+      "epoch": 0.05037974058602653,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.001722,
+      "loss": 3.7896,
+      "step": 574
+    },
+    {
+      "epoch": 0.05046751016892902,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.001725,
+      "loss": 3.8091,
+      "step": 575
+    },
+    {
+      "epoch": 0.0505552797518315,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.001728,
+      "loss": 3.8789,
+      "step": 576
+    },
+    {
+      "epoch": 0.05064304933473399,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.001731,
+      "loss": 3.7671,
+      "step": 577
+    },
+    {
+      "epoch": 0.05073081891763648,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0017339999999999999,
+      "loss": 3.8643,
+      "step": 578
+    },
+    {
+      "epoch": 0.05081858850053896,
+      "grad_norm": 2.609375,
+      "learning_rate": 0.0017369999999999998,
+      "loss": 3.8315,
+      "step": 579
+    },
+    {
+      "epoch": 0.05090635808344145,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00174,
+      "loss": 3.7935,
+      "step": 580
+    },
+    {
+      "epoch": 0.05099412766634393,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.001743,
+      "loss": 3.7778,
+      "step": 581
+    },
+    {
+      "epoch": 0.051081897249246416,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.001746,
+      "loss": 3.8335,
+      "step": 582
+    },
+    {
+      "epoch": 0.0511696668321489,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.001749,
+      "loss": 3.7993,
+      "step": 583
+    },
+    {
+      "epoch": 0.051257436415051386,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.0017519999999999999,
+      "loss": 3.8672,
+      "step": 584
+    },
+    {
+      "epoch": 0.05134520599795387,
+      "grad_norm": 1.984375,
+      "learning_rate": 0.0017549999999999998,
+      "loss": 3.8003,
+      "step": 585
+    },
+    {
+      "epoch": 0.051432975580856355,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.001758,
+      "loss": 3.7607,
+      "step": 586
+    },
+    {
+      "epoch": 0.05152074516375884,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.001761,
+      "loss": 3.8931,
+      "step": 587
+    },
+    {
+      "epoch": 0.051608514746661324,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.001764,
+      "loss": 3.7676,
+      "step": 588
+    },
+    {
+      "epoch": 0.051696284329563816,
+      "grad_norm": 2.5625,
+      "learning_rate": 0.001767,
+      "loss": 3.8403,
+      "step": 589
+    },
+    {
+      "epoch": 0.0517840539124663,
+      "grad_norm": 1.390625,
+      "learning_rate": 0.0017699999999999999,
+      "loss": 3.8208,
+      "step": 590
+    },
+    {
+      "epoch": 0.051871823495368785,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.001773,
+      "loss": 3.8799,
+      "step": 591
+    },
+    {
+      "epoch": 0.05195959307827127,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.001776,
+      "loss": 3.8438,
+      "step": 592
+    },
+    {
+      "epoch": 0.052047362661173754,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.001779,
+      "loss": 3.8594,
+      "step": 593
+    },
+    {
+      "epoch": 0.05213513224407624,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.001782,
+      "loss": 3.8071,
+      "step": 594
+    },
+    {
+      "epoch": 0.052222901826978724,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.001785,
+      "loss": 3.8213,
+      "step": 595
+    },
+    {
+      "epoch": 0.05231067140988121,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.0017879999999999999,
+      "loss": 3.812,
+      "step": 596
+    },
+    {
+      "epoch": 0.05239844099278369,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.001791,
+      "loss": 3.8872,
+      "step": 597
+    },
+    {
+      "epoch": 0.05248621057568618,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.001794,
+      "loss": 3.894,
+      "step": 598
+    },
+    {
+      "epoch": 0.05257398015858866,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.001797,
+      "loss": 3.8413,
+      "step": 599
+    },
+    {
+      "epoch": 0.05266174974149115,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.0018,
+      "loss": 3.9023,
+      "step": 600
+    },
+    {
+      "epoch": 0.05274951932439364,
+      "grad_norm": 6.1875,
+      "learning_rate": 0.001803,
+      "loss": 3.8579,
+      "step": 601
+    },
+    {
+      "epoch": 0.05283728890729612,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0018059999999999999,
+      "loss": 3.7905,
+      "step": 602
+    },
+    {
+      "epoch": 0.05292505849019861,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.001809,
+      "loss": 3.8398,
+      "step": 603
+    },
+    {
+      "epoch": 0.05301282807310109,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.001812,
+      "loss": 3.7539,
+      "step": 604
+    },
+    {
+      "epoch": 0.05310059765600358,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.001815,
+      "loss": 3.8086,
+      "step": 605
+    },
+    {
+      "epoch": 0.05318836723890606,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.001818,
+      "loss": 3.7642,
+      "step": 606
+    },
+    {
+      "epoch": 0.05327613682180855,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.001821,
+      "loss": 3.7954,
+      "step": 607
+    },
+    {
+      "epoch": 0.05336390640471103,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.001824,
+      "loss": 3.7739,
+      "step": 608
+    },
+    {
+      "epoch": 0.053451675987613516,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.001827,
+      "loss": 3.8013,
+      "step": 609
+    },
+    {
+      "epoch": 0.053539445570516,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.00183,
+      "loss": 3.7886,
+      "step": 610
+    },
+    {
+      "epoch": 0.053627215153418485,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.001833,
+      "loss": 3.7124,
+      "step": 611
+    },
+    {
+      "epoch": 0.05371498473632098,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.001836,
+      "loss": 3.8032,
+      "step": 612
+    },
+    {
+      "epoch": 0.05380275431922346,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.001839,
+      "loss": 3.7935,
+      "step": 613
+    },
+    {
+      "epoch": 0.053890523902125946,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.001842,
+      "loss": 3.73,
+      "step": 614
+    },
+    {
+      "epoch": 0.05397829348502843,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.001845,
+      "loss": 3.8667,
+      "step": 615
+    },
+    {
+      "epoch": 0.054066063067930915,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.001848,
+      "loss": 3.8242,
+      "step": 616
+    },
+    {
+      "epoch": 0.0541538326508334,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.001851,
+      "loss": 3.8115,
+      "step": 617
+    },
+    {
+      "epoch": 0.054241602233735885,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.001854,
+      "loss": 3.7793,
+      "step": 618
+    },
+    {
+      "epoch": 0.05432937181663837,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.001857,
+      "loss": 3.7954,
+      "step": 619
+    },
+    {
+      "epoch": 0.054417141399540854,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.00186,
+      "loss": 3.8774,
+      "step": 620
+    },
+    {
+      "epoch": 0.05450491098244334,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.001863,
+      "loss": 3.8413,
+      "step": 621
+    },
+    {
+      "epoch": 0.05459268056534582,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.001866,
+      "loss": 3.8218,
+      "step": 622
+    },
+    {
+      "epoch": 0.05468045014824831,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.001869,
+      "loss": 3.7842,
+      "step": 623
+    },
+    {
+      "epoch": 0.0547682197311508,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.001872,
+      "loss": 3.8037,
+      "step": 624
+    },
+    {
+      "epoch": 0.054855989314053284,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.001875,
+      "loss": 3.7056,
+      "step": 625
+    },
+    {
+      "epoch": 0.05494375889695577,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.0018780000000000001,
+      "loss": 3.8032,
+      "step": 626
+    },
+    {
+      "epoch": 0.055031528479858254,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.001881,
+      "loss": 3.8687,
+      "step": 627
+    },
+    {
+      "epoch": 0.05511929806276074,
+      "grad_norm": 0.345703125,
+      "learning_rate": 0.001884,
+      "loss": 3.7104,
+      "step": 628
+    },
+    {
+      "epoch": 0.05520706764566322,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.001887,
+      "loss": 3.7793,
+      "step": 629
+    },
+    {
+      "epoch": 0.05529483722856571,
+      "grad_norm": 0.31640625,
+      "learning_rate": 0.00189,
+      "loss": 3.8047,
+      "step": 630
+    },
+    {
+      "epoch": 0.05538260681146819,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0018930000000000002,
+      "loss": 3.791,
+      "step": 631
+    },
+    {
+      "epoch": 0.05547037639437068,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.0018960000000000001,
+      "loss": 3.7681,
+      "step": 632
+    },
+    {
+      "epoch": 0.05555814597727316,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.001899,
+      "loss": 3.7451,
+      "step": 633
+    },
+    {
+      "epoch": 0.055645915560175646,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.001902,
+      "loss": 3.772,
+      "step": 634
+    },
+    {
+      "epoch": 0.05573368514307813,
+      "grad_norm": 0.3203125,
+      "learning_rate": 0.001905,
+      "loss": 3.8345,
+      "step": 635
+    },
+    {
+      "epoch": 0.05582145472598062,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.001908,
+      "loss": 3.77,
+      "step": 636
+    },
+    {
+      "epoch": 0.05590922430888311,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.0019110000000000002,
+      "loss": 3.8013,
+      "step": 637
+    },
+    {
+      "epoch": 0.05599699389178559,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.0019140000000000001,
+      "loss": 3.7749,
+      "step": 638
+    },
+    {
+      "epoch": 0.056084763474688076,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.001917,
+      "loss": 3.8413,
+      "step": 639
+    },
+    {
+      "epoch": 0.05617253305759056,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.00192,
+      "loss": 3.7783,
+      "step": 640
+    },
+    {
+      "epoch": 0.056260302640493046,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.001923,
+      "loss": 3.7515,
+      "step": 641
+    },
+    {
+      "epoch": 0.05634807222339553,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.001926,
+      "loss": 3.8101,
+      "step": 642
+    },
+    {
+      "epoch": 0.056435841806298015,
+      "grad_norm": 1.5390625,
+      "learning_rate": 0.0019290000000000002,
+      "loss": 3.8691,
+      "step": 643
+    },
+    {
+      "epoch": 0.0565236113892005,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.0019320000000000001,
+      "loss": 3.8071,
+      "step": 644
+    },
+    {
+      "epoch": 0.056611380972102984,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.001935,
+      "loss": 3.8032,
+      "step": 645
+    },
+    {
+      "epoch": 0.05669915055500547,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.001938,
+      "loss": 3.7886,
+      "step": 646
+    },
+    {
+      "epoch": 0.05678692013790796,
+      "grad_norm": 1.53125,
+      "learning_rate": 0.001941,
+      "loss": 3.7485,
+      "step": 647
+    },
+    {
+      "epoch": 0.056874689720810445,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0019440000000000002,
+      "loss": 3.7437,
+      "step": 648
+    },
+    {
+      "epoch": 0.05696245930371293,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.0019470000000000002,
+      "loss": 3.8135,
+      "step": 649
+    },
+    {
+      "epoch": 0.057050228886615414,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.0019500000000000001,
+      "loss": 3.8784,
+      "step": 650
+    },
+    {
+      "epoch": 0.0571379984695179,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.001953,
+      "loss": 3.7544,
+      "step": 651
+    },
+    {
+      "epoch": 0.057225768052420384,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0019560000000000003,
+      "loss": 3.7856,
+      "step": 652
+    },
+    {
+      "epoch": 0.05731353763532287,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.0019590000000000002,
+      "loss": 3.7559,
+      "step": 653
+    },
+    {
+      "epoch": 0.05740130721822535,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.001962,
+      "loss": 3.6909,
+      "step": 654
+    },
+    {
+      "epoch": 0.05748907680112784,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.001965,
+      "loss": 3.8457,
+      "step": 655
+    },
+    {
+      "epoch": 0.05757684638403032,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.001968,
+      "loss": 3.7324,
+      "step": 656
+    },
+    {
+      "epoch": 0.05766461596693281,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.001971,
+      "loss": 3.7222,
+      "step": 657
+    },
+    {
+      "epoch": 0.05775238554983529,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.001974,
+      "loss": 3.7612,
+      "step": 658
+    },
+    {
+      "epoch": 0.05784015513273778,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.001977,
+      "loss": 3.6855,
+      "step": 659
+    },
+    {
+      "epoch": 0.05792792471564027,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.00198,
+      "loss": 3.7402,
+      "step": 660
+    },
+    {
+      "epoch": 0.05801569429854275,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.001983,
+      "loss": 3.7339,
+      "step": 661
+    },
+    {
+      "epoch": 0.05810346388144524,
+      "grad_norm": 0.31640625,
+      "learning_rate": 0.0019860000000000004,
+      "loss": 3.7827,
+      "step": 662
+    },
+    {
+      "epoch": 0.05819123346434772,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0019890000000000003,
+      "loss": 3.7568,
+      "step": 663
+    },
+    {
+      "epoch": 0.05827900304725021,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.0019920000000000003,
+      "loss": 3.8013,
+      "step": 664
+    },
+    {
+      "epoch": 0.05836677263015269,
+      "grad_norm": 0.421875,
+      "learning_rate": 0.0019950000000000002,
+      "loss": 3.7363,
+      "step": 665
+    },
+    {
+      "epoch": 0.058454542213055176,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.001998,
+      "loss": 3.7886,
+      "step": 666
+    },
+    {
+      "epoch": 0.05854231179595766,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.002001,
+      "loss": 3.7139,
+      "step": 667
+    },
+    {
+      "epoch": 0.058630081378860145,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.002004,
+      "loss": 3.689,
+      "step": 668
+    },
+    {
+      "epoch": 0.05871785096176263,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.002007,
+      "loss": 3.6919,
+      "step": 669
+    },
+    {
+      "epoch": 0.058805620544665115,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00201,
+      "loss": 3.7578,
+      "step": 670
+    },
+    {
+      "epoch": 0.058893390127567606,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.002013,
+      "loss": 3.813,
+      "step": 671
+    },
+    {
+      "epoch": 0.05898115971047009,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.002016,
+      "loss": 3.814,
+      "step": 672
+    },
+    {
+      "epoch": 0.059068929293372575,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.002019,
+      "loss": 3.8003,
+      "step": 673
+    },
+    {
+      "epoch": 0.05915669887627506,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.0020220000000000004,
+      "loss": 3.73,
+      "step": 674
+    },
+    {
+      "epoch": 0.059244468459177545,
+      "grad_norm": 0.93359375,
+      "learning_rate": 0.0020250000000000003,
+      "loss": 3.7837,
+      "step": 675
+    },
+    {
+      "epoch": 0.05933223804208003,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.0020280000000000003,
+      "loss": 3.7627,
+      "step": 676
+    },
+    {
+      "epoch": 0.059420007624982514,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.0020310000000000003,
+      "loss": 3.7266,
+      "step": 677
+    },
+    {
+      "epoch": 0.059507777207885,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.0020340000000000002,
+      "loss": 3.7349,
+      "step": 678
+    },
+    {
+      "epoch": 0.05959554679078748,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.002037,
+      "loss": 3.832,
+      "step": 679
+    },
+    {
+      "epoch": 0.05968331637368997,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00204,
+      "loss": 3.8428,
+      "step": 680
+    },
+    {
+      "epoch": 0.05977108595659245,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.002043,
+      "loss": 3.8057,
+      "step": 681
+    },
+    {
+      "epoch": 0.059858855539494944,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.002046,
+      "loss": 3.7397,
+      "step": 682
+    },
+    {
+      "epoch": 0.05994662512239743,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.002049,
+      "loss": 3.769,
+      "step": 683
+    },
+    {
+      "epoch": 0.060034394705299914,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.002052,
+      "loss": 3.7515,
+      "step": 684
+    },
+    {
+      "epoch": 0.0601221642882024,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.0020550000000000004,
+      "loss": 3.7881,
+      "step": 685
+    },
+    {
+      "epoch": 0.06020993387110488,
+      "grad_norm": 0.326171875,
+      "learning_rate": 0.0020580000000000004,
+      "loss": 3.7822,
+      "step": 686
+    },
+    {
+      "epoch": 0.06029770345400737,
+      "grad_norm": 0.375,
+      "learning_rate": 0.0020610000000000003,
+      "loss": 3.7397,
+      "step": 687
+    },
+    {
+      "epoch": 0.06038547303690985,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.002064,
+      "loss": 3.7529,
+      "step": 688
+    },
+    {
+      "epoch": 0.06047324261981234,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.002067,
+      "loss": 3.6914,
+      "step": 689
+    },
+    {
+      "epoch": 0.06056101220271482,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00207,
+      "loss": 3.7534,
+      "step": 690
+    },
+    {
+      "epoch": 0.060648781785617306,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.0020729999999999998,
+      "loss": 3.731,
+      "step": 691
+    },
+    {
+      "epoch": 0.06073655136851979,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.0020759999999999997,
+      "loss": 3.7261,
+      "step": 692
+    },
+    {
+      "epoch": 0.060824320951422275,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.0020789999999999997,
+      "loss": 3.7568,
+      "step": 693
+    },
+    {
+      "epoch": 0.06091209053432477,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.002082,
+      "loss": 3.7568,
+      "step": 694
+    },
+    {
+      "epoch": 0.06099986011722725,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.002085,
+      "loss": 3.8052,
+      "step": 695
+    },
+    {
+      "epoch": 0.061087629700129736,
+      "grad_norm": 0.46875,
+      "learning_rate": 0.002088,
+      "loss": 3.7393,
+      "step": 696
+    },
+    {
+      "epoch": 0.06117539928303222,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.002091,
+      "loss": 3.7939,
+      "step": 697
+    },
+    {
+      "epoch": 0.061263168865934706,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.002094,
+      "loss": 3.7563,
+      "step": 698
+    },
+    {
+      "epoch": 0.06135093844883719,
+      "grad_norm": 0.8125,
+      "learning_rate": 0.002097,
+      "loss": 3.7168,
+      "step": 699
+    },
+    {
+      "epoch": 0.061438708031739675,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.0021,
+      "loss": 3.7871,
+      "step": 700
+    },
+    {
+      "epoch": 0.06152647761464216,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.002103,
+      "loss": 3.7847,
+      "step": 701
+    },
+    {
+      "epoch": 0.061614247197544644,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.002106,
+      "loss": 3.7852,
+      "step": 702
+    },
+    {
+      "epoch": 0.06170201678044713,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.0021089999999999998,
+      "loss": 3.7314,
+      "step": 703
+    },
+    {
+      "epoch": 0.061789786363349614,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.0021119999999999997,
+      "loss": 3.7412,
+      "step": 704
+    },
+    {
+      "epoch": 0.0618775559462521,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.002115,
+      "loss": 3.7026,
+      "step": 705
+    },
+    {
+      "epoch": 0.06196532552915459,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.002118,
+      "loss": 3.7529,
+      "step": 706
+    },
+    {
+      "epoch": 0.062053095112057075,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.002121,
+      "loss": 3.7627,
+      "step": 707
+    },
+    {
+      "epoch": 0.06214086469495956,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.002124,
+      "loss": 3.7388,
+      "step": 708
+    },
+    {
+      "epoch": 0.062228634277862044,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.002127,
+      "loss": 3.7339,
+      "step": 709
+    },
+    {
+      "epoch": 0.06231640386076453,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00213,
+      "loss": 3.7705,
+      "step": 710
+    },
+    {
+      "epoch": 0.06240417344366701,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.002133,
+      "loss": 3.7134,
+      "step": 711
+    },
+    {
+      "epoch": 0.0624919430265695,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.002136,
+      "loss": 3.7803,
+      "step": 712
+    },
+    {
+      "epoch": 0.06257971260947198,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.002139,
+      "loss": 3.7539,
+      "step": 713
+    },
+    {
+      "epoch": 0.06266748219237447,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.002142,
+      "loss": 3.8052,
+      "step": 714
+    },
+    {
+      "epoch": 0.06275525177527695,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0021449999999999998,
+      "loss": 3.6963,
+      "step": 715
+    },
+    {
+      "epoch": 0.06284302135817944,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.002148,
+      "loss": 3.7109,
+      "step": 716
+    },
+    {
+      "epoch": 0.06293079094108192,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.002151,
+      "loss": 3.665,
+      "step": 717
+    },
+    {
+      "epoch": 0.06301856052398441,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.002154,
+      "loss": 3.71,
+      "step": 718
+    },
+    {
+      "epoch": 0.06310633010688689,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.002157,
+      "loss": 3.7148,
+      "step": 719
+    },
+    {
+      "epoch": 0.06319409968978938,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00216,
+      "loss": 3.6938,
+      "step": 720
+    },
+    {
+      "epoch": 0.06328186927269186,
+      "grad_norm": 6.9375,
+      "learning_rate": 0.002163,
+      "loss": 3.8232,
+      "step": 721
+    },
+    {
+      "epoch": 0.06336963885559435,
+      "grad_norm": 1.28125,
+      "learning_rate": 0.002166,
+      "loss": 3.8701,
+      "step": 722
+    },
+    {
+      "epoch": 0.06345740843849684,
+      "grad_norm": 1.6484375,
+      "learning_rate": 0.002169,
+      "loss": 3.7744,
+      "step": 723
+    },
+    {
+      "epoch": 0.06354517802139932,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.002172,
+      "loss": 3.7949,
+      "step": 724
+    },
+    {
+      "epoch": 0.06363294760430181,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.002175,
+      "loss": 3.7656,
+      "step": 725
+    },
+    {
+      "epoch": 0.06372071718720429,
+      "grad_norm": 0.5,
+      "learning_rate": 0.002178,
+      "loss": 3.7471,
+      "step": 726
+    },
+    {
+      "epoch": 0.06380848677010678,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.0021809999999999998,
+      "loss": 3.686,
+      "step": 727
+    },
+    {
+      "epoch": 0.06389625635300926,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.002184,
+      "loss": 3.7002,
+      "step": 728
+    },
+    {
+      "epoch": 0.06398402593591175,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.002187,
+      "loss": 3.6968,
+      "step": 729
+    },
+    {
+      "epoch": 0.06407179551881423,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.00219,
+      "loss": 3.7095,
+      "step": 730
+    },
+    {
+      "epoch": 0.06415956510171672,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.002193,
+      "loss": 3.7622,
+      "step": 731
+    },
+    {
+      "epoch": 0.0642473346846192,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.002196,
+      "loss": 3.8252,
+      "step": 732
+    },
+    {
+      "epoch": 0.06433510426752169,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.002199,
+      "loss": 3.7144,
+      "step": 733
+    },
+    {
+      "epoch": 0.06442287385042417,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.002202,
+      "loss": 3.7085,
+      "step": 734
+    },
+    {
+      "epoch": 0.06451064343332666,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.002205,
+      "loss": 3.7637,
+      "step": 735
+    },
+    {
+      "epoch": 0.06459841301622915,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.002208,
+      "loss": 3.7861,
+      "step": 736
+    },
+    {
+      "epoch": 0.06468618259913163,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.002211,
+      "loss": 3.7471,
+      "step": 737
+    },
+    {
+      "epoch": 0.06477395218203412,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.002214,
+      "loss": 3.748,
+      "step": 738
+    },
+    {
+      "epoch": 0.0648617217649366,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.0022170000000000002,
+      "loss": 3.7354,
+      "step": 739
+    },
+    {
+      "epoch": 0.06494949134783909,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00222,
+      "loss": 3.6177,
+      "step": 740
+    },
+    {
+      "epoch": 0.06503726093074157,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.002223,
+      "loss": 3.7783,
+      "step": 741
+    },
+    {
+      "epoch": 0.06512503051364406,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.002226,
+      "loss": 3.7266,
+      "step": 742
+    },
+    {
+      "epoch": 0.06521280009654654,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.002229,
+      "loss": 3.7812,
+      "step": 743
+    },
+    {
+      "epoch": 0.06530056967944903,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.002232,
+      "loss": 3.6904,
+      "step": 744
+    },
+    {
+      "epoch": 0.0653883392623515,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.002235,
+      "loss": 3.7261,
+      "step": 745
+    },
+    {
+      "epoch": 0.065476108845254,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.002238,
+      "loss": 3.7495,
+      "step": 746
+    },
+    {
+      "epoch": 0.06556387842815649,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.002241,
+      "loss": 3.7397,
+      "step": 747
+    },
+    {
+      "epoch": 0.06565164801105897,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.002244,
+      "loss": 3.7744,
+      "step": 748
+    },
+    {
+      "epoch": 0.06573941759396146,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.002247,
+      "loss": 3.7305,
+      "step": 749
+    },
+    {
+      "epoch": 0.06582718717686394,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.0022500000000000003,
+      "loss": 3.6714,
+      "step": 750
+    },
+    {
+      "epoch": 0.06591495675976643,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0022530000000000002,
+      "loss": 3.7026,
+      "step": 751
+    },
+    {
+      "epoch": 0.0660027263426689,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.002256,
+      "loss": 3.7227,
+      "step": 752
+    },
+    {
+      "epoch": 0.0660904959255714,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.002259,
+      "loss": 3.6636,
+      "step": 753
+    },
+    {
+      "epoch": 0.06617826550847387,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.002262,
+      "loss": 3.7505,
+      "step": 754
+    },
+    {
+      "epoch": 0.06626603509137637,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.002265,
+      "loss": 3.6899,
+      "step": 755
+    },
+    {
+      "epoch": 0.06635380467427884,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.002268,
+      "loss": 3.7573,
+      "step": 756
+    },
+    {
+      "epoch": 0.06644157425718134,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.002271,
+      "loss": 3.7549,
+      "step": 757
+    },
+    {
+      "epoch": 0.06652934384008383,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.002274,
+      "loss": 3.7002,
+      "step": 758
+    },
+    {
+      "epoch": 0.0666171134229863,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.002277,
+      "loss": 3.7773,
+      "step": 759
+    },
+    {
+      "epoch": 0.0667048830058888,
+      "grad_norm": 0.326171875,
+      "learning_rate": 0.00228,
+      "loss": 3.769,
+      "step": 760
+    },
+    {
+      "epoch": 0.06679265258879127,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.002283,
+      "loss": 3.6855,
+      "step": 761
+    },
+    {
+      "epoch": 0.06688042217169377,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.0022860000000000003,
+      "loss": 3.7285,
+      "step": 762
+    },
+    {
+      "epoch": 0.06696819175459624,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.0022890000000000002,
+      "loss": 3.7085,
+      "step": 763
+    },
+    {
+      "epoch": 0.06705596133749873,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.002292,
+      "loss": 3.7393,
+      "step": 764
+    },
+    {
+      "epoch": 0.06714373092040121,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.002295,
+      "loss": 3.7393,
+      "step": 765
+    },
+    {
+      "epoch": 0.0672315005033037,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.002298,
+      "loss": 3.7456,
+      "step": 766
+    },
+    {
+      "epoch": 0.06731927008620618,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.002301,
+      "loss": 3.686,
+      "step": 767
+    },
+    {
+      "epoch": 0.06740703966910867,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.002304,
+      "loss": 3.7363,
+      "step": 768
+    },
+    {
+      "epoch": 0.06749480925201115,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.002307,
+      "loss": 3.686,
+      "step": 769
+    },
+    {
+      "epoch": 0.06758257883491364,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.00231,
+      "loss": 3.7188,
+      "step": 770
+    },
+    {
+      "epoch": 0.06767034841781613,
+      "grad_norm": 1.2265625,
+      "learning_rate": 0.002313,
+      "loss": 3.6997,
+      "step": 771
+    },
+    {
+      "epoch": 0.06775811800071861,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.002316,
+      "loss": 3.7358,
+      "step": 772
+    },
+    {
+      "epoch": 0.0678458875836211,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.0023190000000000003,
+      "loss": 3.7314,
+      "step": 773
+    },
+    {
+      "epoch": 0.06793365716652358,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.0023220000000000003,
+      "loss": 3.7417,
+      "step": 774
+    },
+    {
+      "epoch": 0.06802142674942607,
+      "grad_norm": 0.625,
+      "learning_rate": 0.0023250000000000002,
+      "loss": 3.7671,
+      "step": 775
+    },
+    {
+      "epoch": 0.06810919633232855,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.002328,
+      "loss": 3.7241,
+      "step": 776
+    },
+    {
+      "epoch": 0.06819696591523104,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.002331,
+      "loss": 3.748,
+      "step": 777
+    },
+    {
+      "epoch": 0.06828473549813352,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.002334,
+      "loss": 3.7295,
+      "step": 778
+    },
+    {
+      "epoch": 0.06837250508103601,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.002337,
+      "loss": 3.7632,
+      "step": 779
+    },
+    {
+      "epoch": 0.06846027466393849,
+      "grad_norm": 0.326171875,
+      "learning_rate": 0.00234,
+      "loss": 3.708,
+      "step": 780
+    },
+    {
+      "epoch": 0.06854804424684098,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.002343,
+      "loss": 3.7207,
+      "step": 781
+    },
+    {
+      "epoch": 0.06863581382974347,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.002346,
+      "loss": 3.6948,
+      "step": 782
+    },
+    {
+      "epoch": 0.06872358341264595,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.002349,
+      "loss": 3.7104,
+      "step": 783
+    },
+    {
+      "epoch": 0.06881135299554844,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.002352,
+      "loss": 3.6846,
+      "step": 784
+    },
+    {
+      "epoch": 0.06889912257845092,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0023550000000000003,
+      "loss": 3.7046,
+      "step": 785
+    },
+    {
+      "epoch": 0.06898689216135341,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.0023580000000000003,
+      "loss": 3.6807,
+      "step": 786
+    },
+    {
+      "epoch": 0.06907466174425589,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0023610000000000003,
+      "loss": 3.7573,
+      "step": 787
+    },
+    {
+      "epoch": 0.06916243132715838,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.002364,
+      "loss": 3.6899,
+      "step": 788
+    },
+    {
+      "epoch": 0.06925020091006086,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.002367,
+      "loss": 3.7749,
+      "step": 789
+    },
+    {
+      "epoch": 0.06933797049296335,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.00237,
+      "loss": 3.6626,
+      "step": 790
+    },
+    {
+      "epoch": 0.06942574007586583,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.002373,
+      "loss": 3.7949,
+      "step": 791
+    },
+    {
+      "epoch": 0.06951350965876832,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.002376,
+      "loss": 3.6807,
+      "step": 792
+    },
+    {
+      "epoch": 0.06960127924167081,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.002379,
+      "loss": 3.6987,
+      "step": 793
+    },
+    {
+      "epoch": 0.06968904882457329,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.002382,
+      "loss": 3.7559,
+      "step": 794
+    },
+    {
+      "epoch": 0.06977681840747578,
+      "grad_norm": 0.375,
+      "learning_rate": 0.002385,
+      "loss": 3.6919,
+      "step": 795
+    },
+    {
+      "epoch": 0.06986458799037826,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.0023880000000000004,
+      "loss": 3.6982,
+      "step": 796
+    },
+    {
+      "epoch": 0.06995235757328075,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.0023910000000000003,
+      "loss": 3.7075,
+      "step": 797
+    },
+    {
+      "epoch": 0.07004012715618323,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.0023940000000000003,
+      "loss": 3.6851,
+      "step": 798
+    },
+    {
+      "epoch": 0.07012789673908572,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.0023970000000000003,
+      "loss": 3.6602,
+      "step": 799
+    },
+    {
+      "epoch": 0.0702156663219882,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.0024000000000000002,
+      "loss": 3.7002,
+      "step": 800
+    },
+    {
+      "epoch": 0.07030343590489069,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.002403,
+      "loss": 3.626,
+      "step": 801
+    },
+    {
+      "epoch": 0.07039120548779317,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.002406,
+      "loss": 3.6548,
+      "step": 802
+    },
+    {
+      "epoch": 0.07047897507069566,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.002409,
+      "loss": 3.73,
+      "step": 803
+    },
+    {
+      "epoch": 0.07056674465359813,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.002412,
+      "loss": 3.6968,
+      "step": 804
+    },
+    {
+      "epoch": 0.07065451423650063,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.002415,
+      "loss": 3.7441,
+      "step": 805
+    },
+    {
+      "epoch": 0.07074228381940312,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.002418,
+      "loss": 3.7603,
+      "step": 806
+    },
+    {
+      "epoch": 0.0708300534023056,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.0024210000000000004,
+      "loss": 3.791,
+      "step": 807
+    },
+    {
+      "epoch": 0.07091782298520809,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.0024240000000000004,
+      "loss": 3.7222,
+      "step": 808
+    },
+    {
+      "epoch": 0.07100559256811056,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.0024270000000000003,
+      "loss": 3.6899,
+      "step": 809
+    },
+    {
+      "epoch": 0.07109336215101306,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.0024300000000000003,
+      "loss": 3.7476,
+      "step": 810
+    },
+    {
+      "epoch": 0.07118113173391553,
+      "grad_norm": 0.8125,
+      "learning_rate": 0.0024330000000000003,
+      "loss": 3.7065,
+      "step": 811
+    },
+    {
+      "epoch": 0.07126890131681803,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.0024360000000000002,
+      "loss": 3.7705,
+      "step": 812
+    },
+    {
+      "epoch": 0.0713566708997205,
+      "grad_norm": 3.5,
+      "learning_rate": 0.0024389999999999998,
+      "loss": 3.7329,
+      "step": 813
+    },
+    {
+      "epoch": 0.071444440482623,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.0024419999999999997,
+      "loss": 3.7852,
+      "step": 814
+    },
+    {
+      "epoch": 0.07153221006552547,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.0024449999999999997,
+      "loss": 3.7646,
+      "step": 815
+    },
+    {
+      "epoch": 0.07161997964842796,
+      "grad_norm": 3.765625,
+      "learning_rate": 0.002448,
+      "loss": 3.7314,
+      "step": 816
+    },
+    {
+      "epoch": 0.07170774923133046,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.002451,
+      "loss": 3.6938,
+      "step": 817
+    },
+    {
+      "epoch": 0.07179551881423293,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.002454,
+      "loss": 3.7383,
+      "step": 818
+    },
+    {
+      "epoch": 0.07188328839713543,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.002457,
+      "loss": 3.7593,
+      "step": 819
+    },
+    {
+      "epoch": 0.0719710579800379,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.00246,
+      "loss": 3.6699,
+      "step": 820
+    },
+    {
+      "epoch": 0.0720588275629404,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.002463,
+      "loss": 3.752,
+      "step": 821
+    },
+    {
+      "epoch": 0.07214659714584287,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.002466,
+      "loss": 3.7798,
+      "step": 822
+    },
+    {
+      "epoch": 0.07223436672874536,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.002469,
+      "loss": 3.7441,
+      "step": 823
+    },
+    {
+      "epoch": 0.07232213631164784,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.002472,
+      "loss": 3.7021,
+      "step": 824
+    },
+    {
+      "epoch": 0.07240990589455033,
+      "grad_norm": 0.443359375,
+      "learning_rate": 0.0024749999999999998,
+      "loss": 3.7612,
+      "step": 825
+    },
+    {
+      "epoch": 0.07249767547745281,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0024779999999999997,
+      "loss": 3.6611,
+      "step": 826
+    },
+    {
+      "epoch": 0.0725854450603553,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.002481,
+      "loss": 3.6909,
+      "step": 827
+    },
+    {
+      "epoch": 0.0726732146432578,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.002484,
+      "loss": 3.7026,
+      "step": 828
+    },
+    {
+      "epoch": 0.07276098422616027,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.002487,
+      "loss": 3.6343,
+      "step": 829
+    },
+    {
+      "epoch": 0.07284875380906276,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.00249,
+      "loss": 3.7246,
+      "step": 830
+    },
+    {
+      "epoch": 0.07293652339196524,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.002493,
+      "loss": 3.6538,
+      "step": 831
+    },
+    {
+      "epoch": 0.07302429297486773,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.002496,
+      "loss": 3.6812,
+      "step": 832
+    },
+    {
+      "epoch": 0.07311206255777021,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.002499,
+      "loss": 3.73,
+      "step": 833
+    },
+    {
+      "epoch": 0.0731998321406727,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.002502,
+      "loss": 3.7378,
+      "step": 834
+    },
+    {
+      "epoch": 0.07328760172357518,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.002505,
+      "loss": 3.7598,
+      "step": 835
+    },
+    {
+      "epoch": 0.07337537130647767,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.002508,
+      "loss": 3.7412,
+      "step": 836
+    },
+    {
+      "epoch": 0.07346314088938015,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.0025109999999999998,
+      "loss": 3.7393,
+      "step": 837
+    },
+    {
+      "epoch": 0.07355091047228264,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0025139999999999997,
+      "loss": 3.7256,
+      "step": 838
+    },
+    {
+      "epoch": 0.07363868005518512,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.002517,
+      "loss": 3.6777,
+      "step": 839
+    },
+    {
+      "epoch": 0.07372644963808761,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.00252,
+      "loss": 3.6694,
+      "step": 840
+    },
+    {
+      "epoch": 0.0738142192209901,
+      "grad_norm": 0.31640625,
+      "learning_rate": 0.002523,
+      "loss": 3.6685,
+      "step": 841
+    },
+    {
+      "epoch": 0.07390198880389258,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.002526,
+      "loss": 3.7007,
+      "step": 842
+    },
+    {
+      "epoch": 0.07398975838679507,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.002529,
+      "loss": 3.7432,
+      "step": 843
+    },
+    {
+      "epoch": 0.07407752796969755,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.002532,
+      "loss": 3.6738,
+      "step": 844
+    },
+    {
+      "epoch": 0.07416529755260004,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.002535,
+      "loss": 3.646,
+      "step": 845
+    },
+    {
+      "epoch": 0.07425306713550252,
+      "grad_norm": 0.35546875,
+      "learning_rate": 0.002538,
+      "loss": 3.6494,
+      "step": 846
+    },
+    {
+      "epoch": 0.07434083671840501,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.002541,
+      "loss": 3.6348,
+      "step": 847
+    },
+    {
+      "epoch": 0.07442860630130749,
+      "grad_norm": 0.3359375,
+      "learning_rate": 0.002544,
+      "loss": 3.7183,
+      "step": 848
+    },
+    {
+      "epoch": 0.07451637588420998,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.002547,
+      "loss": 3.728,
+      "step": 849
+    },
+    {
+      "epoch": 0.07460414546711246,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.00255,
+      "loss": 3.7329,
+      "step": 850
+    },
+    {
+      "epoch": 0.07469191505001495,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.002553,
+      "loss": 3.6992,
+      "step": 851
+    },
+    {
+      "epoch": 0.07477968463291744,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.002556,
+      "loss": 3.626,
+      "step": 852
+    },
+    {
+      "epoch": 0.07486745421581992,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.002559,
+      "loss": 3.6987,
+      "step": 853
+    },
+    {
+      "epoch": 0.07495522379872241,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.002562,
+      "loss": 3.6509,
+      "step": 854
+    },
+    {
+      "epoch": 0.07504299338162489,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.002565,
+      "loss": 3.7505,
+      "step": 855
+    },
+    {
+      "epoch": 0.07513076296452738,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.002568,
+      "loss": 3.7656,
+      "step": 856
+    },
+    {
+      "epoch": 0.07521853254742986,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.002571,
+      "loss": 3.687,
+      "step": 857
+    },
+    {
+      "epoch": 0.07530630213033235,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.002574,
+      "loss": 3.7065,
+      "step": 858
+    },
+    {
+      "epoch": 0.07539407171323483,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.002577,
+      "loss": 3.71,
+      "step": 859
+    },
+    {
+      "epoch": 0.07548184129613732,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00258,
+      "loss": 3.7144,
+      "step": 860
+    },
+    {
+      "epoch": 0.0755696108790398,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.0025830000000000002,
+      "loss": 3.6265,
+      "step": 861
+    },
+    {
+      "epoch": 0.07565738046194229,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.002586,
+      "loss": 3.6665,
+      "step": 862
+    },
+    {
+      "epoch": 0.07574515004484478,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.002589,
+      "loss": 3.6675,
+      "step": 863
+    },
+    {
+      "epoch": 0.07583291962774726,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.002592,
+      "loss": 3.6807,
+      "step": 864
+    },
+    {
+      "epoch": 0.07592068921064975,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.002595,
+      "loss": 3.728,
+      "step": 865
+    },
+    {
+      "epoch": 0.07600845879355222,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.002598,
+      "loss": 3.6689,
+      "step": 866
+    },
+    {
+      "epoch": 0.07609622837645472,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.002601,
+      "loss": 3.6865,
+      "step": 867
+    },
+    {
+      "epoch": 0.0761839979593572,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.002604,
+      "loss": 3.6787,
+      "step": 868
+    },
+    {
+      "epoch": 0.07627176754225969,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.002607,
+      "loss": 3.6592,
+      "step": 869
+    },
+    {
+      "epoch": 0.07635953712516216,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.00261,
+      "loss": 3.7295,
+      "step": 870
+    },
+    {
+      "epoch": 0.07644730670806466,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.002613,
+      "loss": 3.7124,
+      "step": 871
+    },
+    {
+      "epoch": 0.07653507629096713,
+      "grad_norm": 0.5,
+      "learning_rate": 0.002616,
+      "loss": 3.6597,
+      "step": 872
+    },
+    {
+      "epoch": 0.07662284587386962,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.0026190000000000002,
+      "loss": 3.7861,
+      "step": 873
+    },
+    {
+      "epoch": 0.0767106154567721,
+      "grad_norm": 0.31640625,
+      "learning_rate": 0.002622,
+      "loss": 3.6782,
+      "step": 874
+    },
+    {
+      "epoch": 0.0767983850396746,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.002625,
+      "loss": 3.6807,
+      "step": 875
+    },
+    {
+      "epoch": 0.07688615462257709,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.002628,
+      "loss": 3.6914,
+      "step": 876
+    },
+    {
+      "epoch": 0.07697392420547956,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.002631,
+      "loss": 3.6938,
+      "step": 877
+    },
+    {
+      "epoch": 0.07706169378838205,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.002634,
+      "loss": 3.6724,
+      "step": 878
+    },
+    {
+      "epoch": 0.07714946337128453,
+      "grad_norm": 0.49609375,
+      "learning_rate": 0.002637,
+      "loss": 3.6084,
+      "step": 879
+    },
+    {
+      "epoch": 0.07723723295418702,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00264,
+      "loss": 3.6646,
+      "step": 880
+    },
+    {
+      "epoch": 0.0773250025370895,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.002643,
+      "loss": 3.6909,
+      "step": 881
+    },
+    {
+      "epoch": 0.077412772119992,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.002646,
+      "loss": 3.7378,
+      "step": 882
+    },
+    {
+      "epoch": 0.07750054170289447,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.002649,
+      "loss": 3.7207,
+      "step": 883
+    },
+    {
+      "epoch": 0.07758831128579696,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.0026520000000000003,
+      "loss": 3.7061,
+      "step": 884
+    },
+    {
+      "epoch": 0.07767608086869944,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.0026550000000000002,
+      "loss": 3.731,
+      "step": 885
+    },
+    {
+      "epoch": 0.07776385045160193,
+      "grad_norm": 0.326171875,
+      "learning_rate": 0.002658,
+      "loss": 3.6167,
+      "step": 886
+    },
+    {
+      "epoch": 0.07785162003450442,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.002661,
+      "loss": 3.6455,
+      "step": 887
+    },
+    {
+      "epoch": 0.0779393896174069,
+      "grad_norm": 0.3984375,
+      "learning_rate": 0.002664,
+      "loss": 3.6997,
+      "step": 888
+    },
+    {
+      "epoch": 0.07802715920030939,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.002667,
+      "loss": 3.6294,
+      "step": 889
+    },
+    {
+      "epoch": 0.07811492878321187,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00267,
+      "loss": 3.7241,
+      "step": 890
+    },
+    {
+      "epoch": 0.07820269836611436,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.002673,
+      "loss": 3.7808,
+      "step": 891
+    },
+    {
+      "epoch": 0.07829046794901684,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.002676,
+      "loss": 3.5918,
+      "step": 892
+    },
+    {
+      "epoch": 0.07837823753191933,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.002679,
+      "loss": 3.6763,
+      "step": 893
+    },
+    {
+      "epoch": 0.07846600711482181,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.002682,
+      "loss": 3.7134,
+      "step": 894
+    },
+    {
+      "epoch": 0.0785537766977243,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.0026850000000000003,
+      "loss": 3.7212,
+      "step": 895
+    },
+    {
+      "epoch": 0.07864154628062678,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.0026880000000000003,
+      "loss": 3.5791,
+      "step": 896
+    },
+    {
+      "epoch": 0.07872931586352927,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.0026910000000000002,
+      "loss": 3.7041,
+      "step": 897
+    },
+    {
+      "epoch": 0.07881708544643176,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.002694,
+      "loss": 3.7158,
+      "step": 898
+    },
+    {
+      "epoch": 0.07890485502933424,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.002697,
+      "loss": 3.7212,
+      "step": 899
+    },
+    {
+      "epoch": 0.07899262461223673,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0027,
+      "loss": 3.604,
+      "step": 900
+    },
+    {
+      "epoch": 0.07908039419513921,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.002703,
+      "loss": 3.7148,
+      "step": 901
+    },
+    {
+      "epoch": 0.0791681637780417,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.002706,
+      "loss": 3.646,
+      "step": 902
+    },
+    {
+      "epoch": 0.07925593336094418,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.002709,
+      "loss": 3.6235,
+      "step": 903
+    },
+    {
+      "epoch": 0.07934370294384667,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.002712,
+      "loss": 3.668,
+      "step": 904
+    },
+    {
+      "epoch": 0.07943147252674915,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.002715,
+      "loss": 3.7256,
+      "step": 905
+    },
+    {
+      "epoch": 0.07951924210965164,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.002718,
+      "loss": 3.6484,
+      "step": 906
+    },
+    {
+      "epoch": 0.07960701169255412,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.0027210000000000003,
+      "loss": 3.6211,
+      "step": 907
+    },
+    {
+      "epoch": 0.07969478127545661,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0027240000000000003,
+      "loss": 3.7407,
+      "step": 908
+    },
+    {
+      "epoch": 0.07978255085835909,
+      "grad_norm": 0.328125,
+      "learning_rate": 0.0027270000000000003,
+      "loss": 3.6343,
+      "step": 909
+    },
+    {
+      "epoch": 0.07987032044126158,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.0027300000000000002,
+      "loss": 3.6108,
+      "step": 910
+    },
+    {
+      "epoch": 0.07995809002416407,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.002733,
+      "loss": 3.6875,
+      "step": 911
+    },
+    {
+      "epoch": 0.08004585960706655,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.002736,
+      "loss": 3.585,
+      "step": 912
+    },
+    {
+      "epoch": 0.08013362918996904,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.002739,
+      "loss": 3.6724,
+      "step": 913
+    },
+    {
+      "epoch": 0.08022139877287152,
+      "grad_norm": 0.33984375,
+      "learning_rate": 0.002742,
+      "loss": 3.6455,
+      "step": 914
+    },
+    {
+      "epoch": 0.08030916835577401,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.002745,
+      "loss": 3.6523,
+      "step": 915
+    },
+    {
+      "epoch": 0.08039693793867649,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.002748,
+      "loss": 3.6792,
+      "step": 916
+    },
+    {
+      "epoch": 0.08048470752157898,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.002751,
+      "loss": 3.7769,
+      "step": 917
+    },
+    {
+      "epoch": 0.08057247710448145,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.0027540000000000004,
+      "loss": 3.665,
+      "step": 918
+    },
+    {
+      "epoch": 0.08066024668738395,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.0027570000000000003,
+      "loss": 3.6743,
+      "step": 919
+    },
+    {
+      "epoch": 0.08074801627028642,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.0027600000000000003,
+      "loss": 3.6426,
+      "step": 920
+    },
+    {
+      "epoch": 0.08083578585318892,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.0027630000000000003,
+      "loss": 3.6484,
+      "step": 921
+    },
+    {
+      "epoch": 0.08092355543609141,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.0027660000000000002,
+      "loss": 3.6919,
+      "step": 922
+    },
+    {
+      "epoch": 0.08101132501899388,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.002769,
+      "loss": 3.6587,
+      "step": 923
+    },
+    {
+      "epoch": 0.08109909460189638,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.002772,
+      "loss": 3.6606,
+      "step": 924
+    },
+    {
+      "epoch": 0.08118686418479885,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.002775,
+      "loss": 3.7056,
+      "step": 925
+    },
+    {
+      "epoch": 0.08127463376770135,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.002778,
+      "loss": 3.6943,
+      "step": 926
+    },
+    {
+      "epoch": 0.08136240335060382,
+      "grad_norm": 1.828125,
+      "learning_rate": 0.002781,
+      "loss": 3.6484,
+      "step": 927
+    },
+    {
+      "epoch": 0.08145017293350632,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.002784,
+      "loss": 3.6758,
+      "step": 928
+    },
+    {
+      "epoch": 0.08153794251640879,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.0027870000000000004,
+      "loss": 3.7661,
+      "step": 929
+    },
+    {
+      "epoch": 0.08162571209931128,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.0027900000000000004,
+      "loss": 3.7485,
+      "step": 930
+    },
+    {
+      "epoch": 0.08171348168221376,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.0027930000000000003,
+      "loss": 3.7007,
+      "step": 931
+    },
+    {
+      "epoch": 0.08180125126511625,
+      "grad_norm": 0.490234375,
+      "learning_rate": 0.0027960000000000003,
+      "loss": 3.7114,
+      "step": 932
+    },
+    {
+      "epoch": 0.08188902084801875,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.0027990000000000003,
+      "loss": 3.7549,
+      "step": 933
+    },
+    {
+      "epoch": 0.08197679043092122,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.0028020000000000002,
+      "loss": 3.6338,
+      "step": 934
+    },
+    {
+      "epoch": 0.08206456001382371,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.002805,
+      "loss": 3.6816,
+      "step": 935
+    },
+    {
+      "epoch": 0.08215232959672619,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.002808,
+      "loss": 3.6445,
+      "step": 936
+    },
+    {
+      "epoch": 0.08224009917962868,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.002811,
+      "loss": 3.6426,
+      "step": 937
+    },
+    {
+      "epoch": 0.08232786876253116,
+      "grad_norm": 0.34375,
+      "learning_rate": 0.002814,
+      "loss": 3.6396,
+      "step": 938
+    },
+    {
+      "epoch": 0.08241563834543365,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.002817,
+      "loss": 3.6426,
+      "step": 939
+    },
+    {
+      "epoch": 0.08250340792833613,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.00282,
+      "loss": 3.6167,
+      "step": 940
+    },
+    {
+      "epoch": 0.08259117751123862,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.002823,
+      "loss": 3.6108,
+      "step": 941
+    },
+    {
+      "epoch": 0.0826789470941411,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.002826,
+      "loss": 3.689,
+      "step": 942
+    },
+    {
+      "epoch": 0.08276671667704359,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.002829,
+      "loss": 3.6821,
+      "step": 943
+    },
+    {
+      "epoch": 0.08285448625994607,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.002832,
+      "loss": 3.6182,
+      "step": 944
+    },
+    {
+      "epoch": 0.08294225584284856,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.002835,
+      "loss": 3.5884,
+      "step": 945
+    },
+    {
+      "epoch": 0.08303002542575105,
+      "grad_norm": 1.96875,
+      "learning_rate": 0.002838,
+      "loss": 3.7207,
+      "step": 946
+    },
+    {
+      "epoch": 0.08311779500865353,
+      "grad_norm": 17.5,
+      "learning_rate": 0.0028409999999999998,
+      "loss": 4.4404,
+      "step": 947
+    },
+    {
+      "epoch": 0.08320556459155602,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0028439999999999997,
+      "loss": 3.8545,
+      "step": 948
+    },
+    {
+      "epoch": 0.0832933341744585,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.002847,
+      "loss": 3.7534,
+      "step": 949
+    },
+    {
+      "epoch": 0.08338110375736099,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.00285,
+      "loss": 3.8262,
+      "step": 950
+    },
+    {
+      "epoch": 0.08346887334026347,
+      "grad_norm": 1.9921875,
+      "learning_rate": 0.002853,
+      "loss": 3.8135,
+      "step": 951
+    },
+    {
+      "epoch": 0.08355664292316596,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.002856,
+      "loss": 3.7271,
+      "step": 952
+    },
+    {
+      "epoch": 0.08364441250606844,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.002859,
+      "loss": 3.7886,
+      "step": 953
+    },
+    {
+      "epoch": 0.08373218208897093,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.002862,
+      "loss": 3.7295,
+      "step": 954
+    },
+    {
+      "epoch": 0.08381995167187341,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.002865,
+      "loss": 3.6191,
+      "step": 955
+    },
+    {
+      "epoch": 0.0839077212547759,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.002868,
+      "loss": 3.7207,
+      "step": 956
+    },
+    {
+      "epoch": 0.08399549083767839,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.002871,
+      "loss": 3.6973,
+      "step": 957
+    },
+    {
+      "epoch": 0.08408326042058087,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.002874,
+      "loss": 3.7539,
+      "step": 958
+    },
+    {
+      "epoch": 0.08417103000348336,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.002877,
+      "loss": 3.7188,
+      "step": 959
+    },
+    {
+      "epoch": 0.08425879958638584,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.0028799999999999997,
+      "loss": 3.73,
+      "step": 960
+    },
+    {
+      "epoch": 0.08434656916928833,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.002883,
+      "loss": 3.6167,
+      "step": 961
+    },
+    {
+      "epoch": 0.08443433875219081,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.002886,
+      "loss": 3.6514,
+      "step": 962
+    },
+    {
+      "epoch": 0.0845221083350933,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.002889,
+      "loss": 3.7183,
+      "step": 963
+    },
+    {
+      "epoch": 0.08460987791799578,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.002892,
+      "loss": 3.7275,
+      "step": 964
+    },
+    {
+      "epoch": 0.08469764750089827,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.002895,
+      "loss": 3.6123,
+      "step": 965
+    },
+    {
+      "epoch": 0.08478541708380075,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.002898,
+      "loss": 3.687,
+      "step": 966
+    },
+    {
+      "epoch": 0.08487318666670324,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.002901,
+      "loss": 3.6724,
+      "step": 967
+    },
+    {
+      "epoch": 0.08496095624960573,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.002904,
+      "loss": 3.7109,
+      "step": 968
+    },
+    {
+      "epoch": 0.0850487258325082,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.002907,
+      "loss": 3.5679,
+      "step": 969
+    },
+    {
+      "epoch": 0.0851364954154107,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.00291,
+      "loss": 3.6958,
+      "step": 970
+    },
+    {
+      "epoch": 0.08522426499831318,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.002913,
+      "loss": 3.6597,
+      "step": 971
+    },
+    {
+      "epoch": 0.08531203458121567,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.002916,
+      "loss": 3.7212,
+      "step": 972
+    },
+    {
+      "epoch": 0.08539980416411815,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.002919,
+      "loss": 3.6523,
+      "step": 973
+    },
+    {
+      "epoch": 0.08548757374702064,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.002922,
+      "loss": 3.647,
+      "step": 974
+    },
+    {
+      "epoch": 0.08557534332992311,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.002925,
+      "loss": 3.6421,
+      "step": 975
+    },
+    {
+      "epoch": 0.0856631129128256,
+      "grad_norm": 17.0,
+      "learning_rate": 0.002928,
+      "loss": 3.7246,
+      "step": 976
+    },
+    {
+      "epoch": 0.08575088249572808,
+      "grad_norm": 1.125,
+      "learning_rate": 0.002931,
+      "loss": 3.6187,
+      "step": 977
+    },
+    {
+      "epoch": 0.08583865207863058,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.002934,
+      "loss": 3.7295,
+      "step": 978
+    },
+    {
+      "epoch": 0.08592642166153307,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.002937,
+      "loss": 3.7075,
+      "step": 979
+    },
+    {
+      "epoch": 0.08601419124443554,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00294,
+      "loss": 3.7109,
+      "step": 980
+    },
+    {
+      "epoch": 0.08610196082733804,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.002943,
+      "loss": 3.5815,
+      "step": 981
+    },
+    {
+      "epoch": 0.08618973041024051,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.002946,
+      "loss": 3.6768,
+      "step": 982
+    },
+    {
+      "epoch": 0.086277499993143,
+      "grad_norm": 0.3984375,
+      "learning_rate": 0.0029490000000000002,
+      "loss": 3.6567,
+      "step": 983
+    },
+    {
+      "epoch": 0.08636526957604548,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.002952,
+      "loss": 3.6489,
+      "step": 984
+    },
+    {
+      "epoch": 0.08645303915894798,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.002955,
+      "loss": 3.709,
+      "step": 985
+    },
+    {
+      "epoch": 0.08654080874185045,
+      "grad_norm": 0.25,
+      "learning_rate": 0.002958,
+      "loss": 3.6841,
+      "step": 986
+    },
+    {
+      "epoch": 0.08662857832475294,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.002961,
+      "loss": 3.6665,
+      "step": 987
+    },
+    {
+      "epoch": 0.08671634790765542,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.002964,
+      "loss": 3.6299,
+      "step": 988
+    },
+    {
+      "epoch": 0.08680411749055791,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.002967,
+      "loss": 3.5996,
+      "step": 989
+    },
+    {
+      "epoch": 0.08689188707346039,
+      "grad_norm": 24.75,
+      "learning_rate": 0.00297,
+      "loss": 3.7173,
+      "step": 990
+    },
+    {
+      "epoch": 0.08697965665636288,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.002973,
+      "loss": 3.6875,
+      "step": 991
+    },
+    {
+      "epoch": 0.08706742623926537,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.002976,
+      "loss": 3.7153,
+      "step": 992
+    },
+    {
+      "epoch": 0.08715519582216785,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.002979,
+      "loss": 3.6968,
+      "step": 993
+    },
+    {
+      "epoch": 0.08724296540507034,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.002982,
+      "loss": 3.7632,
+      "step": 994
+    },
+    {
+      "epoch": 0.08733073498797282,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.0029850000000000002,
+      "loss": 3.6479,
+      "step": 995
+    },
+    {
+      "epoch": 0.08741850457087531,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.002988,
+      "loss": 3.6816,
+      "step": 996
+    },
+    {
+      "epoch": 0.08750627415377779,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.002991,
+      "loss": 3.6348,
+      "step": 997
+    },
+    {
+      "epoch": 0.08759404373668028,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.002994,
+      "loss": 3.6948,
+      "step": 998
+    },
+    {
+      "epoch": 0.08768181331958276,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.002997,
+      "loss": 3.6421,
+      "step": 999
+    },
+    {
+      "epoch": 0.08776958290248525,
+      "grad_norm": 0.421875,
+      "learning_rate": 0.003,
+      "loss": 3.584,
+      "step": 1000
+    },
+    {
+      "epoch": 0.08776958290248525,
+      "eval_loss": 0.11389437317848206,
+      "eval_runtime": 106.9972,
+      "eval_samples_per_second": 137.845,
+      "eval_steps_per_second": 17.234,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 11393,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.948783792128e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}