diff --git "a/checkpoint-1200/trainer_state.json" "b/checkpoint-1200/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-1200/trainer_state.json"
@@ -0,0 +1,8433 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.012,
+  "eval_steps": 500,
+  "global_step": 1200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 1e-05,
+      "grad_norm": 1.4574347149106512,
+      "learning_rate": 3e-06,
+      "loss": 10.8576,
+      "step": 1
+    },
+    {
+      "epoch": 2e-05,
+      "grad_norm": 1.4482443114713344,
+      "learning_rate": 6e-06,
+      "loss": 10.8575,
+      "step": 2
+    },
+    {
+      "epoch": 3e-05,
+      "grad_norm": 1.4642003089092823,
+      "learning_rate": 9e-06,
+      "loss": 10.8576,
+      "step": 3
+    },
+    {
+      "epoch": 4e-05,
+      "grad_norm": 1.4485584072048276,
+      "learning_rate": 1.2e-05,
+      "loss": 10.8569,
+      "step": 4
+    },
+    {
+      "epoch": 5e-05,
+      "grad_norm": 1.4574321657516995,
+      "learning_rate": 1.5e-05,
+      "loss": 10.8527,
+      "step": 5
+    },
+    {
+      "epoch": 6e-05,
+      "grad_norm": 1.4585537094370684,
+      "learning_rate": 1.8e-05,
+      "loss": 10.8518,
+      "step": 6
+    },
+    {
+      "epoch": 7e-05,
+      "grad_norm": 1.4183780170798466,
+      "learning_rate": 2.1000000000000002e-05,
+      "loss": 10.8383,
+      "step": 7
+    },
+    {
+      "epoch": 8e-05,
+      "grad_norm": 1.286272643239374,
+      "learning_rate": 2.4e-05,
+      "loss": 10.8119,
+      "step": 8
+    },
+    {
+      "epoch": 9e-05,
+      "grad_norm": 1.246364249616181,
+      "learning_rate": 2.7e-05,
+      "loss": 10.8063,
+      "step": 9
+    },
+    {
+      "epoch": 0.0001,
+      "grad_norm": 1.2300728857390288,
+      "learning_rate": 3e-05,
+      "loss": 10.7913,
+      "step": 10
+    },
+    {
+      "epoch": 0.00011,
+      "grad_norm": 1.1639358472437353,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 10.7756,
+      "step": 11
+    },
+    {
+      "epoch": 0.00012,
+      "grad_norm": 1.1455069336623074,
+      "learning_rate": 3.6e-05,
+      "loss": 10.7622,
+      "step": 12
+    },
+    {
+      "epoch": 0.00013,
+      "grad_norm": 1.104711768149659,
+      "learning_rate": 3.9e-05,
+      "loss": 10.7415,
+      "step": 13
+    },
+    {
+      "epoch": 0.00014,
+      "grad_norm": 1.0916409179063882,
+      "learning_rate": 4.2000000000000004e-05,
+      "loss": 10.7296,
+      "step": 14
+    },
+    {
+      "epoch": 0.00015,
+      "grad_norm": 1.0723152562842644,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 10.7183,
+      "step": 15
+    },
+    {
+      "epoch": 0.00016,
+      "grad_norm": 1.0297043497697835,
+      "learning_rate": 4.8e-05,
+      "loss": 10.7,
+      "step": 16
+    },
+    {
+      "epoch": 0.00017,
+      "grad_norm": 0.9882210863590011,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 10.6838,
+      "step": 17
+    },
+    {
+      "epoch": 0.00018,
+      "grad_norm": 0.966791538351231,
+      "learning_rate": 5.4e-05,
+      "loss": 10.6659,
+      "step": 18
+    },
+    {
+      "epoch": 0.00019,
+      "grad_norm": 0.9358769725124266,
+      "learning_rate": 5.7e-05,
+      "loss": 10.6509,
+      "step": 19
+    },
+    {
+      "epoch": 0.0002,
+      "grad_norm": 0.9313245306892226,
+      "learning_rate": 6e-05,
+      "loss": 10.638,
+      "step": 20
+    },
+    {
+      "epoch": 0.00021,
+      "grad_norm": 0.9056933112957348,
+      "learning_rate": 6.3e-05,
+      "loss": 10.6235,
+      "step": 21
+    },
+    {
+      "epoch": 0.00022,
+      "grad_norm": 0.9013277325687112,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 10.6069,
+      "step": 22
+    },
+    {
+      "epoch": 0.00023,
+      "grad_norm": 0.8982860560109955,
+      "learning_rate": 6.9e-05,
+      "loss": 10.5935,
+      "step": 23
+    },
+    {
+      "epoch": 0.00024,
+      "grad_norm": 0.8941575476865218,
+      "learning_rate": 7.2e-05,
+      "loss": 10.5802,
+      "step": 24
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 0.8955811690787305,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 10.5652,
+      "step": 25
+    },
+    {
+      "epoch": 0.00026,
+      "grad_norm": 0.9005417636666915,
+      "learning_rate": 7.8e-05,
+      "loss": 10.5507,
+      "step": 26
+    },
+    {
+      "epoch": 0.00027,
+      "grad_norm": 0.8929394290279239,
+      "learning_rate": 8.1e-05,
+      "loss": 10.5379,
+      "step": 27
+    },
+    {
+      "epoch": 0.00028,
+      "grad_norm": 0.892893092343029,
+      "learning_rate": 8.400000000000001e-05,
+      "loss": 10.5234,
+      "step": 28
+    },
+    {
+      "epoch": 0.00029,
+      "grad_norm": 0.8951984198637418,
+      "learning_rate": 8.7e-05,
+      "loss": 10.5078,
+      "step": 29
+    },
+    {
+      "epoch": 0.0003,
+      "grad_norm": 0.901037080772758,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 10.4913,
+      "step": 30
+    },
+    {
+      "epoch": 0.00031,
+      "grad_norm": 0.899757879512845,
+      "learning_rate": 9.3e-05,
+      "loss": 10.4759,
+      "step": 31
+    },
+    {
+      "epoch": 0.00032,
+      "grad_norm": 0.8963415000423225,
+      "learning_rate": 9.6e-05,
+      "loss": 10.4612,
+      "step": 32
+    },
+    {
+      "epoch": 0.00033,
+      "grad_norm": 0.8971468155029705,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 10.4429,
+      "step": 33
+    },
+    {
+      "epoch": 0.00034,
+      "grad_norm": 0.8964368558533553,
+      "learning_rate": 0.00010200000000000001,
+      "loss": 10.426,
+      "step": 34
+    },
+    {
+      "epoch": 0.00035,
+      "grad_norm": 0.9007340868108898,
+      "learning_rate": 0.00010500000000000002,
+      "loss": 10.4083,
+      "step": 35
+    },
+    {
+      "epoch": 0.00036,
+      "grad_norm": 0.8977475222905443,
+      "learning_rate": 0.000108,
+      "loss": 10.3895,
+      "step": 36
+    },
+    {
+      "epoch": 0.00037,
+      "grad_norm": 0.8929896912849768,
+      "learning_rate": 0.000111,
+      "loss": 10.3721,
+      "step": 37
+    },
+    {
+      "epoch": 0.00038,
+      "grad_norm": 0.8925204062685723,
+      "learning_rate": 0.000114,
+      "loss": 10.3515,
+      "step": 38
+    },
+    {
+      "epoch": 0.00039,
+      "grad_norm": 0.8947925451707294,
+      "learning_rate": 0.000117,
+      "loss": 10.3314,
+      "step": 39
+    },
+    {
+      "epoch": 0.0004,
+      "grad_norm": 0.8990961452390619,
+      "learning_rate": 0.00012,
+      "loss": 10.3088,
+      "step": 40
+    },
+    {
+      "epoch": 0.00041,
+      "grad_norm": 0.8951984584897338,
+      "learning_rate": 0.000123,
+      "loss": 10.2891,
+      "step": 41
+    },
+    {
+      "epoch": 0.00042,
+      "grad_norm": 0.8942493459254965,
+      "learning_rate": 0.000126,
+      "loss": 10.2679,
+      "step": 42
+    },
+    {
+      "epoch": 0.00043,
+      "grad_norm": 0.8983684110980745,
+      "learning_rate": 0.000129,
+      "loss": 10.243,
+      "step": 43
+    },
+    {
+      "epoch": 0.00044,
+      "grad_norm": 0.8985869855625094,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 10.2206,
+      "step": 44
+    },
+    {
+      "epoch": 0.00045,
+      "grad_norm": 0.899962207774676,
+      "learning_rate": 0.000135,
+      "loss": 10.1962,
+      "step": 45
+    },
+    {
+      "epoch": 0.00046,
+      "grad_norm": 0.8914153211826606,
+      "learning_rate": 0.000138,
+      "loss": 10.1735,
+      "step": 46
+    },
+    {
+      "epoch": 0.00047,
+      "grad_norm": 0.8980761529388176,
+      "learning_rate": 0.000141,
+      "loss": 10.146,
+      "step": 47
+    },
+    {
+      "epoch": 0.00048,
+      "grad_norm": 0.9007540011988344,
+      "learning_rate": 0.000144,
+      "loss": 10.1222,
+      "step": 48
+    },
+    {
+      "epoch": 0.00049,
+      "grad_norm": 0.896344611975273,
+      "learning_rate": 0.000147,
+      "loss": 10.0974,
+      "step": 49
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 0.9015948414141589,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 10.0706,
+      "step": 50
+    },
+    {
+      "epoch": 0.00051,
+      "grad_norm": 0.9016102183936469,
+      "learning_rate": 0.000153,
+      "loss": 10.0433,
+      "step": 51
+    },
+    {
+      "epoch": 0.00052,
+      "grad_norm": 0.8993585803645682,
+      "learning_rate": 0.000156,
+      "loss": 10.0158,
+      "step": 52
+    },
+    {
+      "epoch": 0.00053,
+      "grad_norm": 0.8901736410301992,
+      "learning_rate": 0.000159,
+      "loss": 9.9917,
+      "step": 53
+    },
+    {
+      "epoch": 0.00054,
+      "grad_norm": 0.9083328832659304,
+      "learning_rate": 0.000162,
+      "loss": 9.9578,
+      "step": 54
+    },
+    {
+      "epoch": 0.00055,
+      "grad_norm": 0.8975570632788401,
+      "learning_rate": 0.000165,
+      "loss": 9.9337,
+      "step": 55
+    },
+    {
+      "epoch": 0.00056,
+      "grad_norm": 0.8924582131156604,
+      "learning_rate": 0.00016800000000000002,
+      "loss": 9.904,
+      "step": 56
+    },
+    {
+      "epoch": 0.00057,
+      "grad_norm": 0.8937117244788118,
+      "learning_rate": 0.000171,
+      "loss": 9.8765,
+      "step": 57
+    },
+    {
+      "epoch": 0.00058,
+      "grad_norm": 0.8912667557674427,
+      "learning_rate": 0.000174,
+      "loss": 9.8514,
+      "step": 58
+    },
+    {
+      "epoch": 0.00059,
+      "grad_norm": 0.899445317105929,
+      "learning_rate": 0.000177,
+      "loss": 9.819,
+      "step": 59
+    },
+    {
+      "epoch": 0.0006,
+      "grad_norm": 0.8827247644946434,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 9.7927,
+      "step": 60
+    },
+    {
+      "epoch": 0.00061,
+      "grad_norm": 0.8856006114384688,
+      "learning_rate": 0.000183,
+      "loss": 9.7645,
+      "step": 61
+    },
+    {
+      "epoch": 0.00062,
+      "grad_norm": 0.8887861476206323,
+      "learning_rate": 0.000186,
+      "loss": 9.7329,
+      "step": 62
+    },
+    {
+      "epoch": 0.00063,
+      "grad_norm": 0.8870948257177573,
+      "learning_rate": 0.000189,
+      "loss": 9.7044,
+      "step": 63
+    },
+    {
+      "epoch": 0.00064,
+      "grad_norm": 0.8854465518366905,
+      "learning_rate": 0.000192,
+      "loss": 9.6741,
+      "step": 64
+    },
+    {
+      "epoch": 0.00065,
+      "grad_norm": 0.8913964558604793,
+      "learning_rate": 0.00019500000000000002,
+      "loss": 9.6417,
+      "step": 65
+    },
+    {
+      "epoch": 0.00066,
+      "grad_norm": 0.8860459843940278,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 9.6181,
+      "step": 66
+    },
+    {
+      "epoch": 0.00067,
+      "grad_norm": 0.8855602986146246,
+      "learning_rate": 0.000201,
+      "loss": 9.5886,
+      "step": 67
+    },
+    {
+      "epoch": 0.00068,
+      "grad_norm": 0.89182613307124,
+      "learning_rate": 0.00020400000000000003,
+      "loss": 9.5554,
+      "step": 68
+    },
+    {
+      "epoch": 0.00069,
+      "grad_norm": 0.8788147221752699,
+      "learning_rate": 0.00020700000000000002,
+      "loss": 9.5276,
+      "step": 69
+    },
+    {
+      "epoch": 0.0007,
+      "grad_norm": 0.89013480384513,
+      "learning_rate": 0.00021000000000000004,
+      "loss": 9.4911,
+      "step": 70
+    },
+    {
+      "epoch": 0.00071,
+      "grad_norm": 0.8832608390392153,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 9.4637,
+      "step": 71
+    },
+    {
+      "epoch": 0.00072,
+      "grad_norm": 0.8850551460752941,
+      "learning_rate": 0.000216,
+      "loss": 9.4371,
+      "step": 72
+    },
+    {
+      "epoch": 0.00073,
+      "grad_norm": 0.8854430272945265,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 9.4029,
+      "step": 73
+    },
+    {
+      "epoch": 0.00074,
+      "grad_norm": 0.881388508551569,
+      "learning_rate": 0.000222,
+      "loss": 9.3766,
+      "step": 74
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 0.8804734413831777,
+      "learning_rate": 0.000225,
+      "loss": 9.3413,
+      "step": 75
+    },
+    {
+      "epoch": 0.00076,
+      "grad_norm": 0.891470306594604,
+      "learning_rate": 0.000228,
+      "loss": 9.3136,
+      "step": 76
+    },
+    {
+      "epoch": 0.00077,
+      "grad_norm": 0.8835806398314079,
+      "learning_rate": 0.000231,
+      "loss": 9.2843,
+      "step": 77
+    },
+    {
+      "epoch": 0.00078,
+      "grad_norm": 0.8778470434340718,
+      "learning_rate": 0.000234,
+      "loss": 9.2534,
+      "step": 78
+    },
+    {
+      "epoch": 0.00079,
+      "grad_norm": 0.8849142747864956,
+      "learning_rate": 0.00023700000000000001,
+      "loss": 9.2174,
+      "step": 79
+    },
+    {
+      "epoch": 0.0008,
+      "grad_norm": 0.8831859916332898,
+      "learning_rate": 0.00024,
+      "loss": 9.1835,
+      "step": 80
+    },
+    {
+      "epoch": 0.00081,
+      "grad_norm": 0.8883671155155716,
+      "learning_rate": 0.00024300000000000002,
+      "loss": 9.1516,
+      "step": 81
+    },
+    {
+      "epoch": 0.00082,
+      "grad_norm": 0.8919005998406658,
+      "learning_rate": 0.000246,
+      "loss": 9.1278,
+      "step": 82
+    },
+    {
+      "epoch": 0.00083,
+      "grad_norm": 0.8904109612127737,
+      "learning_rate": 0.00024900000000000004,
+      "loss": 9.0926,
+      "step": 83
+    },
+    {
+      "epoch": 0.00084,
+      "grad_norm": 0.8922919840670517,
+      "learning_rate": 0.000252,
+      "loss": 9.0627,
+      "step": 84
+    },
+    {
+      "epoch": 0.00085,
+      "grad_norm": 0.8851770394662277,
+      "learning_rate": 0.000255,
+      "loss": 9.0389,
+      "step": 85
+    },
+    {
+      "epoch": 0.00086,
+      "grad_norm": 0.892287423580011,
+      "learning_rate": 0.000258,
+      "loss": 9.0029,
+      "step": 86
+    },
+    {
+      "epoch": 0.00087,
+      "grad_norm": 0.8819038715566697,
+      "learning_rate": 0.000261,
+      "loss": 8.9819,
+      "step": 87
+    },
+    {
+      "epoch": 0.00088,
+      "grad_norm": 0.8825670273484024,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 8.9452,
+      "step": 88
+    },
+    {
+      "epoch": 0.00089,
+      "grad_norm": 0.87734677010751,
+      "learning_rate": 0.000267,
+      "loss": 8.92,
+      "step": 89
+    },
+    {
+      "epoch": 0.0009,
+      "grad_norm": 0.8800982265590005,
+      "learning_rate": 0.00027,
+      "loss": 8.8935,
+      "step": 90
+    },
+    {
+      "epoch": 0.00091,
+      "grad_norm": 0.8751738333241339,
+      "learning_rate": 0.000273,
+      "loss": 8.8635,
+      "step": 91
+    },
+    {
+      "epoch": 0.00092,
+      "grad_norm": 0.8805112062017327,
+      "learning_rate": 0.000276,
+      "loss": 8.8313,
+      "step": 92
+    },
+    {
+      "epoch": 0.00093,
+      "grad_norm": 0.8782110873043902,
+      "learning_rate": 0.000279,
+      "loss": 8.8018,
+      "step": 93
+    },
+    {
+      "epoch": 0.00094,
+      "grad_norm": 0.8793675998346636,
+      "learning_rate": 0.000282,
+      "loss": 8.7769,
+      "step": 94
+    },
+    {
+      "epoch": 0.00095,
+      "grad_norm": 0.8739437022686155,
+      "learning_rate": 0.000285,
+      "loss": 8.7468,
+      "step": 95
+    },
+    {
+      "epoch": 0.00096,
+      "grad_norm": 0.8792093745975903,
+      "learning_rate": 0.000288,
+      "loss": 8.7137,
+      "step": 96
+    },
+    {
+      "epoch": 0.00097,
+      "grad_norm": 0.8733827683228169,
+      "learning_rate": 0.000291,
+      "loss": 8.6904,
+      "step": 97
+    },
+    {
+      "epoch": 0.00098,
+      "grad_norm": 0.8720585520116785,
+      "learning_rate": 0.000294,
+      "loss": 8.6531,
+      "step": 98
+    },
+    {
+      "epoch": 0.00099,
+      "grad_norm": 0.8706078195741852,
+      "learning_rate": 0.000297,
+      "loss": 8.6333,
+      "step": 99
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 0.868963072522692,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 8.6078,
+      "step": 100
+    },
+    {
+      "epoch": 0.00101,
+      "grad_norm": 0.8733428125714966,
+      "learning_rate": 0.00030300000000000005,
+      "loss": 8.5742,
+      "step": 101
+    },
+    {
+      "epoch": 0.00102,
+      "grad_norm": 0.8594222895452354,
+      "learning_rate": 0.000306,
+      "loss": 8.5555,
+      "step": 102
+    },
+    {
+      "epoch": 0.00103,
+      "grad_norm": 0.8663526773106932,
+      "learning_rate": 0.000309,
+      "loss": 8.5314,
+      "step": 103
+    },
+    {
+      "epoch": 0.00104,
+      "grad_norm": 0.8588890352637412,
+      "learning_rate": 0.000312,
+      "loss": 8.5113,
+      "step": 104
+    },
+    {
+      "epoch": 0.00105,
+      "grad_norm": 0.862875507176608,
+      "learning_rate": 0.000315,
+      "loss": 8.4848,
+      "step": 105
+    },
+    {
+      "epoch": 0.00106,
+      "grad_norm": 0.858707935395144,
+      "learning_rate": 0.000318,
+      "loss": 8.4562,
+      "step": 106
+    },
+    {
+      "epoch": 0.00107,
+      "grad_norm": 0.8477892944676333,
+      "learning_rate": 0.000321,
+      "loss": 8.4313,
+      "step": 107
+    },
+    {
+      "epoch": 0.00108,
+      "grad_norm": 0.8523289020593828,
+      "learning_rate": 0.000324,
+      "loss": 8.4028,
+      "step": 108
+    },
+    {
+      "epoch": 0.00109,
+      "grad_norm": 0.847730490341604,
+      "learning_rate": 0.000327,
+      "loss": 8.3837,
+      "step": 109
+    },
+    {
+      "epoch": 0.0011,
+      "grad_norm": 0.8528691839316361,
+      "learning_rate": 0.00033,
+      "loss": 8.3605,
+      "step": 110
+    },
+    {
+      "epoch": 0.00111,
+      "grad_norm": 0.8490568938398722,
+      "learning_rate": 0.000333,
+      "loss": 8.3302,
+      "step": 111
+    },
+    {
+      "epoch": 0.00112,
+      "grad_norm": 0.8687159568709311,
+      "learning_rate": 0.00033600000000000004,
+      "loss": 8.3067,
+      "step": 112
+    },
+    {
+      "epoch": 0.00113,
+      "grad_norm": 0.9514329048402193,
+      "learning_rate": 0.000339,
+      "loss": 8.2682,
+      "step": 113
+    },
+    {
+      "epoch": 0.00114,
+      "grad_norm": 1.0232734531890686,
+      "learning_rate": 0.000342,
+      "loss": 8.2639,
+      "step": 114
+    },
+    {
+      "epoch": 0.00115,
+      "grad_norm": 0.88782361437861,
+      "learning_rate": 0.00034500000000000004,
+      "loss": 8.2339,
+      "step": 115
+    },
+    {
+      "epoch": 0.00116,
+      "grad_norm": 0.8477580889036649,
+      "learning_rate": 0.000348,
+      "loss": 8.2115,
+      "step": 116
+    },
+    {
+      "epoch": 0.00117,
+      "grad_norm": 0.9025353672223668,
+      "learning_rate": 0.000351,
+      "loss": 8.182,
+      "step": 117
+    },
+    {
+      "epoch": 0.00118,
+      "grad_norm": 0.8194804043714383,
+      "learning_rate": 0.000354,
+      "loss": 8.1632,
+      "step": 118
+    },
+    {
+      "epoch": 0.00119,
+      "grad_norm": 0.8747187640572928,
+      "learning_rate": 0.000357,
+      "loss": 8.1526,
+      "step": 119
+    },
+    {
+      "epoch": 0.0012,
+      "grad_norm": 0.819302051650556,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 8.125,
+      "step": 120
+    },
+    {
+      "epoch": 0.00121,
+      "grad_norm": 0.8357670289489353,
+      "learning_rate": 0.000363,
+      "loss": 8.0977,
+      "step": 121
+    },
+    {
+      "epoch": 0.00122,
+      "grad_norm": 0.8269622916799205,
+      "learning_rate": 0.000366,
+      "loss": 8.0806,
+      "step": 122
+    },
+    {
+      "epoch": 0.00123,
+      "grad_norm": 0.7923909888868983,
+      "learning_rate": 0.000369,
+      "loss": 8.0558,
+      "step": 123
+    },
+    {
+      "epoch": 0.00124,
+      "grad_norm": 0.8405159164509748,
+      "learning_rate": 0.000372,
+      "loss": 8.0354,
+      "step": 124
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 0.7957014508382675,
+      "learning_rate": 0.000375,
+      "loss": 8.0077,
+      "step": 125
+    },
+    {
+      "epoch": 0.00126,
+      "grad_norm": 0.8032762193906265,
+      "learning_rate": 0.000378,
+      "loss": 7.9954,
+      "step": 126
+    },
+    {
+      "epoch": 0.00127,
+      "grad_norm": 0.8534337354315669,
+      "learning_rate": 0.000381,
+      "loss": 7.9712,
+      "step": 127
+    },
+    {
+      "epoch": 0.00128,
+      "grad_norm": 0.7889902179927859,
+      "learning_rate": 0.000384,
+      "loss": 7.9489,
+      "step": 128
+    },
+    {
+      "epoch": 0.00129,
+      "grad_norm": 0.7667804338310586,
+      "learning_rate": 0.00038700000000000003,
+      "loss": 7.936,
+      "step": 129
+    },
+    {
+      "epoch": 0.0013,
+      "grad_norm": 0.9090077886672961,
+      "learning_rate": 0.00039000000000000005,
+      "loss": 7.9113,
+      "step": 130
+    },
+    {
+      "epoch": 0.00131,
+      "grad_norm": 1.1017726636538738,
+      "learning_rate": 0.000393,
+      "loss": 7.8813,
+      "step": 131
+    },
+    {
+      "epoch": 0.00132,
+      "grad_norm": 0.9363739073599702,
+      "learning_rate": 0.00039600000000000003,
+      "loss": 7.8715,
+      "step": 132
+    },
+    {
+      "epoch": 0.00133,
+      "grad_norm": 0.7265974712286192,
+      "learning_rate": 0.00039900000000000005,
+      "loss": 7.8436,
+      "step": 133
+    },
+    {
+      "epoch": 0.00134,
+      "grad_norm": 1.085264808454749,
+      "learning_rate": 0.000402,
+      "loss": 7.8358,
+      "step": 134
+    },
+    {
+      "epoch": 0.00135,
+      "grad_norm": 1.1024504993278263,
+      "learning_rate": 0.00040500000000000003,
+      "loss": 7.8094,
+      "step": 135
+    },
+    {
+      "epoch": 0.00136,
+      "grad_norm": 0.7406778288782102,
+      "learning_rate": 0.00040800000000000005,
+      "loss": 7.7862,
+      "step": 136
+    },
+    {
+      "epoch": 0.00137,
+      "grad_norm": 0.948337541042023,
+      "learning_rate": 0.000411,
+      "loss": 7.7688,
+      "step": 137
+    },
+    {
+      "epoch": 0.00138,
+      "grad_norm": 0.9084991981726261,
+      "learning_rate": 0.00041400000000000003,
+      "loss": 7.7417,
+      "step": 138
+    },
+    {
+      "epoch": 0.00139,
+      "grad_norm": 0.7074494223870315,
+      "learning_rate": 0.00041700000000000005,
+      "loss": 7.7227,
+      "step": 139
+    },
+    {
+      "epoch": 0.0014,
+      "grad_norm": 0.7974190203181878,
+      "learning_rate": 0.00042000000000000007,
+      "loss": 7.7086,
+      "step": 140
+    },
+    {
+      "epoch": 0.00141,
+      "grad_norm": 0.6861992297830735,
+      "learning_rate": 0.000423,
+      "loss": 7.6913,
+      "step": 141
+    },
+    {
+      "epoch": 0.00142,
+      "grad_norm": 0.7349980301116962,
+      "learning_rate": 0.00042599999999999995,
+      "loss": 7.677,
+      "step": 142
+    },
+    {
+      "epoch": 0.00143,
+      "grad_norm": 0.7684994731116016,
+      "learning_rate": 0.00042899999999999997,
+      "loss": 7.6484,
+      "step": 143
+    },
+    {
+      "epoch": 0.00144,
+      "grad_norm": 0.6719380878952201,
+      "learning_rate": 0.000432,
+      "loss": 7.629,
+      "step": 144
+    },
+    {
+      "epoch": 0.00145,
+      "grad_norm": 0.7001362890446781,
+      "learning_rate": 0.000435,
+      "loss": 7.6232,
+      "step": 145
+    },
+    {
+      "epoch": 0.00146,
+      "grad_norm": 0.7384188504614574,
+      "learning_rate": 0.00043799999999999997,
+      "loss": 7.5849,
+      "step": 146
+    },
+    {
+      "epoch": 0.00147,
+      "grad_norm": 0.6536906507985312,
+      "learning_rate": 0.000441,
+      "loss": 7.5755,
+      "step": 147
+    },
+    {
+      "epoch": 0.00148,
+      "grad_norm": 0.6570387947682329,
+      "learning_rate": 0.000444,
+      "loss": 7.5632,
+      "step": 148
+    },
+    {
+      "epoch": 0.00149,
+      "grad_norm": 0.7127206041028632,
+      "learning_rate": 0.00044699999999999997,
+      "loss": 7.5571,
+      "step": 149
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 0.7308289900379658,
+      "learning_rate": 0.00045,
+      "loss": 7.5253,
+      "step": 150
+    },
+    {
+      "epoch": 0.00151,
+      "grad_norm": 0.8664894863114535,
+      "learning_rate": 0.000453,
+      "loss": 7.505,
+      "step": 151
+    },
+    {
+      "epoch": 0.00152,
+      "grad_norm": 0.8895841422117233,
+      "learning_rate": 0.000456,
+      "loss": 7.5034,
+      "step": 152
+    },
+    {
+      "epoch": 0.00153,
+      "grad_norm": 0.662621907628313,
+      "learning_rate": 0.000459,
+      "loss": 7.4855,
+      "step": 153
+    },
+    {
+      "epoch": 0.00154,
+      "grad_norm": 0.7176449004695503,
+      "learning_rate": 0.000462,
+      "loss": 7.4655,
+      "step": 154
+    },
+    {
+      "epoch": 0.00155,
+      "grad_norm": 0.7407325792933065,
+      "learning_rate": 0.000465,
+      "loss": 7.4508,
+      "step": 155
+    },
+    {
+      "epoch": 0.00156,
+      "grad_norm": 0.7037717820949844,
+      "learning_rate": 0.000468,
+      "loss": 7.4265,
+      "step": 156
+    },
+    {
+      "epoch": 0.00157,
+      "grad_norm": 0.5688508632010355,
+      "learning_rate": 0.000471,
+      "loss": 7.4159,
+      "step": 157
+    },
+    {
+      "epoch": 0.00158,
+      "grad_norm": 0.812094320370736,
+      "learning_rate": 0.00047400000000000003,
+      "loss": 7.4015,
+      "step": 158
+    },
+    {
+      "epoch": 0.00159,
+      "grad_norm": 0.6851465337071334,
+      "learning_rate": 0.000477,
+      "loss": 7.3879,
+      "step": 159
+    },
+    {
+      "epoch": 0.0016,
+      "grad_norm": 0.9553203809874583,
+      "learning_rate": 0.00048,
+      "loss": 7.3739,
+      "step": 160
+    },
+    {
+      "epoch": 0.00161,
+      "grad_norm": 0.7328141993841643,
+      "learning_rate": 0.00048300000000000003,
+      "loss": 7.3497,
+      "step": 161
+    },
+    {
+      "epoch": 0.00162,
+      "grad_norm": 0.6781537584586433,
+      "learning_rate": 0.00048600000000000005,
+      "loss": 7.3535,
+      "step": 162
+    },
+    {
+      "epoch": 0.00163,
+      "grad_norm": 0.5807738383940755,
+      "learning_rate": 0.0004890000000000001,
+      "loss": 7.3306,
+      "step": 163
+    },
+    {
+      "epoch": 0.00164,
+      "grad_norm": 0.6466438270814104,
+      "learning_rate": 0.000492,
+      "loss": 7.3371,
+      "step": 164
+    },
+    {
+      "epoch": 0.00165,
+      "grad_norm": 0.4365334647721832,
+      "learning_rate": 0.000495,
+      "loss": 7.3038,
+      "step": 165
+    },
+    {
+      "epoch": 0.00166,
+      "grad_norm": 0.7387692677149859,
+      "learning_rate": 0.0004980000000000001,
+      "loss": 7.2849,
+      "step": 166
+    },
+    {
+      "epoch": 0.00167,
+      "grad_norm": 0.6489238814400348,
+      "learning_rate": 0.000501,
+      "loss": 7.2765,
+      "step": 167
+    },
+    {
+      "epoch": 0.00168,
+      "grad_norm": 0.4943076416069544,
+      "learning_rate": 0.000504,
+      "loss": 7.2512,
+      "step": 168
+    },
+    {
+      "epoch": 0.00169,
+      "grad_norm": 0.6623047646201438,
+      "learning_rate": 0.0005070000000000001,
+      "loss": 7.2485,
+      "step": 169
+    },
+    {
+      "epoch": 0.0017,
+      "grad_norm": 0.5841690806336316,
+      "learning_rate": 0.00051,
+      "loss": 7.2331,
+      "step": 170
+    },
+    {
+      "epoch": 0.00171,
+      "grad_norm": 0.455313434659774,
+      "learning_rate": 0.000513,
+      "loss": 7.2132,
+      "step": 171
+    },
+    {
+      "epoch": 0.00172,
+      "grad_norm": 0.6012704724036779,
+      "learning_rate": 0.000516,
+      "loss": 7.214,
+      "step": 172
+    },
+    {
+      "epoch": 0.00173,
+      "grad_norm": 0.5433283029678282,
+      "learning_rate": 0.0005189999999999999,
+      "loss": 7.1993,
+      "step": 173
+    },
+    {
+      "epoch": 0.00174,
+      "grad_norm": 0.38251135313099477,
+      "learning_rate": 0.000522,
+      "loss": 7.2121,
+      "step": 174
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 0.6483519243845758,
+      "learning_rate": 0.000525,
+      "loss": 7.1796,
+      "step": 175
+    },
+    {
+      "epoch": 0.00176,
+      "grad_norm": 0.5106681215905619,
+      "learning_rate": 0.0005279999999999999,
+      "loss": 7.1531,
+      "step": 176
+    },
+    {
+      "epoch": 0.00177,
+      "grad_norm": 0.3867365673342131,
+      "learning_rate": 0.000531,
+      "loss": 7.1566,
+      "step": 177
+    },
+    {
+      "epoch": 0.00178,
+      "grad_norm": 0.6803005272577738,
+      "learning_rate": 0.000534,
+      "loss": 7.1525,
+      "step": 178
+    },
+    {
+      "epoch": 0.00179,
+      "grad_norm": 0.44918531151657043,
+      "learning_rate": 0.000537,
+      "loss": 7.1407,
+      "step": 179
+    },
+    {
+      "epoch": 0.0018,
+      "grad_norm": 0.39944612038572996,
+      "learning_rate": 0.00054,
+      "loss": 7.1121,
+      "step": 180
+    },
+    {
+      "epoch": 0.00181,
+      "grad_norm": 0.5937704312371144,
+      "learning_rate": 0.000543,
+      "loss": 7.1037,
+      "step": 181
+    },
+    {
+      "epoch": 0.00182,
+      "grad_norm": 0.3956001398507633,
+      "learning_rate": 0.000546,
+      "loss": 7.0976,
+      "step": 182
+    },
+    {
+      "epoch": 0.00183,
+      "grad_norm": 0.36202282333499824,
+      "learning_rate": 0.000549,
+      "loss": 7.0864,
+      "step": 183
+    },
+    {
+      "epoch": 0.00184,
+      "grad_norm": 0.564760920006622,
+      "learning_rate": 0.000552,
+      "loss": 7.0785,
+      "step": 184
+    },
+    {
+      "epoch": 0.00185,
+      "grad_norm": 0.36333201524042624,
+      "learning_rate": 0.000555,
+      "loss": 7.0515,
+      "step": 185
+    },
+    {
+      "epoch": 0.00186,
+      "grad_norm": 0.3517854453126823,
+      "learning_rate": 0.000558,
+      "loss": 7.0614,
+      "step": 186
+    },
+    {
+      "epoch": 0.00187,
+      "grad_norm": 0.5413049623390372,
+      "learning_rate": 0.000561,
+      "loss": 7.0394,
+      "step": 187
+    },
+    {
+      "epoch": 0.00188,
+      "grad_norm": 0.4452202297805183,
+      "learning_rate": 0.000564,
+      "loss": 7.0168,
+      "step": 188
+    },
+    {
+      "epoch": 0.00189,
+      "grad_norm": 0.2888404630990515,
+      "learning_rate": 0.000567,
+      "loss": 7.0108,
+      "step": 189
+    },
+    {
+      "epoch": 0.0019,
+      "grad_norm": 0.5069984229642092,
+      "learning_rate": 0.00057,
+      "loss": 6.9967,
+      "step": 190
+    },
+    {
+      "epoch": 0.00191,
+      "grad_norm": 0.5049159939798347,
+      "learning_rate": 0.000573,
+      "loss": 6.9897,
+      "step": 191
+    },
+    {
+      "epoch": 0.00192,
+      "grad_norm": 0.353459237780404,
+      "learning_rate": 0.000576,
+      "loss": 6.9784,
+      "step": 192
+    },
+    {
+      "epoch": 0.00193,
+      "grad_norm": 0.4583730098712965,
+      "learning_rate": 0.000579,
+      "loss": 6.9798,
+      "step": 193
+    },
+    {
+      "epoch": 0.00194,
+      "grad_norm": 0.5934016188318005,
+      "learning_rate": 0.000582,
+      "loss": 6.9757,
+      "step": 194
+    },
+    {
+      "epoch": 0.00195,
+      "grad_norm": 0.6611510755360627,
+      "learning_rate": 0.000585,
+      "loss": 6.9556,
+      "step": 195
+    },
+    {
+      "epoch": 0.00196,
+      "grad_norm": 0.6794801564658052,
+      "learning_rate": 0.000588,
+      "loss": 6.9455,
+      "step": 196
+    },
+    {
+      "epoch": 0.00197,
+      "grad_norm": 0.6965025324131076,
+      "learning_rate": 0.000591,
+      "loss": 6.9424,
+      "step": 197
+    },
+    {
+      "epoch": 0.00198,
+      "grad_norm": 1.0010424094015926,
+      "learning_rate": 0.000594,
+      "loss": 6.9428,
+      "step": 198
+    },
+    {
+      "epoch": 0.00199,
+      "grad_norm": 0.8473959375492113,
+      "learning_rate": 0.0005970000000000001,
+      "loss": 6.9257,
+      "step": 199
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 0.4674164352666973,
+      "learning_rate": 0.0006000000000000001,
+      "loss": 6.9135,
+      "step": 200
+    },
+    {
+      "epoch": 0.00201,
+      "grad_norm": 0.6130101305251902,
+      "learning_rate": 0.000603,
+      "loss": 6.8907,
+      "step": 201
+    },
+    {
+      "epoch": 0.00202,
+      "grad_norm": 0.5766204915712545,
+      "learning_rate": 0.0006060000000000001,
+      "loss": 6.8786,
+      "step": 202
+    },
+    {
+      "epoch": 0.00203,
+      "grad_norm": 0.42726729176690026,
+      "learning_rate": 0.0006090000000000001,
+      "loss": 6.8819,
+      "step": 203
+    },
+    {
+      "epoch": 0.00204,
+      "grad_norm": 0.5161441207128264,
+      "learning_rate": 0.000612,
+      "loss": 6.8675,
+      "step": 204
+    },
+    {
+      "epoch": 0.00205,
+      "grad_norm": 0.44005998133884483,
+      "learning_rate": 0.000615,
+      "loss": 6.852,
+      "step": 205
+    },
+    {
+      "epoch": 0.00206,
+      "grad_norm": 0.4214710856814876,
+      "learning_rate": 0.000618,
+      "loss": 6.8455,
+      "step": 206
+    },
+    {
+      "epoch": 0.00207,
+      "grad_norm": 0.3835330802523119,
+      "learning_rate": 0.000621,
+      "loss": 6.8358,
+      "step": 207
+    },
+    {
+      "epoch": 0.00208,
+      "grad_norm": 0.4118282878888594,
+      "learning_rate": 0.000624,
+      "loss": 6.8325,
+      "step": 208
+    },
+    {
+      "epoch": 0.00209,
+      "grad_norm": 0.35207765117623135,
+      "learning_rate": 0.000627,
+      "loss": 6.8108,
+      "step": 209
+    },
+    {
+      "epoch": 0.0021,
+      "grad_norm": 0.35165038308098223,
+      "learning_rate": 0.00063,
+      "loss": 6.8001,
+      "step": 210
+    },
+    {
+      "epoch": 0.00211,
+      "grad_norm": 0.2907998960376576,
+      "learning_rate": 0.000633,
+      "loss": 6.794,
+      "step": 211
+    },
+    {
+      "epoch": 0.00212,
+      "grad_norm": 0.34183459291083,
+      "learning_rate": 0.000636,
+      "loss": 6.7864,
+      "step": 212
+    },
+    {
+      "epoch": 0.00213,
+      "grad_norm": 0.31625023606106545,
+      "learning_rate": 0.000639,
+      "loss": 6.7664,
+      "step": 213
+    },
+    {
+      "epoch": 0.00214,
+      "grad_norm": 0.3389167462098675,
+      "learning_rate": 0.000642,
+      "loss": 6.7645,
+      "step": 214
+    },
+    {
+      "epoch": 0.00215,
+      "grad_norm": 0.4689783367804683,
+      "learning_rate": 0.000645,
+      "loss": 6.7617,
+      "step": 215
+    },
+    {
+      "epoch": 0.00216,
+      "grad_norm": 0.7554849005001304,
+      "learning_rate": 0.000648,
+      "loss": 6.763,
+      "step": 216
+    },
+    {
+      "epoch": 0.00217,
+      "grad_norm": 1.3852957930548755,
+      "learning_rate": 0.000651,
+      "loss": 6.7541,
+      "step": 217
+    },
+    {
+      "epoch": 0.00218,
+      "grad_norm": 0.6607509792225358,
+      "learning_rate": 0.000654,
+      "loss": 6.7405,
+      "step": 218
+    },
+    {
+      "epoch": 0.00219,
+      "grad_norm": 0.6480761266032634,
+      "learning_rate": 0.000657,
+      "loss": 6.7122,
+      "step": 219
+    },
+    {
+      "epoch": 0.0022,
+      "grad_norm": 1.2227203355321279,
+      "learning_rate": 0.00066,
+      "loss": 6.7471,
+      "step": 220
+    },
+    {
+      "epoch": 0.00221,
+      "grad_norm": 1.1390371544106608,
+      "learning_rate": 0.0006630000000000001,
+      "loss": 6.7248,
+      "step": 221
+    },
+    {
+      "epoch": 0.00222,
+      "grad_norm": 0.7354052143767025,
+      "learning_rate": 0.000666,
+      "loss": 6.7013,
+      "step": 222
+    },
+    {
+      "epoch": 0.00223,
+      "grad_norm": 0.6134943537271429,
+      "learning_rate": 0.000669,
+      "loss": 6.6937,
+      "step": 223
+    },
+    {
+      "epoch": 0.00224,
+      "grad_norm": 0.5502679770385809,
+      "learning_rate": 0.0006720000000000001,
+      "loss": 6.6934,
+      "step": 224
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 0.47738132510350817,
+      "learning_rate": 0.000675,
+      "loss": 6.6778,
+      "step": 225
+    },
+    {
+      "epoch": 0.00226,
+      "grad_norm": 0.507114065421453,
+      "learning_rate": 0.000678,
+      "loss": 6.6712,
+      "step": 226
+    },
+    {
+      "epoch": 0.00227,
+      "grad_norm": 0.4403172481189424,
+      "learning_rate": 0.0006810000000000001,
+      "loss": 6.646,
+      "step": 227
+    },
+    {
+      "epoch": 0.00228,
+      "grad_norm": 0.34083753361811386,
+      "learning_rate": 0.000684,
+      "loss": 6.6465,
+      "step": 228
+    },
+    {
+      "epoch": 0.00229,
+      "grad_norm": 0.3502366193988861,
+      "learning_rate": 0.000687,
+      "loss": 6.6418,
+      "step": 229
+    },
+    {
+      "epoch": 0.0023,
+      "grad_norm": 0.40867670062411404,
+      "learning_rate": 0.0006900000000000001,
+      "loss": 6.6218,
+      "step": 230
+    },
+    {
+      "epoch": 0.00231,
+      "grad_norm": 0.3512396474179769,
+      "learning_rate": 0.000693,
+      "loss": 6.605,
+      "step": 231
+    },
+    {
+      "epoch": 0.00232,
+      "grad_norm": 0.37061772246720376,
+      "learning_rate": 0.000696,
+      "loss": 6.6164,
+      "step": 232
+    },
+    {
+      "epoch": 0.00233,
+      "grad_norm": 0.30975845835339183,
+      "learning_rate": 0.0006990000000000001,
+      "loss": 6.6026,
+      "step": 233
+    },
+    {
+      "epoch": 0.00234,
+      "grad_norm": 0.31056463949410484,
+      "learning_rate": 0.000702,
+      "loss": 6.5934,
+      "step": 234
+    },
+    {
+      "epoch": 0.00235,
+      "grad_norm": 0.3416829938754332,
+      "learning_rate": 0.000705,
+      "loss": 6.5798,
+      "step": 235
+    },
+    {
+      "epoch": 0.00236,
+      "grad_norm": 0.3580449551775802,
+      "learning_rate": 0.000708,
+      "loss": 6.5618,
+      "step": 236
+    },
+    {
+      "epoch": 0.00237,
+      "grad_norm": 0.31186166958910727,
+      "learning_rate": 0.0007109999999999999,
+      "loss": 6.5709,
+      "step": 237
+    },
+    {
+      "epoch": 0.00238,
+      "grad_norm": 0.35712039334009765,
+      "learning_rate": 0.000714,
+      "loss": 6.5601,
+      "step": 238
+    },
+    {
+      "epoch": 0.00239,
+      "grad_norm": 0.4689874670209286,
+      "learning_rate": 0.000717,
+      "loss": 6.5431,
+      "step": 239
+    },
+    {
+      "epoch": 0.0024,
+      "grad_norm": 0.8385556457838154,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 6.5652,
+      "step": 240
+    },
+    {
+      "epoch": 0.00241,
+      "grad_norm": 1.2597764356570753,
+      "learning_rate": 0.000723,
+      "loss": 6.5721,
+      "step": 241
+    },
+    {
+      "epoch": 0.00242,
+      "grad_norm": 0.7018547016449656,
+      "learning_rate": 0.000726,
+      "loss": 6.5258,
+      "step": 242
+    },
+    {
+      "epoch": 0.00243,
+      "grad_norm": 0.9679840331331718,
+      "learning_rate": 0.000729,
+      "loss": 6.5235,
+      "step": 243
+    },
+    {
+      "epoch": 0.00244,
+      "grad_norm": 0.8151244219176279,
+      "learning_rate": 0.000732,
+      "loss": 6.5289,
+      "step": 244
+    },
+    {
+      "epoch": 0.00245,
+      "grad_norm": 0.7408239035858821,
+      "learning_rate": 0.000735,
+      "loss": 6.5159,
+      "step": 245
+    },
+    {
+      "epoch": 0.00246,
+      "grad_norm": 0.7269888029650857,
+      "learning_rate": 0.000738,
+      "loss": 6.5006,
+      "step": 246
+    },
+    {
+      "epoch": 0.00247,
+      "grad_norm": 0.6715573633422388,
+      "learning_rate": 0.000741,
+      "loss": 6.4916,
+      "step": 247
+    },
+    {
+      "epoch": 0.00248,
+      "grad_norm": 0.5355440416613977,
+      "learning_rate": 0.000744,
+      "loss": 6.4965,
+      "step": 248
+    },
+    {
+      "epoch": 0.00249,
+      "grad_norm": 0.47837466185087324,
+      "learning_rate": 0.000747,
+      "loss": 6.474,
+      "step": 249
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 0.5421674013984196,
+      "learning_rate": 0.00075,
+      "loss": 6.4781,
+      "step": 250
+    },
+    {
+      "epoch": 0.00251,
+      "grad_norm": 0.5046285559759315,
+      "learning_rate": 0.000753,
+      "loss": 6.4394,
+      "step": 251
+    },
+    {
+      "epoch": 0.00252,
+      "grad_norm": 0.4902399046559828,
+      "learning_rate": 0.000756,
+      "loss": 6.4478,
+      "step": 252
+    },
+    {
+      "epoch": 0.00253,
+      "grad_norm": 0.6026763597870729,
+      "learning_rate": 0.000759,
+      "loss": 6.4417,
+      "step": 253
+    },
+    {
+      "epoch": 0.00254,
+      "grad_norm": 0.6534629043130656,
+      "learning_rate": 0.000762,
+      "loss": 6.4357,
+      "step": 254
+    },
+    {
+      "epoch": 0.00255,
+      "grad_norm": 0.5879568865224861,
+      "learning_rate": 0.0007650000000000001,
+      "loss": 6.4331,
+      "step": 255
+    },
+    {
+      "epoch": 0.00256,
+      "grad_norm": 0.587549512889128,
+      "learning_rate": 0.000768,
+      "loss": 6.4121,
+      "step": 256
+    },
+    {
+      "epoch": 0.00257,
+      "grad_norm": 0.4948417519702818,
+      "learning_rate": 0.000771,
+      "loss": 6.4144,
+      "step": 257
+    },
+    {
+      "epoch": 0.00258,
+      "grad_norm": 0.500640645286767,
+      "learning_rate": 0.0007740000000000001,
+      "loss": 6.3915,
+      "step": 258
+    },
+    {
+      "epoch": 0.00259,
+      "grad_norm": 0.46096026590523803,
+      "learning_rate": 0.000777,
+      "loss": 6.3904,
+      "step": 259
+    },
+    {
+      "epoch": 0.0026,
+      "grad_norm": 0.36710644781296853,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 6.3738,
+      "step": 260
+    },
+    {
+      "epoch": 0.00261,
+      "grad_norm": 0.3228962429333065,
+      "learning_rate": 0.0007830000000000001,
+      "loss": 6.3839,
+      "step": 261
+    },
+    {
+      "epoch": 0.00262,
+      "grad_norm": 0.33723485694460476,
+      "learning_rate": 0.000786,
+      "loss": 6.3601,
+      "step": 262
+    },
+    {
+      "epoch": 0.00263,
+      "grad_norm": 0.3792844357509868,
+      "learning_rate": 0.0007890000000000001,
+      "loss": 6.3664,
+      "step": 263
+    },
+    {
+      "epoch": 0.00264,
+      "grad_norm": 0.41999467297545995,
+      "learning_rate": 0.0007920000000000001,
+      "loss": 6.3505,
+      "step": 264
+    },
+    {
+      "epoch": 0.00265,
+      "grad_norm": 0.5058327981670172,
+      "learning_rate": 0.000795,
+      "loss": 6.3427,
+      "step": 265
+    },
+    {
+      "epoch": 0.00266,
+      "grad_norm": 0.728666013298328,
+      "learning_rate": 0.0007980000000000001,
+      "loss": 6.3445,
+      "step": 266
+    },
+    {
+      "epoch": 0.00267,
+      "grad_norm": 1.051597167429085,
+      "learning_rate": 0.0008010000000000001,
+      "loss": 6.3504,
+      "step": 267
+    },
+    {
+      "epoch": 0.00268,
+      "grad_norm": 0.8184600443662068,
+      "learning_rate": 0.000804,
+      "loss": 6.3425,
+      "step": 268
+    },
+    {
+      "epoch": 0.00269,
+      "grad_norm": 0.9044150906220578,
+      "learning_rate": 0.0008070000000000001,
+      "loss": 6.3249,
+      "step": 269
+    },
+    {
+      "epoch": 0.0027,
+      "grad_norm": 1.128589514160359,
+      "learning_rate": 0.0008100000000000001,
+      "loss": 6.3418,
+      "step": 270
+    },
+    {
+      "epoch": 0.00271,
+      "grad_norm": 0.8269681270889044,
+      "learning_rate": 0.000813,
+      "loss": 6.3324,
+      "step": 271
+    },
+    {
+      "epoch": 0.00272,
+      "grad_norm": 0.9605488749490486,
+      "learning_rate": 0.0008160000000000001,
+      "loss": 6.3229,
+      "step": 272
+    },
+    {
+      "epoch": 0.00273,
+      "grad_norm": 1.4256959145221717,
+      "learning_rate": 0.0008190000000000001,
+      "loss": 6.3332,
+      "step": 273
+    },
+    {
+      "epoch": 0.00274,
+      "grad_norm": 0.8746481700595169,
+      "learning_rate": 0.000822,
+      "loss": 6.301,
+      "step": 274
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 0.7483839590071554,
+      "learning_rate": 0.0008250000000000001,
+      "loss": 6.3047,
+      "step": 275
+    },
+    {
+      "epoch": 0.00276,
+      "grad_norm": 0.9683696395132253,
+      "learning_rate": 0.0008280000000000001,
+      "loss": 6.2965,
+      "step": 276
+    },
+    {
+      "epoch": 0.00277,
+      "grad_norm": 0.9445364353770574,
+      "learning_rate": 0.0008310000000000001,
+      "loss": 6.3021,
+      "step": 277
+    },
+    {
+      "epoch": 0.00278,
+      "grad_norm": 0.5637393663463164,
+      "learning_rate": 0.0008340000000000001,
+      "loss": 6.2711,
+      "step": 278
+    },
+    {
+      "epoch": 0.00279,
+      "grad_norm": 0.6474632709262491,
+      "learning_rate": 0.0008370000000000001,
+      "loss": 6.2722,
+      "step": 279
+    },
+    {
+      "epoch": 0.0028,
+      "grad_norm": 0.5477535984985658,
+      "learning_rate": 0.0008400000000000001,
+      "loss": 6.2548,
+      "step": 280
+    },
+    {
+      "epoch": 0.00281,
+      "grad_norm": 0.5235139271800053,
+      "learning_rate": 0.0008430000000000001,
+      "loss": 6.2428,
+      "step": 281
+    },
+    {
+      "epoch": 0.00282,
+      "grad_norm": 0.4309641347762167,
+      "learning_rate": 0.000846,
+      "loss": 6.2382,
+      "step": 282
+    },
+    {
+      "epoch": 0.00283,
+      "grad_norm": 0.47598469419389666,
+      "learning_rate": 0.0008489999999999999,
+      "loss": 6.2321,
+      "step": 283
+    },
+    {
+      "epoch": 0.00284,
+      "grad_norm": 0.41791767912211497,
+      "learning_rate": 0.0008519999999999999,
+      "loss": 6.2199,
+      "step": 284
+    },
+    {
+      "epoch": 0.00285,
+      "grad_norm": 0.4068034669968041,
+      "learning_rate": 0.000855,
+      "loss": 6.2226,
+      "step": 285
+    },
+    {
+      "epoch": 0.00286,
+      "grad_norm": 0.461449105404332,
+      "learning_rate": 0.0008579999999999999,
+      "loss": 6.2004,
+      "step": 286
+    },
+    {
+      "epoch": 0.00287,
+      "grad_norm": 0.3833700707849029,
+      "learning_rate": 0.000861,
+      "loss": 6.2013,
+      "step": 287
+    },
+    {
+      "epoch": 0.00288,
+      "grad_norm": 0.3960645710165419,
+      "learning_rate": 0.000864,
+      "loss": 6.1818,
+      "step": 288
+    },
+    {
+      "epoch": 0.00289,
+      "grad_norm": 0.44431212101949524,
+      "learning_rate": 0.0008669999999999999,
+      "loss": 6.1961,
+      "step": 289
+    },
+    {
+      "epoch": 0.0029,
+      "grad_norm": 0.5861155579490576,
+      "learning_rate": 0.00087,
+      "loss": 6.1786,
+      "step": 290
+    },
+    {
+      "epoch": 0.00291,
+      "grad_norm": 0.6639450218120423,
+      "learning_rate": 0.000873,
+      "loss": 6.1739,
+      "step": 291
+    },
+    {
+      "epoch": 0.00292,
+      "grad_norm": 0.8465664162859037,
+      "learning_rate": 0.0008759999999999999,
+      "loss": 6.1756,
+      "step": 292
+    },
+    {
+      "epoch": 0.00293,
+      "grad_norm": 0.9291009677838105,
+      "learning_rate": 0.000879,
+      "loss": 6.1618,
+      "step": 293
+    },
+    {
+      "epoch": 0.00294,
+      "grad_norm": 1.3082966000676708,
+      "learning_rate": 0.000882,
+      "loss": 6.1833,
+      "step": 294
+    },
+    {
+      "epoch": 0.00295,
+      "grad_norm": 1.1893342417992165,
+      "learning_rate": 0.0008849999999999999,
+      "loss": 6.1586,
+      "step": 295
+    },
+    {
+      "epoch": 0.00296,
+      "grad_norm": 0.6546841726146672,
+      "learning_rate": 0.000888,
+      "loss": 6.1551,
+      "step": 296
+    },
+    {
+      "epoch": 0.00297,
+      "grad_norm": 0.8416245966488266,
+      "learning_rate": 0.000891,
+      "loss": 6.1315,
+      "step": 297
+    },
+    {
+      "epoch": 0.00298,
+      "grad_norm": 0.806301942306892,
+      "learning_rate": 0.0008939999999999999,
+      "loss": 6.1452,
+      "step": 298
+    },
+    {
+      "epoch": 0.00299,
+      "grad_norm": 1.0994469943740992,
+      "learning_rate": 0.000897,
+      "loss": 6.1288,
+      "step": 299
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 0.7425213774845364,
+      "learning_rate": 0.0009,
+      "loss": 6.123,
+      "step": 300
+    },
+    {
+      "epoch": 0.00301,
+      "grad_norm": 0.6433032250398036,
+      "learning_rate": 0.0009029999999999999,
+      "loss": 6.1149,
+      "step": 301
+    },
+    {
+      "epoch": 0.00302,
+      "grad_norm": 0.5449990789285414,
+      "learning_rate": 0.000906,
+      "loss": 6.0952,
+      "step": 302
+    },
+    {
+      "epoch": 0.00303,
+      "grad_norm": 0.5356674479383384,
+      "learning_rate": 0.000909,
+      "loss": 6.1067,
+      "step": 303
+    },
+    {
+      "epoch": 0.00304,
+      "grad_norm": 0.5049891146020773,
+      "learning_rate": 0.000912,
+      "loss": 6.1013,
+      "step": 304
+    },
+    {
+      "epoch": 0.00305,
+      "grad_norm": 0.48068374440179407,
+      "learning_rate": 0.000915,
+      "loss": 6.0985,
+      "step": 305
+    },
+    {
+      "epoch": 0.00306,
+      "grad_norm": 0.39305660615027715,
+      "learning_rate": 0.000918,
+      "loss": 6.0835,
+      "step": 306
+    },
+    {
+      "epoch": 0.00307,
+      "grad_norm": 0.4044411514506039,
+      "learning_rate": 0.000921,
+      "loss": 6.0774,
+      "step": 307
+    },
+    {
+      "epoch": 0.00308,
+      "grad_norm": 0.3317383414417003,
+      "learning_rate": 0.000924,
+      "loss": 6.0578,
+      "step": 308
+    },
+    {
+      "epoch": 0.00309,
+      "grad_norm": 0.4170081186564571,
+      "learning_rate": 0.000927,
+      "loss": 6.0741,
+      "step": 309
+    },
+    {
+      "epoch": 0.0031,
+      "grad_norm": 0.3537372632819937,
+      "learning_rate": 0.00093,
+      "loss": 6.0544,
+      "step": 310
+    },
+    {
+      "epoch": 0.00311,
+      "grad_norm": 0.3807795877969161,
+      "learning_rate": 0.000933,
+      "loss": 6.0455,
+      "step": 311
+    },
+    {
+      "epoch": 0.00312,
+      "grad_norm": 0.36301395919450175,
+      "learning_rate": 0.000936,
+      "loss": 6.025,
+      "step": 312
+    },
+    {
+      "epoch": 0.00313,
+      "grad_norm": 0.456154645430038,
+      "learning_rate": 0.0009390000000000001,
+      "loss": 6.0399,
+      "step": 313
+    },
+    {
+      "epoch": 0.00314,
+      "grad_norm": 0.6252307526263491,
+      "learning_rate": 0.000942,
+      "loss": 6.048,
+      "step": 314
+    },
+    {
+      "epoch": 0.00315,
+      "grad_norm": 1.0054536709858048,
+      "learning_rate": 0.000945,
+      "loss": 6.0332,
+      "step": 315
+    },
+    {
+      "epoch": 0.00316,
+      "grad_norm": 1.2671801420052267,
+      "learning_rate": 0.0009480000000000001,
+      "loss": 6.0322,
+      "step": 316
+    },
+    {
+      "epoch": 0.00317,
+      "grad_norm": 0.5267566088823273,
+      "learning_rate": 0.000951,
+      "loss": 6.0198,
+      "step": 317
+    },
+    {
+      "epoch": 0.00318,
+      "grad_norm": 1.2005145600955376,
+      "learning_rate": 0.000954,
+      "loss": 6.041,
+      "step": 318
+    },
+    {
+      "epoch": 0.00319,
+      "grad_norm": 1.004603435172909,
+      "learning_rate": 0.0009570000000000001,
+      "loss": 6.025,
+      "step": 319
+    },
+    {
+      "epoch": 0.0032,
+      "grad_norm": 1.1201777687880106,
+      "learning_rate": 0.00096,
+      "loss": 6.0356,
+      "step": 320
+    },
+    {
+      "epoch": 0.00321,
+      "grad_norm": 1.0167944206203925,
+      "learning_rate": 0.000963,
+      "loss": 6.012,
+      "step": 321
+    },
+    {
+      "epoch": 0.00322,
+      "grad_norm": 1.374478470085335,
+      "learning_rate": 0.0009660000000000001,
+      "loss": 6.025,
+      "step": 322
+    },
+    {
+      "epoch": 0.00323,
+      "grad_norm": 0.7765654772896999,
+      "learning_rate": 0.000969,
+      "loss": 6.0057,
+      "step": 323
+    },
+    {
+      "epoch": 0.00324,
+      "grad_norm": 0.6987857474960232,
+      "learning_rate": 0.0009720000000000001,
+      "loss": 6.0035,
+      "step": 324
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 0.7564069291902213,
+      "learning_rate": 0.0009750000000000001,
+      "loss": 5.9742,
+      "step": 325
+    },
+    {
+      "epoch": 0.00326,
+      "grad_norm": 0.7275905653375951,
+      "learning_rate": 0.0009780000000000001,
+      "loss": 5.9877,
+      "step": 326
+    },
+    {
+      "epoch": 0.00327,
+      "grad_norm": 0.6984989438722892,
+      "learning_rate": 0.000981,
+      "loss": 5.9857,
+      "step": 327
+    },
+    {
+      "epoch": 0.00328,
+      "grad_norm": 0.721382136086564,
+      "learning_rate": 0.000984,
+      "loss": 5.9627,
+      "step": 328
+    },
+    {
+      "epoch": 0.00329,
+      "grad_norm": 0.7841656285031653,
+      "learning_rate": 0.000987,
+      "loss": 5.9649,
+      "step": 329
+    },
+    {
+      "epoch": 0.0033,
+      "grad_norm": 0.8334397806028965,
+      "learning_rate": 0.00099,
+      "loss": 5.9568,
+      "step": 330
+    },
+    {
+      "epoch": 0.00331,
+      "grad_norm": 1.070828806370832,
+      "learning_rate": 0.0009930000000000002,
+      "loss": 5.9642,
+      "step": 331
+    },
+    {
+      "epoch": 0.00332,
+      "grad_norm": 1.0826703064607723,
+      "learning_rate": 0.0009960000000000001,
+      "loss": 5.9538,
+      "step": 332
+    },
+    {
+      "epoch": 0.00333,
+      "grad_norm": 0.9458285226462658,
+      "learning_rate": 0.000999,
+      "loss": 5.9361,
+      "step": 333
+    },
+    {
+      "epoch": 0.00334,
+      "grad_norm": 0.8783572499928525,
+      "learning_rate": 0.001002,
+      "loss": 5.9443,
+      "step": 334
+    },
+    {
+      "epoch": 0.00335,
+      "grad_norm": 0.5654549281403035,
+      "learning_rate": 0.001005,
+      "loss": 5.9354,
+      "step": 335
+    },
+    {
+      "epoch": 0.00336,
+      "grad_norm": 0.6516350129184051,
+      "learning_rate": 0.001008,
+      "loss": 5.9285,
+      "step": 336
+    },
+    {
+      "epoch": 0.00337,
+      "grad_norm": 0.5946351510133966,
+      "learning_rate": 0.0010110000000000002,
+      "loss": 5.9026,
+      "step": 337
+    },
+    {
+      "epoch": 0.00338,
+      "grad_norm": 0.46824568572142555,
+      "learning_rate": 0.0010140000000000001,
+      "loss": 5.9165,
+      "step": 338
+    },
+    {
+      "epoch": 0.00339,
+      "grad_norm": 0.5098867793571775,
+      "learning_rate": 0.0010170000000000001,
+      "loss": 5.9215,
+      "step": 339
+    },
+    {
+      "epoch": 0.0034,
+      "grad_norm": 0.5202026348670868,
+      "learning_rate": 0.00102,
+      "loss": 5.8881,
+      "step": 340
+    },
+    {
+      "epoch": 0.00341,
+      "grad_norm": 0.4748523902202147,
+      "learning_rate": 0.001023,
+      "loss": 5.8705,
+      "step": 341
+    },
+    {
+      "epoch": 0.00342,
+      "grad_norm": 0.42065780247182966,
+      "learning_rate": 0.001026,
+      "loss": 5.8732,
+      "step": 342
+    },
+    {
+      "epoch": 0.00343,
+      "grad_norm": 0.4385189981740614,
+      "learning_rate": 0.0010290000000000002,
+      "loss": 5.8762,
+      "step": 343
+    },
+    {
+      "epoch": 0.00344,
+      "grad_norm": 0.44131626809067537,
+      "learning_rate": 0.001032,
+      "loss": 5.8646,
+      "step": 344
+    },
+    {
+      "epoch": 0.00345,
+      "grad_norm": 0.5602404326473228,
+      "learning_rate": 0.001035,
+      "loss": 5.8576,
+      "step": 345
+    },
+    {
+      "epoch": 0.00346,
+      "grad_norm": 0.7338782884674983,
+      "learning_rate": 0.0010379999999999999,
+      "loss": 5.8684,
+      "step": 346
+    },
+    {
+      "epoch": 0.00347,
+      "grad_norm": 0.8871217127911775,
+      "learning_rate": 0.001041,
+      "loss": 5.8429,
+      "step": 347
+    },
+    {
+      "epoch": 0.00348,
+      "grad_norm": 0.7294005585257834,
+      "learning_rate": 0.001044,
+      "loss": 5.8498,
+      "step": 348
+    },
+    {
+      "epoch": 0.00349,
+      "grad_norm": 0.672488440407315,
+      "learning_rate": 0.001047,
+      "loss": 5.8268,
+      "step": 349
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 0.944274771030096,
+      "learning_rate": 0.00105,
+      "loss": 5.8574,
+      "step": 350
+    },
+    {
+      "epoch": 0.00351,
+      "grad_norm": 0.8431353927439175,
+      "learning_rate": 0.001053,
+      "loss": 5.8586,
+      "step": 351
+    },
+    {
+      "epoch": 0.00352,
+      "grad_norm": 0.8223595555288745,
+      "learning_rate": 0.0010559999999999999,
+      "loss": 5.8326,
+      "step": 352
+    },
+    {
+      "epoch": 0.00353,
+      "grad_norm": 0.7412573085023916,
+      "learning_rate": 0.001059,
+      "loss": 5.8336,
+      "step": 353
+    },
+    {
+      "epoch": 0.00354,
+      "grad_norm": 0.7219192917744488,
+      "learning_rate": 0.001062,
+      "loss": 5.8047,
+      "step": 354
+    },
+    {
+      "epoch": 0.00355,
+      "grad_norm": 0.7332482461779326,
+      "learning_rate": 0.001065,
+      "loss": 5.8213,
+      "step": 355
+    },
+    {
+      "epoch": 0.00356,
+      "grad_norm": 0.6300796792438244,
+      "learning_rate": 0.001068,
+      "loss": 5.8096,
+      "step": 356
+    },
+    {
+      "epoch": 0.00357,
+      "grad_norm": 0.7796926368866208,
+      "learning_rate": 0.001071,
+      "loss": 5.802,
+      "step": 357
+    },
+    {
+      "epoch": 0.00358,
+      "grad_norm": 1.048118676098695,
+      "learning_rate": 0.001074,
+      "loss": 5.8089,
+      "step": 358
+    },
+    {
+      "epoch": 0.00359,
+      "grad_norm": 0.8469283052653666,
+      "learning_rate": 0.001077,
+      "loss": 5.7908,
+      "step": 359
+    },
+    {
+      "epoch": 0.0036,
+      "grad_norm": 0.824969051981288,
+      "learning_rate": 0.00108,
+      "loss": 5.7948,
+      "step": 360
+    },
+    {
+      "epoch": 0.00361,
+      "grad_norm": 0.8007424119602606,
+      "learning_rate": 0.001083,
+      "loss": 5.7876,
+      "step": 361
+    },
+    {
+      "epoch": 0.00362,
+      "grad_norm": 0.7740442530973947,
+      "learning_rate": 0.001086,
+      "loss": 5.7713,
+      "step": 362
+    },
+    {
+      "epoch": 0.00363,
+      "grad_norm": 0.8217790100779416,
+      "learning_rate": 0.001089,
+      "loss": 5.7673,
+      "step": 363
+    },
+    {
+      "epoch": 0.00364,
+      "grad_norm": 0.9222078053619298,
+      "learning_rate": 0.001092,
+      "loss": 5.7813,
+      "step": 364
+    },
+    {
+      "epoch": 0.00365,
+      "grad_norm": 1.3725693861639678,
+      "learning_rate": 0.001095,
+      "loss": 5.801,
+      "step": 365
+    },
+    {
+      "epoch": 0.00366,
+      "grad_norm": 0.7809021294214551,
+      "learning_rate": 0.001098,
+      "loss": 5.7587,
+      "step": 366
+    },
+    {
+      "epoch": 0.00367,
+      "grad_norm": 1.1843193874441642,
+      "learning_rate": 0.001101,
+      "loss": 5.7731,
+      "step": 367
+    },
+    {
+      "epoch": 0.00368,
+      "grad_norm": 0.7875897064576857,
+      "learning_rate": 0.001104,
+      "loss": 5.7636,
+      "step": 368
+    },
+    {
+      "epoch": 0.00369,
+      "grad_norm": 0.6858717859751121,
+      "learning_rate": 0.001107,
+      "loss": 5.7637,
+      "step": 369
+    },
+    {
+      "epoch": 0.0037,
+      "grad_norm": 0.6252688126791845,
+      "learning_rate": 0.00111,
+      "loss": 5.7428,
+      "step": 370
+    },
+    {
+      "epoch": 0.00371,
+      "grad_norm": 0.6219765776345385,
+      "learning_rate": 0.001113,
+      "loss": 5.7116,
+      "step": 371
+    },
+    {
+      "epoch": 0.00372,
+      "grad_norm": 0.5998751494739524,
+      "learning_rate": 0.001116,
+      "loss": 5.7331,
+      "step": 372
+    },
+    {
+      "epoch": 0.00373,
+      "grad_norm": 0.5714053922949759,
+      "learning_rate": 0.001119,
+      "loss": 5.72,
+      "step": 373
+    },
+    {
+      "epoch": 0.00374,
+      "grad_norm": 0.628102864915935,
+      "learning_rate": 0.001122,
+      "loss": 5.7262,
+      "step": 374
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 0.5909991857975568,
+      "learning_rate": 0.0011250000000000001,
+      "loss": 5.7276,
+      "step": 375
+    },
+    {
+      "epoch": 0.00376,
+      "grad_norm": 0.5216592986380127,
+      "learning_rate": 0.001128,
+      "loss": 5.72,
+      "step": 376
+    },
+    {
+      "epoch": 0.00377,
+      "grad_norm": 0.5469626358369448,
+      "learning_rate": 0.001131,
+      "loss": 5.7145,
+      "step": 377
+    },
+    {
+      "epoch": 0.00378,
+      "grad_norm": 0.6291207985788981,
+      "learning_rate": 0.001134,
+      "loss": 5.7123,
+      "step": 378
+    },
+    {
+      "epoch": 0.00379,
+      "grad_norm": 0.6445779804368097,
+      "learning_rate": 0.001137,
+      "loss": 5.698,
+      "step": 379
+    },
+    {
+      "epoch": 0.0038,
+      "grad_norm": 0.7034810438425685,
+      "learning_rate": 0.00114,
+      "loss": 5.7177,
+      "step": 380
+    },
+    {
+      "epoch": 0.00381,
+      "grad_norm": 0.9330449756794958,
+      "learning_rate": 0.0011430000000000001,
+      "loss": 5.6924,
+      "step": 381
+    },
+    {
+      "epoch": 0.00382,
+      "grad_norm": 0.9449582673655755,
+      "learning_rate": 0.001146,
+      "loss": 5.6863,
+      "step": 382
+    },
+    {
+      "epoch": 0.00383,
+      "grad_norm": 0.6715143882341864,
+      "learning_rate": 0.001149,
+      "loss": 5.6846,
+      "step": 383
+    },
+    {
+      "epoch": 0.00384,
+      "grad_norm": 0.8662082921806001,
+      "learning_rate": 0.001152,
+      "loss": 5.699,
+      "step": 384
+    },
+    {
+      "epoch": 0.00385,
+      "grad_norm": 0.8482542645994808,
+      "learning_rate": 0.001155,
+      "loss": 5.6818,
+      "step": 385
+    },
+    {
+      "epoch": 0.00386,
+      "grad_norm": 0.8694120663581818,
+      "learning_rate": 0.001158,
+      "loss": 5.6914,
+      "step": 386
+    },
+    {
+      "epoch": 0.00387,
+      "grad_norm": 1.382803938865885,
+      "learning_rate": 0.0011610000000000001,
+      "loss": 5.6813,
+      "step": 387
+    },
+    {
+      "epoch": 0.00388,
+      "grad_norm": 0.9644228724095026,
+      "learning_rate": 0.001164,
+      "loss": 5.6689,
+      "step": 388
+    },
+    {
+      "epoch": 0.00389,
+      "grad_norm": 0.7396617808819147,
+      "learning_rate": 0.001167,
+      "loss": 5.6709,
+      "step": 389
+    },
+    {
+      "epoch": 0.0039,
+      "grad_norm": 0.5609185173455993,
+      "learning_rate": 0.00117,
+      "loss": 5.648,
+      "step": 390
+    },
+    {
+      "epoch": 0.00391,
+      "grad_norm": 0.735416794156057,
+      "learning_rate": 0.001173,
+      "loss": 5.6589,
+      "step": 391
+    },
+    {
+      "epoch": 0.00392,
+      "grad_norm": 0.7564046131732848,
+      "learning_rate": 0.001176,
+      "loss": 5.6662,
+      "step": 392
+    },
+    {
+      "epoch": 0.00393,
+      "grad_norm": 0.7233997204719024,
+      "learning_rate": 0.0011790000000000001,
+      "loss": 5.6377,
+      "step": 393
+    },
+    {
+      "epoch": 0.00394,
+      "grad_norm": 0.6288246845478384,
+      "learning_rate": 0.001182,
+      "loss": 5.6239,
+      "step": 394
+    },
+    {
+      "epoch": 0.00395,
+      "grad_norm": 0.6696319293361586,
+      "learning_rate": 0.001185,
+      "loss": 5.6102,
+      "step": 395
+    },
+    {
+      "epoch": 0.00396,
+      "grad_norm": 0.7214553632515296,
+      "learning_rate": 0.001188,
+      "loss": 5.6276,
+      "step": 396
+    },
+    {
+      "epoch": 0.00397,
+      "grad_norm": 0.7104651338358826,
+      "learning_rate": 0.001191,
+      "loss": 5.619,
+      "step": 397
+    },
+    {
+      "epoch": 0.00398,
+      "grad_norm": 0.5955487966253655,
+      "learning_rate": 0.0011940000000000002,
+      "loss": 5.6195,
+      "step": 398
+    },
+    {
+      "epoch": 0.00399,
+      "grad_norm": 0.6061151049974988,
+      "learning_rate": 0.0011970000000000001,
+      "loss": 5.6211,
+      "step": 399
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 0.747975845260549,
+      "learning_rate": 0.0012000000000000001,
+      "loss": 5.6119,
+      "step": 400
+    },
+    {
+      "epoch": 0.00401,
+      "grad_norm": 1.066378470212407,
+      "learning_rate": 0.001203,
+      "loss": 5.6055,
+      "step": 401
+    },
+    {
+      "epoch": 0.00402,
+      "grad_norm": 1.1521859370487337,
+      "learning_rate": 0.001206,
+      "loss": 5.6144,
+      "step": 402
+    },
+    {
+      "epoch": 0.00403,
+      "grad_norm": 1.0137206411467539,
+      "learning_rate": 0.001209,
+      "loss": 5.6121,
+      "step": 403
+    },
+    {
+      "epoch": 0.00404,
+      "grad_norm": 1.1663989264954397,
+      "learning_rate": 0.0012120000000000002,
+      "loss": 5.6177,
+      "step": 404
+    },
+    {
+      "epoch": 0.00405,
+      "grad_norm": 1.1363050593119097,
+      "learning_rate": 0.0012150000000000002,
+      "loss": 5.6098,
+      "step": 405
+    },
+    {
+      "epoch": 0.00406,
+      "grad_norm": 1.0087910495777368,
+      "learning_rate": 0.0012180000000000001,
+      "loss": 5.5952,
+      "step": 406
+    },
+    {
+      "epoch": 0.00407,
+      "grad_norm": 0.9520738669869687,
+      "learning_rate": 0.0012209999999999999,
+      "loss": 5.5915,
+      "step": 407
+    },
+    {
+      "epoch": 0.00408,
+      "grad_norm": 1.0377478362348644,
+      "learning_rate": 0.001224,
+      "loss": 5.6129,
+      "step": 408
+    },
+    {
+      "epoch": 0.00409,
+      "grad_norm": 1.1016437762309168,
+      "learning_rate": 0.001227,
+      "loss": 5.5927,
+      "step": 409
+    },
+    {
+      "epoch": 0.0041,
+      "grad_norm": 0.9709841842795465,
+      "learning_rate": 0.00123,
+      "loss": 5.5883,
+      "step": 410
+    },
+    {
+      "epoch": 0.00411,
+      "grad_norm": 0.8240534275873757,
+      "learning_rate": 0.001233,
+      "loss": 5.5977,
+      "step": 411
+    },
+    {
+      "epoch": 0.00412,
+      "grad_norm": 0.8077831494606339,
+      "learning_rate": 0.001236,
+      "loss": 5.5894,
+      "step": 412
+    },
+    {
+      "epoch": 0.00413,
+      "grad_norm": 0.7433003641205871,
+      "learning_rate": 0.0012389999999999999,
+      "loss": 5.5815,
+      "step": 413
+    },
+    {
+      "epoch": 0.00414,
+      "grad_norm": 0.615571577509929,
+      "learning_rate": 0.001242,
+      "loss": 5.5582,
+      "step": 414
+    },
+    {
+      "epoch": 0.00415,
+      "grad_norm": 0.656841702591939,
+      "learning_rate": 0.001245,
+      "loss": 5.5606,
+      "step": 415
+    },
+    {
+      "epoch": 0.00416,
+      "grad_norm": 0.613196865035469,
+      "learning_rate": 0.001248,
+      "loss": 5.5418,
+      "step": 416
+    },
+    {
+      "epoch": 0.00417,
+      "grad_norm": 0.48631699887544017,
+      "learning_rate": 0.001251,
+      "loss": 5.5683,
+      "step": 417
+    },
+    {
+      "epoch": 0.00418,
+      "grad_norm": 0.4611320175617472,
+      "learning_rate": 0.001254,
+      "loss": 5.5401,
+      "step": 418
+    },
+    {
+      "epoch": 0.00419,
+      "grad_norm": 0.5824396705507784,
+      "learning_rate": 0.0012569999999999999,
+      "loss": 5.5305,
+      "step": 419
+    },
+    {
+      "epoch": 0.0042,
+      "grad_norm": 0.8055087238093066,
+      "learning_rate": 0.00126,
+      "loss": 5.5467,
+      "step": 420
+    },
+    {
+      "epoch": 0.00421,
+      "grad_norm": 0.9831917562206494,
+      "learning_rate": 0.001263,
+      "loss": 5.5503,
+      "step": 421
+    },
+    {
+      "epoch": 0.00422,
+      "grad_norm": 0.8124592037579013,
+      "learning_rate": 0.001266,
+      "loss": 5.5357,
+      "step": 422
+    },
+    {
+      "epoch": 0.00423,
+      "grad_norm": 1.0116811247233115,
+      "learning_rate": 0.001269,
+      "loss": 5.5337,
+      "step": 423
+    },
+    {
+      "epoch": 0.00424,
+      "grad_norm": 1.368200107021336,
+      "learning_rate": 0.001272,
+      "loss": 5.5595,
+      "step": 424
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 0.9353292964204464,
+      "learning_rate": 0.001275,
+      "loss": 5.5445,
+      "step": 425
+    },
+    {
+      "epoch": 0.00426,
+      "grad_norm": 0.9604600896045714,
+      "learning_rate": 0.001278,
+      "loss": 5.5493,
+      "step": 426
+    },
+    {
+      "epoch": 0.00427,
+      "grad_norm": 0.9684632680550875,
+      "learning_rate": 0.001281,
+      "loss": 5.5328,
+      "step": 427
+    },
+    {
+      "epoch": 0.00428,
+      "grad_norm": 0.9533788870858234,
+      "learning_rate": 0.001284,
+      "loss": 5.5058,
+      "step": 428
+    },
+    {
+      "epoch": 0.00429,
+      "grad_norm": 1.113271451738206,
+      "learning_rate": 0.001287,
+      "loss": 5.5322,
+      "step": 429
+    },
+    {
+      "epoch": 0.0043,
+      "grad_norm": 0.9492520989573756,
+      "learning_rate": 0.00129,
+      "loss": 5.5226,
+      "step": 430
+    },
+    {
+      "epoch": 0.00431,
+      "grad_norm": 1.1309185966069144,
+      "learning_rate": 0.001293,
+      "loss": 5.5118,
+      "step": 431
+    },
+    {
+      "epoch": 0.00432,
+      "grad_norm": 1.2851571092183034,
+      "learning_rate": 0.001296,
+      "loss": 5.5367,
+      "step": 432
+    },
+    {
+      "epoch": 0.00433,
+      "grad_norm": 0.7219415163940603,
+      "learning_rate": 0.001299,
+      "loss": 5.5106,
+      "step": 433
+    },
+    {
+      "epoch": 0.00434,
+      "grad_norm": 0.7943786167833424,
+      "learning_rate": 0.001302,
+      "loss": 5.5109,
+      "step": 434
+    },
+    {
+      "epoch": 0.00435,
+      "grad_norm": 0.7514833526497924,
+      "learning_rate": 0.001305,
+      "loss": 5.5051,
+      "step": 435
+    },
+    {
+      "epoch": 0.00436,
+      "grad_norm": 0.5499552719714149,
+      "learning_rate": 0.001308,
+      "loss": 5.4758,
+      "step": 436
+    },
+    {
+      "epoch": 0.00437,
+      "grad_norm": 0.5803132978630802,
+      "learning_rate": 0.001311,
+      "loss": 5.494,
+      "step": 437
+    },
+    {
+      "epoch": 0.00438,
+      "grad_norm": 0.4343754614294673,
+      "learning_rate": 0.001314,
+      "loss": 5.4701,
+      "step": 438
+    },
+    {
+      "epoch": 0.00439,
+      "grad_norm": 0.49450539118744047,
+      "learning_rate": 0.001317,
+      "loss": 5.4997,
+      "step": 439
+    },
+    {
+      "epoch": 0.0044,
+      "grad_norm": 0.3670496641121159,
+      "learning_rate": 0.00132,
+      "loss": 5.4633,
+      "step": 440
+    },
+    {
+      "epoch": 0.00441,
+      "grad_norm": 0.44023069828614625,
+      "learning_rate": 0.001323,
+      "loss": 5.4624,
+      "step": 441
+    },
+    {
+      "epoch": 0.00442,
+      "grad_norm": 0.45124022563783683,
+      "learning_rate": 0.0013260000000000001,
+      "loss": 5.4541,
+      "step": 442
+    },
+    {
+      "epoch": 0.00443,
+      "grad_norm": 0.5170657075032803,
+      "learning_rate": 0.001329,
+      "loss": 5.4572,
+      "step": 443
+    },
+    {
+      "epoch": 0.00444,
+      "grad_norm": 0.6406930762525409,
+      "learning_rate": 0.001332,
+      "loss": 5.4608,
+      "step": 444
+    },
+    {
+      "epoch": 0.00445,
+      "grad_norm": 0.7431205120059805,
+      "learning_rate": 0.001335,
+      "loss": 5.4462,
+      "step": 445
+    },
+    {
+      "epoch": 0.00446,
+      "grad_norm": 0.9849435734941417,
+      "learning_rate": 0.001338,
+      "loss": 5.4578,
+      "step": 446
+    },
+    {
+      "epoch": 0.00447,
+      "grad_norm": 1.1031854356255555,
+      "learning_rate": 0.001341,
+      "loss": 5.469,
+      "step": 447
+    },
+    {
+      "epoch": 0.00448,
+      "grad_norm": 0.8550264287932019,
+      "learning_rate": 0.0013440000000000001,
+      "loss": 5.4411,
+      "step": 448
+    },
+    {
+      "epoch": 0.00449,
+      "grad_norm": 0.8569858280927948,
+      "learning_rate": 0.001347,
+      "loss": 5.4491,
+      "step": 449
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 0.766625402040655,
+      "learning_rate": 0.00135,
+      "loss": 5.4461,
+      "step": 450
+    },
+    {
+      "epoch": 0.00451,
+      "grad_norm": 0.8992860281272513,
+      "learning_rate": 0.001353,
+      "loss": 5.4411,
+      "step": 451
+    },
+    {
+      "epoch": 0.00452,
+      "grad_norm": 0.9587820940422671,
+      "learning_rate": 0.001356,
+      "loss": 5.4273,
+      "step": 452
+    },
+    {
+      "epoch": 0.00453,
+      "grad_norm": 0.9706881368414635,
+      "learning_rate": 0.001359,
+      "loss": 5.4204,
+      "step": 453
+    },
+    {
+      "epoch": 0.00454,
+      "grad_norm": 1.1942708782025127,
+      "learning_rate": 0.0013620000000000001,
+      "loss": 5.4481,
+      "step": 454
+    },
+    {
+      "epoch": 0.00455,
+      "grad_norm": 1.0651260215062406,
+      "learning_rate": 0.0013650000000000001,
+      "loss": 5.4483,
+      "step": 455
+    },
+    {
+      "epoch": 0.00456,
+      "grad_norm": 1.0894309440916736,
+      "learning_rate": 0.001368,
+      "loss": 5.4405,
+      "step": 456
+    },
+    {
+      "epoch": 0.00457,
+      "grad_norm": 1.3632436316792311,
+      "learning_rate": 0.001371,
+      "loss": 5.442,
+      "step": 457
+    },
+    {
+      "epoch": 0.00458,
+      "grad_norm": 0.9802910013598249,
+      "learning_rate": 0.001374,
+      "loss": 5.4423,
+      "step": 458
+    },
+    {
+      "epoch": 0.00459,
+      "grad_norm": 0.9681577835390196,
+      "learning_rate": 0.0013770000000000002,
+      "loss": 5.4314,
+      "step": 459
+    },
+    {
+      "epoch": 0.0046,
+      "grad_norm": 0.955881976726698,
+      "learning_rate": 0.0013800000000000002,
+      "loss": 5.4146,
+      "step": 460
+    },
+    {
+      "epoch": 0.00461,
+      "grad_norm": 1.072138092727722,
+      "learning_rate": 0.0013830000000000001,
+      "loss": 5.4186,
+      "step": 461
+    },
+    {
+      "epoch": 0.00462,
+      "grad_norm": 0.8672451154734739,
+      "learning_rate": 0.001386,
+      "loss": 5.4193,
+      "step": 462
+    },
+    {
+      "epoch": 0.00463,
+      "grad_norm": 0.9908998654710571,
+      "learning_rate": 0.001389,
+      "loss": 5.4011,
+      "step": 463
+    },
+    {
+      "epoch": 0.00464,
+      "grad_norm": 1.1599842863140275,
+      "learning_rate": 0.001392,
+      "loss": 5.4321,
+      "step": 464
+    },
+    {
+      "epoch": 0.00465,
+      "grad_norm": 0.698212811337685,
+      "learning_rate": 0.0013950000000000002,
+      "loss": 5.3861,
+      "step": 465
+    },
+    {
+      "epoch": 0.00466,
+      "grad_norm": 0.6209828526031017,
+      "learning_rate": 0.0013980000000000002,
+      "loss": 5.4004,
+      "step": 466
+    },
+    {
+      "epoch": 0.00467,
+      "grad_norm": 0.626908730655191,
+      "learning_rate": 0.0014010000000000001,
+      "loss": 5.3894,
+      "step": 467
+    },
+    {
+      "epoch": 0.00468,
+      "grad_norm": 0.6046317726003113,
+      "learning_rate": 0.001404,
+      "loss": 5.3787,
+      "step": 468
+    },
+    {
+      "epoch": 0.00469,
+      "grad_norm": 0.5283854127913247,
+      "learning_rate": 0.001407,
+      "loss": 5.3767,
+      "step": 469
+    },
+    {
+      "epoch": 0.0047,
+      "grad_norm": 0.5494342704278756,
+      "learning_rate": 0.00141,
+      "loss": 5.3859,
+      "step": 470
+    },
+    {
+      "epoch": 0.00471,
+      "grad_norm": 0.6548172150181142,
+      "learning_rate": 0.001413,
+      "loss": 5.3653,
+      "step": 471
+    },
+    {
+      "epoch": 0.00472,
+      "grad_norm": 0.8352552132751145,
+      "learning_rate": 0.001416,
+      "loss": 5.3799,
+      "step": 472
+    },
+    {
+      "epoch": 0.00473,
+      "grad_norm": 1.04775859687285,
+      "learning_rate": 0.001419,
+      "loss": 5.3985,
+      "step": 473
+    },
+    {
+      "epoch": 0.00474,
+      "grad_norm": 0.9808295146566943,
+      "learning_rate": 0.0014219999999999999,
+      "loss": 5.3787,
+      "step": 474
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 1.0065096729700158,
+      "learning_rate": 0.001425,
+      "loss": 5.3725,
+      "step": 475
+    },
+    {
+      "epoch": 0.00476,
+      "grad_norm": 1.0604383614131039,
+      "learning_rate": 0.001428,
+      "loss": 5.4172,
+      "step": 476
+    },
+    {
+      "epoch": 0.00477,
+      "grad_norm": 0.5551620314639825,
+      "learning_rate": 0.001431,
+      "loss": 5.3484,
+      "step": 477
+    },
+    {
+      "epoch": 0.00478,
+      "grad_norm": 0.6446758952809472,
+      "learning_rate": 0.001434,
+      "loss": 5.3819,
+      "step": 478
+    },
+    {
+      "epoch": 0.00479,
+      "grad_norm": 0.6811842486215005,
+      "learning_rate": 0.001437,
+      "loss": 5.3551,
+      "step": 479
+    },
+    {
+      "epoch": 0.0048,
+      "grad_norm": 0.6985415352827566,
+      "learning_rate": 0.0014399999999999999,
+      "loss": 5.3502,
+      "step": 480
+    },
+    {
+      "epoch": 0.00481,
+      "grad_norm": 0.7430626707063542,
+      "learning_rate": 0.001443,
+      "loss": 5.3537,
+      "step": 481
+    },
+    {
+      "epoch": 0.00482,
+      "grad_norm": 0.7883555304691023,
+      "learning_rate": 0.001446,
+      "loss": 5.3281,
+      "step": 482
+    },
+    {
+      "epoch": 0.00483,
+      "grad_norm": 0.9604272864715826,
+      "learning_rate": 0.001449,
+      "loss": 5.3455,
+      "step": 483
+    },
+    {
+      "epoch": 0.00484,
+      "grad_norm": 0.9435245256880889,
+      "learning_rate": 0.001452,
+      "loss": 5.3494,
+      "step": 484
+    },
+    {
+      "epoch": 0.00485,
+      "grad_norm": 0.8042883678590446,
+      "learning_rate": 0.001455,
+      "loss": 5.3511,
+      "step": 485
+    },
+    {
+      "epoch": 0.00486,
+      "grad_norm": 0.7958734942371152,
+      "learning_rate": 0.001458,
+      "loss": 5.3247,
+      "step": 486
+    },
+    {
+      "epoch": 0.00487,
+      "grad_norm": 0.7735191187819994,
+      "learning_rate": 0.001461,
+      "loss": 5.3397,
+      "step": 487
+    },
+    {
+      "epoch": 0.00488,
+      "grad_norm": 1.012081783792506,
+      "learning_rate": 0.001464,
+      "loss": 5.3265,
+      "step": 488
+    },
+    {
+      "epoch": 0.00489,
+      "grad_norm": 0.8905045147335564,
+      "learning_rate": 0.001467,
+      "loss": 5.3131,
+      "step": 489
+    },
+    {
+      "epoch": 0.0049,
+      "grad_norm": 0.8898357576674187,
+      "learning_rate": 0.00147,
+      "loss": 5.3259,
+      "step": 490
+    },
+    {
+      "epoch": 0.00491,
+      "grad_norm": 0.9415043482581119,
+      "learning_rate": 0.001473,
+      "loss": 5.3269,
+      "step": 491
+    },
+    {
+      "epoch": 0.00492,
+      "grad_norm": 0.8320755719970162,
+      "learning_rate": 0.001476,
+      "loss": 5.3196,
+      "step": 492
+    },
+    {
+      "epoch": 0.00493,
+      "grad_norm": 0.8445852574519659,
+      "learning_rate": 0.001479,
+      "loss": 5.3145,
+      "step": 493
+    },
+    {
+      "epoch": 0.00494,
+      "grad_norm": 1.055556747775556,
+      "learning_rate": 0.001482,
+      "loss": 5.3274,
+      "step": 494
+    },
+    {
+      "epoch": 0.00495,
+      "grad_norm": 0.8469107680360277,
+      "learning_rate": 0.001485,
+      "loss": 5.3268,
+      "step": 495
+    },
+    {
+      "epoch": 0.00496,
+      "grad_norm": 0.7635320378283988,
+      "learning_rate": 0.001488,
+      "loss": 5.3083,
+      "step": 496
+    },
+    {
+      "epoch": 0.00497,
+      "grad_norm": 0.8508130432624411,
+      "learning_rate": 0.001491,
+      "loss": 5.2927,
+      "step": 497
+    },
+    {
+      "epoch": 0.00498,
+      "grad_norm": 0.6847944056515368,
+      "learning_rate": 0.001494,
+      "loss": 5.3134,
+      "step": 498
+    },
+    {
+      "epoch": 0.00499,
+      "grad_norm": 0.73441383592788,
+      "learning_rate": 0.001497,
+      "loss": 5.2894,
+      "step": 499
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 0.6599139353534871,
+      "learning_rate": 0.0015,
+      "loss": 5.2898,
+      "step": 500
+    },
+    {
+      "epoch": 0.00501,
+      "grad_norm": 0.5852764030687648,
+      "learning_rate": 0.001503,
+      "loss": 5.2881,
+      "step": 501
+    },
+    {
+      "epoch": 0.00502,
+      "grad_norm": 0.6853021422001514,
+      "learning_rate": 0.001506,
+      "loss": 5.2946,
+      "step": 502
+    },
+    {
+      "epoch": 0.00503,
+      "grad_norm": 0.720141729806561,
+      "learning_rate": 0.0015090000000000001,
+      "loss": 5.2988,
+      "step": 503
+    },
+    {
+      "epoch": 0.00504,
+      "grad_norm": 0.7571759373525393,
+      "learning_rate": 0.001512,
+      "loss": 5.2829,
+      "step": 504
+    },
+    {
+      "epoch": 0.00505,
+      "grad_norm": 0.7892633543044816,
+      "learning_rate": 0.001515,
+      "loss": 5.2795,
+      "step": 505
+    },
+    {
+      "epoch": 0.00506,
+      "grad_norm": 0.7628738296895545,
+      "learning_rate": 0.001518,
+      "loss": 5.2778,
+      "step": 506
+    },
+    {
+      "epoch": 0.00507,
+      "grad_norm": 0.7334519380566521,
+      "learning_rate": 0.001521,
+      "loss": 5.2722,
+      "step": 507
+    },
+    {
+      "epoch": 0.00508,
+      "grad_norm": 0.7350882061617702,
+      "learning_rate": 0.001524,
+      "loss": 5.2693,
+      "step": 508
+    },
+    {
+      "epoch": 0.00509,
+      "grad_norm": 0.7407389039658853,
+      "learning_rate": 0.0015270000000000001,
+      "loss": 5.2562,
+      "step": 509
+    },
+    {
+      "epoch": 0.0051,
+      "grad_norm": 0.7104400166661097,
+      "learning_rate": 0.0015300000000000001,
+      "loss": 5.2696,
+      "step": 510
+    },
+    {
+      "epoch": 0.00511,
+      "grad_norm": 0.8484400275698938,
+      "learning_rate": 0.001533,
+      "loss": 5.2541,
+      "step": 511
+    },
+    {
+      "epoch": 0.00512,
+      "grad_norm": 0.7909533536283434,
+      "learning_rate": 0.001536,
+      "loss": 5.2475,
+      "step": 512
+    },
+    {
+      "epoch": 0.00513,
+      "grad_norm": 0.9481517521047255,
+      "learning_rate": 0.001539,
+      "loss": 5.2684,
+      "step": 513
+    },
+    {
+      "epoch": 0.00514,
+      "grad_norm": 1.405665610624422,
+      "learning_rate": 0.001542,
+      "loss": 5.2742,
+      "step": 514
+    },
+    {
+      "epoch": 0.00515,
+      "grad_norm": 0.9237050984729354,
+      "learning_rate": 0.0015450000000000001,
+      "loss": 5.2636,
+      "step": 515
+    },
+    {
+      "epoch": 0.00516,
+      "grad_norm": 1.1634719669204763,
+      "learning_rate": 0.0015480000000000001,
+      "loss": 5.2755,
+      "step": 516
+    },
+    {
+      "epoch": 0.00517,
+      "grad_norm": 0.8585760471561034,
+      "learning_rate": 0.001551,
+      "loss": 5.2577,
+      "step": 517
+    },
+    {
+      "epoch": 0.00518,
+      "grad_norm": 0.728905665920263,
+      "learning_rate": 0.001554,
+      "loss": 5.2422,
+      "step": 518
+    },
+    {
+      "epoch": 0.00519,
+      "grad_norm": 0.841003643450708,
+      "learning_rate": 0.001557,
+      "loss": 5.2472,
+      "step": 519
+    },
+    {
+      "epoch": 0.0052,
+      "grad_norm": 0.8542329179478354,
+      "learning_rate": 0.0015600000000000002,
+      "loss": 5.2573,
+      "step": 520
+    },
+    {
+      "epoch": 0.00521,
+      "grad_norm": 0.9136344705581151,
+      "learning_rate": 0.0015630000000000002,
+      "loss": 5.2327,
+      "step": 521
+    },
+    {
+      "epoch": 0.00522,
+      "grad_norm": 1.1142541660381047,
+      "learning_rate": 0.0015660000000000001,
+      "loss": 5.2435,
+      "step": 522
+    },
+    {
+      "epoch": 0.00523,
+      "grad_norm": 1.2424782927197504,
+      "learning_rate": 0.001569,
+      "loss": 5.2642,
+      "step": 523
+    },
+    {
+      "epoch": 0.00524,
+      "grad_norm": 0.9564855927553343,
+      "learning_rate": 0.001572,
+      "loss": 5.264,
+      "step": 524
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 0.8662515649518988,
+      "learning_rate": 0.001575,
+      "loss": 5.2537,
+      "step": 525
+    },
+    {
+      "epoch": 0.00526,
+      "grad_norm": 0.72079723918141,
+      "learning_rate": 0.0015780000000000002,
+      "loss": 5.2224,
+      "step": 526
+    },
+    {
+      "epoch": 0.00527,
+      "grad_norm": 0.6822789926865414,
+      "learning_rate": 0.0015810000000000002,
+      "loss": 5.2353,
+      "step": 527
+    },
+    {
+      "epoch": 0.00528,
+      "grad_norm": 0.6986034229049856,
+      "learning_rate": 0.0015840000000000001,
+      "loss": 5.2128,
+      "step": 528
+    },
+    {
+      "epoch": 0.00529,
+      "grad_norm": 0.6935526974121139,
+      "learning_rate": 0.001587,
+      "loss": 5.2236,
+      "step": 529
+    },
+    {
+      "epoch": 0.0053,
+      "grad_norm": 0.6635846542220551,
+      "learning_rate": 0.00159,
+      "loss": 5.2302,
+      "step": 530
+    },
+    {
+      "epoch": 0.00531,
+      "grad_norm": 0.7197123783695031,
+      "learning_rate": 0.001593,
+      "loss": 5.2167,
+      "step": 531
+    },
+    {
+      "epoch": 0.00532,
+      "grad_norm": 0.6862921713046577,
+      "learning_rate": 0.0015960000000000002,
+      "loss": 5.2205,
+      "step": 532
+    },
+    {
+      "epoch": 0.00533,
+      "grad_norm": 0.672254585940336,
+      "learning_rate": 0.0015990000000000002,
+      "loss": 5.2165,
+      "step": 533
+    },
+    {
+      "epoch": 0.00534,
+      "grad_norm": 0.6180247915886188,
+      "learning_rate": 0.0016020000000000001,
+      "loss": 5.2018,
+      "step": 534
+    },
+    {
+      "epoch": 0.00535,
+      "grad_norm": 0.7076887136252737,
+      "learning_rate": 0.001605,
+      "loss": 5.2099,
+      "step": 535
+    },
+    {
+      "epoch": 0.00536,
+      "grad_norm": 0.8627381010586813,
+      "learning_rate": 0.001608,
+      "loss": 5.2158,
+      "step": 536
+    },
+    {
+      "epoch": 0.00537,
+      "grad_norm": 0.9890089503230703,
+      "learning_rate": 0.0016110000000000002,
+      "loss": 5.2125,
+      "step": 537
+    },
+    {
+      "epoch": 0.00538,
+      "grad_norm": 1.0890684861329838,
+      "learning_rate": 0.0016140000000000002,
+      "loss": 5.1997,
+      "step": 538
+    },
+    {
+      "epoch": 0.00539,
+      "grad_norm": 0.7898695514456295,
+      "learning_rate": 0.0016170000000000002,
+      "loss": 5.1885,
+      "step": 539
+    },
+    {
+      "epoch": 0.0054,
+      "grad_norm": 0.7877858658336557,
+      "learning_rate": 0.0016200000000000001,
+      "loss": 5.1979,
+      "step": 540
+    },
+    {
+      "epoch": 0.00541,
+      "grad_norm": 0.8613625325852945,
+      "learning_rate": 0.001623,
+      "loss": 5.1884,
+      "step": 541
+    },
+    {
+      "epoch": 0.00542,
+      "grad_norm": 0.9310959295325639,
+      "learning_rate": 0.001626,
+      "loss": 5.2135,
+      "step": 542
+    },
+    {
+      "epoch": 0.00543,
+      "grad_norm": 1.0650490028229627,
+      "learning_rate": 0.0016290000000000002,
+      "loss": 5.1821,
+      "step": 543
+    },
+    {
+      "epoch": 0.00544,
+      "grad_norm": 1.058155288535174,
+      "learning_rate": 0.0016320000000000002,
+      "loss": 5.2106,
+      "step": 544
+    },
+    {
+      "epoch": 0.00545,
+      "grad_norm": 0.7888748411290551,
+      "learning_rate": 0.0016350000000000002,
+      "loss": 5.2093,
+      "step": 545
+    },
+    {
+      "epoch": 0.00546,
+      "grad_norm": 0.8327623405474364,
+      "learning_rate": 0.0016380000000000001,
+      "loss": 5.1821,
+      "step": 546
+    },
+    {
+      "epoch": 0.00547,
+      "grad_norm": 0.9359889846839972,
+      "learning_rate": 0.001641,
+      "loss": 5.1867,
+      "step": 547
+    },
+    {
+      "epoch": 0.00548,
+      "grad_norm": 0.8250520922796094,
+      "learning_rate": 0.001644,
+      "loss": 5.1781,
+      "step": 548
+    },
+    {
+      "epoch": 0.00549,
+      "grad_norm": 0.8346002681162655,
+      "learning_rate": 0.0016470000000000002,
+      "loss": 5.1786,
+      "step": 549
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 0.8307775357404767,
+      "learning_rate": 0.0016500000000000002,
+      "loss": 5.1901,
+      "step": 550
+    },
+    {
+      "epoch": 0.00551,
+      "grad_norm": 0.8868587049067305,
+      "learning_rate": 0.0016530000000000002,
+      "loss": 5.1882,
+      "step": 551
+    },
+    {
+      "epoch": 0.00552,
+      "grad_norm": 1.0021721179116465,
+      "learning_rate": 0.0016560000000000001,
+      "loss": 5.1887,
+      "step": 552
+    },
+    {
+      "epoch": 0.00553,
+      "grad_norm": 1.0435640829012027,
+      "learning_rate": 0.001659,
+      "loss": 5.1828,
+      "step": 553
+    },
+    {
+      "epoch": 0.00554,
+      "grad_norm": 1.091952614784726,
+      "learning_rate": 0.0016620000000000003,
+      "loss": 5.1843,
+      "step": 554
+    },
+    {
+      "epoch": 0.00555,
+      "grad_norm": 0.7530351004511261,
+      "learning_rate": 0.0016650000000000002,
+      "loss": 5.1804,
+      "step": 555
+    },
+    {
+      "epoch": 0.00556,
+      "grad_norm": 0.7840106393058217,
+      "learning_rate": 0.0016680000000000002,
+      "loss": 5.1664,
+      "step": 556
+    },
+    {
+      "epoch": 0.00557,
+      "grad_norm": 0.6999816254444311,
+      "learning_rate": 0.0016710000000000002,
+      "loss": 5.1437,
+      "step": 557
+    },
+    {
+      "epoch": 0.00558,
+      "grad_norm": 0.7845980119871422,
+      "learning_rate": 0.0016740000000000001,
+      "loss": 5.1792,
+      "step": 558
+    },
+    {
+      "epoch": 0.00559,
+      "grad_norm": 1.002237738003299,
+      "learning_rate": 0.001677,
+      "loss": 5.165,
+      "step": 559
+    },
+    {
+      "epoch": 0.0056,
+      "grad_norm": 1.1590360602458978,
+      "learning_rate": 0.0016800000000000003,
+      "loss": 5.15,
+      "step": 560
+    },
+    {
+      "epoch": 0.00561,
+      "grad_norm": 0.9693350319936842,
+      "learning_rate": 0.0016830000000000003,
+      "loss": 5.1673,
+      "step": 561
+    },
+    {
+      "epoch": 0.00562,
+      "grad_norm": 0.9337806686381701,
+      "learning_rate": 0.0016860000000000002,
+      "loss": 5.1656,
+      "step": 562
+    },
+    {
+      "epoch": 0.00563,
+      "grad_norm": 0.7021371152666548,
+      "learning_rate": 0.001689,
+      "loss": 5.1554,
+      "step": 563
+    },
+    {
+      "epoch": 0.00564,
+      "grad_norm": 0.761762323285238,
+      "learning_rate": 0.001692,
+      "loss": 5.1364,
+      "step": 564
+    },
+    {
+      "epoch": 0.00565,
+      "grad_norm": 0.9962272803832337,
+      "learning_rate": 0.001695,
+      "loss": 5.1465,
+      "step": 565
+    },
+    {
+      "epoch": 0.00566,
+      "grad_norm": 0.9270663941212276,
+      "learning_rate": 0.0016979999999999999,
+      "loss": 5.1454,
+      "step": 566
+    },
+    {
+      "epoch": 0.00567,
+      "grad_norm": 1.0296751705367089,
+      "learning_rate": 0.0017009999999999998,
+      "loss": 5.1403,
+      "step": 567
+    },
+    {
+      "epoch": 0.00568,
+      "grad_norm": 1.0921308967550072,
+      "learning_rate": 0.0017039999999999998,
+      "loss": 5.1647,
+      "step": 568
+    },
+    {
+      "epoch": 0.00569,
+      "grad_norm": 0.9680118526613578,
+      "learning_rate": 0.001707,
+      "loss": 5.1376,
+      "step": 569
+    },
+    {
+      "epoch": 0.0057,
+      "grad_norm": 1.1373894554942883,
+      "learning_rate": 0.00171,
+      "loss": 5.1667,
+      "step": 570
+    },
+    {
+      "epoch": 0.00571,
+      "grad_norm": 0.9491793536820188,
+      "learning_rate": 0.001713,
+      "loss": 5.1618,
+      "step": 571
+    },
+    {
+      "epoch": 0.00572,
+      "grad_norm": 1.0938883440367575,
+      "learning_rate": 0.0017159999999999999,
+      "loss": 5.1499,
+      "step": 572
+    },
+    {
+      "epoch": 0.00573,
+      "grad_norm": 0.8953824806877001,
+      "learning_rate": 0.0017189999999999998,
+      "loss": 5.1459,
+      "step": 573
+    },
+    {
+      "epoch": 0.00574,
+      "grad_norm": 0.8777825611555061,
+      "learning_rate": 0.001722,
+      "loss": 5.1467,
+      "step": 574
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 0.9427806830230203,
+      "learning_rate": 0.001725,
+      "loss": 5.132,
+      "step": 575
+    },
+    {
+      "epoch": 0.00576,
+      "grad_norm": 0.8846520959631657,
+      "learning_rate": 0.001728,
+      "loss": 5.1269,
+      "step": 576
+    },
+    {
+      "epoch": 0.00577,
+      "grad_norm": 0.6344552018416748,
+      "learning_rate": 0.001731,
+      "loss": 5.1395,
+      "step": 577
+    },
+    {
+      "epoch": 0.00578,
+      "grad_norm": 0.6771922101340231,
+      "learning_rate": 0.0017339999999999999,
+      "loss": 5.1199,
+      "step": 578
+    },
+    {
+      "epoch": 0.00579,
+      "grad_norm": 0.5381676919488529,
+      "learning_rate": 0.0017369999999999998,
+      "loss": 5.1001,
+      "step": 579
+    },
+    {
+      "epoch": 0.0058,
+      "grad_norm": 0.5416857271346956,
+      "learning_rate": 0.00174,
+      "loss": 5.1179,
+      "step": 580
+    },
+    {
+      "epoch": 0.00581,
+      "grad_norm": 0.4886409554358658,
+      "learning_rate": 0.001743,
+      "loss": 5.1131,
+      "step": 581
+    },
+    {
+      "epoch": 0.00582,
+      "grad_norm": 0.3953796837190132,
+      "learning_rate": 0.001746,
+      "loss": 5.0842,
+      "step": 582
+    },
+    {
+      "epoch": 0.00583,
+      "grad_norm": 0.3816231764982968,
+      "learning_rate": 0.001749,
+      "loss": 5.1076,
+      "step": 583
+    },
+    {
+      "epoch": 0.00584,
+      "grad_norm": 0.39062356918599367,
+      "learning_rate": 0.0017519999999999999,
+      "loss": 5.1003,
+      "step": 584
+    },
+    {
+      "epoch": 0.00585,
+      "grad_norm": 0.3915858898605007,
+      "learning_rate": 0.0017549999999999998,
+      "loss": 5.0837,
+      "step": 585
+    },
+    {
+      "epoch": 0.00586,
+      "grad_norm": 0.4063811781565053,
+      "learning_rate": 0.001758,
+      "loss": 5.0866,
+      "step": 586
+    },
+    {
+      "epoch": 0.00587,
+      "grad_norm": 0.4274699383917251,
+      "learning_rate": 0.001761,
+      "loss": 5.0709,
+      "step": 587
+    },
+    {
+      "epoch": 0.00588,
+      "grad_norm": 0.5008029306674315,
+      "learning_rate": 0.001764,
+      "loss": 5.0767,
+      "step": 588
+    },
+    {
+      "epoch": 0.00589,
+      "grad_norm": 0.7229655060897183,
+      "learning_rate": 0.001767,
+      "loss": 5.084,
+      "step": 589
+    },
+    {
+      "epoch": 0.0059,
+      "grad_norm": 0.9281114393382421,
+      "learning_rate": 0.0017699999999999999,
+      "loss": 5.0779,
+      "step": 590
+    },
+    {
+      "epoch": 0.00591,
+      "grad_norm": 0.9689787780419555,
+      "learning_rate": 0.001773,
+      "loss": 5.0756,
+      "step": 591
+    },
+    {
+      "epoch": 0.00592,
+      "grad_norm": 0.8435580831880815,
+      "learning_rate": 0.001776,
+      "loss": 5.0975,
+      "step": 592
+    },
+    {
+      "epoch": 0.00593,
+      "grad_norm": 0.9986679613531879,
+      "learning_rate": 0.001779,
+      "loss": 5.098,
+      "step": 593
+    },
+    {
+      "epoch": 0.00594,
+      "grad_norm": 1.0996840470415932,
+      "learning_rate": 0.001782,
+      "loss": 5.1014,
+      "step": 594
+    },
+    {
+      "epoch": 0.00595,
+      "grad_norm": 0.9601369241773534,
+      "learning_rate": 0.001785,
+      "loss": 5.0957,
+      "step": 595
+    },
+    {
+      "epoch": 0.00596,
+      "grad_norm": 0.8606352953891906,
+      "learning_rate": 0.0017879999999999999,
+      "loss": 5.0932,
+      "step": 596
+    },
+    {
+      "epoch": 0.00597,
+      "grad_norm": 0.8716680293105395,
+      "learning_rate": 0.001791,
+      "loss": 5.066,
+      "step": 597
+    },
+    {
+      "epoch": 0.00598,
+      "grad_norm": 0.9954805607751368,
+      "learning_rate": 0.001794,
+      "loss": 5.0837,
+      "step": 598
+    },
+    {
+      "epoch": 0.00599,
+      "grad_norm": 1.0139742121604893,
+      "learning_rate": 0.001797,
+      "loss": 5.0892,
+      "step": 599
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 0.8994983668637855,
+      "learning_rate": 0.0018,
+      "loss": 5.083,
+      "step": 600
+    },
+    {
+      "epoch": 0.00601,
+      "grad_norm": 1.012817382509115,
+      "learning_rate": 0.001803,
+      "loss": 5.0872,
+      "step": 601
+    },
+    {
+      "epoch": 0.00602,
+      "grad_norm": 0.829792766990954,
+      "learning_rate": 0.0018059999999999999,
+      "loss": 5.0655,
+      "step": 602
+    },
+    {
+      "epoch": 0.00603,
+      "grad_norm": 0.8978641621726422,
+      "learning_rate": 0.001809,
+      "loss": 5.0896,
+      "step": 603
+    },
+    {
+      "epoch": 0.00604,
+      "grad_norm": 1.006526980400353,
+      "learning_rate": 0.001812,
+      "loss": 5.0733,
+      "step": 604
+    },
+    {
+      "epoch": 0.00605,
+      "grad_norm": 0.9664384705752951,
+      "learning_rate": 0.001815,
+      "loss": 5.0828,
+      "step": 605
+    },
+    {
+      "epoch": 0.00606,
+      "grad_norm": 1.0962791606856168,
+      "learning_rate": 0.001818,
+      "loss": 5.0888,
+      "step": 606
+    },
+    {
+      "epoch": 0.00607,
+      "grad_norm": 0.9313257227556363,
+      "learning_rate": 0.001821,
+      "loss": 5.0876,
+      "step": 607
+    },
+    {
+      "epoch": 0.00608,
+      "grad_norm": 0.9207616431206566,
+      "learning_rate": 0.001824,
+      "loss": 5.068,
+      "step": 608
+    },
+    {
+      "epoch": 0.00609,
+      "grad_norm": 0.9552257534317795,
+      "learning_rate": 0.001827,
+      "loss": 5.0751,
+      "step": 609
+    },
+    {
+      "epoch": 0.0061,
+      "grad_norm": 1.1338881088580717,
+      "learning_rate": 0.00183,
+      "loss": 5.062,
+      "step": 610
+    },
+    {
+      "epoch": 0.00611,
+      "grad_norm": 0.907481169345242,
+      "learning_rate": 0.001833,
+      "loss": 5.0576,
+      "step": 611
+    },
+    {
+      "epoch": 0.00612,
+      "grad_norm": 1.04757168088542,
+      "learning_rate": 0.001836,
+      "loss": 5.0686,
+      "step": 612
+    },
+    {
+      "epoch": 0.00613,
+      "grad_norm": 0.9021638869008188,
+      "learning_rate": 0.001839,
+      "loss": 5.0559,
+      "step": 613
+    },
+    {
+      "epoch": 0.00614,
+      "grad_norm": 0.7877209937196055,
+      "learning_rate": 0.001842,
+      "loss": 5.0768,
+      "step": 614
+    },
+    {
+      "epoch": 0.00615,
+      "grad_norm": 0.7456491069500266,
+      "learning_rate": 0.001845,
+      "loss": 5.0572,
+      "step": 615
+    },
+    {
+      "epoch": 0.00616,
+      "grad_norm": 0.7643816652567319,
+      "learning_rate": 0.001848,
+      "loss": 5.043,
+      "step": 616
+    },
+    {
+      "epoch": 0.00617,
+      "grad_norm": 0.8920948602141958,
+      "learning_rate": 0.001851,
+      "loss": 5.0555,
+      "step": 617
+    },
+    {
+      "epoch": 0.00618,
+      "grad_norm": 1.0257178323795717,
+      "learning_rate": 0.001854,
+      "loss": 5.0507,
+      "step": 618
+    },
+    {
+      "epoch": 0.00619,
+      "grad_norm": 0.7092119630036949,
+      "learning_rate": 0.001857,
+      "loss": 5.0429,
+      "step": 619
+    },
+    {
+      "epoch": 0.0062,
+      "grad_norm": 0.6092469798525071,
+      "learning_rate": 0.00186,
+      "loss": 5.0262,
+      "step": 620
+    },
+    {
+      "epoch": 0.00621,
+      "grad_norm": 0.7552706683211996,
+      "learning_rate": 0.001863,
+      "loss": 5.032,
+      "step": 621
+    },
+    {
+      "epoch": 0.00622,
+      "grad_norm": 0.6988650940233329,
+      "learning_rate": 0.001866,
+      "loss": 5.0245,
+      "step": 622
+    },
+    {
+      "epoch": 0.00623,
+      "grad_norm": 0.6062593525325608,
+      "learning_rate": 0.001869,
+      "loss": 5.0091,
+      "step": 623
+    },
+    {
+      "epoch": 0.00624,
+      "grad_norm": 0.555479795278756,
+      "learning_rate": 0.001872,
+      "loss": 5.0161,
+      "step": 624
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 0.6339519418385815,
+      "learning_rate": 0.001875,
+      "loss": 5.0276,
+      "step": 625
+    },
+    {
+      "epoch": 0.00626,
+      "grad_norm": 0.7820017260467441,
+      "learning_rate": 0.0018780000000000001,
+      "loss": 5.0084,
+      "step": 626
+    },
+    {
+      "epoch": 0.00627,
+      "grad_norm": 0.8263278266088275,
+      "learning_rate": 0.001881,
+      "loss": 5.0003,
+      "step": 627
+    },
+    {
+      "epoch": 0.00628,
+      "grad_norm": 0.5717634931424201,
+      "learning_rate": 0.001884,
+      "loss": 5.0204,
+      "step": 628
+    },
+    {
+      "epoch": 0.00629,
+      "grad_norm": 0.5505525631139665,
+      "learning_rate": 0.001887,
+      "loss": 4.9928,
+      "step": 629
+    },
+    {
+      "epoch": 0.0063,
+      "grad_norm": 0.47331922258372455,
+      "learning_rate": 0.00189,
+      "loss": 4.9837,
+      "step": 630
+    },
+    {
+      "epoch": 0.00631,
+      "grad_norm": 0.4173073508747504,
+      "learning_rate": 0.0018930000000000002,
+      "loss": 4.9757,
+      "step": 631
+    },
+    {
+      "epoch": 0.00632,
+      "grad_norm": 0.42808966750972455,
+      "learning_rate": 0.0018960000000000001,
+      "loss": 4.9928,
+      "step": 632
+    },
+    {
+      "epoch": 0.00633,
+      "grad_norm": 0.46750467774391197,
+      "learning_rate": 0.001899,
+      "loss": 4.9879,
+      "step": 633
+    },
+    {
+      "epoch": 0.00634,
+      "grad_norm": 0.5696215613123979,
+      "learning_rate": 0.001902,
+      "loss": 4.9763,
+      "step": 634
+    },
+    {
+      "epoch": 0.00635,
+      "grad_norm": 0.7138673203190975,
+      "learning_rate": 0.001905,
+      "loss": 4.9686,
+      "step": 635
+    },
+    {
+      "epoch": 0.00636,
+      "grad_norm": 0.8836310721952346,
+      "learning_rate": 0.001908,
+      "loss": 4.9828,
+      "step": 636
+    },
+    {
+      "epoch": 0.00637,
+      "grad_norm": 0.9755328249694639,
+      "learning_rate": 0.0019110000000000002,
+      "loss": 5.002,
+      "step": 637
+    },
+    {
+      "epoch": 0.00638,
+      "grad_norm": 1.3272264449165134,
+      "learning_rate": 0.0019140000000000001,
+      "loss": 5.0051,
+      "step": 638
+    },
+    {
+      "epoch": 0.00639,
+      "grad_norm": 0.9140856571338883,
+      "learning_rate": 0.001917,
+      "loss": 4.9726,
+      "step": 639
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 1.0466736994412218,
+      "learning_rate": 0.00192,
+      "loss": 4.9869,
+      "step": 640
+    },
+    {
+      "epoch": 0.00641,
+      "grad_norm": 1.1161814084223103,
+      "learning_rate": 0.001923,
+      "loss": 5.0003,
+      "step": 641
+    },
+    {
+      "epoch": 0.00642,
+      "grad_norm": 1.1352153221992676,
+      "learning_rate": 0.001926,
+      "loss": 4.992,
+      "step": 642
+    },
+    {
+      "epoch": 0.00643,
+      "grad_norm": 1.2268384891507862,
+      "learning_rate": 0.0019290000000000002,
+      "loss": 5.008,
+      "step": 643
+    },
+    {
+      "epoch": 0.00644,
+      "grad_norm": 0.9564122465750423,
+      "learning_rate": 0.0019320000000000001,
+      "loss": 4.9857,
+      "step": 644
+    },
+    {
+      "epoch": 0.00645,
+      "grad_norm": 0.9066234784688915,
+      "learning_rate": 0.001935,
+      "loss": 4.9889,
+      "step": 645
+    },
+    {
+      "epoch": 0.00646,
+      "grad_norm": 0.894776309426942,
+      "learning_rate": 0.001938,
+      "loss": 4.995,
+      "step": 646
+    },
+    {
+      "epoch": 0.00647,
+      "grad_norm": 1.036514260058091,
+      "learning_rate": 0.001941,
+      "loss": 5.0081,
+      "step": 647
+    },
+    {
+      "epoch": 0.00648,
+      "grad_norm": 1.0669688261896326,
+      "learning_rate": 0.0019440000000000002,
+      "loss": 4.9991,
+      "step": 648
+    },
+    {
+      "epoch": 0.00649,
+      "grad_norm": 1.0027602192299327,
+      "learning_rate": 0.0019470000000000002,
+      "loss": 4.9834,
+      "step": 649
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 1.2453243505592877,
+      "learning_rate": 0.0019500000000000001,
+      "loss": 5.0183,
+      "step": 650
+    },
+    {
+      "epoch": 0.00651,
+      "grad_norm": 0.9810465710779535,
+      "learning_rate": 0.001953,
+      "loss": 4.9899,
+      "step": 651
+    },
+    {
+      "epoch": 0.00652,
+      "grad_norm": 1.0529607937914427,
+      "learning_rate": 0.0019560000000000003,
+      "loss": 4.9855,
+      "step": 652
+    },
+    {
+      "epoch": 0.00653,
+      "grad_norm": 0.8367844014470341,
+      "learning_rate": 0.0019590000000000002,
+      "loss": 4.9719,
+      "step": 653
+    },
+    {
+      "epoch": 0.00654,
+      "grad_norm": 0.9556352679930769,
+      "learning_rate": 0.001962,
+      "loss": 4.96,
+      "step": 654
+    },
+    {
+      "epoch": 0.00655,
+      "grad_norm": 1.0712859822115042,
+      "learning_rate": 0.001965,
+      "loss": 4.9811,
+      "step": 655
+    },
+    {
+      "epoch": 0.00656,
+      "grad_norm": 0.8381525542412449,
+      "learning_rate": 0.001968,
+      "loss": 4.9628,
+      "step": 656
+    },
+    {
+      "epoch": 0.00657,
+      "grad_norm": 0.853522104902103,
+      "learning_rate": 0.001971,
+      "loss": 4.9486,
+      "step": 657
+    },
+    {
+      "epoch": 0.00658,
+      "grad_norm": 0.9276507218825019,
+      "learning_rate": 0.001974,
+      "loss": 4.9746,
+      "step": 658
+    },
+    {
+      "epoch": 0.00659,
+      "grad_norm": 1.0041122671115763,
+      "learning_rate": 0.001977,
+      "loss": 4.9552,
+      "step": 659
+    },
+    {
+      "epoch": 0.0066,
+      "grad_norm": 0.925609681342879,
+      "learning_rate": 0.00198,
+      "loss": 4.9516,
+      "step": 660
+    },
+    {
+      "epoch": 0.00661,
+      "grad_norm": 0.9340897694356347,
+      "learning_rate": 0.001983,
+      "loss": 4.9301,
+      "step": 661
+    },
+    {
+      "epoch": 0.00662,
+      "grad_norm": 1.1039924911609775,
+      "learning_rate": 0.0019860000000000004,
+      "loss": 4.9353,
+      "step": 662
+    },
+    {
+      "epoch": 0.00663,
+      "grad_norm": 0.842271096957419,
+      "learning_rate": 0.0019890000000000003,
+      "loss": 4.942,
+      "step": 663
+    },
+    {
+      "epoch": 0.00664,
+      "grad_norm": 0.9255021185692508,
+      "learning_rate": 0.0019920000000000003,
+      "loss": 4.9294,
+      "step": 664
+    },
+    {
+      "epoch": 0.00665,
+      "grad_norm": 0.9624840652606003,
+      "learning_rate": 0.0019950000000000002,
+      "loss": 4.9033,
+      "step": 665
+    },
+    {
+      "epoch": 0.00666,
+      "grad_norm": 0.8706912476713129,
+      "learning_rate": 0.001998,
+      "loss": 4.9275,
+      "step": 666
+    },
+    {
+      "epoch": 0.00667,
+      "grad_norm": 0.9355886153293348,
+      "learning_rate": 0.002001,
+      "loss": 4.9186,
+      "step": 667
+    },
+    {
+      "epoch": 0.00668,
+      "grad_norm": 0.9525697321085718,
+      "learning_rate": 0.002004,
+      "loss": 4.9422,
+      "step": 668
+    },
+    {
+      "epoch": 0.00669,
+      "grad_norm": 0.9357519892866003,
+      "learning_rate": 0.002007,
+      "loss": 4.9423,
+      "step": 669
+    },
+    {
+      "epoch": 0.0067,
+      "grad_norm": 0.8227655724451596,
+      "learning_rate": 0.00201,
+      "loss": 4.9287,
+      "step": 670
+    },
+    {
+      "epoch": 0.00671,
+      "grad_norm": 0.6363703099565691,
+      "learning_rate": 0.002013,
+      "loss": 4.8975,
+      "step": 671
+    },
+    {
+      "epoch": 0.00672,
+      "grad_norm": 0.5774852286623988,
+      "learning_rate": 0.002016,
+      "loss": 4.894,
+      "step": 672
+    },
+    {
+      "epoch": 0.00673,
+      "grad_norm": 0.5472152449551199,
+      "learning_rate": 0.002019,
+      "loss": 4.8881,
+      "step": 673
+    },
+    {
+      "epoch": 0.00674,
+      "grad_norm": 0.5267919584843693,
+      "learning_rate": 0.0020220000000000004,
+      "loss": 4.8857,
+      "step": 674
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 0.4823533534454447,
+      "learning_rate": 0.0020250000000000003,
+      "loss": 4.8922,
+      "step": 675
+    },
+    {
+      "epoch": 0.00676,
+      "grad_norm": 0.5135326218761617,
+      "learning_rate": 0.0020280000000000003,
+      "loss": 4.8739,
+      "step": 676
+    },
+    {
+      "epoch": 0.00677,
+      "grad_norm": 0.5629070673817519,
+      "learning_rate": 0.0020310000000000003,
+      "loss": 4.8753,
+      "step": 677
+    },
+    {
+      "epoch": 0.00678,
+      "grad_norm": 0.6636518437686225,
+      "learning_rate": 0.0020340000000000002,
+      "loss": 4.8844,
+      "step": 678
+    },
+    {
+      "epoch": 0.00679,
+      "grad_norm": 0.7939150938071041,
+      "learning_rate": 0.002037,
+      "loss": 4.8694,
+      "step": 679
+    },
+    {
+      "epoch": 0.0068,
+      "grad_norm": 0.6961050241971515,
+      "learning_rate": 0.00204,
+      "loss": 4.8681,
+      "step": 680
+    },
+    {
+      "epoch": 0.00681,
+      "grad_norm": 0.55281231514776,
+      "learning_rate": 0.002043,
+      "loss": 4.854,
+      "step": 681
+    },
+    {
+      "epoch": 0.00682,
+      "grad_norm": 0.6966375314025457,
+      "learning_rate": 0.002046,
+      "loss": 4.8606,
+      "step": 682
+    },
+    {
+      "epoch": 0.00683,
+      "grad_norm": 0.7919918537287608,
+      "learning_rate": 0.002049,
+      "loss": 4.8437,
+      "step": 683
+    },
+    {
+      "epoch": 0.00684,
+      "grad_norm": 1.014505658611748,
+      "learning_rate": 0.002052,
+      "loss": 4.868,
+      "step": 684
+    },
+    {
+      "epoch": 0.00685,
+      "grad_norm": 1.1979460789898289,
+      "learning_rate": 0.0020550000000000004,
+      "loss": 4.8824,
+      "step": 685
+    },
+    {
+      "epoch": 0.00686,
+      "grad_norm": 1.0209197434565493,
+      "learning_rate": 0.0020580000000000004,
+      "loss": 4.8555,
+      "step": 686
+    },
+    {
+      "epoch": 0.00687,
+      "grad_norm": 0.8666785819902827,
+      "learning_rate": 0.0020610000000000003,
+      "loss": 4.8457,
+      "step": 687
+    },
+    {
+      "epoch": 0.00688,
+      "grad_norm": 1.066277682185938,
+      "learning_rate": 0.002064,
+      "loss": 4.8576,
+      "step": 688
+    },
+    {
+      "epoch": 0.00689,
+      "grad_norm": 0.9564373481813865,
+      "learning_rate": 0.002067,
+      "loss": 4.8511,
+      "step": 689
+    },
+    {
+      "epoch": 0.0069,
+      "grad_norm": 1.0921039434762347,
+      "learning_rate": 0.00207,
+      "loss": 4.854,
+      "step": 690
+    },
+    {
+      "epoch": 0.00691,
+      "grad_norm": 0.9682767243250711,
+      "learning_rate": 0.0020729999999999998,
+      "loss": 4.8265,
+      "step": 691
+    },
+    {
+      "epoch": 0.00692,
+      "grad_norm": 1.0182149348151321,
+      "learning_rate": 0.0020759999999999997,
+      "loss": 4.8523,
+      "step": 692
+    },
+    {
+      "epoch": 0.00693,
+      "grad_norm": 1.1559582715656997,
+      "learning_rate": 0.0020789999999999997,
+      "loss": 4.8626,
+      "step": 693
+    },
+    {
+      "epoch": 0.00694,
+      "grad_norm": 0.8731253854999043,
+      "learning_rate": 0.002082,
+      "loss": 4.8314,
+      "step": 694
+    },
+    {
+      "epoch": 0.00695,
+      "grad_norm": 1.02924772696324,
+      "learning_rate": 0.002085,
+      "loss": 4.8388,
+      "step": 695
+    },
+    {
+      "epoch": 0.00696,
+      "grad_norm": 1.1148665982097032,
+      "learning_rate": 0.002088,
+      "loss": 4.8532,
+      "step": 696
+    },
+    {
+      "epoch": 0.00697,
+      "grad_norm": 1.1306884538795905,
+      "learning_rate": 0.002091,
+      "loss": 4.8556,
+      "step": 697
+    },
+    {
+      "epoch": 0.00698,
+      "grad_norm": 1.267565149046305,
+      "learning_rate": 0.002094,
+      "loss": 4.8474,
+      "step": 698
+    },
+    {
+      "epoch": 0.00699,
+      "grad_norm": 0.9067921371830103,
+      "learning_rate": 0.002097,
+      "loss": 4.8312,
+      "step": 699
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 0.8797095491461238,
+      "learning_rate": 0.0021,
+      "loss": 4.8378,
+      "step": 700
+    },
+    {
+      "epoch": 0.00701,
+      "grad_norm": 0.9764684557107473,
+      "learning_rate": 0.002103,
+      "loss": 4.8231,
+      "step": 701
+    },
+    {
+      "epoch": 0.00702,
+      "grad_norm": 0.9975561372265659,
+      "learning_rate": 0.002106,
+      "loss": 4.8269,
+      "step": 702
+    },
+    {
+      "epoch": 0.00703,
+      "grad_norm": 0.8409706105195134,
+      "learning_rate": 0.0021089999999999998,
+      "loss": 4.8046,
+      "step": 703
+    },
+    {
+      "epoch": 0.00704,
+      "grad_norm": 0.8602349489288845,
+      "learning_rate": 0.0021119999999999997,
+      "loss": 4.8112,
+      "step": 704
+    },
+    {
+      "epoch": 0.00705,
+      "grad_norm": 0.810283974291816,
+      "learning_rate": 0.002115,
+      "loss": 4.8189,
+      "step": 705
+    },
+    {
+      "epoch": 0.00706,
+      "grad_norm": 0.7904979262360543,
+      "learning_rate": 0.002118,
+      "loss": 4.8127,
+      "step": 706
+    },
+    {
+      "epoch": 0.00707,
+      "grad_norm": 0.8638007527609218,
+      "learning_rate": 0.002121,
+      "loss": 4.7902,
+      "step": 707
+    },
+    {
+      "epoch": 0.00708,
+      "grad_norm": 0.9274642295068019,
+      "learning_rate": 0.002124,
+      "loss": 4.7754,
+      "step": 708
+    },
+    {
+      "epoch": 0.00709,
+      "grad_norm": 0.9265048264631317,
+      "learning_rate": 0.002127,
+      "loss": 4.8051,
+      "step": 709
+    },
+    {
+      "epoch": 0.0071,
+      "grad_norm": 1.0021235226233682,
+      "learning_rate": 0.00213,
+      "loss": 4.8021,
+      "step": 710
+    },
+    {
+      "epoch": 0.00711,
+      "grad_norm": 0.8616225643918476,
+      "learning_rate": 0.002133,
+      "loss": 4.7687,
+      "step": 711
+    },
+    {
+      "epoch": 0.00712,
+      "grad_norm": 1.1509118831082872,
+      "learning_rate": 0.002136,
+      "loss": 4.8063,
+      "step": 712
+    },
+    {
+      "epoch": 0.00713,
+      "grad_norm": 0.8979386810595619,
+      "learning_rate": 0.002139,
+      "loss": 4.8085,
+      "step": 713
+    },
+    {
+      "epoch": 0.00714,
+      "grad_norm": 0.9672478380991173,
+      "learning_rate": 0.002142,
+      "loss": 4.7999,
+      "step": 714
+    },
+    {
+      "epoch": 0.00715,
+      "grad_norm": 1.0283238664767786,
+      "learning_rate": 0.0021449999999999998,
+      "loss": 4.7746,
+      "step": 715
+    },
+    {
+      "epoch": 0.00716,
+      "grad_norm": 0.8031836486660412,
+      "learning_rate": 0.002148,
+      "loss": 4.7611,
+      "step": 716
+    },
+    {
+      "epoch": 0.00717,
+      "grad_norm": 0.8079202930068127,
+      "learning_rate": 0.002151,
+      "loss": 4.7802,
+      "step": 717
+    },
+    {
+      "epoch": 0.00718,
+      "grad_norm": 0.7625878769693025,
+      "learning_rate": 0.002154,
+      "loss": 4.7607,
+      "step": 718
+    },
+    {
+      "epoch": 0.00719,
+      "grad_norm": 0.8833410128202536,
+      "learning_rate": 0.002157,
+      "loss": 4.7767,
+      "step": 719
+    },
+    {
+      "epoch": 0.0072,
+      "grad_norm": 1.0099718860880083,
+      "learning_rate": 0.00216,
+      "loss": 4.7733,
+      "step": 720
+    },
+    {
+      "epoch": 0.00721,
+      "grad_norm": 1.0037155368349988,
+      "learning_rate": 0.002163,
+      "loss": 4.78,
+      "step": 721
+    },
+    {
+      "epoch": 0.00722,
+      "grad_norm": 0.8678480066188063,
+      "learning_rate": 0.002166,
+      "loss": 4.7474,
+      "step": 722
+    },
+    {
+      "epoch": 0.00723,
+      "grad_norm": 0.9399210002572385,
+      "learning_rate": 0.002169,
+      "loss": 4.7657,
+      "step": 723
+    },
+    {
+      "epoch": 0.00724,
+      "grad_norm": 0.8184829699796181,
+      "learning_rate": 0.002172,
+      "loss": 4.7533,
+      "step": 724
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 0.8676839571587074,
+      "learning_rate": 0.002175,
+      "loss": 4.7513,
+      "step": 725
+    },
+    {
+      "epoch": 0.00726,
+      "grad_norm": 0.9799992988904748,
+      "learning_rate": 0.002178,
+      "loss": 4.7626,
+      "step": 726
+    },
+    {
+      "epoch": 0.00727,
+      "grad_norm": 1.2281779129682024,
+      "learning_rate": 0.0021809999999999998,
+      "loss": 4.7581,
+      "step": 727
+    },
+    {
+      "epoch": 0.00728,
+      "grad_norm": 1.082945747060172,
+      "learning_rate": 0.002184,
+      "loss": 4.7657,
+      "step": 728
+    },
+    {
+      "epoch": 0.00729,
+      "grad_norm": 1.0915510364818644,
+      "learning_rate": 0.002187,
+      "loss": 4.7617,
+      "step": 729
+    },
+    {
+      "epoch": 0.0073,
+      "grad_norm": 1.0738468909531949,
+      "learning_rate": 0.00219,
+      "loss": 4.7676,
+      "step": 730
+    },
+    {
+      "epoch": 0.00731,
+      "grad_norm": 1.0774407965183543,
+      "learning_rate": 0.002193,
+      "loss": 4.7572,
+      "step": 731
+    },
+    {
+      "epoch": 0.00732,
+      "grad_norm": 0.9732910355796593,
+      "learning_rate": 0.002196,
+      "loss": 4.7638,
+      "step": 732
+    },
+    {
+      "epoch": 0.00733,
+      "grad_norm": 1.02567545008427,
+      "learning_rate": 0.002199,
+      "loss": 4.7467,
+      "step": 733
+    },
+    {
+      "epoch": 0.00734,
+      "grad_norm": 0.9058765241181546,
+      "learning_rate": 0.002202,
+      "loss": 4.7686,
+      "step": 734
+    },
+    {
+      "epoch": 0.00735,
+      "grad_norm": 0.9682392169542167,
+      "learning_rate": 0.002205,
+      "loss": 4.7759,
+      "step": 735
+    },
+    {
+      "epoch": 0.00736,
+      "grad_norm": 0.9938811207200824,
+      "learning_rate": 0.002208,
+      "loss": 4.7615,
+      "step": 736
+    },
+    {
+      "epoch": 0.00737,
+      "grad_norm": 1.0964297344539389,
+      "learning_rate": 0.002211,
+      "loss": 4.7347,
+      "step": 737
+    },
+    {
+      "epoch": 0.00738,
+      "grad_norm": 0.8707349769325928,
+      "learning_rate": 0.002214,
+      "loss": 4.7342,
+      "step": 738
+    },
+    {
+      "epoch": 0.00739,
+      "grad_norm": 0.7399818233744658,
+      "learning_rate": 0.0022170000000000002,
+      "loss": 4.717,
+      "step": 739
+    },
+    {
+      "epoch": 0.0074,
+      "grad_norm": 0.9377539124718652,
+      "learning_rate": 0.00222,
+      "loss": 4.7301,
+      "step": 740
+    },
+    {
+      "epoch": 0.00741,
+      "grad_norm": 0.8596400455739317,
+      "learning_rate": 0.002223,
+      "loss": 4.699,
+      "step": 741
+    },
+    {
+      "epoch": 0.00742,
+      "grad_norm": 0.6578432901740889,
+      "learning_rate": 0.002226,
+      "loss": 4.7249,
+      "step": 742
+    },
+    {
+      "epoch": 0.00743,
+      "grad_norm": 0.6175884285032084,
+      "learning_rate": 0.002229,
+      "loss": 4.6843,
+      "step": 743
+    },
+    {
+      "epoch": 0.00744,
+      "grad_norm": 0.6615203369086972,
+      "learning_rate": 0.002232,
+      "loss": 4.6918,
+      "step": 744
+    },
+    {
+      "epoch": 0.00745,
+      "grad_norm": 0.6999197355703424,
+      "learning_rate": 0.002235,
+      "loss": 4.7005,
+      "step": 745
+    },
+    {
+      "epoch": 0.00746,
+      "grad_norm": 0.7056349857734648,
+      "learning_rate": 0.002238,
+      "loss": 4.6964,
+      "step": 746
+    },
+    {
+      "epoch": 0.00747,
+      "grad_norm": 0.6678161149510893,
+      "learning_rate": 0.002241,
+      "loss": 4.6817,
+      "step": 747
+    },
+    {
+      "epoch": 0.00748,
+      "grad_norm": 0.6725287147155753,
+      "learning_rate": 0.002244,
+      "loss": 4.6915,
+      "step": 748
+    },
+    {
+      "epoch": 0.00749,
+      "grad_norm": 0.7368138262221237,
+      "learning_rate": 0.002247,
+      "loss": 4.6725,
+      "step": 749
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 0.7550252977049275,
+      "learning_rate": 0.0022500000000000003,
+      "loss": 4.7124,
+      "step": 750
+    },
+    {
+      "epoch": 0.00751,
+      "grad_norm": 0.6461697616177359,
+      "learning_rate": 0.0022530000000000002,
+      "loss": 4.6948,
+      "step": 751
+    },
+    {
+      "epoch": 0.00752,
+      "grad_norm": 0.6473605328229959,
+      "learning_rate": 0.002256,
+      "loss": 4.648,
+      "step": 752
+    },
+    {
+      "epoch": 0.00753,
+      "grad_norm": 0.8172272904356894,
+      "learning_rate": 0.002259,
+      "loss": 4.6929,
+      "step": 753
+    },
+    {
+      "epoch": 0.00754,
+      "grad_norm": 0.8400684755887758,
+      "learning_rate": 0.002262,
+      "loss": 4.7068,
+      "step": 754
+    },
+    {
+      "epoch": 0.00755,
+      "grad_norm": 0.7078832518794317,
+      "learning_rate": 0.002265,
+      "loss": 4.6656,
+      "step": 755
+    },
+    {
+      "epoch": 0.00756,
+      "grad_norm": 0.6076060117836831,
+      "learning_rate": 0.002268,
+      "loss": 4.6484,
+      "step": 756
+    },
+    {
+      "epoch": 0.00757,
+      "grad_norm": 0.7133093934008413,
+      "learning_rate": 0.002271,
+      "loss": 4.6658,
+      "step": 757
+    },
+    {
+      "epoch": 0.00758,
+      "grad_norm": 0.7661771348142844,
+      "learning_rate": 0.002274,
+      "loss": 4.6521,
+      "step": 758
+    },
+    {
+      "epoch": 0.00759,
+      "grad_norm": 0.9250288948777622,
+      "learning_rate": 0.002277,
+      "loss": 4.6753,
+      "step": 759
+    },
+    {
+      "epoch": 0.0076,
+      "grad_norm": 1.037174236565274,
+      "learning_rate": 0.00228,
+      "loss": 4.669,
+      "step": 760
+    },
+    {
+      "epoch": 0.00761,
+      "grad_norm": 0.9678315157211191,
+      "learning_rate": 0.002283,
+      "loss": 4.6392,
+      "step": 761
+    },
+    {
+      "epoch": 0.00762,
+      "grad_norm": 1.3728001530688312,
+      "learning_rate": 0.0022860000000000003,
+      "loss": 4.6453,
+      "step": 762
+    },
+    {
+      "epoch": 0.00763,
+      "grad_norm": 1.0284727877786697,
+      "learning_rate": 0.0022890000000000002,
+      "loss": 4.6793,
+      "step": 763
+    },
+    {
+      "epoch": 0.00764,
+      "grad_norm": 0.9914794664489192,
+      "learning_rate": 0.002292,
+      "loss": 4.6942,
+      "step": 764
+    },
+    {
+      "epoch": 0.00765,
+      "grad_norm": 0.984322504117537,
+      "learning_rate": 0.002295,
+      "loss": 4.6765,
+      "step": 765
+    },
+    {
+      "epoch": 0.00766,
+      "grad_norm": 0.9320893698991433,
+      "learning_rate": 0.002298,
+      "loss": 4.6792,
+      "step": 766
+    },
+    {
+      "epoch": 0.00767,
+      "grad_norm": 1.0651442494276249,
+      "learning_rate": 0.002301,
+      "loss": 4.6823,
+      "step": 767
+    },
+    {
+      "epoch": 0.00768,
+      "grad_norm": 0.9179111624711317,
+      "learning_rate": 0.002304,
+      "loss": 4.6817,
+      "step": 768
+    },
+    {
+      "epoch": 0.00769,
+      "grad_norm": 1.026483766110404,
+      "learning_rate": 0.002307,
+      "loss": 4.6987,
+      "step": 769
+    },
+    {
+      "epoch": 0.0077,
+      "grad_norm": 1.1653119571960542,
+      "learning_rate": 0.00231,
+      "loss": 4.6621,
+      "step": 770
+    },
+    {
+      "epoch": 0.00771,
+      "grad_norm": 0.8477238808348645,
+      "learning_rate": 0.002313,
+      "loss": 4.689,
+      "step": 771
+    },
+    {
+      "epoch": 0.00772,
+      "grad_norm": 0.6519421622488206,
+      "learning_rate": 0.002316,
+      "loss": 4.6631,
+      "step": 772
+    },
+    {
+      "epoch": 0.00773,
+      "grad_norm": 0.6177861857364649,
+      "learning_rate": 0.0023190000000000003,
+      "loss": 4.6627,
+      "step": 773
+    },
+    {
+      "epoch": 0.00774,
+      "grad_norm": 0.6901665734497584,
+      "learning_rate": 0.0023220000000000003,
+      "loss": 4.6775,
+      "step": 774
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 0.7356087164350898,
+      "learning_rate": 0.0023250000000000002,
+      "loss": 4.6725,
+      "step": 775
+    },
+    {
+      "epoch": 0.00776,
+      "grad_norm": 0.8693432194982287,
+      "learning_rate": 0.002328,
+      "loss": 4.6722,
+      "step": 776
+    },
+    {
+      "epoch": 0.00777,
+      "grad_norm": 1.099570573598906,
+      "learning_rate": 0.002331,
+      "loss": 4.6596,
+      "step": 777
+    },
+    {
+      "epoch": 0.00778,
+      "grad_norm": 1.0694357671416344,
+      "learning_rate": 0.002334,
+      "loss": 4.6725,
+      "step": 778
+    },
+    {
+      "epoch": 0.00779,
+      "grad_norm": 1.0144407661707453,
+      "learning_rate": 0.002337,
+      "loss": 4.6345,
+      "step": 779
+    },
+    {
+      "epoch": 0.0078,
+      "grad_norm": 0.9392788681741788,
+      "learning_rate": 0.00234,
+      "loss": 4.6579,
+      "step": 780
+    },
+    {
+      "epoch": 0.00781,
+      "grad_norm": 0.9039044158767507,
+      "learning_rate": 0.002343,
+      "loss": 4.6528,
+      "step": 781
+    },
+    {
+      "epoch": 0.00782,
+      "grad_norm": 0.9671545635863801,
+      "learning_rate": 0.002346,
+      "loss": 4.6552,
+      "step": 782
+    },
+    {
+      "epoch": 0.00783,
+      "grad_norm": 0.9789758043675277,
+      "learning_rate": 0.002349,
+      "loss": 4.6404,
+      "step": 783
+    },
+    {
+      "epoch": 0.00784,
+      "grad_norm": 0.7674708275760124,
+      "learning_rate": 0.002352,
+      "loss": 4.6449,
+      "step": 784
+    },
+    {
+      "epoch": 0.00785,
+      "grad_norm": 0.6509064102772842,
+      "learning_rate": 0.0023550000000000003,
+      "loss": 4.6391,
+      "step": 785
+    },
+    {
+      "epoch": 0.00786,
+      "grad_norm": 0.6282839162170526,
+      "learning_rate": 0.0023580000000000003,
+      "loss": 4.6356,
+      "step": 786
+    },
+    {
+      "epoch": 0.00787,
+      "grad_norm": 0.6520823986249177,
+      "learning_rate": 0.0023610000000000003,
+      "loss": 4.6384,
+      "step": 787
+    },
+    {
+      "epoch": 0.00788,
+      "grad_norm": 0.6693545084259133,
+      "learning_rate": 0.002364,
+      "loss": 4.6342,
+      "step": 788
+    },
+    {
+      "epoch": 0.00789,
+      "grad_norm": 0.7061838607079715,
+      "learning_rate": 0.002367,
+      "loss": 4.6597,
+      "step": 789
+    },
+    {
+      "epoch": 0.0079,
+      "grad_norm": 0.7001074671969121,
+      "learning_rate": 0.00237,
+      "loss": 4.6333,
+      "step": 790
+    },
+    {
+      "epoch": 0.00791,
+      "grad_norm": 0.6934841557036142,
+      "learning_rate": 0.002373,
+      "loss": 4.5873,
+      "step": 791
+    },
+    {
+      "epoch": 0.00792,
+      "grad_norm": 0.6555126375785874,
+      "learning_rate": 0.002376,
+      "loss": 4.6158,
+      "step": 792
+    },
+    {
+      "epoch": 0.00793,
+      "grad_norm": 0.6198619755446345,
+      "learning_rate": 0.002379,
+      "loss": 4.6082,
+      "step": 793
+    },
+    {
+      "epoch": 0.00794,
+      "grad_norm": 0.5540734864838481,
+      "learning_rate": 0.002382,
+      "loss": 4.615,
+      "step": 794
+    },
+    {
+      "epoch": 0.00795,
+      "grad_norm": 0.5756469670820633,
+      "learning_rate": 0.002385,
+      "loss": 4.5927,
+      "step": 795
+    },
+    {
+      "epoch": 0.00796,
+      "grad_norm": 0.5912602735893169,
+      "learning_rate": 0.0023880000000000004,
+      "loss": 4.5839,
+      "step": 796
+    },
+    {
+      "epoch": 0.00797,
+      "grad_norm": 0.6477479187436139,
+      "learning_rate": 0.0023910000000000003,
+      "loss": 4.628,
+      "step": 797
+    },
+    {
+      "epoch": 0.00798,
+      "grad_norm": 0.6568610896012951,
+      "learning_rate": 0.0023940000000000003,
+      "loss": 4.5975,
+      "step": 798
+    },
+    {
+      "epoch": 0.00799,
+      "grad_norm": 0.8964867071559416,
+      "learning_rate": 0.0023970000000000003,
+      "loss": 4.6327,
+      "step": 799
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 1.1130272400690795,
+      "learning_rate": 0.0024000000000000002,
+      "loss": 4.6127,
+      "step": 800
+    },
+    {
+      "epoch": 0.00801,
+      "grad_norm": 0.8936330318178004,
+      "learning_rate": 0.002403,
+      "loss": 4.624,
+      "step": 801
+    },
+    {
+      "epoch": 0.00802,
+      "grad_norm": 1.279259611288336,
+      "learning_rate": 0.002406,
+      "loss": 4.6431,
+      "step": 802
+    },
+    {
+      "epoch": 0.00803,
+      "grad_norm": 0.7588568023085343,
+      "learning_rate": 0.002409,
+      "loss": 4.5967,
+      "step": 803
+    },
+    {
+      "epoch": 0.00804,
+      "grad_norm": 0.9911172738466991,
+      "learning_rate": 0.002412,
+      "loss": 4.6083,
+      "step": 804
+    },
+    {
+      "epoch": 0.00805,
+      "grad_norm": 1.0297855025384943,
+      "learning_rate": 0.002415,
+      "loss": 4.6095,
+      "step": 805
+    },
+    {
+      "epoch": 0.00806,
+      "grad_norm": 0.8656546102408385,
+      "learning_rate": 0.002418,
+      "loss": 4.6231,
+      "step": 806
+    },
+    {
+      "epoch": 0.00807,
+      "grad_norm": 0.7922998530369505,
+      "learning_rate": 0.0024210000000000004,
+      "loss": 4.6144,
+      "step": 807
+    },
+    {
+      "epoch": 0.00808,
+      "grad_norm": 0.7119238411669042,
+      "learning_rate": 0.0024240000000000004,
+      "loss": 4.5933,
+      "step": 808
+    },
+    {
+      "epoch": 0.00809,
+      "grad_norm": 0.6923611180056076,
+      "learning_rate": 0.0024270000000000003,
+      "loss": 4.5855,
+      "step": 809
+    },
+    {
+      "epoch": 0.0081,
+      "grad_norm": 0.759382946117684,
+      "learning_rate": 0.0024300000000000003,
+      "loss": 4.6149,
+      "step": 810
+    },
+    {
+      "epoch": 0.00811,
+      "grad_norm": 0.9629068291697402,
+      "learning_rate": 0.0024330000000000003,
+      "loss": 4.5955,
+      "step": 811
+    },
+    {
+      "epoch": 0.00812,
+      "grad_norm": 0.9897740547487952,
+      "learning_rate": 0.0024360000000000002,
+      "loss": 4.5793,
+      "step": 812
+    },
+    {
+      "epoch": 0.00813,
+      "grad_norm": 1.0343029805479964,
+      "learning_rate": 0.0024389999999999998,
+      "loss": 4.6192,
+      "step": 813
+    },
+    {
+      "epoch": 0.00814,
+      "grad_norm": 1.188458156645309,
+      "learning_rate": 0.0024419999999999997,
+      "loss": 4.6409,
+      "step": 814
+    },
+    {
+      "epoch": 0.00815,
+      "grad_norm": 0.960728927074141,
+      "learning_rate": 0.0024449999999999997,
+      "loss": 4.6319,
+      "step": 815
+    },
+    {
+      "epoch": 0.00816,
+      "grad_norm": 1.0099133377105225,
+      "learning_rate": 0.002448,
+      "loss": 4.6186,
+      "step": 816
+    },
+    {
+      "epoch": 0.00817,
+      "grad_norm": 0.9311887346236664,
+      "learning_rate": 0.002451,
+      "loss": 4.6246,
+      "step": 817
+    },
+    {
+      "epoch": 0.00818,
+      "grad_norm": 1.2609797944179002,
+      "learning_rate": 0.002454,
+      "loss": 4.625,
+      "step": 818
+    },
+    {
+      "epoch": 0.00819,
+      "grad_norm": 1.0447566978787928,
+      "learning_rate": 0.002457,
+      "loss": 4.6401,
+      "step": 819
+    },
+    {
+      "epoch": 0.0082,
+      "grad_norm": 1.1000322233283122,
+      "learning_rate": 0.00246,
+      "loss": 4.6239,
+      "step": 820
+    },
+    {
+      "epoch": 0.00821,
+      "grad_norm": 0.8928816997822231,
+      "learning_rate": 0.002463,
+      "loss": 4.5932,
+      "step": 821
+    },
+    {
+      "epoch": 0.00822,
+      "grad_norm": 0.9105337296182261,
+      "learning_rate": 0.002466,
+      "loss": 4.5996,
+      "step": 822
+    },
+    {
+      "epoch": 0.00823,
+      "grad_norm": 0.6537296194510775,
+      "learning_rate": 0.002469,
+      "loss": 4.5905,
+      "step": 823
+    },
+    {
+      "epoch": 0.00824,
+      "grad_norm": 0.6615102644633621,
+      "learning_rate": 0.002472,
+      "loss": 4.61,
+      "step": 824
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 0.7605760557018463,
+      "learning_rate": 0.0024749999999999998,
+      "loss": 4.6148,
+      "step": 825
+    },
+    {
+      "epoch": 0.00826,
+      "grad_norm": 0.9288808111624368,
+      "learning_rate": 0.0024779999999999997,
+      "loss": 4.5806,
+      "step": 826
+    },
+    {
+      "epoch": 0.00827,
+      "grad_norm": 0.8099205132023622,
+      "learning_rate": 0.002481,
+      "loss": 4.6084,
+      "step": 827
+    },
+    {
+      "epoch": 0.00828,
+      "grad_norm": 0.7444084275861881,
+      "learning_rate": 0.002484,
+      "loss": 4.5965,
+      "step": 828
+    },
+    {
+      "epoch": 0.00829,
+      "grad_norm": 0.6438605280155706,
+      "learning_rate": 0.002487,
+      "loss": 4.5891,
+      "step": 829
+    },
+    {
+      "epoch": 0.0083,
+      "grad_norm": 0.6242869251575957,
+      "learning_rate": 0.00249,
+      "loss": 4.5955,
+      "step": 830
+    },
+    {
+      "epoch": 0.00831,
+      "grad_norm": 0.6705073637500226,
+      "learning_rate": 0.002493,
+      "loss": 4.576,
+      "step": 831
+    },
+    {
+      "epoch": 0.00832,
+      "grad_norm": 0.7603459804613621,
+      "learning_rate": 0.002496,
+      "loss": 4.5759,
+      "step": 832
+    },
+    {
+      "epoch": 0.00833,
+      "grad_norm": 0.7662808437283888,
+      "learning_rate": 0.002499,
+      "loss": 4.5696,
+      "step": 833
+    },
+    {
+      "epoch": 0.00834,
+      "grad_norm": 0.8012380275176963,
+      "learning_rate": 0.002502,
+      "loss": 4.5743,
+      "step": 834
+    },
+    {
+      "epoch": 0.00835,
+      "grad_norm": 1.0075156101089233,
+      "learning_rate": 0.002505,
+      "loss": 4.5783,
+      "step": 835
+    },
+    {
+      "epoch": 0.00836,
+      "grad_norm": 1.0847073328294785,
+      "learning_rate": 0.002508,
+      "loss": 4.6155,
+      "step": 836
+    },
+    {
+      "epoch": 0.00837,
+      "grad_norm": 0.8849400794535106,
+      "learning_rate": 0.0025109999999999998,
+      "loss": 4.5603,
+      "step": 837
+    },
+    {
+      "epoch": 0.00838,
+      "grad_norm": 0.8832989831172253,
+      "learning_rate": 0.0025139999999999997,
+      "loss": 4.5816,
+      "step": 838
+    },
+    {
+      "epoch": 0.00839,
+      "grad_norm": 0.9511730781538094,
+      "learning_rate": 0.002517,
+      "loss": 4.5843,
+      "step": 839
+    },
+    {
+      "epoch": 0.0084,
+      "grad_norm": 1.037817208386755,
+      "learning_rate": 0.00252,
+      "loss": 4.5831,
+      "step": 840
+    },
+    {
+      "epoch": 0.00841,
+      "grad_norm": 0.8472466492390959,
+      "learning_rate": 0.002523,
+      "loss": 4.5624,
+      "step": 841
+    },
+    {
+      "epoch": 0.00842,
+      "grad_norm": 0.9269217528832409,
+      "learning_rate": 0.002526,
+      "loss": 4.5661,
+      "step": 842
+    },
+    {
+      "epoch": 0.00843,
+      "grad_norm": 0.8716247137682855,
+      "learning_rate": 0.002529,
+      "loss": 4.5688,
+      "step": 843
+    },
+    {
+      "epoch": 0.00844,
+      "grad_norm": 0.7416972066179122,
+      "learning_rate": 0.002532,
+      "loss": 4.585,
+      "step": 844
+    },
+    {
+      "epoch": 0.00845,
+      "grad_norm": 0.6177824387251759,
+      "learning_rate": 0.002535,
+      "loss": 4.5442,
+      "step": 845
+    },
+    {
+      "epoch": 0.00846,
+      "grad_norm": 0.5865348849804463,
+      "learning_rate": 0.002538,
+      "loss": 4.5656,
+      "step": 846
+    },
+    {
+      "epoch": 0.00847,
+      "grad_norm": 0.5324615561823309,
+      "learning_rate": 0.002541,
+      "loss": 4.5442,
+      "step": 847
+    },
+    {
+      "epoch": 0.00848,
+      "grad_norm": 0.5568583953268653,
+      "learning_rate": 0.002544,
+      "loss": 4.5407,
+      "step": 848
+    },
+    {
+      "epoch": 0.00849,
+      "grad_norm": 0.600307108588631,
+      "learning_rate": 0.002547,
+      "loss": 4.5371,
+      "step": 849
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 0.5343909127282478,
+      "learning_rate": 0.00255,
+      "loss": 4.5435,
+      "step": 850
+    },
+    {
+      "epoch": 0.00851,
+      "grad_norm": 0.5790732432599173,
+      "learning_rate": 0.002553,
+      "loss": 4.5588,
+      "step": 851
+    },
+    {
+      "epoch": 0.00852,
+      "grad_norm": 0.6406298666409433,
+      "learning_rate": 0.002556,
+      "loss": 4.5503,
+      "step": 852
+    },
+    {
+      "epoch": 0.00853,
+      "grad_norm": 0.6711876835719848,
+      "learning_rate": 0.002559,
+      "loss": 4.5298,
+      "step": 853
+    },
+    {
+      "epoch": 0.00854,
+      "grad_norm": 0.643558144142947,
+      "learning_rate": 0.002562,
+      "loss": 4.5219,
+      "step": 854
+    },
+    {
+      "epoch": 0.00855,
+      "grad_norm": 0.621905866188772,
+      "learning_rate": 0.002565,
+      "loss": 4.5026,
+      "step": 855
+    },
+    {
+      "epoch": 0.00856,
+      "grad_norm": 0.7503391003054042,
+      "learning_rate": 0.002568,
+      "loss": 4.5375,
+      "step": 856
+    },
+    {
+      "epoch": 0.00857,
+      "grad_norm": 0.90802719077466,
+      "learning_rate": 0.002571,
+      "loss": 4.5344,
+      "step": 857
+    },
+    {
+      "epoch": 0.00858,
+      "grad_norm": 0.9157789056738207,
+      "learning_rate": 0.002574,
+      "loss": 4.5308,
+      "step": 858
+    },
+    {
+      "epoch": 0.00859,
+      "grad_norm": 0.8455467899517649,
+      "learning_rate": 0.002577,
+      "loss": 4.5208,
+      "step": 859
+    },
+    {
+      "epoch": 0.0086,
+      "grad_norm": 0.929986585155377,
+      "learning_rate": 0.00258,
+      "loss": 4.5557,
+      "step": 860
+    },
+    {
+      "epoch": 0.00861,
+      "grad_norm": 1.062443351155347,
+      "learning_rate": 0.0025830000000000002,
+      "loss": 4.5815,
+      "step": 861
+    },
+    {
+      "epoch": 0.00862,
+      "grad_norm": 1.0476479971551458,
+      "learning_rate": 0.002586,
+      "loss": 4.5382,
+      "step": 862
+    },
+    {
+      "epoch": 0.00863,
+      "grad_norm": 0.9516272804079478,
+      "learning_rate": 0.002589,
+      "loss": 4.5686,
+      "step": 863
+    },
+    {
+      "epoch": 0.00864,
+      "grad_norm": 1.0775847362135182,
+      "learning_rate": 0.002592,
+      "loss": 4.5222,
+      "step": 864
+    },
+    {
+      "epoch": 0.00865,
+      "grad_norm": 0.902924351244226,
+      "learning_rate": 0.002595,
+      "loss": 4.5357,
+      "step": 865
+    },
+    {
+      "epoch": 0.00866,
+      "grad_norm": 0.889409265471235,
+      "learning_rate": 0.002598,
+      "loss": 4.5623,
+      "step": 866
+    },
+    {
+      "epoch": 0.00867,
+      "grad_norm": 1.1312843257696636,
+      "learning_rate": 0.002601,
+      "loss": 4.5565,
+      "step": 867
+    },
+    {
+      "epoch": 0.00868,
+      "grad_norm": 1.0195609890673947,
+      "learning_rate": 0.002604,
+      "loss": 4.5479,
+      "step": 868
+    },
+    {
+      "epoch": 0.00869,
+      "grad_norm": 0.8378311515658349,
+      "learning_rate": 0.002607,
+      "loss": 4.5492,
+      "step": 869
+    },
+    {
+      "epoch": 0.0087,
+      "grad_norm": 0.93569973985326,
+      "learning_rate": 0.00261,
+      "loss": 4.5413,
+      "step": 870
+    },
+    {
+      "epoch": 0.00871,
+      "grad_norm": 0.9947247250751194,
+      "learning_rate": 0.002613,
+      "loss": 4.5608,
+      "step": 871
+    },
+    {
+      "epoch": 0.00872,
+      "grad_norm": 0.9456334146879876,
+      "learning_rate": 0.002616,
+      "loss": 4.5489,
+      "step": 872
+    },
+    {
+      "epoch": 0.00873,
+      "grad_norm": 0.9088952463307589,
+      "learning_rate": 0.0026190000000000002,
+      "loss": 4.5587,
+      "step": 873
+    },
+    {
+      "epoch": 0.00874,
+      "grad_norm": 0.7636008668853458,
+      "learning_rate": 0.002622,
+      "loss": 4.5702,
+      "step": 874
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 0.8446171091890929,
+      "learning_rate": 0.002625,
+      "loss": 4.5191,
+      "step": 875
+    },
+    {
+      "epoch": 0.00876,
+      "grad_norm": 0.8731048122579586,
+      "learning_rate": 0.002628,
+      "loss": 4.551,
+      "step": 876
+    },
+    {
+      "epoch": 0.00877,
+      "grad_norm": 0.9444127454444267,
+      "learning_rate": 0.002631,
+      "loss": 4.5701,
+      "step": 877
+    },
+    {
+      "epoch": 0.00878,
+      "grad_norm": 0.9863621265162925,
+      "learning_rate": 0.002634,
+      "loss": 4.5108,
+      "step": 878
+    },
+    {
+      "epoch": 0.00879,
+      "grad_norm": 0.7746522972030878,
+      "learning_rate": 0.002637,
+      "loss": 4.5293,
+      "step": 879
+    },
+    {
+      "epoch": 0.0088,
+      "grad_norm": 0.7150869014843059,
+      "learning_rate": 0.00264,
+      "loss": 4.5368,
+      "step": 880
+    },
+    {
+      "epoch": 0.00881,
+      "grad_norm": 0.866142425879982,
+      "learning_rate": 0.002643,
+      "loss": 4.5448,
+      "step": 881
+    },
+    {
+      "epoch": 0.00882,
+      "grad_norm": 1.0565004714930601,
+      "learning_rate": 0.002646,
+      "loss": 4.5522,
+      "step": 882
+    },
+    {
+      "epoch": 0.00883,
+      "grad_norm": 0.88323512650563,
+      "learning_rate": 0.002649,
+      "loss": 4.5382,
+      "step": 883
+    },
+    {
+      "epoch": 0.00884,
+      "grad_norm": 0.8377440854137777,
+      "learning_rate": 0.0026520000000000003,
+      "loss": 4.5227,
+      "step": 884
+    },
+    {
+      "epoch": 0.00885,
+      "grad_norm": 0.8587569363150891,
+      "learning_rate": 0.0026550000000000002,
+      "loss": 4.5189,
+      "step": 885
+    },
+    {
+      "epoch": 0.00886,
+      "grad_norm": 0.9455034420832737,
+      "learning_rate": 0.002658,
+      "loss": 4.5539,
+      "step": 886
+    },
+    {
+      "epoch": 0.00887,
+      "grad_norm": 0.8258368480900744,
+      "learning_rate": 0.002661,
+      "loss": 4.536,
+      "step": 887
+    },
+    {
+      "epoch": 0.00888,
+      "grad_norm": 0.927023832062946,
+      "learning_rate": 0.002664,
+      "loss": 4.5392,
+      "step": 888
+    },
+    {
+      "epoch": 0.00889,
+      "grad_norm": 0.9905400773231482,
+      "learning_rate": 0.002667,
+      "loss": 4.5077,
+      "step": 889
+    },
+    {
+      "epoch": 0.0089,
+      "grad_norm": 1.0181625806478707,
+      "learning_rate": 0.00267,
+      "loss": 4.551,
+      "step": 890
+    },
+    {
+      "epoch": 0.00891,
+      "grad_norm": 1.0618776306697646,
+      "learning_rate": 0.002673,
+      "loss": 4.5446,
+      "step": 891
+    },
+    {
+      "epoch": 0.00892,
+      "grad_norm": 0.9464629097549706,
+      "learning_rate": 0.002676,
+      "loss": 4.5493,
+      "step": 892
+    },
+    {
+      "epoch": 0.00893,
+      "grad_norm": 1.02959831042168,
+      "learning_rate": 0.002679,
+      "loss": 4.5321,
+      "step": 893
+    },
+    {
+      "epoch": 0.00894,
+      "grad_norm": 0.8717588229222071,
+      "learning_rate": 0.002682,
+      "loss": 4.5126,
+      "step": 894
+    },
+    {
+      "epoch": 0.00895,
+      "grad_norm": 0.7766302283006379,
+      "learning_rate": 0.0026850000000000003,
+      "loss": 4.5191,
+      "step": 895
+    },
+    {
+      "epoch": 0.00896,
+      "grad_norm": 0.7210349653989065,
+      "learning_rate": 0.0026880000000000003,
+      "loss": 4.5061,
+      "step": 896
+    },
+    {
+      "epoch": 0.00897,
+      "grad_norm": 0.7840425119747009,
+      "learning_rate": 0.0026910000000000002,
+      "loss": 4.5249,
+      "step": 897
+    },
+    {
+      "epoch": 0.00898,
+      "grad_norm": 0.7907811575022647,
+      "learning_rate": 0.002694,
+      "loss": 4.518,
+      "step": 898
+    },
+    {
+      "epoch": 0.00899,
+      "grad_norm": 0.7866169386193732,
+      "learning_rate": 0.002697,
+      "loss": 4.5195,
+      "step": 899
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 0.7303478296484299,
+      "learning_rate": 0.0027,
+      "loss": 4.5449,
+      "step": 900
+    },
+    {
+      "epoch": 0.00901,
+      "grad_norm": 0.6444049592088867,
+      "learning_rate": 0.002703,
+      "loss": 4.4943,
+      "step": 901
+    },
+    {
+      "epoch": 0.00902,
+      "grad_norm": 0.5222361304775145,
+      "learning_rate": 0.002706,
+      "loss": 4.498,
+      "step": 902
+    },
+    {
+      "epoch": 0.00903,
+      "grad_norm": 0.6152635474063327,
+      "learning_rate": 0.002709,
+      "loss": 4.4967,
+      "step": 903
+    },
+    {
+      "epoch": 0.00904,
+      "grad_norm": 0.6931791624762859,
+      "learning_rate": 0.002712,
+      "loss": 4.4818,
+      "step": 904
+    },
+    {
+      "epoch": 0.00905,
+      "grad_norm": 0.6467753244335184,
+      "learning_rate": 0.002715,
+      "loss": 4.4929,
+      "step": 905
+    },
+    {
+      "epoch": 0.00906,
+      "grad_norm": 0.6638188976320862,
+      "learning_rate": 0.002718,
+      "loss": 4.5049,
+      "step": 906
+    },
+    {
+      "epoch": 0.00907,
+      "grad_norm": 0.503631484987646,
+      "learning_rate": 0.0027210000000000003,
+      "loss": 4.5121,
+      "step": 907
+    },
+    {
+      "epoch": 0.00908,
+      "grad_norm": 0.4745734069368428,
+      "learning_rate": 0.0027240000000000003,
+      "loss": 4.481,
+      "step": 908
+    },
+    {
+      "epoch": 0.00909,
+      "grad_norm": 0.5648420216556126,
+      "learning_rate": 0.0027270000000000003,
+      "loss": 4.5013,
+      "step": 909
+    },
+    {
+      "epoch": 0.0091,
+      "grad_norm": 0.5471966544704259,
+      "learning_rate": 0.0027300000000000002,
+      "loss": 4.5027,
+      "step": 910
+    },
+    {
+      "epoch": 0.00911,
+      "grad_norm": 0.6580432739675335,
+      "learning_rate": 0.002733,
+      "loss": 4.4619,
+      "step": 911
+    },
+    {
+      "epoch": 0.00912,
+      "grad_norm": 0.652804172410743,
+      "learning_rate": 0.002736,
+      "loss": 4.4678,
+      "step": 912
+    },
+    {
+      "epoch": 0.00913,
+      "grad_norm": 0.6108886726833864,
+      "learning_rate": 0.002739,
+      "loss": 4.4601,
+      "step": 913
+    },
+    {
+      "epoch": 0.00914,
+      "grad_norm": 0.5246452629450868,
+      "learning_rate": 0.002742,
+      "loss": 4.4865,
+      "step": 914
+    },
+    {
+      "epoch": 0.00915,
+      "grad_norm": 0.6046562570851716,
+      "learning_rate": 0.002745,
+      "loss": 4.4501,
+      "step": 915
+    },
+    {
+      "epoch": 0.00916,
+      "grad_norm": 0.6470672223067736,
+      "learning_rate": 0.002748,
+      "loss": 4.4304,
+      "step": 916
+    },
+    {
+      "epoch": 0.00917,
+      "grad_norm": 0.6458845430146066,
+      "learning_rate": 0.002751,
+      "loss": 4.4431,
+      "step": 917
+    },
+    {
+      "epoch": 0.00918,
+      "grad_norm": 0.6107710101855935,
+      "learning_rate": 0.0027540000000000004,
+      "loss": 4.4622,
+      "step": 918
+    },
+    {
+      "epoch": 0.00919,
+      "grad_norm": 0.6056157888916294,
+      "learning_rate": 0.0027570000000000003,
+      "loss": 4.4436,
+      "step": 919
+    },
+    {
+      "epoch": 0.0092,
+      "grad_norm": 0.6858065732895877,
+      "learning_rate": 0.0027600000000000003,
+      "loss": 4.462,
+      "step": 920
+    },
+    {
+      "epoch": 0.00921,
+      "grad_norm": 0.9391926745722488,
+      "learning_rate": 0.0027630000000000003,
+      "loss": 4.4556,
+      "step": 921
+    },
+    {
+      "epoch": 0.00922,
+      "grad_norm": 1.1348542218598812,
+      "learning_rate": 0.0027660000000000002,
+      "loss": 4.5063,
+      "step": 922
+    },
+    {
+      "epoch": 0.00923,
+      "grad_norm": 1.0421503736233508,
+      "learning_rate": 0.002769,
+      "loss": 4.4713,
+      "step": 923
+    },
+    {
+      "epoch": 0.00924,
+      "grad_norm": 1.0588478925879097,
+      "learning_rate": 0.002772,
+      "loss": 4.4959,
+      "step": 924
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 0.9781832929515508,
+      "learning_rate": 0.002775,
+      "loss": 4.4806,
+      "step": 925
+    },
+    {
+      "epoch": 0.00926,
+      "grad_norm": 0.7776497655560727,
+      "learning_rate": 0.002778,
+      "loss": 4.488,
+      "step": 926
+    },
+    {
+      "epoch": 0.00927,
+      "grad_norm": 0.7034519817521436,
+      "learning_rate": 0.002781,
+      "loss": 4.4448,
+      "step": 927
+    },
+    {
+      "epoch": 0.00928,
+      "grad_norm": 0.9270286412247504,
+      "learning_rate": 0.002784,
+      "loss": 4.4965,
+      "step": 928
+    },
+    {
+      "epoch": 0.00929,
+      "grad_norm": 1.0728170784278697,
+      "learning_rate": 0.0027870000000000004,
+      "loss": 4.4869,
+      "step": 929
+    },
+    {
+      "epoch": 0.0093,
+      "grad_norm": 0.9492127377122095,
+      "learning_rate": 0.0027900000000000004,
+      "loss": 4.4732,
+      "step": 930
+    },
+    {
+      "epoch": 0.00931,
+      "grad_norm": 0.9513900088751025,
+      "learning_rate": 0.0027930000000000003,
+      "loss": 4.4833,
+      "step": 931
+    },
+    {
+      "epoch": 0.00932,
+      "grad_norm": 1.4536440294804005,
+      "learning_rate": 0.0027960000000000003,
+      "loss": 4.5168,
+      "step": 932
+    },
+    {
+      "epoch": 0.00933,
+      "grad_norm": 1.132862147568844,
+      "learning_rate": 0.0027990000000000003,
+      "loss": 4.4769,
+      "step": 933
+    },
+    {
+      "epoch": 0.00934,
+      "grad_norm": 0.8513409648274727,
+      "learning_rate": 0.0028020000000000002,
+      "loss": 4.4941,
+      "step": 934
+    },
+    {
+      "epoch": 0.00935,
+      "grad_norm": 0.8677938211616196,
+      "learning_rate": 0.002805,
+      "loss": 4.5057,
+      "step": 935
+    },
+    {
+      "epoch": 0.00936,
+      "grad_norm": 0.8298971693271944,
+      "learning_rate": 0.002808,
+      "loss": 4.5081,
+      "step": 936
+    },
+    {
+      "epoch": 0.00937,
+      "grad_norm": 0.8363829119527492,
+      "learning_rate": 0.002811,
+      "loss": 4.521,
+      "step": 937
+    },
+    {
+      "epoch": 0.00938,
+      "grad_norm": 1.0036953395826609,
+      "learning_rate": 0.002814,
+      "loss": 4.5006,
+      "step": 938
+    },
+    {
+      "epoch": 0.00939,
+      "grad_norm": 1.0054078613176451,
+      "learning_rate": 0.002817,
+      "loss": 4.5073,
+      "step": 939
+    },
+    {
+      "epoch": 0.0094,
+      "grad_norm": 0.8667374621688471,
+      "learning_rate": 0.00282,
+      "loss": 4.4958,
+      "step": 940
+    },
+    {
+      "epoch": 0.00941,
+      "grad_norm": 0.9205808599892458,
+      "learning_rate": 0.002823,
+      "loss": 4.4733,
+      "step": 941
+    },
+    {
+      "epoch": 0.00942,
+      "grad_norm": 0.9778408651584425,
+      "learning_rate": 0.002826,
+      "loss": 4.5008,
+      "step": 942
+    },
+    {
+      "epoch": 0.00943,
+      "grad_norm": 1.082550194860624,
+      "learning_rate": 0.002829,
+      "loss": 4.4958,
+      "step": 943
+    },
+    {
+      "epoch": 0.00944,
+      "grad_norm": 1.355725245571646,
+      "learning_rate": 0.002832,
+      "loss": 4.5179,
+      "step": 944
+    },
+    {
+      "epoch": 0.00945,
+      "grad_norm": 0.8345488861943434,
+      "learning_rate": 0.002835,
+      "loss": 4.5015,
+      "step": 945
+    },
+    {
+      "epoch": 0.00946,
+      "grad_norm": 1.0009931674850154,
+      "learning_rate": 0.002838,
+      "loss": 4.5193,
+      "step": 946
+    },
+    {
+      "epoch": 0.00947,
+      "grad_norm": 1.0176650405493206,
+      "learning_rate": 0.0028409999999999998,
+      "loss": 4.4912,
+      "step": 947
+    },
+    {
+      "epoch": 0.00948,
+      "grad_norm": 0.836896216552642,
+      "learning_rate": 0.0028439999999999997,
+      "loss": 4.472,
+      "step": 948
+    },
+    {
+      "epoch": 0.00949,
+      "grad_norm": 0.6881971993105954,
+      "learning_rate": 0.002847,
+      "loss": 4.5057,
+      "step": 949
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 0.8926787293989068,
+      "learning_rate": 0.00285,
+      "loss": 4.5027,
+      "step": 950
+    },
+    {
+      "epoch": 0.00951,
+      "grad_norm": 1.0437772730376889,
+      "learning_rate": 0.002853,
+      "loss": 4.4861,
+      "step": 951
+    },
+    {
+      "epoch": 0.00952,
+      "grad_norm": 0.8745743913439339,
+      "learning_rate": 0.002856,
+      "loss": 4.5041,
+      "step": 952
+    },
+    {
+      "epoch": 0.00953,
+      "grad_norm": 0.7847706213592531,
+      "learning_rate": 0.002859,
+      "loss": 4.4446,
+      "step": 953
+    },
+    {
+      "epoch": 0.00954,
+      "grad_norm": 0.6236105730880978,
+      "learning_rate": 0.002862,
+      "loss": 4.4945,
+      "step": 954
+    },
+    {
+      "epoch": 0.00955,
+      "grad_norm": 0.5696186051972435,
+      "learning_rate": 0.002865,
+      "loss": 4.475,
+      "step": 955
+    },
+    {
+      "epoch": 0.00956,
+      "grad_norm": 0.5459272314199634,
+      "learning_rate": 0.002868,
+      "loss": 4.4518,
+      "step": 956
+    },
+    {
+      "epoch": 0.00957,
+      "grad_norm": 0.5100325019322003,
+      "learning_rate": 0.002871,
+      "loss": 4.4721,
+      "step": 957
+    },
+    {
+      "epoch": 0.00958,
+      "grad_norm": 0.6617022302690957,
+      "learning_rate": 0.002874,
+      "loss": 4.4567,
+      "step": 958
+    },
+    {
+      "epoch": 0.00959,
+      "grad_norm": 0.7948420381771908,
+      "learning_rate": 0.002877,
+      "loss": 4.4693,
+      "step": 959
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 1.0277128972108451,
+      "learning_rate": 0.0028799999999999997,
+      "loss": 4.4747,
+      "step": 960
+    },
+    {
+      "epoch": 0.00961,
+      "grad_norm": 0.9925275083373442,
+      "learning_rate": 0.002883,
+      "loss": 4.4507,
+      "step": 961
+    },
+    {
+      "epoch": 0.00962,
+      "grad_norm": 1.0177847800658486,
+      "learning_rate": 0.002886,
+      "loss": 4.4861,
+      "step": 962
+    },
+    {
+      "epoch": 0.00963,
+      "grad_norm": 1.0118802628275685,
+      "learning_rate": 0.002889,
+      "loss": 4.4865,
+      "step": 963
+    },
+    {
+      "epoch": 0.00964,
+      "grad_norm": 1.1856958623453784,
+      "learning_rate": 0.002892,
+      "loss": 4.4868,
+      "step": 964
+    },
+    {
+      "epoch": 0.00965,
+      "grad_norm": 0.8341279969334199,
+      "learning_rate": 0.002895,
+      "loss": 4.4524,
+      "step": 965
+    },
+    {
+      "epoch": 0.00966,
+      "grad_norm": 0.6756655743310646,
+      "learning_rate": 0.002898,
+      "loss": 4.4552,
+      "step": 966
+    },
+    {
+      "epoch": 0.00967,
+      "grad_norm": 0.6634929266596646,
+      "learning_rate": 0.002901,
+      "loss": 4.4933,
+      "step": 967
+    },
+    {
+      "epoch": 0.00968,
+      "grad_norm": 0.6850072615364151,
+      "learning_rate": 0.002904,
+      "loss": 4.4564,
+      "step": 968
+    },
+    {
+      "epoch": 0.00969,
+      "grad_norm": 0.6166935294692494,
+      "learning_rate": 0.002907,
+      "loss": 4.4862,
+      "step": 969
+    },
+    {
+      "epoch": 0.0097,
+      "grad_norm": 0.6616720297110108,
+      "learning_rate": 0.00291,
+      "loss": 4.4401,
+      "step": 970
+    },
+    {
+      "epoch": 0.00971,
+      "grad_norm": 0.5463010292381552,
+      "learning_rate": 0.002913,
+      "loss": 4.4415,
+      "step": 971
+    },
+    {
+      "epoch": 0.00972,
+      "grad_norm": 0.5115285821904162,
+      "learning_rate": 0.002916,
+      "loss": 4.4454,
+      "step": 972
+    },
+    {
+      "epoch": 0.00973,
+      "grad_norm": 0.543781975527911,
+      "learning_rate": 0.002919,
+      "loss": 4.4402,
+      "step": 973
+    },
+    {
+      "epoch": 0.00974,
+      "grad_norm": 0.5167263203162235,
+      "learning_rate": 0.002922,
+      "loss": 4.4246,
+      "step": 974
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 0.4676730441900605,
+      "learning_rate": 0.002925,
+      "loss": 4.4434,
+      "step": 975
+    },
+    {
+      "epoch": 0.00976,
+      "grad_norm": 0.4483140513716565,
+      "learning_rate": 0.002928,
+      "loss": 4.4259,
+      "step": 976
+    },
+    {
+      "epoch": 0.00977,
+      "grad_norm": 0.45300675248114197,
+      "learning_rate": 0.002931,
+      "loss": 4.4582,
+      "step": 977
+    },
+    {
+      "epoch": 0.00978,
+      "grad_norm": 0.524704391185419,
+      "learning_rate": 0.002934,
+      "loss": 4.4335,
+      "step": 978
+    },
+    {
+      "epoch": 0.00979,
+      "grad_norm": 0.6643809565846889,
+      "learning_rate": 0.002937,
+      "loss": 4.4266,
+      "step": 979
+    },
+    {
+      "epoch": 0.0098,
+      "grad_norm": 0.9159629342753403,
+      "learning_rate": 0.00294,
+      "loss": 4.433,
+      "step": 980
+    },
+    {
+      "epoch": 0.00981,
+      "grad_norm": 1.0134318026936866,
+      "learning_rate": 0.002943,
+      "loss": 4.4516,
+      "step": 981
+    },
+    {
+      "epoch": 0.00982,
+      "grad_norm": 0.7532241089289973,
+      "learning_rate": 0.002946,
+      "loss": 4.4387,
+      "step": 982
+    },
+    {
+      "epoch": 0.00983,
+      "grad_norm": 0.7539130672753217,
+      "learning_rate": 0.0029490000000000002,
+      "loss": 4.4372,
+      "step": 983
+    },
+    {
+      "epoch": 0.00984,
+      "grad_norm": 0.7045823796689694,
+      "learning_rate": 0.002952,
+      "loss": 4.4213,
+      "step": 984
+    },
+    {
+      "epoch": 0.00985,
+      "grad_norm": 0.6478785171714704,
+      "learning_rate": 0.002955,
+      "loss": 4.444,
+      "step": 985
+    },
+    {
+      "epoch": 0.00986,
+      "grad_norm": 0.5640139479823427,
+      "learning_rate": 0.002958,
+      "loss": 4.3997,
+      "step": 986
+    },
+    {
+      "epoch": 0.00987,
+      "grad_norm": 0.6145958247621988,
+      "learning_rate": 0.002961,
+      "loss": 4.4141,
+      "step": 987
+    },
+    {
+      "epoch": 0.00988,
+      "grad_norm": 0.7121366087401472,
+      "learning_rate": 0.002964,
+      "loss": 4.4512,
+      "step": 988
+    },
+    {
+      "epoch": 0.00989,
+      "grad_norm": 0.7474063595618389,
+      "learning_rate": 0.002967,
+      "loss": 4.3897,
+      "step": 989
+    },
+    {
+      "epoch": 0.0099,
+      "grad_norm": 0.7608901784540225,
+      "learning_rate": 0.00297,
+      "loss": 4.4296,
+      "step": 990
+    },
+    {
+      "epoch": 0.00991,
+      "grad_norm": 0.8136541225519112,
+      "learning_rate": 0.002973,
+      "loss": 4.4314,
+      "step": 991
+    },
+    {
+      "epoch": 0.00992,
+      "grad_norm": 0.7865701920195308,
+      "learning_rate": 0.002976,
+      "loss": 4.4266,
+      "step": 992
+    },
+    {
+      "epoch": 0.00993,
+      "grad_norm": 0.8315737176917932,
+      "learning_rate": 0.002979,
+      "loss": 4.4267,
+      "step": 993
+    },
+    {
+      "epoch": 0.00994,
+      "grad_norm": 0.7821802322270756,
+      "learning_rate": 0.002982,
+      "loss": 4.4281,
+      "step": 994
+    },
+    {
+      "epoch": 0.00995,
+      "grad_norm": 0.8705272973695986,
+      "learning_rate": 0.0029850000000000002,
+      "loss": 4.4331,
+      "step": 995
+    },
+    {
+      "epoch": 0.00996,
+      "grad_norm": 0.9954881536889274,
+      "learning_rate": 0.002988,
+      "loss": 4.4408,
+      "step": 996
+    },
+    {
+      "epoch": 0.00997,
+      "grad_norm": 0.9720370341600497,
+      "learning_rate": 0.002991,
+      "loss": 4.4354,
+      "step": 997
+    },
+    {
+      "epoch": 0.00998,
+      "grad_norm": 0.7522042168889891,
+      "learning_rate": 0.002994,
+      "loss": 4.4331,
+      "step": 998
+    },
+    {
+      "epoch": 0.00999,
+      "grad_norm": 0.9425882614336212,
+      "learning_rate": 0.002997,
+      "loss": 4.4261,
+      "step": 999
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.9689827453070083,
+      "learning_rate": 0.003,
+      "loss": 4.458,
+      "step": 1000
+    },
+    {
+      "epoch": 0.01001,
+      "grad_norm": 0.9802085236972514,
+      "learning_rate": 0.003,
+      "loss": 4.4569,
+      "step": 1001
+    },
+    {
+      "epoch": 0.01002,
+      "grad_norm": 0.9742392022619593,
+      "learning_rate": 0.003,
+      "loss": 4.4262,
+      "step": 1002
+    },
+    {
+      "epoch": 0.01003,
+      "grad_norm": 0.9115330327806416,
+      "learning_rate": 0.003,
+      "loss": 4.4513,
+      "step": 1003
+    },
+    {
+      "epoch": 0.01004,
+      "grad_norm": 0.8562148232052564,
+      "learning_rate": 0.003,
+      "loss": 4.4397,
+      "step": 1004
+    },
+    {
+      "epoch": 0.01005,
+      "grad_norm": 0.9105663755998641,
+      "learning_rate": 0.003,
+      "loss": 4.4154,
+      "step": 1005
+    },
+    {
+      "epoch": 0.01006,
+      "grad_norm": 1.018045955439956,
+      "learning_rate": 0.003,
+      "loss": 4.433,
+      "step": 1006
+    },
+    {
+      "epoch": 0.01007,
+      "grad_norm": 0.9357911521230117,
+      "learning_rate": 0.003,
+      "loss": 4.4713,
+      "step": 1007
+    },
+    {
+      "epoch": 0.01008,
+      "grad_norm": 0.9681416399703533,
+      "learning_rate": 0.003,
+      "loss": 4.4846,
+      "step": 1008
+    },
+    {
+      "epoch": 0.01009,
+      "grad_norm": 1.030244827646589,
+      "learning_rate": 0.003,
+      "loss": 4.4619,
+      "step": 1009
+    },
+    {
+      "epoch": 0.0101,
+      "grad_norm": 1.0010362528613534,
+      "learning_rate": 0.003,
+      "loss": 4.4688,
+      "step": 1010
+    },
+    {
+      "epoch": 0.01011,
+      "grad_norm": 1.0284858383133122,
+      "learning_rate": 0.003,
+      "loss": 4.4881,
+      "step": 1011
+    },
+    {
+      "epoch": 0.01012,
+      "grad_norm": 0.9778345429910184,
+      "learning_rate": 0.003,
+      "loss": 4.4674,
+      "step": 1012
+    },
+    {
+      "epoch": 0.01013,
+      "grad_norm": 0.8705611016855861,
+      "learning_rate": 0.003,
+      "loss": 4.4919,
+      "step": 1013
+    },
+    {
+      "epoch": 0.01014,
+      "grad_norm": 0.8083144078788229,
+      "learning_rate": 0.003,
+      "loss": 4.4268,
+      "step": 1014
+    },
+    {
+      "epoch": 0.01015,
+      "grad_norm": 0.7155205086193928,
+      "learning_rate": 0.003,
+      "loss": 4.4657,
+      "step": 1015
+    },
+    {
+      "epoch": 0.01016,
+      "grad_norm": 0.5844948538982189,
+      "learning_rate": 0.003,
+      "loss": 4.466,
+      "step": 1016
+    },
+    {
+      "epoch": 0.01017,
+      "grad_norm": 0.50419422786756,
+      "learning_rate": 0.003,
+      "loss": 4.4019,
+      "step": 1017
+    },
+    {
+      "epoch": 0.01018,
+      "grad_norm": 0.5273528303530755,
+      "learning_rate": 0.003,
+      "loss": 4.4545,
+      "step": 1018
+    },
+    {
+      "epoch": 0.01019,
+      "grad_norm": 0.5329688106940915,
+      "learning_rate": 0.003,
+      "loss": 4.4315,
+      "step": 1019
+    },
+    {
+      "epoch": 0.0102,
+      "grad_norm": 0.6285459843623249,
+      "learning_rate": 0.003,
+      "loss": 4.4292,
+      "step": 1020
+    },
+    {
+      "epoch": 0.01021,
+      "grad_norm": 0.6444525489483212,
+      "learning_rate": 0.003,
+      "loss": 4.412,
+      "step": 1021
+    },
+    {
+      "epoch": 0.01022,
+      "grad_norm": 0.6609826768689684,
+      "learning_rate": 0.003,
+      "loss": 4.4194,
+      "step": 1022
+    },
+    {
+      "epoch": 0.01023,
+      "grad_norm": 0.6479610287689606,
+      "learning_rate": 0.003,
+      "loss": 4.4151,
+      "step": 1023
+    },
+    {
+      "epoch": 0.01024,
+      "grad_norm": 0.7091931447524652,
+      "learning_rate": 0.003,
+      "loss": 4.3798,
+      "step": 1024
+    },
+    {
+      "epoch": 0.01025,
+      "grad_norm": 0.7026391214213478,
+      "learning_rate": 0.003,
+      "loss": 4.4116,
+      "step": 1025
+    },
+    {
+      "epoch": 0.01026,
+      "grad_norm": 0.6731332289892269,
+      "learning_rate": 0.003,
+      "loss": 4.3991,
+      "step": 1026
+    },
+    {
+      "epoch": 0.01027,
+      "grad_norm": 0.6590644472165706,
+      "learning_rate": 0.003,
+      "loss": 4.4177,
+      "step": 1027
+    },
+    {
+      "epoch": 0.01028,
+      "grad_norm": 0.7485101036485022,
+      "learning_rate": 0.003,
+      "loss": 4.4014,
+      "step": 1028
+    },
+    {
+      "epoch": 0.01029,
+      "grad_norm": 0.7198477689690366,
+      "learning_rate": 0.003,
+      "loss": 4.3803,
+      "step": 1029
+    },
+    {
+      "epoch": 0.0103,
+      "grad_norm": 0.5542753635749327,
+      "learning_rate": 0.003,
+      "loss": 4.4023,
+      "step": 1030
+    },
+    {
+      "epoch": 0.01031,
+      "grad_norm": 0.5292390629019561,
+      "learning_rate": 0.003,
+      "loss": 4.4144,
+      "step": 1031
+    },
+    {
+      "epoch": 0.01032,
+      "grad_norm": 0.500926566427153,
+      "learning_rate": 0.003,
+      "loss": 4.3837,
+      "step": 1032
+    },
+    {
+      "epoch": 0.01033,
+      "grad_norm": 0.49056326531128164,
+      "learning_rate": 0.003,
+      "loss": 4.4201,
+      "step": 1033
+    },
+    {
+      "epoch": 0.01034,
+      "grad_norm": 0.5309292795236984,
+      "learning_rate": 0.003,
+      "loss": 4.3865,
+      "step": 1034
+    },
+    {
+      "epoch": 0.01035,
+      "grad_norm": 0.6084296376545847,
+      "learning_rate": 0.003,
+      "loss": 4.3892,
+      "step": 1035
+    },
+    {
+      "epoch": 0.01036,
+      "grad_norm": 0.8615895093148164,
+      "learning_rate": 0.003,
+      "loss": 4.3883,
+      "step": 1036
+    },
+    {
+      "epoch": 0.01037,
+      "grad_norm": 0.8936988458454226,
+      "learning_rate": 0.003,
+      "loss": 4.4336,
+      "step": 1037
+    },
+    {
+      "epoch": 0.01038,
+      "grad_norm": 0.6948017038229403,
+      "learning_rate": 0.003,
+      "loss": 4.4116,
+      "step": 1038
+    },
+    {
+      "epoch": 0.01039,
+      "grad_norm": 0.9114790426144561,
+      "learning_rate": 0.003,
+      "loss": 4.4152,
+      "step": 1039
+    },
+    {
+      "epoch": 0.0104,
+      "grad_norm": 0.8448966818619524,
+      "learning_rate": 0.003,
+      "loss": 4.4288,
+      "step": 1040
+    },
+    {
+      "epoch": 0.01041,
+      "grad_norm": 0.7995140521375168,
+      "learning_rate": 0.003,
+      "loss": 4.404,
+      "step": 1041
+    },
+    {
+      "epoch": 0.01042,
+      "grad_norm": 0.8979223927667839,
+      "learning_rate": 0.003,
+      "loss": 4.4357,
+      "step": 1042
+    },
+    {
+      "epoch": 0.01043,
+      "grad_norm": 0.7233892787514891,
+      "learning_rate": 0.003,
+      "loss": 4.3775,
+      "step": 1043
+    },
+    {
+      "epoch": 0.01044,
+      "grad_norm": 0.782866039824708,
+      "learning_rate": 0.003,
+      "loss": 4.4144,
+      "step": 1044
+    },
+    {
+      "epoch": 0.01045,
+      "grad_norm": 0.841755171402396,
+      "learning_rate": 0.003,
+      "loss": 4.4353,
+      "step": 1045
+    },
+    {
+      "epoch": 0.01046,
+      "grad_norm": 0.780348729633882,
+      "learning_rate": 0.003,
+      "loss": 4.4043,
+      "step": 1046
+    },
+    {
+      "epoch": 0.01047,
+      "grad_norm": 0.8623775519511051,
+      "learning_rate": 0.003,
+      "loss": 4.4135,
+      "step": 1047
+    },
+    {
+      "epoch": 0.01048,
+      "grad_norm": 0.9009461265034386,
+      "learning_rate": 0.003,
+      "loss": 4.4049,
+      "step": 1048
+    },
+    {
+      "epoch": 0.01049,
+      "grad_norm": 0.739794637514069,
+      "learning_rate": 0.003,
+      "loss": 4.3786,
+      "step": 1049
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 0.6259101207085414,
+      "learning_rate": 0.003,
+      "loss": 4.4154,
+      "step": 1050
+    },
+    {
+      "epoch": 0.01051,
+      "grad_norm": 0.6694575857098493,
+      "learning_rate": 0.003,
+      "loss": 4.393,
+      "step": 1051
+    },
+    {
+      "epoch": 0.01052,
+      "grad_norm": 0.7141380180592181,
+      "learning_rate": 0.003,
+      "loss": 4.4052,
+      "step": 1052
+    },
+    {
+      "epoch": 0.01053,
+      "grad_norm": 0.7943106393078887,
+      "learning_rate": 0.003,
+      "loss": 4.4143,
+      "step": 1053
+    },
+    {
+      "epoch": 0.01054,
+      "grad_norm": 0.7478655479661611,
+      "learning_rate": 0.003,
+      "loss": 4.4077,
+      "step": 1054
+    },
+    {
+      "epoch": 0.01055,
+      "grad_norm": 0.707966249364238,
+      "learning_rate": 0.003,
+      "loss": 4.3861,
+      "step": 1055
+    },
+    {
+      "epoch": 0.01056,
+      "grad_norm": 0.6358438963919717,
+      "learning_rate": 0.003,
+      "loss": 4.3823,
+      "step": 1056
+    },
+    {
+      "epoch": 0.01057,
+      "grad_norm": 0.5691063929797074,
+      "learning_rate": 0.003,
+      "loss": 4.3849,
+      "step": 1057
+    },
+    {
+      "epoch": 0.01058,
+      "grad_norm": 0.5109096197008457,
+      "learning_rate": 0.003,
+      "loss": 4.3737,
+      "step": 1058
+    },
+    {
+      "epoch": 0.01059,
+      "grad_norm": 0.419033279294021,
+      "learning_rate": 0.003,
+      "loss": 4.3749,
+      "step": 1059
+    },
+    {
+      "epoch": 0.0106,
+      "grad_norm": 0.4506837031966603,
+      "learning_rate": 0.003,
+      "loss": 4.3654,
+      "step": 1060
+    },
+    {
+      "epoch": 0.01061,
+      "grad_norm": 0.42662194113596513,
+      "learning_rate": 0.003,
+      "loss": 4.361,
+      "step": 1061
+    },
+    {
+      "epoch": 0.01062,
+      "grad_norm": 0.5018523046776621,
+      "learning_rate": 0.003,
+      "loss": 4.3568,
+      "step": 1062
+    },
+    {
+      "epoch": 0.01063,
+      "grad_norm": 0.6580287488917459,
+      "learning_rate": 0.003,
+      "loss": 4.3552,
+      "step": 1063
+    },
+    {
+      "epoch": 0.01064,
+      "grad_norm": 0.831793270479749,
+      "learning_rate": 0.003,
+      "loss": 4.3844,
+      "step": 1064
+    },
+    {
+      "epoch": 0.01065,
+      "grad_norm": 1.0701105543701324,
+      "learning_rate": 0.003,
+      "loss": 4.3903,
+      "step": 1065
+    },
+    {
+      "epoch": 0.01066,
+      "grad_norm": 0.943731953453328,
+      "learning_rate": 0.003,
+      "loss": 4.4118,
+      "step": 1066
+    },
+    {
+      "epoch": 0.01067,
+      "grad_norm": 0.9236025360095688,
+      "learning_rate": 0.003,
+      "loss": 4.4136,
+      "step": 1067
+    },
+    {
+      "epoch": 0.01068,
+      "grad_norm": 1.1282330525862354,
+      "learning_rate": 0.003,
+      "loss": 4.4118,
+      "step": 1068
+    },
+    {
+      "epoch": 0.01069,
+      "grad_norm": 0.840410887973158,
+      "learning_rate": 0.003,
+      "loss": 4.4054,
+      "step": 1069
+    },
+    {
+      "epoch": 0.0107,
+      "grad_norm": 0.733631994312363,
+      "learning_rate": 0.003,
+      "loss": 4.4566,
+      "step": 1070
+    },
+    {
+      "epoch": 0.01071,
+      "grad_norm": 0.7799052277350242,
+      "learning_rate": 0.003,
+      "loss": 4.3774,
+      "step": 1071
+    },
+    {
+      "epoch": 0.01072,
+      "grad_norm": 0.7960071687741883,
+      "learning_rate": 0.003,
+      "loss": 4.4136,
+      "step": 1072
+    },
+    {
+      "epoch": 0.01073,
+      "grad_norm": 0.6975430946737453,
+      "learning_rate": 0.003,
+      "loss": 4.4079,
+      "step": 1073
+    },
+    {
+      "epoch": 0.01074,
+      "grad_norm": 0.6813769264433432,
+      "learning_rate": 0.003,
+      "loss": 4.37,
+      "step": 1074
+    },
+    {
+      "epoch": 0.01075,
+      "grad_norm": 0.5971783233576602,
+      "learning_rate": 0.003,
+      "loss": 4.4063,
+      "step": 1075
+    },
+    {
+      "epoch": 0.01076,
+      "grad_norm": 0.5510713740534053,
+      "learning_rate": 0.003,
+      "loss": 4.3967,
+      "step": 1076
+    },
+    {
+      "epoch": 0.01077,
+      "grad_norm": 0.5595097233894012,
+      "learning_rate": 0.003,
+      "loss": 4.3917,
+      "step": 1077
+    },
+    {
+      "epoch": 0.01078,
+      "grad_norm": 0.6230934184673731,
+      "learning_rate": 0.003,
+      "loss": 4.3491,
+      "step": 1078
+    },
+    {
+      "epoch": 0.01079,
+      "grad_norm": 0.7178707958532615,
+      "learning_rate": 0.003,
+      "loss": 4.3825,
+      "step": 1079
+    },
+    {
+      "epoch": 0.0108,
+      "grad_norm": 0.7484447643520958,
+      "learning_rate": 0.003,
+      "loss": 4.3665,
+      "step": 1080
+    },
+    {
+      "epoch": 0.01081,
+      "grad_norm": 0.780034137982597,
+      "learning_rate": 0.003,
+      "loss": 4.3727,
+      "step": 1081
+    },
+    {
+      "epoch": 0.01082,
+      "grad_norm": 0.7403701975238451,
+      "learning_rate": 0.003,
+      "loss": 4.3894,
+      "step": 1082
+    },
+    {
+      "epoch": 0.01083,
+      "grad_norm": 0.6305537452051372,
+      "learning_rate": 0.003,
+      "loss": 4.3733,
+      "step": 1083
+    },
+    {
+      "epoch": 0.01084,
+      "grad_norm": 0.7435200213630843,
+      "learning_rate": 0.003,
+      "loss": 4.3825,
+      "step": 1084
+    },
+    {
+      "epoch": 0.01085,
+      "grad_norm": 0.7405129376016774,
+      "learning_rate": 0.003,
+      "loss": 4.3735,
+      "step": 1085
+    },
+    {
+      "epoch": 0.01086,
+      "grad_norm": 0.6037526750507048,
+      "learning_rate": 0.003,
+      "loss": 4.3449,
+      "step": 1086
+    },
+    {
+      "epoch": 0.01087,
+      "grad_norm": 0.5349574840042758,
+      "learning_rate": 0.003,
+      "loss": 4.3549,
+      "step": 1087
+    },
+    {
+      "epoch": 0.01088,
+      "grad_norm": 0.5555341529780033,
+      "learning_rate": 0.003,
+      "loss": 4.3917,
+      "step": 1088
+    },
+    {
+      "epoch": 0.01089,
+      "grad_norm": 0.6254527345250118,
+      "learning_rate": 0.003,
+      "loss": 4.3599,
+      "step": 1089
+    },
+    {
+      "epoch": 0.0109,
+      "grad_norm": 0.6828156926551208,
+      "learning_rate": 0.003,
+      "loss": 4.3884,
+      "step": 1090
+    },
+    {
+      "epoch": 0.01091,
+      "grad_norm": 0.5753021180468998,
+      "learning_rate": 0.003,
+      "loss": 4.3549,
+      "step": 1091
+    },
+    {
+      "epoch": 0.01092,
+      "grad_norm": 0.4077341561870379,
+      "learning_rate": 0.003,
+      "loss": 4.3436,
+      "step": 1092
+    },
+    {
+      "epoch": 0.01093,
+      "grad_norm": 0.47615819716416974,
+      "learning_rate": 0.003,
+      "loss": 4.3664,
+      "step": 1093
+    },
+    {
+      "epoch": 0.01094,
+      "grad_norm": 0.5845002574061751,
+      "learning_rate": 0.003,
+      "loss": 4.3553,
+      "step": 1094
+    },
+    {
+      "epoch": 0.01095,
+      "grad_norm": 0.8583788887394415,
+      "learning_rate": 0.003,
+      "loss": 4.3944,
+      "step": 1095
+    },
+    {
+      "epoch": 0.01096,
+      "grad_norm": 1.0479503783625854,
+      "learning_rate": 0.003,
+      "loss": 4.3824,
+      "step": 1096
+    },
+    {
+      "epoch": 0.01097,
+      "grad_norm": 0.9584408455564232,
+      "learning_rate": 0.003,
+      "loss": 4.3695,
+      "step": 1097
+    },
+    {
+      "epoch": 0.01098,
+      "grad_norm": 1.2545731298718157,
+      "learning_rate": 0.003,
+      "loss": 4.4068,
+      "step": 1098
+    },
+    {
+      "epoch": 0.01099,
+      "grad_norm": 0.8305012272413557,
+      "learning_rate": 0.003,
+      "loss": 4.3706,
+      "step": 1099
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 0.7311868902259275,
+      "learning_rate": 0.003,
+      "loss": 4.3677,
+      "step": 1100
+    },
+    {
+      "epoch": 0.01101,
+      "grad_norm": 0.6742597336847759,
+      "learning_rate": 0.003,
+      "loss": 4.3752,
+      "step": 1101
+    },
+    {
+      "epoch": 0.01102,
+      "grad_norm": 0.6773795133108548,
+      "learning_rate": 0.003,
+      "loss": 4.3802,
+      "step": 1102
+    },
+    {
+      "epoch": 0.01103,
+      "grad_norm": 0.6414092553075785,
+      "learning_rate": 0.003,
+      "loss": 4.3701,
+      "step": 1103
+    },
+    {
+      "epoch": 0.01104,
+      "grad_norm": 0.6378763268399013,
+      "learning_rate": 0.003,
+      "loss": 4.3589,
+      "step": 1104
+    },
+    {
+      "epoch": 0.01105,
+      "grad_norm": 0.6837675260878959,
+      "learning_rate": 0.003,
+      "loss": 4.3856,
+      "step": 1105
+    },
+    {
+      "epoch": 0.01106,
+      "grad_norm": 0.667950958790211,
+      "learning_rate": 0.003,
+      "loss": 4.3574,
+      "step": 1106
+    },
+    {
+      "epoch": 0.01107,
+      "grad_norm": 0.6082482122734244,
+      "learning_rate": 0.003,
+      "loss": 4.3623,
+      "step": 1107
+    },
+    {
+      "epoch": 0.01108,
+      "grad_norm": 0.5983800302348057,
+      "learning_rate": 0.003,
+      "loss": 4.3313,
+      "step": 1108
+    },
+    {
+      "epoch": 0.01109,
+      "grad_norm": 0.6046679697278923,
+      "learning_rate": 0.003,
+      "loss": 4.3476,
+      "step": 1109
+    },
+    {
+      "epoch": 0.0111,
+      "grad_norm": 0.6353880163110637,
+      "learning_rate": 0.003,
+      "loss": 4.3675,
+      "step": 1110
+    },
+    {
+      "epoch": 0.01111,
+      "grad_norm": 0.7049628214149826,
+      "learning_rate": 0.003,
+      "loss": 4.3473,
+      "step": 1111
+    },
+    {
+      "epoch": 0.01112,
+      "grad_norm": 0.758960503794264,
+      "learning_rate": 0.003,
+      "loss": 4.369,
+      "step": 1112
+    },
+    {
+      "epoch": 0.01113,
+      "grad_norm": 0.6091686772357772,
+      "learning_rate": 0.003,
+      "loss": 4.3335,
+      "step": 1113
+    },
+    {
+      "epoch": 0.01114,
+      "grad_norm": 0.628169460357088,
+      "learning_rate": 0.003,
+      "loss": 4.3462,
+      "step": 1114
+    },
+    {
+      "epoch": 0.01115,
+      "grad_norm": 0.7453432538813878,
+      "learning_rate": 0.003,
+      "loss": 4.3779,
+      "step": 1115
+    },
+    {
+      "epoch": 0.01116,
+      "grad_norm": 0.9959163456029771,
+      "learning_rate": 0.003,
+      "loss": 4.3771,
+      "step": 1116
+    },
+    {
+      "epoch": 0.01117,
+      "grad_norm": 0.9566018042437344,
+      "learning_rate": 0.003,
+      "loss": 4.3915,
+      "step": 1117
+    },
+    {
+      "epoch": 0.01118,
+      "grad_norm": 0.6794219206934775,
+      "learning_rate": 0.003,
+      "loss": 4.372,
+      "step": 1118
+    },
+    {
+      "epoch": 0.01119,
+      "grad_norm": 0.6027310157842346,
+      "learning_rate": 0.003,
+      "loss": 4.375,
+      "step": 1119
+    },
+    {
+      "epoch": 0.0112,
+      "grad_norm": 0.6064414406921254,
+      "learning_rate": 0.003,
+      "loss": 4.3588,
+      "step": 1120
+    },
+    {
+      "epoch": 0.01121,
+      "grad_norm": 0.643465354938861,
+      "learning_rate": 0.003,
+      "loss": 4.3691,
+      "step": 1121
+    },
+    {
+      "epoch": 0.01122,
+      "grad_norm": 0.8439491151148678,
+      "learning_rate": 0.003,
+      "loss": 4.3539,
+      "step": 1122
+    },
+    {
+      "epoch": 0.01123,
+      "grad_norm": 0.857123487934385,
+      "learning_rate": 0.003,
+      "loss": 4.345,
+      "step": 1123
+    },
+    {
+      "epoch": 0.01124,
+      "grad_norm": 0.7412296015988188,
+      "learning_rate": 0.003,
+      "loss": 4.3865,
+      "step": 1124
+    },
+    {
+      "epoch": 0.01125,
+      "grad_norm": 0.8143761816040683,
+      "learning_rate": 0.003,
+      "loss": 4.3583,
+      "step": 1125
+    },
+    {
+      "epoch": 0.01126,
+      "grad_norm": 0.741767094028342,
+      "learning_rate": 0.003,
+      "loss": 4.3618,
+      "step": 1126
+    },
+    {
+      "epoch": 0.01127,
+      "grad_norm": 0.737420855232079,
+      "learning_rate": 0.003,
+      "loss": 4.3948,
+      "step": 1127
+    },
+    {
+      "epoch": 0.01128,
+      "grad_norm": 0.7656500975813709,
+      "learning_rate": 0.003,
+      "loss": 4.372,
+      "step": 1128
+    },
+    {
+      "epoch": 0.01129,
+      "grad_norm": 0.8248659664813461,
+      "learning_rate": 0.003,
+      "loss": 4.3685,
+      "step": 1129
+    },
+    {
+      "epoch": 0.0113,
+      "grad_norm": 0.7756542452356111,
+      "learning_rate": 0.003,
+      "loss": 4.3447,
+      "step": 1130
+    },
+    {
+      "epoch": 0.01131,
+      "grad_norm": 0.7965849423705411,
+      "learning_rate": 0.003,
+      "loss": 4.3615,
+      "step": 1131
+    },
+    {
+      "epoch": 0.01132,
+      "grad_norm": 0.718692737781693,
+      "learning_rate": 0.003,
+      "loss": 4.3824,
+      "step": 1132
+    },
+    {
+      "epoch": 0.01133,
+      "grad_norm": 0.6500905111846983,
+      "learning_rate": 0.003,
+      "loss": 4.34,
+      "step": 1133
+    },
+    {
+      "epoch": 0.01134,
+      "grad_norm": 0.6052834019155894,
+      "learning_rate": 0.003,
+      "loss": 4.3525,
+      "step": 1134
+    },
+    {
+      "epoch": 0.01135,
+      "grad_norm": 0.57728723025885,
+      "learning_rate": 0.003,
+      "loss": 4.3767,
+      "step": 1135
+    },
+    {
+      "epoch": 0.01136,
+      "grad_norm": 0.6622614298653483,
+      "learning_rate": 0.003,
+      "loss": 4.3514,
+      "step": 1136
+    },
+    {
+      "epoch": 0.01137,
+      "grad_norm": 0.6945355696134266,
+      "learning_rate": 0.003,
+      "loss": 4.3784,
+      "step": 1137
+    },
+    {
+      "epoch": 0.01138,
+      "grad_norm": 0.7423434054596434,
+      "learning_rate": 0.003,
+      "loss": 4.3361,
+      "step": 1138
+    },
+    {
+      "epoch": 0.01139,
+      "grad_norm": 0.7678127530736037,
+      "learning_rate": 0.003,
+      "loss": 4.3463,
+      "step": 1139
+    },
+    {
+      "epoch": 0.0114,
+      "grad_norm": 0.7470822118949633,
+      "learning_rate": 0.003,
+      "loss": 4.3322,
+      "step": 1140
+    },
+    {
+      "epoch": 0.01141,
+      "grad_norm": 0.903198095826591,
+      "learning_rate": 0.003,
+      "loss": 4.3608,
+      "step": 1141
+    },
+    {
+      "epoch": 0.01142,
+      "grad_norm": 1.0797547624524322,
+      "learning_rate": 0.003,
+      "loss": 4.3612,
+      "step": 1142
+    },
+    {
+      "epoch": 0.01143,
+      "grad_norm": 0.7877548466868329,
+      "learning_rate": 0.003,
+      "loss": 4.3542,
+      "step": 1143
+    },
+    {
+      "epoch": 0.01144,
+      "grad_norm": 0.8147882944795127,
+      "learning_rate": 0.003,
+      "loss": 4.3253,
+      "step": 1144
+    },
+    {
+      "epoch": 0.01145,
+      "grad_norm": 0.7148112106554612,
+      "learning_rate": 0.003,
+      "loss": 4.3655,
+      "step": 1145
+    },
+    {
+      "epoch": 0.01146,
+      "grad_norm": 0.6580426381015894,
+      "learning_rate": 0.003,
+      "loss": 4.3445,
+      "step": 1146
+    },
+    {
+      "epoch": 0.01147,
+      "grad_norm": 0.7448256416265617,
+      "learning_rate": 0.003,
+      "loss": 4.3367,
+      "step": 1147
+    },
+    {
+      "epoch": 0.01148,
+      "grad_norm": 0.6968832253815664,
+      "learning_rate": 0.003,
+      "loss": 4.3386,
+      "step": 1148
+    },
+    {
+      "epoch": 0.01149,
+      "grad_norm": 0.627190656091196,
+      "learning_rate": 0.003,
+      "loss": 4.3527,
+      "step": 1149
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 0.6739971655272956,
+      "learning_rate": 0.003,
+      "loss": 4.3445,
+      "step": 1150
+    },
+    {
+      "epoch": 0.01151,
+      "grad_norm": 0.7518825421443525,
+      "learning_rate": 0.003,
+      "loss": 4.3445,
+      "step": 1151
+    },
+    {
+      "epoch": 0.01152,
+      "grad_norm": 0.7628403894732813,
+      "learning_rate": 0.003,
+      "loss": 4.3529,
+      "step": 1152
+    },
+    {
+      "epoch": 0.01153,
+      "grad_norm": 0.726109999889872,
+      "learning_rate": 0.003,
+      "loss": 4.3668,
+      "step": 1153
+    },
+    {
+      "epoch": 0.01154,
+      "grad_norm": 0.7747132516080261,
+      "learning_rate": 0.003,
+      "loss": 4.3438,
+      "step": 1154
+    },
+    {
+      "epoch": 0.01155,
+      "grad_norm": 0.8180276502719757,
+      "learning_rate": 0.003,
+      "loss": 4.3369,
+      "step": 1155
+    },
+    {
+      "epoch": 0.01156,
+      "grad_norm": 0.7757994684934312,
+      "learning_rate": 0.003,
+      "loss": 4.3405,
+      "step": 1156
+    },
+    {
+      "epoch": 0.01157,
+      "grad_norm": 0.7233155240278132,
+      "learning_rate": 0.003,
+      "loss": 4.3659,
+      "step": 1157
+    },
+    {
+      "epoch": 0.01158,
+      "grad_norm": 0.6552894912204377,
+      "learning_rate": 0.003,
+      "loss": 4.332,
+      "step": 1158
+    },
+    {
+      "epoch": 0.01159,
+      "grad_norm": 0.7654422010962233,
+      "learning_rate": 0.003,
+      "loss": 4.3316,
+      "step": 1159
+    },
+    {
+      "epoch": 0.0116,
+      "grad_norm": 0.7129248873282286,
+      "learning_rate": 0.003,
+      "loss": 4.3462,
+      "step": 1160
+    },
+    {
+      "epoch": 0.01161,
+      "grad_norm": 0.728340877307682,
+      "learning_rate": 0.003,
+      "loss": 4.3213,
+      "step": 1161
+    },
+    {
+      "epoch": 0.01162,
+      "grad_norm": 0.6672912532139038,
+      "learning_rate": 0.003,
+      "loss": 4.3344,
+      "step": 1162
+    },
+    {
+      "epoch": 0.01163,
+      "grad_norm": 0.6087688372398936,
+      "learning_rate": 0.003,
+      "loss": 4.3512,
+      "step": 1163
+    },
+    {
+      "epoch": 0.01164,
+      "grad_norm": 0.5012921787632111,
+      "learning_rate": 0.003,
+      "loss": 4.3329,
+      "step": 1164
+    },
+    {
+      "epoch": 0.01165,
+      "grad_norm": 0.4972192156782153,
+      "learning_rate": 0.003,
+      "loss": 4.3277,
+      "step": 1165
+    },
+    {
+      "epoch": 0.01166,
+      "grad_norm": 0.48804976467805194,
+      "learning_rate": 0.003,
+      "loss": 4.3262,
+      "step": 1166
+    },
+    {
+      "epoch": 0.01167,
+      "grad_norm": 0.5481538887564317,
+      "learning_rate": 0.003,
+      "loss": 4.2952,
+      "step": 1167
+    },
+    {
+      "epoch": 0.01168,
+      "grad_norm": 0.6090792597366076,
+      "learning_rate": 0.003,
+      "loss": 4.3026,
+      "step": 1168
+    },
+    {
+      "epoch": 0.01169,
+      "grad_norm": 0.653085333922602,
+      "learning_rate": 0.003,
+      "loss": 4.347,
+      "step": 1169
+    },
+    {
+      "epoch": 0.0117,
+      "grad_norm": 0.62436544437129,
+      "learning_rate": 0.003,
+      "loss": 4.3309,
+      "step": 1170
+    },
+    {
+      "epoch": 0.01171,
+      "grad_norm": 0.654802486268078,
+      "learning_rate": 0.003,
+      "loss": 4.3143,
+      "step": 1171
+    },
+    {
+      "epoch": 0.01172,
+      "grad_norm": 0.6246624927886145,
+      "learning_rate": 0.003,
+      "loss": 4.3038,
+      "step": 1172
+    },
+    {
+      "epoch": 0.01173,
+      "grad_norm": 0.6813303457409844,
+      "learning_rate": 0.003,
+      "loss": 4.3382,
+      "step": 1173
+    },
+    {
+      "epoch": 0.01174,
+      "grad_norm": 0.7721136771646261,
+      "learning_rate": 0.003,
+      "loss": 4.3203,
+      "step": 1174
+    },
+    {
+      "epoch": 0.01175,
+      "grad_norm": 0.823782632958531,
+      "learning_rate": 0.003,
+      "loss": 4.3152,
+      "step": 1175
+    },
+    {
+      "epoch": 0.01176,
+      "grad_norm": 0.9763749462159719,
+      "learning_rate": 0.003,
+      "loss": 4.3509,
+      "step": 1176
+    },
+    {
+      "epoch": 0.01177,
+      "grad_norm": 1.0894015492716702,
+      "learning_rate": 0.003,
+      "loss": 4.3444,
+      "step": 1177
+    },
+    {
+      "epoch": 0.01178,
+      "grad_norm": 0.8156537670820067,
+      "learning_rate": 0.003,
+      "loss": 4.3473,
+      "step": 1178
+    },
+    {
+      "epoch": 0.01179,
+      "grad_norm": 0.8529362760715462,
+      "learning_rate": 0.003,
+      "loss": 4.3276,
+      "step": 1179
+    },
+    {
+      "epoch": 0.0118,
+      "grad_norm": 0.857392382910556,
+      "learning_rate": 0.003,
+      "loss": 4.395,
+      "step": 1180
+    },
+    {
+      "epoch": 0.01181,
+      "grad_norm": 0.8128436446210868,
+      "learning_rate": 0.003,
+      "loss": 4.3809,
+      "step": 1181
+    },
+    {
+      "epoch": 0.01182,
+      "grad_norm": 0.911584010746237,
+      "learning_rate": 0.003,
+      "loss": 4.3433,
+      "step": 1182
+    },
+    {
+      "epoch": 0.01183,
+      "grad_norm": 0.8460613119287651,
+      "learning_rate": 0.003,
+      "loss": 4.337,
+      "step": 1183
+    },
+    {
+      "epoch": 0.01184,
+      "grad_norm": 0.7730717809982315,
+      "learning_rate": 0.003,
+      "loss": 4.3753,
+      "step": 1184
+    },
+    {
+      "epoch": 0.01185,
+      "grad_norm": 0.7718355908910234,
+      "learning_rate": 0.003,
+      "loss": 4.3449,
+      "step": 1185
+    },
+    {
+      "epoch": 0.01186,
+      "grad_norm": 0.749856576054878,
+      "learning_rate": 0.003,
+      "loss": 4.3669,
+      "step": 1186
+    },
+    {
+      "epoch": 0.01187,
+      "grad_norm": 0.6703031442863907,
+      "learning_rate": 0.003,
+      "loss": 4.33,
+      "step": 1187
+    },
+    {
+      "epoch": 0.01188,
+      "grad_norm": 0.6597409636847196,
+      "learning_rate": 0.003,
+      "loss": 4.3387,
+      "step": 1188
+    },
+    {
+      "epoch": 0.01189,
+      "grad_norm": 0.7528540145175824,
+      "learning_rate": 0.003,
+      "loss": 4.3431,
+      "step": 1189
+    },
+    {
+      "epoch": 0.0119,
+      "grad_norm": 0.7990197433249405,
+      "learning_rate": 0.003,
+      "loss": 4.3273,
+      "step": 1190
+    },
+    {
+      "epoch": 0.01191,
+      "grad_norm": 0.8012185403960035,
+      "learning_rate": 0.003,
+      "loss": 4.3081,
+      "step": 1191
+    },
+    {
+      "epoch": 0.01192,
+      "grad_norm": 0.7571323847445259,
+      "learning_rate": 0.003,
+      "loss": 4.3381,
+      "step": 1192
+    },
+    {
+      "epoch": 0.01193,
+      "grad_norm": 0.6624809456477357,
+      "learning_rate": 0.003,
+      "loss": 4.3371,
+      "step": 1193
+    },
+    {
+      "epoch": 0.01194,
+      "grad_norm": 0.7303795208901569,
+      "learning_rate": 0.003,
+      "loss": 4.3395,
+      "step": 1194
+    },
+    {
+      "epoch": 0.01195,
+      "grad_norm": 0.7841704264970735,
+      "learning_rate": 0.003,
+      "loss": 4.3211,
+      "step": 1195
+    },
+    {
+      "epoch": 0.01196,
+      "grad_norm": 0.8218420512118195,
+      "learning_rate": 0.003,
+      "loss": 4.3286,
+      "step": 1196
+    },
+    {
+      "epoch": 0.01197,
+      "grad_norm": 0.6346054129298183,
+      "learning_rate": 0.003,
+      "loss": 4.2874,
+      "step": 1197
+    },
+    {
+      "epoch": 0.01198,
+      "grad_norm": 0.648198494061273,
+      "learning_rate": 0.003,
+      "loss": 4.3575,
+      "step": 1198
+    },
+    {
+      "epoch": 0.01199,
+      "grad_norm": 0.719551428234481,
+      "learning_rate": 0.003,
+      "loss": 4.3378,
+      "step": 1199
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 0.7355036739818444,
+      "learning_rate": 0.003,
+      "loss": 4.3295,
+      "step": 1200
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 100000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 25,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.75723714265088e+16,
+  "train_batch_size": 256,
+  "trial_name": null,
+  "trial_params": null
+}