diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,9018 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 15.04,
+  "eval_steps": 500000000,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "crossentropy": 2.7819188833236694,
+      "epoch": 0.001,
+      "grad_norm": 2.296875,
+      "learning_rate": 2e-05,
+      "loss": 55.4633,
+      "step": 1
+    },
+    {
+      "crossentropy": 2.886364698410034,
+      "epoch": 0.002,
+      "grad_norm": 2.4375,
+      "learning_rate": 4e-05,
+      "loss": 57.6593,
+      "step": 2
+    },
+    {
+      "crossentropy": 2.610915184020996,
+      "epoch": 0.003,
+      "grad_norm": 2.265625,
+      "learning_rate": 6e-05,
+      "loss": 54.4112,
+      "step": 3
+    },
+    {
+      "crossentropy": 2.8315508365631104,
+      "epoch": 0.004,
+      "grad_norm": 2.3125,
+      "learning_rate": 8e-05,
+      "loss": 57.7924,
+      "step": 4
+    },
+    {
+      "crossentropy": 2.8231977224349976,
+      "epoch": 0.005,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.0001,
+      "loss": 57.4897,
+      "step": 5
+    },
+    {
+      "crossentropy": 2.8636255264282227,
+      "epoch": 0.006,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.00012,
+      "loss": 57.0583,
+      "step": 6
+    },
+    {
+      "crossentropy": 2.8325681686401367,
+      "epoch": 0.007,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.00014000000000000001,
+      "loss": 56.1569,
+      "step": 7
+    },
+    {
+      "crossentropy": 2.788058638572693,
+      "epoch": 0.008,
+      "grad_norm": 1.734375,
+      "learning_rate": 0.00016,
+      "loss": 56.0723,
+      "step": 8
+    },
+    {
+      "crossentropy": 2.860862612724304,
+      "epoch": 0.009,
+      "grad_norm": 1.5703125,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 56.7063,
+      "step": 9
+    },
+    {
+      "crossentropy": 2.876722812652588,
+      "epoch": 0.01,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.0002,
+      "loss": 56.7833,
+      "step": 10
+    },
+    {
+      "crossentropy": 2.8933022022247314,
+      "epoch": 0.011,
+      "grad_norm": 1.25,
+      "learning_rate": 0.00022,
+      "loss": 57.416,
+      "step": 11
+    },
+    {
+      "crossentropy": 2.9026139974594116,
+      "epoch": 0.012,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.00024,
+      "loss": 57.1726,
+      "step": 12
+    },
+    {
+      "crossentropy": 2.8697686195373535,
+      "epoch": 0.013,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00026000000000000003,
+      "loss": 57.0761,
+      "step": 13
+    },
+    {
+      "crossentropy": 2.823047637939453,
+      "epoch": 0.014,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00028000000000000003,
+      "loss": 57.1636,
+      "step": 14
+    },
+    {
+      "crossentropy": 2.81307590007782,
+      "epoch": 0.015,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0003,
+      "loss": 56.68,
+      "step": 15
+    },
+    {
+      "crossentropy": 2.893505334854126,
+      "epoch": 0.016,
+      "grad_norm": 0.953125,
+      "grad_norm_var": 0.33952433268229165,
+      "learning_rate": 0.00032,
+      "loss": 56.6504,
+      "step": 16
+    },
+    {
+      "crossentropy": 2.7856497764587402,
+      "epoch": 0.017,
+      "grad_norm": 0.9375,
+      "grad_norm_var": 0.33678080240885416,
+      "learning_rate": 0.00034,
+      "loss": 56.4876,
+      "step": 17
+    },
+    {
+      "crossentropy": 2.87225878238678,
+      "epoch": 0.018,
+      "grad_norm": 0.86328125,
+      "grad_norm_var": 0.3073923110961914,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 57.2457,
+      "step": 18
+    },
+    {
+      "crossentropy": 2.8507845401763916,
+      "epoch": 0.019,
+      "grad_norm": 0.87890625,
+      "grad_norm_var": 0.27884089152018227,
+      "learning_rate": 0.00038,
+      "loss": 56.671,
+      "step": 19
+    },
+    {
+      "crossentropy": 2.923235774040222,
+      "epoch": 0.02,
+      "grad_norm": 0.85546875,
+      "grad_norm_var": 0.2293008804321289,
+      "learning_rate": 0.0004,
+      "loss": 57.6122,
+      "step": 20
+    },
+    {
+      "crossentropy": 2.7433438301086426,
+      "epoch": 0.021,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.1661752700805664,
+      "learning_rate": 0.00042,
+      "loss": 55.3607,
+      "step": 21
+    },
+    {
+      "crossentropy": 2.8193527460098267,
+      "epoch": 0.022,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.11674779256184896,
+      "learning_rate": 0.00044,
+      "loss": 55.6994,
+      "step": 22
+    },
+    {
+      "crossentropy": 2.7718470096588135,
+      "epoch": 0.023,
+      "grad_norm": 0.8125,
+      "grad_norm_var": 0.07935994466145833,
+      "learning_rate": 0.00046,
+      "loss": 56.1888,
+      "step": 23
+    },
+    {
+      "crossentropy": 2.887826681137085,
+      "epoch": 0.024,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.047818756103515624,
+      "learning_rate": 0.00048,
+      "loss": 58.3357,
+      "step": 24
+    },
+    {
+      "crossentropy": 2.8159505128860474,
+      "epoch": 0.025,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.02467625935872396,
+      "learning_rate": 0.0005,
+      "loss": 56.4528,
+      "step": 25
+    },
+    {
+      "crossentropy": 2.786083936691284,
+      "epoch": 0.026,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.01730187733968099,
+      "learning_rate": 0.0005200000000000001,
+      "loss": 55.865,
+      "step": 26
+    },
+    {
+      "crossentropy": 2.878562092781067,
+      "epoch": 0.027,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.009134928385416666,
+      "learning_rate": 0.00054,
+      "loss": 57.3279,
+      "step": 27
+    },
+    {
+      "crossentropy": 2.8583030700683594,
+      "epoch": 0.028,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.005572954813639323,
+      "learning_rate": 0.0005600000000000001,
+      "loss": 56.4947,
+      "step": 28
+    },
+    {
+      "crossentropy": 2.867052912712097,
+      "epoch": 0.029,
+      "grad_norm": 0.828125,
+      "grad_norm_var": 0.003748003641764323,
+      "learning_rate": 0.00058,
+      "loss": 58.2363,
+      "step": 29
+    },
+    {
+      "crossentropy": 2.880504012107849,
+      "epoch": 0.03,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0036615371704101563,
+      "learning_rate": 0.0006,
+      "loss": 57.2681,
+      "step": 30
+    },
+    {
+      "crossentropy": 2.8287140130996704,
+      "epoch": 0.031,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.003290239969889323,
+      "learning_rate": 0.00062,
+      "loss": 56.6969,
+      "step": 31
+    },
+    {
+      "crossentropy": 2.8504087924957275,
+      "epoch": 0.032,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.002397600809733073,
+      "learning_rate": 0.00064,
+      "loss": 57.0567,
+      "step": 32
+    },
+    {
+      "crossentropy": 2.830446481704712,
+      "epoch": 0.033,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0013933817545572916,
+      "learning_rate": 0.00066,
+      "loss": 55.9364,
+      "step": 33
+    },
+    {
+      "crossentropy": 2.805998682975769,
+      "epoch": 0.034,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0011288960774739583,
+      "learning_rate": 0.00068,
+      "loss": 57.0045,
+      "step": 34
+    },
+    {
+      "crossentropy": 2.8322499990463257,
+      "epoch": 0.035,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0006845474243164062,
+      "learning_rate": 0.0007,
+      "loss": 56.6746,
+      "step": 35
+    },
+    {
+      "crossentropy": 2.978684902191162,
+      "epoch": 0.036,
+      "grad_norm": 0.83203125,
+      "grad_norm_var": 0.0005121231079101562,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 58.4334,
+      "step": 36
+    },
+    {
+      "crossentropy": 2.847362518310547,
+      "epoch": 0.037,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.000498199462890625,
+      "learning_rate": 0.00074,
+      "loss": 56.7639,
+      "step": 37
+    },
+    {
+      "crossentropy": 2.8229883909225464,
+      "epoch": 0.038,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0005136489868164063,
+      "learning_rate": 0.00076,
+      "loss": 55.6415,
+      "step": 38
+    },
+    {
+      "crossentropy": 2.813721776008606,
+      "epoch": 0.039,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005645116170247395,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 58.0355,
+      "step": 39
+    },
+    {
+      "crossentropy": 2.755218267440796,
+      "epoch": 0.04,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006589253743489583,
+      "learning_rate": 0.0008,
+      "loss": 54.858,
+      "step": 40
+    },
+    {
+      "crossentropy": 2.842182993888855,
+      "epoch": 0.041,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0006825129191080729,
+      "learning_rate": 0.00082,
+      "loss": 56.0784,
+      "step": 41
+    },
+    {
+      "crossentropy": 2.9310812950134277,
+      "epoch": 0.042,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0006772359212239583,
+      "learning_rate": 0.00084,
+      "loss": 58.2136,
+      "step": 42
+    },
+    {
+      "crossentropy": 2.890253186225891,
+      "epoch": 0.043,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006548563639322916,
+      "learning_rate": 0.00086,
+      "loss": 56.9909,
+      "step": 43
+    },
+    {
+      "crossentropy": 2.967428684234619,
+      "epoch": 0.044,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0007023493448893229,
+      "learning_rate": 0.00088,
+      "loss": 56.9546,
+      "step": 44
+    },
+    {
+      "crossentropy": 2.779818534851074,
+      "epoch": 0.045,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005294164021809896,
+      "learning_rate": 0.0009000000000000001,
+      "loss": 56.4747,
+      "step": 45
+    },
+    {
+      "crossentropy": 2.9322011470794678,
+      "epoch": 0.046,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0005736668904622396,
+      "learning_rate": 0.00092,
+      "loss": 58.0048,
+      "step": 46
+    },
+    {
+      "crossentropy": 2.796338438987732,
+      "epoch": 0.047,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005777994791666667,
+      "learning_rate": 0.00094,
+      "loss": 56.074,
+      "step": 47
+    },
+    {
+      "crossentropy": 2.8237509727478027,
+      "epoch": 0.048,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005355199178059896,
+      "learning_rate": 0.00096,
+      "loss": 56.0959,
+      "step": 48
+    },
+    {
+      "crossentropy": 2.9482001066207886,
+      "epoch": 0.049,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005673726399739583,
+      "learning_rate": 0.00098,
+      "loss": 57.4375,
+      "step": 49
+    },
+    {
+      "crossentropy": 2.8411340713500977,
+      "epoch": 0.05,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005731582641601562,
+      "learning_rate": 0.001,
+      "loss": 56.8031,
+      "step": 50
+    },
+    {
+      "crossentropy": 2.8085217475891113,
+      "epoch": 0.051,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.0006284077962239583,
+      "learning_rate": 0.0009989473684210526,
+      "loss": 56.6429,
+      "step": 51
+    },
+    {
+      "crossentropy": 2.884308099746704,
+      "epoch": 0.052,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0004330952962239583,
+      "learning_rate": 0.0009978947368421054,
+      "loss": 57.8023,
+      "step": 52
+    },
+    {
+      "crossentropy": 2.89777672290802,
+      "epoch": 0.053,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00043277740478515626,
+      "learning_rate": 0.000996842105263158,
+      "loss": 58.1485,
+      "step": 53
+    },
+    {
+      "crossentropy": 2.833405017852783,
+      "epoch": 0.054,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00036869049072265627,
+      "learning_rate": 0.0009957894736842105,
+      "loss": 56.4035,
+      "step": 54
+    },
+    {
+      "crossentropy": 2.898597002029419,
+      "epoch": 0.055,
+      "grad_norm": 0.8125,
+      "grad_norm_var": 0.00042362213134765624,
+      "learning_rate": 0.000994736842105263,
+      "loss": 58.3541,
+      "step": 55
+    },
+    {
+      "crossentropy": 2.83421790599823,
+      "epoch": 0.056,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003679911295572917,
+      "learning_rate": 0.0009936842105263159,
+      "loss": 56.4192,
+      "step": 56
+    },
+    {
+      "crossentropy": 2.8999528884887695,
+      "epoch": 0.057,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0003794352213541667,
+      "learning_rate": 0.0009926315789473685,
+      "loss": 58.0858,
+      "step": 57
+    },
+    {
+      "crossentropy": 2.992220401763916,
+      "epoch": 0.058,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00040384928385416664,
+      "learning_rate": 0.000991578947368421,
+      "loss": 59.5004,
+      "step": 58
+    },
+    {
+      "crossentropy": 2.8451712131500244,
+      "epoch": 0.059,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00046564737955729165,
+      "learning_rate": 0.0009905263157894738,
+      "loss": 56.6674,
+      "step": 59
+    },
+    {
+      "crossentropy": 2.8718665838241577,
+      "epoch": 0.06,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.00041478474934895834,
+      "learning_rate": 0.0009894736842105264,
+      "loss": 56.9806,
+      "step": 60
+    },
+    {
+      "crossentropy": 2.9184417724609375,
+      "epoch": 0.061,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004170735677083333,
+      "learning_rate": 0.000988421052631579,
+      "loss": 57.6043,
+      "step": 61
+    },
+    {
+      "crossentropy": 2.834872841835022,
+      "epoch": 0.062,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00044854482014973957,
+      "learning_rate": 0.0009873684210526315,
+      "loss": 55.8379,
+      "step": 62
+    },
+    {
+      "crossentropy": 2.9252291917800903,
+      "epoch": 0.063,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004607518513997396,
+      "learning_rate": 0.0009863157894736843,
+      "loss": 57.7051,
+      "step": 63
+    },
+    {
+      "crossentropy": 2.7089247703552246,
+      "epoch": 0.064,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0004636128743489583,
+      "learning_rate": 0.000985263157894737,
+      "loss": 55.8377,
+      "step": 64
+    },
+    {
+      "crossentropy": 2.8316437005996704,
+      "epoch": 1.001,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00044040679931640626,
+      "learning_rate": 0.0009842105263157895,
+      "loss": 56.5235,
+      "step": 65
+    },
+    {
+      "crossentropy": 2.881982922554016,
+      "epoch": 1.002,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.00037333170572916664,
+      "learning_rate": 0.000983157894736842,
+      "loss": 56.9903,
+      "step": 66
+    },
+    {
+      "crossentropy": 2.8200029134750366,
+      "epoch": 1.003,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0004017512003580729,
+      "learning_rate": 0.0009821052631578948,
+      "loss": 56.6703,
+      "step": 67
+    },
+    {
+      "crossentropy": 2.7705549001693726,
+      "epoch": 1.004,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0004012425740559896,
+      "learning_rate": 0.0009810526315789474,
+      "loss": 55.9015,
+      "step": 68
+    },
+    {
+      "crossentropy": 2.8487346172332764,
+      "epoch": 1.005,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005208333333333333,
+      "learning_rate": 0.00098,
+      "loss": 56.4856,
+      "step": 69
+    },
+    {
+      "crossentropy": 2.809660792350769,
+      "epoch": 1.006,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00041802724202473957,
+      "learning_rate": 0.0009789473684210528,
+      "loss": 57.3916,
+      "step": 70
+    },
+    {
+      "crossentropy": 2.8564740419387817,
+      "epoch": 1.007,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.00041681925455729166,
+      "learning_rate": 0.0009778947368421053,
+      "loss": 55.8133,
+      "step": 71
+    },
+    {
+      "crossentropy": 2.8648808002471924,
+      "epoch": 1.008,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.00038859049479166664,
+      "learning_rate": 0.000976842105263158,
+      "loss": 57.4218,
+      "step": 72
+    },
+    {
+      "crossentropy": 2.7834339141845703,
+      "epoch": 1.009,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0003794352213541667,
+      "learning_rate": 0.0009757894736842106,
+      "loss": 56.7101,
+      "step": 73
+    },
+    {
+      "crossentropy": 2.9197006225585938,
+      "epoch": 1.01,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0003371556599934896,
+      "learning_rate": 0.0009747368421052632,
+      "loss": 58.2303,
+      "step": 74
+    },
+    {
+      "crossentropy": 2.8555904626846313,
+      "epoch": 1.011,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0003509521484375,
+      "learning_rate": 0.0009736842105263158,
+      "loss": 56.6662,
+      "step": 75
+    },
+    {
+      "crossentropy": 2.833241581916809,
+      "epoch": 1.012,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00035444895426432293,
+      "learning_rate": 0.0009726315789473684,
+      "loss": 56.8305,
+      "step": 76
+    },
+    {
+      "crossentropy": 2.913132429122925,
+      "epoch": 1.013,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.0003814697265625,
+      "learning_rate": 0.0009715789473684211,
+      "loss": 57.7847,
+      "step": 77
+    },
+    {
+      "crossentropy": 2.794333338737488,
+      "epoch": 1.014,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00044854482014973957,
+      "learning_rate": 0.0009705263157894737,
+      "loss": 55.5186,
+      "step": 78
+    },
+    {
+      "crossentropy": 2.7918150424957275,
+      "epoch": 1.015,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00044530232747395834,
+      "learning_rate": 0.0009694736842105263,
+      "loss": 54.9685,
+      "step": 79
+    },
+    {
+      "crossentropy": 2.9973483085632324,
+      "epoch": 1.016,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0004668553670247396,
+      "learning_rate": 0.000968421052631579,
+      "loss": 57.5043,
+      "step": 80
+    },
+    {
+      "crossentropy": 2.6424334049224854,
+      "epoch": 1.017,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004668553670247396,
+      "learning_rate": 0.0009673684210526316,
+      "loss": 53.7855,
+      "step": 81
+    },
+    {
+      "crossentropy": 2.901629686355591,
+      "epoch": 1.018,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0005828221638997396,
+      "learning_rate": 0.0009663157894736843,
+      "loss": 56.3773,
+      "step": 82
+    },
+    {
+      "crossentropy": 2.9043585062026978,
+      "epoch": 1.019,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005416234334309895,
+      "learning_rate": 0.0009652631578947368,
+      "loss": 56.4689,
+      "step": 83
+    },
+    {
+      "crossentropy": 2.89982807636261,
+      "epoch": 1.02,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0005360921223958333,
+      "learning_rate": 0.0009642105263157895,
+      "loss": 55.5952,
+      "step": 84
+    },
+    {
+      "crossentropy": 2.795522093772888,
+      "epoch": 1.021,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00044530232747395834,
+      "learning_rate": 0.0009631578947368421,
+      "loss": 55.937,
+      "step": 85
+    },
+    {
+      "crossentropy": 2.879517436027527,
+      "epoch": 1.022,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004439671834309896,
+      "learning_rate": 0.0009621052631578947,
+      "loss": 57.3097,
+      "step": 86
+    },
+    {
+      "crossentropy": 2.977717161178589,
+      "epoch": 1.023,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00044530232747395834,
+      "learning_rate": 0.0009610526315789475,
+      "loss": 58.7542,
+      "step": 87
+    },
+    {
+      "crossentropy": 2.6625332832336426,
+      "epoch": 1.024,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00043919881184895836,
+      "learning_rate": 0.00096,
+      "loss": 55.1611,
+      "step": 88
+    },
+    {
+      "crossentropy": 2.8213740587234497,
+      "epoch": 1.025,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0004577000935872396,
+      "learning_rate": 0.0009589473684210527,
+      "loss": 56.8185,
+      "step": 89
+    },
+    {
+      "crossentropy": 2.746438980102539,
+      "epoch": 1.026,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0004750569661458333,
+      "learning_rate": 0.0009578947368421053,
+      "loss": 56.6947,
+      "step": 90
+    },
+    {
+      "crossentropy": 2.6565908193588257,
+      "epoch": 1.027,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004607518513997396,
+      "learning_rate": 0.000956842105263158,
+      "loss": 54.1427,
+      "step": 91
+    },
+    {
+      "crossentropy": 2.88510262966156,
+      "epoch": 1.028,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0005299886067708333,
+      "learning_rate": 0.0009557894736842105,
+      "loss": 57.024,
+      "step": 92
+    },
+    {
+      "crossentropy": 2.8993020057678223,
+      "epoch": 1.029,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.00044498443603515627,
+      "learning_rate": 0.0009547368421052631,
+      "loss": 57.8516,
+      "step": 93
+    },
+    {
+      "crossentropy": 2.856235146522522,
+      "epoch": 1.03,
+      "grad_norm": 0.8203125,
+      "grad_norm_var": 0.000510406494140625,
+      "learning_rate": 0.0009536842105263158,
+      "loss": 57.8481,
+      "step": 94
+    },
+    {
+      "crossentropy": 2.766387939453125,
+      "epoch": 1.031,
+      "grad_norm": 0.8359375,
+      "grad_norm_var": 0.0007364273071289063,
+      "learning_rate": 0.0009526315789473684,
+      "loss": 57.7393,
+      "step": 95
+    },
+    {
+      "crossentropy": 2.7507987022399902,
+      "epoch": 1.032,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0007262547810872396,
+      "learning_rate": 0.0009515789473684211,
+      "loss": 55.9701,
+      "step": 96
+    },
+    {
+      "crossentropy": 2.862091898918152,
+      "epoch": 1.033,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0006955464680989584,
+      "learning_rate": 0.0009505263157894737,
+      "loss": 57.3552,
+      "step": 97
+    },
+    {
+      "crossentropy": 2.942552924156189,
+      "epoch": 1.034,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0004943211873372396,
+      "learning_rate": 0.0009494736842105264,
+      "loss": 56.6328,
+      "step": 98
+    },
+    {
+      "crossentropy": 2.8581180572509766,
+      "epoch": 1.035,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005063374837239583,
+      "learning_rate": 0.000948421052631579,
+      "loss": 57.5714,
+      "step": 99
+    },
+    {
+      "crossentropy": 2.811803102493286,
+      "epoch": 1.036,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005233128865559896,
+      "learning_rate": 0.0009473684210526315,
+      "loss": 56.7149,
+      "step": 100
+    },
+    {
+      "crossentropy": 2.8827518224716187,
+      "epoch": 1.037,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0005368550618489583,
+      "learning_rate": 0.0009463157894736842,
+      "loss": 56.9192,
+      "step": 101
+    },
+    {
+      "crossentropy": 2.907082676887512,
+      "epoch": 1.038,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0004973729451497396,
+      "learning_rate": 0.0009452631578947368,
+      "loss": 57.5872,
+      "step": 102
+    },
+    {
+      "crossentropy": 2.859190344810486,
+      "epoch": 1.039,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005360921223958333,
+      "learning_rate": 0.0009442105263157895,
+      "loss": 55.8119,
+      "step": 103
+    },
+    {
+      "crossentropy": 2.740812063217163,
+      "epoch": 1.04,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004851659138997396,
+      "learning_rate": 0.0009431578947368421,
+      "loss": 55.4499,
+      "step": 104
+    },
+    {
+      "crossentropy": 2.786709427833557,
+      "epoch": 1.041,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005792617797851562,
+      "learning_rate": 0.0009421052631578948,
+      "loss": 55.1188,
+      "step": 105
+    },
+    {
+      "crossentropy": 2.8872324228286743,
+      "epoch": 1.042,
+      "grad_norm": 0.828125,
+      "grad_norm_var": 0.0007100423177083333,
+      "learning_rate": 0.0009410526315789474,
+      "loss": 58.2194,
+      "step": 106
+    },
+    {
+      "crossentropy": 2.7634663581848145,
+      "epoch": 1.043,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0008664449055989583,
+      "learning_rate": 0.00094,
+      "loss": 55.4621,
+      "step": 107
+    },
+    {
+      "crossentropy": 2.874852418899536,
+      "epoch": 1.044,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0008664449055989583,
+      "learning_rate": 0.0009389473684210527,
+      "loss": 57.0353,
+      "step": 108
+    },
+    {
+      "crossentropy": 2.775438666343689,
+      "epoch": 1.045,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0008864720662434896,
+      "learning_rate": 0.0009378947368421052,
+      "loss": 56.4358,
+      "step": 109
+    },
+    {
+      "crossentropy": 2.8654950857162476,
+      "epoch": 1.046,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0008015950520833333,
+      "learning_rate": 0.0009368421052631579,
+      "loss": 57.3299,
+      "step": 110
+    },
+    {
+      "crossentropy": 2.761886715888977,
+      "epoch": 1.047,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005812962849934896,
+      "learning_rate": 0.0009357894736842105,
+      "loss": 57.1592,
+      "step": 111
+    },
+    {
+      "crossentropy": 2.8443450927734375,
+      "epoch": 1.048,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0005812962849934896,
+      "learning_rate": 0.0009347368421052633,
+      "loss": 56.3252,
+      "step": 112
+    },
+    {
+      "crossentropy": 2.7667864561080933,
+      "epoch": 1.049,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005856831868489583,
+      "learning_rate": 0.0009336842105263158,
+      "loss": 56.8155,
+      "step": 113
+    },
+    {
+      "crossentropy": 2.8398473262786865,
+      "epoch": 1.05,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0005757649739583333,
+      "learning_rate": 0.0009326315789473684,
+      "loss": 56.5225,
+      "step": 114
+    },
+    {
+      "crossentropy": 2.79670786857605,
+      "epoch": 1.051,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005757013956705729,
+      "learning_rate": 0.0009315789473684211,
+      "loss": 57.254,
+      "step": 115
+    },
+    {
+      "crossentropy": 2.774913787841797,
+      "epoch": 1.052,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0006113052368164062,
+      "learning_rate": 0.0009305263157894737,
+      "loss": 57.1821,
+      "step": 116
+    },
+    {
+      "crossentropy": 2.73682963848114,
+      "epoch": 1.053,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005757013956705729,
+      "learning_rate": 0.0009294736842105263,
+      "loss": 56.0187,
+      "step": 117
+    },
+    {
+      "crossentropy": 2.802118182182312,
+      "epoch": 1.054,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005655924479166667,
+      "learning_rate": 0.0009284210526315789,
+      "loss": 57.0246,
+      "step": 118
+    },
+    {
+      "crossentropy": 2.833091616630554,
+      "epoch": 1.055,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0007252375284830729,
+      "learning_rate": 0.0009273684210526316,
+      "loss": 56.1985,
+      "step": 119
+    },
+    {
+      "crossentropy": 2.892102599143982,
+      "epoch": 1.056,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00072021484375,
+      "learning_rate": 0.0009263157894736843,
+      "loss": 58.5873,
+      "step": 120
+    },
+    {
+      "crossentropy": 2.8181869983673096,
+      "epoch": 1.057,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000748443603515625,
+      "learning_rate": 0.0009252631578947368,
+      "loss": 55.6273,
+      "step": 121
+    },
+    {
+      "crossentropy": 2.8839467763900757,
+      "epoch": 1.058,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0005633036295572916,
+      "learning_rate": 0.0009242105263157895,
+      "loss": 57.4793,
+      "step": 122
+    },
+    {
+      "crossentropy": 2.82876455783844,
+      "epoch": 1.059,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004954020182291667,
+      "learning_rate": 0.0009231578947368421,
+      "loss": 56.5058,
+      "step": 123
+    },
+    {
+      "crossentropy": 2.8746414184570312,
+      "epoch": 1.06,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0005222956339518229,
+      "learning_rate": 0.0009221052631578948,
+      "loss": 56.7531,
+      "step": 124
+    },
+    {
+      "crossentropy": 2.8503822088241577,
+      "epoch": 1.061,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005222956339518229,
+      "learning_rate": 0.0009210526315789473,
+      "loss": 56.2036,
+      "step": 125
+    },
+    {
+      "crossentropy": 2.736844062805176,
+      "epoch": 1.062,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005497614542643229,
+      "learning_rate": 0.00092,
+      "loss": 56.6791,
+      "step": 126
+    },
+    {
+      "crossentropy": 2.8095351457595825,
+      "epoch": 1.063,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005406061808268229,
+      "learning_rate": 0.0009189473684210526,
+      "loss": 57.0038,
+      "step": 127
+    },
+    {
+      "crossentropy": 2.8585771322250366,
+      "epoch": 1.064,
+      "grad_norm": 0.82421875,
+      "grad_norm_var": 0.0007313410441080729,
+      "learning_rate": 0.0009178947368421053,
+      "loss": 57.2698,
+      "step": 128
+    },
+    {
+      "crossentropy": 2.827934741973877,
+      "epoch": 2.001,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000925445556640625,
+      "learning_rate": 0.000916842105263158,
+      "loss": 55.3524,
+      "step": 129
+    },
+    {
+      "crossentropy": 2.834007978439331,
+      "epoch": 2.002,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.00096435546875,
+      "learning_rate": 0.0009157894736842105,
+      "loss": 55.6883,
+      "step": 130
+    },
+    {
+      "crossentropy": 2.752839684486389,
+      "epoch": 2.003,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000937652587890625,
+      "learning_rate": 0.0009147368421052632,
+      "loss": 56.9396,
+      "step": 131
+    },
+    {
+      "crossentropy": 2.780048131942749,
+      "epoch": 2.004,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000937652587890625,
+      "learning_rate": 0.0009136842105263158,
+      "loss": 55.8962,
+      "step": 132
+    },
+    {
+      "crossentropy": 2.7902239561080933,
+      "epoch": 2.005,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.000937652587890625,
+      "learning_rate": 0.0009126315789473685,
+      "loss": 56.2156,
+      "step": 133
+    },
+    {
+      "crossentropy": 2.856339693069458,
+      "epoch": 2.006,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0007939020792643229,
+      "learning_rate": 0.000911578947368421,
+      "loss": 57.4976,
+      "step": 134
+    },
+    {
+      "crossentropy": 2.8085330724716187,
+      "epoch": 2.007,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0007862726847330729,
+      "learning_rate": 0.0009105263157894737,
+      "loss": 56.5116,
+      "step": 135
+    },
+    {
+      "crossentropy": 2.7982712984085083,
+      "epoch": 2.008,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0006998062133789062,
+      "learning_rate": 0.0009094736842105264,
+      "loss": 56.9157,
+      "step": 136
+    },
+    {
+      "crossentropy": 2.8740073442459106,
+      "epoch": 2.009,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000748443603515625,
+      "learning_rate": 0.000908421052631579,
+      "loss": 56.9175,
+      "step": 137
+    },
+    {
+      "crossentropy": 2.849576234817505,
+      "epoch": 2.01,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0007593154907226563,
+      "learning_rate": 0.0009073684210526316,
+      "loss": 56.2422,
+      "step": 138
+    },
+    {
+      "crossentropy": 2.764275550842285,
+      "epoch": 2.011,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.000742022196451823,
+      "learning_rate": 0.0009063157894736842,
+      "loss": 55.6254,
+      "step": 139
+    },
+    {
+      "crossentropy": 2.868152141571045,
+      "epoch": 2.012,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0007593154907226563,
+      "learning_rate": 0.0009052631578947369,
+      "loss": 55.5984,
+      "step": 140
+    },
+    {
+      "crossentropy": 2.7295836210250854,
+      "epoch": 2.013,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0006392796834309896,
+      "learning_rate": 0.0009042105263157895,
+      "loss": 55.8961,
+      "step": 141
+    },
+    {
+      "crossentropy": 2.8953765630722046,
+      "epoch": 2.014,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0007932027180989583,
+      "learning_rate": 0.0009031578947368422,
+      "loss": 57.4738,
+      "step": 142
+    },
+    {
+      "crossentropy": 2.8890058994293213,
+      "epoch": 2.015,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0006528218587239583,
+      "learning_rate": 0.0009021052631578947,
+      "loss": 56.5425,
+      "step": 143
+    },
+    {
+      "crossentropy": 2.7584519386291504,
+      "epoch": 2.016,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.000510406494140625,
+      "learning_rate": 0.0009010526315789473,
+      "loss": 55.3745,
+      "step": 144
+    },
+    {
+      "crossentropy": 2.9137370586395264,
+      "epoch": 2.017,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005365371704101563,
+      "learning_rate": 0.0009000000000000001,
+      "loss": 57.0554,
+      "step": 145
+    },
+    {
+      "crossentropy": 2.836198091506958,
+      "epoch": 2.018,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00048160552978515625,
+      "learning_rate": 0.0008989473684210527,
+      "loss": 55.6138,
+      "step": 146
+    },
+    {
+      "crossentropy": 2.8125778436660767,
+      "epoch": 2.019,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00048160552978515625,
+      "learning_rate": 0.0008978947368421053,
+      "loss": 56.4268,
+      "step": 147
+    },
+    {
+      "crossentropy": 2.9599599838256836,
+      "epoch": 2.02,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0004943211873372396,
+      "learning_rate": 0.0008968421052631579,
+      "loss": 57.8099,
+      "step": 148
+    },
+    {
+      "crossentropy": 2.9571365118026733,
+      "epoch": 2.021,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00046971638997395836,
+      "learning_rate": 0.0008957894736842106,
+      "loss": 58.1597,
+      "step": 149
+    },
+    {
+      "crossentropy": 2.8628395795822144,
+      "epoch": 2.022,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0005035400390625,
+      "learning_rate": 0.0008947368421052632,
+      "loss": 58.4094,
+      "step": 150
+    },
+    {
+      "crossentropy": 2.8677598237991333,
+      "epoch": 2.023,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005136489868164063,
+      "learning_rate": 0.0008936842105263157,
+      "loss": 57.5826,
+      "step": 151
+    },
+    {
+      "crossentropy": 2.8962247371673584,
+      "epoch": 2.024,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005185445149739583,
+      "learning_rate": 0.0008926315789473684,
+      "loss": 57.425,
+      "step": 152
+    },
+    {
+      "crossentropy": 2.8157013654708862,
+      "epoch": 2.025,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00044956207275390623,
+      "learning_rate": 0.0008915789473684211,
+      "loss": 54.5755,
+      "step": 153
+    },
+    {
+      "crossentropy": 2.874367117881775,
+      "epoch": 2.026,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00045413970947265624,
+      "learning_rate": 0.0008905263157894738,
+      "loss": 55.1411,
+      "step": 154
+    },
+    {
+      "crossentropy": 2.78195583820343,
+      "epoch": 2.027,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00045166015625,
+      "learning_rate": 0.0008894736842105263,
+      "loss": 55.6419,
+      "step": 155
+    },
+    {
+      "crossentropy": 2.7602696418762207,
+      "epoch": 2.028,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0004150390625,
+      "learning_rate": 0.000888421052631579,
+      "loss": 55.7835,
+      "step": 156
+    },
+    {
+      "crossentropy": 2.785650372505188,
+      "epoch": 2.029,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00037740071614583335,
+      "learning_rate": 0.0008873684210526316,
+      "loss": 56.6855,
+      "step": 157
+    },
+    {
+      "crossentropy": 2.9064029455184937,
+      "epoch": 2.03,
+      "grad_norm": 0.8671875,
+      "grad_norm_var": 0.0008626302083333333,
+      "learning_rate": 0.0008863157894736842,
+      "loss": 57.8286,
+      "step": 158
+    },
+    {
+      "crossentropy": 2.751049518585205,
+      "epoch": 2.031,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0008773167928059895,
+      "learning_rate": 0.0008852631578947368,
+      "loss": 55.0993,
+      "step": 159
+    },
+    {
+      "crossentropy": 2.914052367210388,
+      "epoch": 2.032,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0008298238118489583,
+      "learning_rate": 0.0008842105263157894,
+      "loss": 57.5343,
+      "step": 160
+    },
+    {
+      "crossentropy": 2.799962878227234,
+      "epoch": 2.033,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0009286880493164062,
+      "learning_rate": 0.0008831578947368422,
+      "loss": 55.8085,
+      "step": 161
+    },
+    {
+      "crossentropy": 2.844335198402405,
+      "epoch": 2.034,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0009297688802083333,
+      "learning_rate": 0.0008821052631578948,
+      "loss": 57.7179,
+      "step": 162
+    },
+    {
+      "crossentropy": 2.914214015007019,
+      "epoch": 2.035,
+      "grad_norm": 0.8125,
+      "grad_norm_var": 0.0010027567545572917,
+      "learning_rate": 0.0008810526315789475,
+      "loss": 56.2728,
+      "step": 163
+    },
+    {
+      "crossentropy": 2.8970110416412354,
+      "epoch": 2.036,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.001065508524576823,
+      "learning_rate": 0.00088,
+      "loss": 57.5116,
+      "step": 164
+    },
+    {
+      "crossentropy": 2.953248143196106,
+      "epoch": 2.037,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0010759989420572917,
+      "learning_rate": 0.0008789473684210526,
+      "loss": 59.0574,
+      "step": 165
+    },
+    {
+      "crossentropy": 2.8602688312530518,
+      "epoch": 2.038,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0010390599568684896,
+      "learning_rate": 0.0008778947368421053,
+      "loss": 56.9093,
+      "step": 166
+    },
+    {
+      "crossentropy": 2.8289544582366943,
+      "epoch": 2.039,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0010238011678059896,
+      "learning_rate": 0.0008768421052631579,
+      "loss": 55.8569,
+      "step": 167
+    },
+    {
+      "crossentropy": 2.809374690055847,
+      "epoch": 2.04,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0010355631510416666,
+      "learning_rate": 0.0008757894736842105,
+      "loss": 57.3663,
+      "step": 168
+    },
+    {
+      "crossentropy": 2.753124713897705,
+      "epoch": 2.041,
+      "grad_norm": 0.82421875,
+      "grad_norm_var": 0.0011738459269205728,
+      "learning_rate": 0.0008747368421052632,
+      "loss": 55.8763,
+      "step": 169
+    },
+    {
+      "crossentropy": 2.842271566390991,
+      "epoch": 2.042,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0010965983072916666,
+      "learning_rate": 0.0008736842105263159,
+      "loss": 56.9834,
+      "step": 170
+    },
+    {
+      "crossentropy": 2.672803044319153,
+      "epoch": 2.043,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0011281331380208334,
+      "learning_rate": 0.0008726315789473685,
+      "loss": 54.2653,
+      "step": 171
+    },
+    {
+      "crossentropy": 2.887640357017517,
+      "epoch": 2.044,
+      "grad_norm": 0.83984375,
+      "grad_norm_var": 0.0013163248697916667,
+      "learning_rate": 0.000871578947368421,
+      "loss": 57.3367,
+      "step": 172
+    },
+    {
+      "crossentropy": 2.95772647857666,
+      "epoch": 2.045,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0013132095336914062,
+      "learning_rate": 0.0008705263157894737,
+      "loss": 58.1996,
+      "step": 173
+    },
+    {
+      "crossentropy": 2.9062867164611816,
+      "epoch": 2.046,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0008646647135416666,
+      "learning_rate": 0.0008694736842105263,
+      "loss": 56.2754,
+      "step": 174
+    },
+    {
+      "crossentropy": 2.816689133644104,
+      "epoch": 2.047,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0008359273274739583,
+      "learning_rate": 0.000868421052631579,
+      "loss": 56.8248,
+      "step": 175
+    },
+    {
+      "crossentropy": 2.7304184436798096,
+      "epoch": 2.048,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0008989969889322916,
+      "learning_rate": 0.0008673684210526315,
+      "loss": 56.0621,
+      "step": 176
+    },
+    {
+      "crossentropy": 2.8292678594589233,
+      "epoch": 2.049,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0006998062133789062,
+      "learning_rate": 0.0008663157894736843,
+      "loss": 57.5406,
+      "step": 177
+    },
+    {
+      "crossentropy": 2.7578535079956055,
+      "epoch": 2.05,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0007293701171875,
+      "learning_rate": 0.0008652631578947369,
+      "loss": 55.5626,
+      "step": 178
+    },
+    {
+      "crossentropy": 2.8397985696792603,
+      "epoch": 2.051,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0006955464680989584,
+      "learning_rate": 0.0008642105263157895,
+      "loss": 56.7329,
+      "step": 179
+    },
+    {
+      "crossentropy": 2.8631927967071533,
+      "epoch": 2.052,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0006467183430989583,
+      "learning_rate": 0.0008631578947368422,
+      "loss": 56.4729,
+      "step": 180
+    },
+    {
+      "crossentropy": 2.829651117324829,
+      "epoch": 2.053,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0006398518880208333,
+      "learning_rate": 0.0008621052631578947,
+      "loss": 55.3689,
+      "step": 181
+    },
+    {
+      "crossentropy": 2.866907000541687,
+      "epoch": 2.054,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0006650288899739584,
+      "learning_rate": 0.0008610526315789474,
+      "loss": 57.1882,
+      "step": 182
+    },
+    {
+      "crossentropy": 2.833262324333191,
+      "epoch": 2.055,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0007375081380208334,
+      "learning_rate": 0.00086,
+      "loss": 57.2365,
+      "step": 183
+    },
+    {
+      "crossentropy": 2.846835970878601,
+      "epoch": 2.056,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0007222493489583334,
+      "learning_rate": 0.0008589473684210527,
+      "loss": 57.233,
+      "step": 184
+    },
+    {
+      "crossentropy": 2.7716782093048096,
+      "epoch": 2.057,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005690892537434896,
+      "learning_rate": 0.0008578947368421052,
+      "loss": 55.5457,
+      "step": 185
+    },
+    {
+      "crossentropy": 2.8875041007995605,
+      "epoch": 2.058,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0005666097005208333,
+      "learning_rate": 0.0008568421052631579,
+      "loss": 57.5646,
+      "step": 186
+    },
+    {
+      "crossentropy": 2.9292314052581787,
+      "epoch": 2.059,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005492528279622395,
+      "learning_rate": 0.0008557894736842106,
+      "loss": 58.1659,
+      "step": 187
+    },
+    {
+      "crossentropy": 2.877217411994934,
+      "epoch": 2.06,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00025424957275390627,
+      "learning_rate": 0.0008547368421052632,
+      "loss": 57.4952,
+      "step": 188
+    },
+    {
+      "crossentropy": 2.9632803201675415,
+      "epoch": 2.061,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0002684911092122396,
+      "learning_rate": 0.0008536842105263158,
+      "loss": 56.5015,
+      "step": 189
+    },
+    {
+      "crossentropy": 2.754631280899048,
+      "epoch": 2.062,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.000286865234375,
+      "learning_rate": 0.0008526315789473684,
+      "loss": 56.8266,
+      "step": 190
+    },
+    {
+      "crossentropy": 2.7981419563293457,
+      "epoch": 2.063,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0003631591796875,
+      "learning_rate": 0.0008515789473684211,
+      "loss": 55.2874,
+      "step": 191
+    },
+    {
+      "crossentropy": 2.779536008834839,
+      "epoch": 2.064,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00033512115478515623,
+      "learning_rate": 0.0008505263157894737,
+      "loss": 56.5862,
+      "step": 192
+    },
+    {
+      "crossentropy": 2.8000186681747437,
+      "epoch": 3.001,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0003203709920247396,
+      "learning_rate": 0.0008494736842105262,
+      "loss": 55.9334,
+      "step": 193
+    },
+    {
+      "crossentropy": 2.8197896480560303,
+      "epoch": 3.002,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0003615697224934896,
+      "learning_rate": 0.000848421052631579,
+      "loss": 56.534,
+      "step": 194
+    },
+    {
+      "crossentropy": 2.788080930709839,
+      "epoch": 3.003,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00039265950520833335,
+      "learning_rate": 0.0008473684210526316,
+      "loss": 55.8295,
+      "step": 195
+    },
+    {
+      "crossentropy": 2.658084750175476,
+      "epoch": 3.004,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00046183268229166665,
+      "learning_rate": 0.0008463157894736843,
+      "loss": 54.5686,
+      "step": 196
+    },
+    {
+      "crossentropy": 2.9385114908218384,
+      "epoch": 3.005,
+      "grad_norm": 0.828125,
+      "grad_norm_var": 0.0006205240885416666,
+      "learning_rate": 0.0008452631578947369,
+      "loss": 59.5615,
+      "step": 197
+    },
+    {
+      "crossentropy": 2.836456775665283,
+      "epoch": 3.006,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005666097005208333,
+      "learning_rate": 0.0008442105263157895,
+      "loss": 55.7351,
+      "step": 198
+    },
+    {
+      "crossentropy": 2.8391683101654053,
+      "epoch": 3.007,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005879084269205729,
+      "learning_rate": 0.0008431578947368421,
+      "loss": 56.6525,
+      "step": 199
+    },
+    {
+      "crossentropy": 2.840343952178955,
+      "epoch": 3.008,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0007548014322916667,
+      "learning_rate": 0.0008421052631578947,
+      "loss": 56.1,
+      "step": 200
+    },
+    {
+      "crossentropy": 2.7855753898620605,
+      "epoch": 3.009,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0007525126139322917,
+      "learning_rate": 0.0008410526315789474,
+      "loss": 55.2499,
+      "step": 201
+    },
+    {
+      "crossentropy": 2.915839195251465,
+      "epoch": 3.01,
+      "grad_norm": 0.8359375,
+      "grad_norm_var": 0.001015154520670573,
+      "learning_rate": 0.00084,
+      "loss": 58.2427,
+      "step": 202
+    },
+    {
+      "crossentropy": 2.785078763961792,
+      "epoch": 3.011,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.001015154520670573,
+      "learning_rate": 0.0008389473684210527,
+      "loss": 57.127,
+      "step": 203
+    },
+    {
+      "crossentropy": 2.7886747121810913,
+      "epoch": 3.012,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0010027567545572917,
+      "learning_rate": 0.0008378947368421053,
+      "loss": 55.473,
+      "step": 204
+    },
+    {
+      "crossentropy": 2.8934919834136963,
+      "epoch": 3.013,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0009052912394205729,
+      "learning_rate": 0.000836842105263158,
+      "loss": 56.9954,
+      "step": 205
+    },
+    {
+      "crossentropy": 2.8888269662857056,
+      "epoch": 3.014,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0008697509765625,
+      "learning_rate": 0.0008357894736842105,
+      "loss": 56.4975,
+      "step": 206
+    },
+    {
+      "crossentropy": 2.8736512660980225,
+      "epoch": 3.015,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.000888824462890625,
+      "learning_rate": 0.0008347368421052631,
+      "loss": 57.8767,
+      "step": 207
+    },
+    {
+      "crossentropy": 2.808267593383789,
+      "epoch": 3.016,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0008992513020833333,
+      "learning_rate": 0.0008336842105263158,
+      "loss": 55.929,
+      "step": 208
+    },
+    {
+      "crossentropy": 2.927086591720581,
+      "epoch": 3.017,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0009684244791666667,
+      "learning_rate": 0.0008326315789473684,
+      "loss": 57.214,
+      "step": 209
+    },
+    {
+      "crossentropy": 2.742948889732361,
+      "epoch": 3.018,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0009012222290039062,
+      "learning_rate": 0.0008315789473684212,
+      "loss": 55.9185,
+      "step": 210
+    },
+    {
+      "crossentropy": 2.879304528236389,
+      "epoch": 3.019,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.000946044921875,
+      "learning_rate": 0.0008305263157894737,
+      "loss": 57.5414,
+      "step": 211
+    },
+    {
+      "crossentropy": 2.866540312767029,
+      "epoch": 3.02,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0009027481079101563,
+      "learning_rate": 0.0008294736842105264,
+      "loss": 55.6362,
+      "step": 212
+    },
+    {
+      "crossentropy": 2.8433879613876343,
+      "epoch": 3.021,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0007104873657226562,
+      "learning_rate": 0.000828421052631579,
+      "loss": 56.9735,
+      "step": 213
+    },
+    {
+      "crossentropy": 2.8319772481918335,
+      "epoch": 3.022,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0007104873657226562,
+      "learning_rate": 0.0008273684210526315,
+      "loss": 56.2473,
+      "step": 214
+    },
+    {
+      "crossentropy": 2.781643509864807,
+      "epoch": 3.023,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0006955464680989584,
+      "learning_rate": 0.0008263157894736842,
+      "loss": 55.011,
+      "step": 215
+    },
+    {
+      "crossentropy": 2.8087105751037598,
+      "epoch": 3.024,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.0005696614583333334,
+      "learning_rate": 0.0008252631578947368,
+      "loss": 56.3954,
+      "step": 216
+    },
+    {
+      "crossentropy": 3.0052173137664795,
+      "epoch": 3.025,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0005900065104166666,
+      "learning_rate": 0.0008242105263157895,
+      "loss": 58.9994,
+      "step": 217
+    },
+    {
+      "crossentropy": 2.844808340072632,
+      "epoch": 3.026,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004018147786458333,
+      "learning_rate": 0.0008231578947368422,
+      "loss": 56.4327,
+      "step": 218
+    },
+    {
+      "crossentropy": 2.7618932723999023,
+      "epoch": 3.027,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005330403645833333,
+      "learning_rate": 0.0008221052631578948,
+      "loss": 55.3659,
+      "step": 219
+    },
+    {
+      "crossentropy": 2.800921320915222,
+      "epoch": 3.028,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006103515625,
+      "learning_rate": 0.0008210526315789474,
+      "loss": 55.6769,
+      "step": 220
+    },
+    {
+      "crossentropy": 2.724760413169861,
+      "epoch": 3.029,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0006243387858072916,
+      "learning_rate": 0.00082,
+      "loss": 54.9201,
+      "step": 221
+    },
+    {
+      "crossentropy": 2.781510591506958,
+      "epoch": 3.03,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006039937337239584,
+      "learning_rate": 0.0008189473684210527,
+      "loss": 55.7819,
+      "step": 222
+    },
+    {
+      "crossentropy": 2.776448607444763,
+      "epoch": 3.031,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0006408055623372395,
+      "learning_rate": 0.0008178947368421052,
+      "loss": 55.0119,
+      "step": 223
+    },
+    {
+      "crossentropy": 2.7627722024917603,
+      "epoch": 3.032,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0006276448567708333,
+      "learning_rate": 0.0008168421052631579,
+      "loss": 55.1523,
+      "step": 224
+    },
+    {
+      "crossentropy": 2.800988793373108,
+      "epoch": 3.033,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.000595855712890625,
+      "learning_rate": 0.0008157894736842105,
+      "loss": 56.915,
+      "step": 225
+    },
+    {
+      "crossentropy": 2.8703144788742065,
+      "epoch": 3.034,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0006692886352539062,
+      "learning_rate": 0.0008147368421052633,
+      "loss": 56.9158,
+      "step": 226
+    },
+    {
+      "crossentropy": 2.82321560382843,
+      "epoch": 3.035,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005971272786458333,
+      "learning_rate": 0.0008136842105263158,
+      "loss": 56.0276,
+      "step": 227
+    },
+    {
+      "crossentropy": 2.7239447832107544,
+      "epoch": 3.036,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005777994791666667,
+      "learning_rate": 0.0008126315789473684,
+      "loss": 55.3433,
+      "step": 228
+    },
+    {
+      "crossentropy": 2.8542327880859375,
+      "epoch": 3.037,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005971272786458333,
+      "learning_rate": 0.0008115789473684211,
+      "loss": 56.0984,
+      "step": 229
+    },
+    {
+      "crossentropy": 2.6775788068771362,
+      "epoch": 3.038,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005818684895833333,
+      "learning_rate": 0.0008105263157894737,
+      "loss": 56.2637,
+      "step": 230
+    },
+    {
+      "crossentropy": 2.818870425224304,
+      "epoch": 3.039,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006266276041666667,
+      "learning_rate": 0.0008094736842105264,
+      "loss": 56.7463,
+      "step": 231
+    },
+    {
+      "crossentropy": 2.831796646118164,
+      "epoch": 3.04,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0005772272745768229,
+      "learning_rate": 0.0008084210526315789,
+      "loss": 57.8099,
+      "step": 232
+    },
+    {
+      "crossentropy": 2.789393901824951,
+      "epoch": 3.041,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005451838175455729,
+      "learning_rate": 0.0008073684210526316,
+      "loss": 55.6014,
+      "step": 233
+    },
+    {
+      "crossentropy": 2.873471736907959,
+      "epoch": 3.042,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005388895670572917,
+      "learning_rate": 0.0008063157894736842,
+      "loss": 55.5173,
+      "step": 234
+    },
+    {
+      "crossentropy": 2.758022665977478,
+      "epoch": 3.043,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004557291666666667,
+      "learning_rate": 0.0008052631578947369,
+      "loss": 55.6085,
+      "step": 235
+    },
+    {
+      "crossentropy": 2.940276026725769,
+      "epoch": 3.044,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0003865559895833333,
+      "learning_rate": 0.0008042105263157895,
+      "loss": 57.4984,
+      "step": 236
+    },
+    {
+      "crossentropy": 2.8459670543670654,
+      "epoch": 3.045,
+      "grad_norm": 0.8203125,
+      "grad_norm_var": 0.0005065282185872396,
+      "learning_rate": 0.0008031578947368421,
+      "loss": 57.8657,
+      "step": 237
+    },
+    {
+      "crossentropy": 2.82193660736084,
+      "epoch": 3.046,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.000504302978515625,
+      "learning_rate": 0.0008021052631578948,
+      "loss": 57.3414,
+      "step": 238
+    },
+    {
+      "crossentropy": 2.692602515220642,
+      "epoch": 3.047,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00048770904541015623,
+      "learning_rate": 0.0008010526315789474,
+      "loss": 54.8326,
+      "step": 239
+    },
+    {
+      "crossentropy": 2.7393832206726074,
+      "epoch": 3.048,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0005136489868164063,
+      "learning_rate": 0.0008,
+      "loss": 55.2464,
+      "step": 240
+    },
+    {
+      "crossentropy": 2.8725425004959106,
+      "epoch": 3.049,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0006184895833333333,
+      "learning_rate": 0.0007989473684210526,
+      "loss": 56.4413,
+      "step": 241
+    },
+    {
+      "crossentropy": 3.001668334007263,
+      "epoch": 3.05,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005551656087239583,
+      "learning_rate": 0.0007978947368421052,
+      "loss": 57.9582,
+      "step": 242
+    },
+    {
+      "crossentropy": 2.92042338848114,
+      "epoch": 3.051,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005411783854166667,
+      "learning_rate": 0.000796842105263158,
+      "loss": 58.8934,
+      "step": 243
+    },
+    {
+      "crossentropy": 2.869771957397461,
+      "epoch": 3.052,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.000547027587890625,
+      "learning_rate": 0.0007957894736842105,
+      "loss": 56.5323,
+      "step": 244
+    },
+    {
+      "crossentropy": 3.0153883695602417,
+      "epoch": 3.053,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005278905232747396,
+      "learning_rate": 0.0007947368421052632,
+      "loss": 57.5091,
+      "step": 245
+    },
+    {
+      "crossentropy": 2.887974262237549,
+      "epoch": 3.054,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0005429585774739584,
+      "learning_rate": 0.0007936842105263158,
+      "loss": 57.8712,
+      "step": 246
+    },
+    {
+      "crossentropy": 2.864806890487671,
+      "epoch": 3.055,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005044937133789062,
+      "learning_rate": 0.0007926315789473685,
+      "loss": 56.7738,
+      "step": 247
+    },
+    {
+      "crossentropy": 2.8489774465560913,
+      "epoch": 3.056,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00045871734619140625,
+      "learning_rate": 0.000791578947368421,
+      "loss": 57.5012,
+      "step": 248
+    },
+    {
+      "crossentropy": 2.867731213569641,
+      "epoch": 3.057,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00043074289957682293,
+      "learning_rate": 0.0007905263157894736,
+      "loss": 57.2708,
+      "step": 249
+    },
+    {
+      "crossentropy": 2.8339585065841675,
+      "epoch": 3.058,
+      "grad_norm": 0.81640625,
+      "grad_norm_var": 0.000551287333170573,
+      "learning_rate": 0.0007894736842105263,
+      "loss": 56.8616,
+      "step": 250
+    },
+    {
+      "crossentropy": 2.9042646884918213,
+      "epoch": 3.059,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0005238215128580729,
+      "learning_rate": 0.000788421052631579,
+      "loss": 57.7641,
+      "step": 251
+    },
+    {
+      "crossentropy": 2.86965274810791,
+      "epoch": 3.06,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0005197525024414062,
+      "learning_rate": 0.0007873684210526317,
+      "loss": 56.565,
+      "step": 252
+    },
+    {
+      "crossentropy": 2.9171911478042603,
+      "epoch": 3.061,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0003916422526041667,
+      "learning_rate": 0.0007863157894736842,
+      "loss": 59.3709,
+      "step": 253
+    },
+    {
+      "crossentropy": 2.7907087802886963,
+      "epoch": 3.062,
+      "grad_norm": 0.82421875,
+      "grad_norm_var": 0.0005462010701497396,
+      "learning_rate": 0.0007852631578947369,
+      "loss": 56.1077,
+      "step": 254
+    },
+    {
+      "crossentropy": 2.810297727584839,
+      "epoch": 3.063,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0004821141560872396,
+      "learning_rate": 0.0007842105263157895,
+      "loss": 57.7457,
+      "step": 255
+    },
+    {
+      "crossentropy": 2.8475112915039062,
+      "epoch": 3.064,
+      "grad_norm": 0.82421875,
+      "grad_norm_var": 0.0005938212076822916,
+      "learning_rate": 0.000783157894736842,
+      "loss": 55.2254,
+      "step": 256
+    },
+    {
+      "crossentropy": 2.777912735939026,
+      "epoch": 4.001,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005144755045572917,
+      "learning_rate": 0.0007821052631578947,
+      "loss": 56.0005,
+      "step": 257
+    },
+    {
+      "crossentropy": 2.8968677520751953,
+      "epoch": 4.002,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000540924072265625,
+      "learning_rate": 0.0007810526315789473,
+      "loss": 55.7583,
+      "step": 258
+    },
+    {
+      "crossentropy": 2.9626487493515015,
+      "epoch": 4.003,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00050048828125,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 57.9443,
+      "step": 259
+    },
+    {
+      "crossentropy": 2.8472225666046143,
+      "epoch": 4.004,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0004628499348958333,
+      "learning_rate": 0.0007789473684210527,
+      "loss": 56.6347,
+      "step": 260
+    },
+    {
+      "crossentropy": 2.9540131092071533,
+      "epoch": 4.005,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005538304646809896,
+      "learning_rate": 0.0007778947368421053,
+      "loss": 57.5122,
+      "step": 261
+    },
+    {
+      "crossentropy": 2.801051616668701,
+      "epoch": 4.006,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005787531534830729,
+      "learning_rate": 0.0007768421052631579,
+      "loss": 56.3798,
+      "step": 262
+    },
+    {
+      "crossentropy": 2.816064238548279,
+      "epoch": 4.007,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005950291951497395,
+      "learning_rate": 0.0007757894736842105,
+      "loss": 55.7209,
+      "step": 263
+    },
+    {
+      "crossentropy": 2.7781593799591064,
+      "epoch": 4.008,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0006285985310872396,
+      "learning_rate": 0.0007747368421052632,
+      "loss": 56.1839,
+      "step": 264
+    },
+    {
+      "crossentropy": 2.8623517751693726,
+      "epoch": 4.009,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0006123860677083333,
+      "learning_rate": 0.0007736842105263157,
+      "loss": 57.0193,
+      "step": 265
+    },
+    {
+      "crossentropy": 2.950491189956665,
+      "epoch": 4.01,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0006209691365559896,
+      "learning_rate": 0.0007726315789473684,
+      "loss": 58.5925,
+      "step": 266
+    },
+    {
+      "crossentropy": 2.855417013168335,
+      "epoch": 4.011,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0006650288899739584,
+      "learning_rate": 0.000771578947368421,
+      "loss": 56.9977,
+      "step": 267
+    },
+    {
+      "crossentropy": 2.7262661457061768,
+      "epoch": 4.012,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0006621678670247396,
+      "learning_rate": 0.0007705263157894738,
+      "loss": 55.8765,
+      "step": 268
+    },
+    {
+      "crossentropy": 2.7855740785598755,
+      "epoch": 4.013,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005935033162434896,
+      "learning_rate": 0.0007694736842105264,
+      "loss": 55.4868,
+      "step": 269
+    },
+    {
+      "crossentropy": 2.743911862373352,
+      "epoch": 4.014,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006820042928059896,
+      "learning_rate": 0.0007684210526315789,
+      "loss": 56.2341,
+      "step": 270
+    },
+    {
+      "crossentropy": 2.7292779684066772,
+      "epoch": 4.015,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005055745442708333,
+      "learning_rate": 0.0007673684210526316,
+      "loss": 54.8426,
+      "step": 271
+    },
+    {
+      "crossentropy": 2.8533518314361572,
+      "epoch": 4.016,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00030307769775390626,
+      "learning_rate": 0.0007663157894736842,
+      "loss": 56.84,
+      "step": 272
+    },
+    {
+      "crossentropy": 2.7164812088012695,
+      "epoch": 4.017,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0003387451171875,
+      "learning_rate": 0.0007652631578947369,
+      "loss": 54.748,
+      "step": 273
+    },
+    {
+      "crossentropy": 2.742064118385315,
+      "epoch": 4.018,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0003819147745768229,
+      "learning_rate": 0.0007642105263157894,
+      "loss": 57.1381,
+      "step": 274
+    },
+    {
+      "crossentropy": 2.8081696033477783,
+      "epoch": 4.019,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00039265950520833335,
+      "learning_rate": 0.0007631578947368421,
+      "loss": 54.7163,
+      "step": 275
+    },
+    {
+      "crossentropy": 2.995293974876404,
+      "epoch": 4.02,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0005034764607747395,
+      "learning_rate": 0.0007621052631578948,
+      "loss": 58.3995,
+      "step": 276
+    },
+    {
+      "crossentropy": 2.789522409439087,
+      "epoch": 4.021,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004984537760416666,
+      "learning_rate": 0.0007610526315789474,
+      "loss": 56.2995,
+      "step": 277
+    },
+    {
+      "crossentropy": 2.9243152141571045,
+      "epoch": 4.022,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005063374837239583,
+      "learning_rate": 0.00076,
+      "loss": 56.1988,
+      "step": 278
+    },
+    {
+      "crossentropy": 2.75484561920166,
+      "epoch": 4.023,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005269368489583333,
+      "learning_rate": 0.0007589473684210526,
+      "loss": 55.6807,
+      "step": 279
+    },
+    {
+      "crossentropy": 2.753888249397278,
+      "epoch": 4.024,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005456924438476563,
+      "learning_rate": 0.0007578947368421053,
+      "loss": 55.653,
+      "step": 280
+    },
+    {
+      "crossentropy": 2.889884352684021,
+      "epoch": 4.025,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005533218383789063,
+      "learning_rate": 0.0007568421052631579,
+      "loss": 56.7904,
+      "step": 281
+    },
+    {
+      "crossentropy": 2.790767788887024,
+      "epoch": 4.026,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0006194432576497396,
+      "learning_rate": 0.0007557894736842105,
+      "loss": 54.6775,
+      "step": 282
+    },
+    {
+      "crossentropy": 2.7210811376571655,
+      "epoch": 4.027,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00047092437744140626,
+      "learning_rate": 0.0007547368421052631,
+      "loss": 54.3834,
+      "step": 283
+    },
+    {
+      "crossentropy": 2.862104892730713,
+      "epoch": 4.028,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004124323527018229,
+      "learning_rate": 0.0007536842105263158,
+      "loss": 56.5897,
+      "step": 284
+    },
+    {
+      "crossentropy": 2.861825942993164,
+      "epoch": 4.029,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0004261652628580729,
+      "learning_rate": 0.0007526315789473685,
+      "loss": 57.199,
+      "step": 285
+    },
+    {
+      "crossentropy": 2.738445281982422,
+      "epoch": 4.03,
+      "grad_norm": 0.828125,
+      "grad_norm_var": 0.0007405598958333333,
+      "learning_rate": 0.000751578947368421,
+      "loss": 56.5317,
+      "step": 286
+    },
+    {
+      "crossentropy": 2.8645286560058594,
+      "epoch": 4.031,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0007445653279622396,
+      "learning_rate": 0.0007505263157894737,
+      "loss": 56.3045,
+      "step": 287
+    },
+    {
+      "crossentropy": 2.903935194015503,
+      "epoch": 4.032,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0007369359334309896,
+      "learning_rate": 0.0007494736842105263,
+      "loss": 56.0398,
+      "step": 288
+    },
+    {
+      "crossentropy": 2.8562066555023193,
+      "epoch": 4.033,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0007196426391601563,
+      "learning_rate": 0.000748421052631579,
+      "loss": 55.6453,
+      "step": 289
+    },
+    {
+      "crossentropy": 2.7645561695098877,
+      "epoch": 4.034,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0006916681925455729,
+      "learning_rate": 0.0007473684210526316,
+      "loss": 55.2077,
+      "step": 290
+    },
+    {
+      "crossentropy": 2.8362536430358887,
+      "epoch": 4.035,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006792704264322917,
+      "learning_rate": 0.0007463157894736842,
+      "loss": 56.4424,
+      "step": 291
+    },
+    {
+      "crossentropy": 2.9669147729873657,
+      "epoch": 4.036,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0006368001302083333,
+      "learning_rate": 0.0007452631578947369,
+      "loss": 58.4369,
+      "step": 292
+    },
+    {
+      "crossentropy": 2.751739501953125,
+      "epoch": 4.037,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0006408055623372395,
+      "learning_rate": 0.0007442105263157895,
+      "loss": 55.4938,
+      "step": 293
+    },
+    {
+      "crossentropy": 2.7845678329467773,
+      "epoch": 4.038,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0006621678670247396,
+      "learning_rate": 0.0007431578947368422,
+      "loss": 57.6227,
+      "step": 294
+    },
+    {
+      "crossentropy": 2.894593834877014,
+      "epoch": 4.039,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0006357828776041666,
+      "learning_rate": 0.0007421052631578947,
+      "loss": 56.1408,
+      "step": 295
+    },
+    {
+      "crossentropy": 2.8750534057617188,
+      "epoch": 4.04,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006260553995768229,
+      "learning_rate": 0.0007410526315789474,
+      "loss": 57.3143,
+      "step": 296
+    },
+    {
+      "crossentropy": 2.828782796859741,
+      "epoch": 4.041,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0006260553995768229,
+      "learning_rate": 0.00074,
+      "loss": 56.8303,
+      "step": 297
+    },
+    {
+      "crossentropy": 2.830629348754883,
+      "epoch": 4.042,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005670547485351563,
+      "learning_rate": 0.0007389473684210527,
+      "loss": 57.1108,
+      "step": 298
+    },
+    {
+      "crossentropy": 2.8372979164123535,
+      "epoch": 4.043,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005551656087239583,
+      "learning_rate": 0.0007378947368421052,
+      "loss": 56.3579,
+      "step": 299
+    },
+    {
+      "crossentropy": 2.7618184089660645,
+      "epoch": 4.044,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005538304646809896,
+      "learning_rate": 0.0007368421052631579,
+      "loss": 56.3516,
+      "step": 300
+    },
+    {
+      "crossentropy": 2.8434700965881348,
+      "epoch": 4.045,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005694071451822917,
+      "learning_rate": 0.0007357894736842106,
+      "loss": 57.8224,
+      "step": 301
+    },
+    {
+      "crossentropy": 2.845301866531372,
+      "epoch": 4.046,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002886454264322917,
+      "learning_rate": 0.0007347368421052632,
+      "loss": 56.0516,
+      "step": 302
+    },
+    {
+      "crossentropy": 2.902227759361267,
+      "epoch": 4.047,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.00038426717122395834,
+      "learning_rate": 0.0007336842105263159,
+      "loss": 57.105,
+      "step": 303
+    },
+    {
+      "crossentropy": 2.8356798887252808,
+      "epoch": 4.048,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00037740071614583335,
+      "learning_rate": 0.0007326315789473684,
+      "loss": 57.0535,
+      "step": 304
+    },
+    {
+      "crossentropy": 2.9245163202285767,
+      "epoch": 4.049,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0004470189412434896,
+      "learning_rate": 0.0007315789473684211,
+      "loss": 58.2829,
+      "step": 305
+    },
+    {
+      "crossentropy": 2.773668646812439,
+      "epoch": 4.05,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0005462646484375,
+      "learning_rate": 0.0007305263157894737,
+      "loss": 56.9855,
+      "step": 306
+    },
+    {
+      "crossentropy": 2.974271535873413,
+      "epoch": 4.051,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0005711237589518229,
+      "learning_rate": 0.0007294736842105262,
+      "loss": 58.2544,
+      "step": 307
+    },
+    {
+      "crossentropy": 2.7676045894622803,
+      "epoch": 4.052,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005304336547851563,
+      "learning_rate": 0.000728421052631579,
+      "loss": 55.4997,
+      "step": 308
+    },
+    {
+      "crossentropy": 2.8458242416381836,
+      "epoch": 4.053,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.000540924072265625,
+      "learning_rate": 0.0007273684210526316,
+      "loss": 56.7489,
+      "step": 309
+    },
+    {
+      "crossentropy": 2.9176676273345947,
+      "epoch": 4.054,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005395889282226562,
+      "learning_rate": 0.0007263157894736843,
+      "loss": 57.7019,
+      "step": 310
+    },
+    {
+      "crossentropy": 2.7517309188842773,
+      "epoch": 4.055,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005381266276041667,
+      "learning_rate": 0.0007252631578947369,
+      "loss": 56.0364,
+      "step": 311
+    },
+    {
+      "crossentropy": 2.8117847442626953,
+      "epoch": 4.056,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0005755106608072917,
+      "learning_rate": 0.0007242105263157895,
+      "loss": 57.0578,
+      "step": 312
+    },
+    {
+      "crossentropy": 2.7207109928131104,
+      "epoch": 4.057,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005904515584309896,
+      "learning_rate": 0.0007231578947368421,
+      "loss": 55.5513,
+      "step": 313
+    },
+    {
+      "crossentropy": 2.798499345779419,
+      "epoch": 4.058,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006144205729166667,
+      "learning_rate": 0.0007221052631578947,
+      "loss": 56.0931,
+      "step": 314
+    },
+    {
+      "crossentropy": 2.8496850728988647,
+      "epoch": 4.059,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005877176920572917,
+      "learning_rate": 0.0007210526315789474,
+      "loss": 56.5729,
+      "step": 315
+    },
+    {
+      "crossentropy": 2.739889979362488,
+      "epoch": 4.06,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005879084269205729,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 57.048,
+      "step": 316
+    },
+    {
+      "crossentropy": 2.8727529048919678,
+      "epoch": 4.061,
+      "grad_norm": 0.8203125,
+      "grad_norm_var": 0.0007298151652018229,
+      "learning_rate": 0.0007189473684210527,
+      "loss": 57.4175,
+      "step": 317
+    },
+    {
+      "crossentropy": 2.9316142797470093,
+      "epoch": 4.062,
+      "grad_norm": 0.83203125,
+      "grad_norm_var": 0.0009470621744791667,
+      "learning_rate": 0.0007178947368421053,
+      "loss": 57.4141,
+      "step": 318
+    },
+    {
+      "crossentropy": 2.8334540128707886,
+      "epoch": 4.063,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000909868876139323,
+      "learning_rate": 0.000716842105263158,
+      "loss": 56.4528,
+      "step": 319
+    },
+    {
+      "crossentropy": 2.8760459423065186,
+      "epoch": 4.064,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0009195963541666667,
+      "learning_rate": 0.0007157894736842105,
+      "loss": 57.3916,
+      "step": 320
+    },
+    {
+      "crossentropy": 2.7978386878967285,
+      "epoch": 5.001,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0007181803385416667,
+      "learning_rate": 0.0007147368421052631,
+      "loss": 56.2145,
+      "step": 321
+    },
+    {
+      "crossentropy": 2.744502544403076,
+      "epoch": 5.002,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0006683349609375,
+      "learning_rate": 0.0007136842105263158,
+      "loss": 56.6854,
+      "step": 322
+    },
+    {
+      "crossentropy": 2.8497555255889893,
+      "epoch": 5.003,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006540298461914062,
+      "learning_rate": 0.0007126315789473684,
+      "loss": 55.9165,
+      "step": 323
+    },
+    {
+      "crossentropy": 2.7810473442077637,
+      "epoch": 5.004,
+      "grad_norm": 0.8515625,
+      "grad_norm_var": 0.000968170166015625,
+      "learning_rate": 0.000711578947368421,
+      "loss": 56.7632,
+      "step": 324
+    },
+    {
+      "crossentropy": 2.9091150760650635,
+      "epoch": 5.005,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0010171890258789062,
+      "learning_rate": 0.0007105263157894737,
+      "loss": 57.0648,
+      "step": 325
+    },
+    {
+      "crossentropy": 2.9585464000701904,
+      "epoch": 5.006,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.001034990946451823,
+      "learning_rate": 0.0007094736842105264,
+      "loss": 57.6818,
+      "step": 326
+    },
+    {
+      "crossentropy": 2.7995084524154663,
+      "epoch": 5.007,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0010457356770833333,
+      "learning_rate": 0.000708421052631579,
+      "loss": 55.2677,
+      "step": 327
+    },
+    {
+      "crossentropy": 2.8381704092025757,
+      "epoch": 5.008,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0010207494099934895,
+      "learning_rate": 0.0007073684210526316,
+      "loss": 56.1405,
+      "step": 328
+    },
+    {
+      "crossentropy": 2.8964773416519165,
+      "epoch": 5.009,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0009571711222330729,
+      "learning_rate": 0.0007063157894736842,
+      "loss": 57.4197,
+      "step": 329
+    },
+    {
+      "crossentropy": 2.8280014991760254,
+      "epoch": 5.01,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0009571711222330729,
+      "learning_rate": 0.0007052631578947368,
+      "loss": 56.3693,
+      "step": 330
+    },
+    {
+      "crossentropy": 2.792478084564209,
+      "epoch": 5.011,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0009600321451822916,
+      "learning_rate": 0.0007042105263157895,
+      "loss": 56.2133,
+      "step": 331
+    },
+    {
+      "crossentropy": 2.8384552001953125,
+      "epoch": 5.012,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0009052912394205729,
+      "learning_rate": 0.0007031578947368421,
+      "loss": 57.0165,
+      "step": 332
+    },
+    {
+      "crossentropy": 2.888237237930298,
+      "epoch": 5.013,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0006998697916666667,
+      "learning_rate": 0.0007021052631578948,
+      "loss": 58.966,
+      "step": 333
+    },
+    {
+      "crossentropy": 2.834162950515747,
+      "epoch": 5.014,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0007338841756184896,
+      "learning_rate": 0.0007010526315789474,
+      "loss": 55.0934,
+      "step": 334
+    },
+    {
+      "crossentropy": 2.7016130685806274,
+      "epoch": 5.015,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0007382710774739583,
+      "learning_rate": 0.0007,
+      "loss": 55.4662,
+      "step": 335
+    },
+    {
+      "crossentropy": 2.894522786140442,
+      "epoch": 5.016,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0007369359334309896,
+      "learning_rate": 0.0006989473684210527,
+      "loss": 55.544,
+      "step": 336
+    },
+    {
+      "crossentropy": 2.891346573829651,
+      "epoch": 5.017,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0007375081380208334,
+      "learning_rate": 0.0006978947368421052,
+      "loss": 57.4768,
+      "step": 337
+    },
+    {
+      "crossentropy": 2.8023122549057007,
+      "epoch": 5.018,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0007491429646809896,
+      "learning_rate": 0.0006968421052631579,
+      "loss": 56.9822,
+      "step": 338
+    },
+    {
+      "crossentropy": 2.893723964691162,
+      "epoch": 5.019,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0007659276326497396,
+      "learning_rate": 0.0006957894736842105,
+      "loss": 56.84,
+      "step": 339
+    },
+    {
+      "crossentropy": 2.8015278577804565,
+      "epoch": 5.02,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0003326416015625,
+      "learning_rate": 0.0006947368421052632,
+      "loss": 55.3613,
+      "step": 340
+    },
+    {
+      "crossentropy": 2.868039131164551,
+      "epoch": 5.021,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00030517578125,
+      "learning_rate": 0.0006936842105263159,
+      "loss": 57.3593,
+      "step": 341
+    },
+    {
+      "crossentropy": 2.800724148750305,
+      "epoch": 5.022,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000390625,
+      "learning_rate": 0.0006926315789473684,
+      "loss": 55.3344,
+      "step": 342
+    },
+    {
+      "crossentropy": 2.791234016418457,
+      "epoch": 5.023,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00039768218994140625,
+      "learning_rate": 0.0006915789473684211,
+      "loss": 55.2978,
+      "step": 343
+    },
+    {
+      "crossentropy": 2.718875527381897,
+      "epoch": 5.024,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0004042943318684896,
+      "learning_rate": 0.0006905263157894737,
+      "loss": 54.9774,
+      "step": 344
+    },
+    {
+      "crossentropy": 2.888947367668152,
+      "epoch": 5.025,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004025777180989583,
+      "learning_rate": 0.0006894736842105264,
+      "loss": 56.9891,
+      "step": 345
+    },
+    {
+      "crossentropy": 2.7386454343795776,
+      "epoch": 5.026,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0004241307576497396,
+      "learning_rate": 0.0006884210526315789,
+      "loss": 55.8273,
+      "step": 346
+    },
+    {
+      "crossentropy": 2.703773021697998,
+      "epoch": 5.027,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000421142578125,
+      "learning_rate": 0.0006873684210526316,
+      "loss": 55.5817,
+      "step": 347
+    },
+    {
+      "crossentropy": 2.821714162826538,
+      "epoch": 5.028,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.000384521484375,
+      "learning_rate": 0.0006863157894736842,
+      "loss": 55.9298,
+      "step": 348
+    },
+    {
+      "crossentropy": 2.8079909086227417,
+      "epoch": 5.029,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.00042012532552083336,
+      "learning_rate": 0.0006852631578947368,
+      "loss": 55.0177,
+      "step": 349
+    },
+    {
+      "crossentropy": 2.694362759590149,
+      "epoch": 5.03,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00042057037353515625,
+      "learning_rate": 0.0006842105263157895,
+      "loss": 54.1136,
+      "step": 350
+    },
+    {
+      "crossentropy": 2.8693727254867554,
+      "epoch": 5.031,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004302342732747396,
+      "learning_rate": 0.0006831578947368421,
+      "loss": 57.0075,
+      "step": 351
+    },
+    {
+      "crossentropy": 2.8346846103668213,
+      "epoch": 5.032,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.000433349609375,
+      "learning_rate": 0.0006821052631578948,
+      "loss": 56.9684,
+      "step": 352
+    },
+    {
+      "crossentropy": 2.9195793867111206,
+      "epoch": 5.033,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00044193267822265623,
+      "learning_rate": 0.0006810526315789474,
+      "loss": 57.0331,
+      "step": 353
+    },
+    {
+      "crossentropy": 2.8132762908935547,
+      "epoch": 5.034,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00038547515869140624,
+      "learning_rate": 0.00068,
+      "loss": 55.508,
+      "step": 354
+    },
+    {
+      "crossentropy": 2.7499265670776367,
+      "epoch": 5.035,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00035037994384765623,
+      "learning_rate": 0.0006789473684210526,
+      "loss": 56.0205,
+      "step": 355
+    },
+    {
+      "crossentropy": 2.800133466720581,
+      "epoch": 5.036,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0003720601399739583,
+      "learning_rate": 0.0006778947368421052,
+      "loss": 55.9033,
+      "step": 356
+    },
+    {
+      "crossentropy": 2.820741653442383,
+      "epoch": 5.037,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0003720601399739583,
+      "learning_rate": 0.0006768421052631579,
+      "loss": 55.9092,
+      "step": 357
+    },
+    {
+      "crossentropy": 2.781237006187439,
+      "epoch": 5.038,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003102620442708333,
+      "learning_rate": 0.0006757894736842106,
+      "loss": 56.8959,
+      "step": 358
+    },
+    {
+      "crossentropy": 2.8900152444839478,
+      "epoch": 5.039,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0003280003865559896,
+      "learning_rate": 0.0006747368421052632,
+      "loss": 56.5975,
+      "step": 359
+    },
+    {
+      "crossentropy": 2.6693739891052246,
+      "epoch": 5.04,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003280003865559896,
+      "learning_rate": 0.0006736842105263158,
+      "loss": 55.8081,
+      "step": 360
+    },
+    {
+      "crossentropy": 2.8285491466522217,
+      "epoch": 5.041,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003280003865559896,
+      "learning_rate": 0.0006726315789473685,
+      "loss": 55.7024,
+      "step": 361
+    },
+    {
+      "crossentropy": 2.970876455307007,
+      "epoch": 5.042,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.000299072265625,
+      "learning_rate": 0.0006715789473684211,
+      "loss": 57.7093,
+      "step": 362
+    },
+    {
+      "crossentropy": 2.902941942214966,
+      "epoch": 5.043,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0003265380859375,
+      "learning_rate": 0.0006705263157894736,
+      "loss": 58.8785,
+      "step": 363
+    },
+    {
+      "crossentropy": 2.888626217842102,
+      "epoch": 5.044,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.000321197509765625,
+      "learning_rate": 0.0006694736842105263,
+      "loss": 56.7212,
+      "step": 364
+    },
+    {
+      "crossentropy": 2.9147382974624634,
+      "epoch": 5.045,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000199127197265625,
+      "learning_rate": 0.0006684210526315789,
+      "loss": 57.0614,
+      "step": 365
+    },
+    {
+      "crossentropy": 2.816585063934326,
+      "epoch": 5.046,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.00019474029541015624,
+      "learning_rate": 0.0006673684210526317,
+      "loss": 57.5533,
+      "step": 366
+    },
+    {
+      "crossentropy": 2.801294684410095,
+      "epoch": 5.047,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00018761952718098957,
+      "learning_rate": 0.0006663157894736842,
+      "loss": 56.8426,
+      "step": 367
+    },
+    {
+      "crossentropy": 2.8105653524398804,
+      "epoch": 5.048,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00017439524332682292,
+      "learning_rate": 0.0006652631578947369,
+      "loss": 56.7316,
+      "step": 368
+    },
+    {
+      "crossentropy": 2.8542263507843018,
+      "epoch": 5.049,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00018202463785807292,
+      "learning_rate": 0.0006642105263157895,
+      "loss": 57.1404,
+      "step": 369
+    },
+    {
+      "crossentropy": 2.75753116607666,
+      "epoch": 5.05,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00019321441650390624,
+      "learning_rate": 0.0006631578947368421,
+      "loss": 56.3615,
+      "step": 370
+    },
+    {
+      "crossentropy": 2.7114880084991455,
+      "epoch": 5.051,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00014896392822265624,
+      "learning_rate": 0.0006621052631578947,
+      "loss": 55.6979,
+      "step": 371
+    },
+    {
+      "crossentropy": 2.8140069246292114,
+      "epoch": 5.052,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0001983006795247396,
+      "learning_rate": 0.0006610526315789473,
+      "loss": 57.4654,
+      "step": 372
+    },
+    {
+      "crossentropy": 2.9149471521377563,
+      "epoch": 5.053,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0002517064412434896,
+      "learning_rate": 0.00066,
+      "loss": 58.3265,
+      "step": 373
+    },
+    {
+      "crossentropy": 2.922295570373535,
+      "epoch": 5.054,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00023167928059895834,
+      "learning_rate": 0.0006589473684210527,
+      "loss": 58.1646,
+      "step": 374
+    },
+    {
+      "crossentropy": 2.7302639484405518,
+      "epoch": 5.055,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0002459208170572917,
+      "learning_rate": 0.0006578947368421054,
+      "loss": 56.4477,
+      "step": 375
+    },
+    {
+      "crossentropy": 2.849370002746582,
+      "epoch": 5.056,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00022627512613932293,
+      "learning_rate": 0.0006568421052631579,
+      "loss": 56.5778,
+      "step": 376
+    },
+    {
+      "crossentropy": 2.7923754453659058,
+      "epoch": 5.057,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000284576416015625,
+      "learning_rate": 0.0006557894736842105,
+      "loss": 56.2568,
+      "step": 377
+    },
+    {
+      "crossentropy": 2.807488203048706,
+      "epoch": 5.058,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002827326456705729,
+      "learning_rate": 0.0006547368421052632,
+      "loss": 55.3052,
+      "step": 378
+    },
+    {
+      "crossentropy": 2.947903275489807,
+      "epoch": 5.059,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00027872721354166665,
+      "learning_rate": 0.0006536842105263158,
+      "loss": 58.4199,
+      "step": 379
+    },
+    {
+      "crossentropy": 2.838424563407898,
+      "epoch": 5.06,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0002812067667643229,
+      "learning_rate": 0.0006526315789473684,
+      "loss": 55.4688,
+      "step": 380
+    },
+    {
+      "crossentropy": 2.931127429008484,
+      "epoch": 5.061,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00028578440348307293,
+      "learning_rate": 0.000651578947368421,
+      "loss": 57.8073,
+      "step": 381
+    },
+    {
+      "crossentropy": 2.776180386543274,
+      "epoch": 5.062,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002848307291666667,
+      "learning_rate": 0.0006505263157894738,
+      "loss": 55.5867,
+      "step": 382
+    },
+    {
+      "crossentropy": 2.95345675945282,
+      "epoch": 5.063,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002848307291666667,
+      "learning_rate": 0.0006494736842105264,
+      "loss": 57.5454,
+      "step": 383
+    },
+    {
+      "crossentropy": 2.890448570251465,
+      "epoch": 5.064,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00027815500895182293,
+      "learning_rate": 0.0006484210526315789,
+      "loss": 57.2167,
+      "step": 384
+    },
+    {
+      "crossentropy": 2.896027088165283,
+      "epoch": 6.001,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0002486546834309896,
+      "learning_rate": 0.0006473684210526316,
+      "loss": 56.383,
+      "step": 385
+    },
+    {
+      "crossentropy": 2.7579907178878784,
+      "epoch": 6.002,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002471923828125,
+      "learning_rate": 0.0006463157894736842,
+      "loss": 54.8391,
+      "step": 386
+    },
+    {
+      "crossentropy": 2.7551032304763794,
+      "epoch": 6.003,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00018253326416015626,
+      "learning_rate": 0.0006452631578947369,
+      "loss": 56.4615,
+      "step": 387
+    },
+    {
+      "crossentropy": 2.7534666061401367,
+      "epoch": 6.004,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.000140380859375,
+      "learning_rate": 0.0006442105263157894,
+      "loss": 55.3748,
+      "step": 388
+    },
+    {
+      "crossentropy": 2.878955602645874,
+      "epoch": 6.005,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.000131988525390625,
+      "learning_rate": 0.0006431578947368421,
+      "loss": 56.7231,
+      "step": 389
+    },
+    {
+      "crossentropy": 2.7627967596054077,
+      "epoch": 6.006,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.00014216105143229166,
+      "learning_rate": 0.0006421052631578948,
+      "loss": 55.1437,
+      "step": 390
+    },
+    {
+      "crossentropy": 2.7477593421936035,
+      "epoch": 6.007,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0001545588175455729,
+      "learning_rate": 0.0006410526315789474,
+      "loss": 55.0148,
+      "step": 391
+    },
+    {
+      "crossentropy": 2.919530153274536,
+      "epoch": 6.008,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0001891454060872396,
+      "learning_rate": 0.00064,
+      "loss": 57.6885,
+      "step": 392
+    },
+    {
+      "crossentropy": 2.819299101829529,
+      "epoch": 6.009,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0001891454060872396,
+      "learning_rate": 0.0006389473684210526,
+      "loss": 56.0235,
+      "step": 393
+    },
+    {
+      "crossentropy": 2.9445995092391968,
+      "epoch": 6.01,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.00022761027018229166,
+      "learning_rate": 0.0006378947368421053,
+      "loss": 58.1254,
+      "step": 394
+    },
+    {
+      "crossentropy": 2.777245879173279,
+      "epoch": 6.011,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0002726236979166667,
+      "learning_rate": 0.0006368421052631579,
+      "loss": 55.7731,
+      "step": 395
+    },
+    {
+      "crossentropy": 2.983623743057251,
+      "epoch": 6.012,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00026950836181640627,
+      "learning_rate": 0.0006357894736842106,
+      "loss": 58.7228,
+      "step": 396
+    },
+    {
+      "crossentropy": 2.9414658546447754,
+      "epoch": 6.013,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0002899169921875,
+      "learning_rate": 0.0006347368421052631,
+      "loss": 56.5824,
+      "step": 397
+    },
+    {
+      "crossentropy": 2.8643269538879395,
+      "epoch": 6.014,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00033156077067057293,
+      "learning_rate": 0.0006336842105263157,
+      "loss": 57.9013,
+      "step": 398
+    },
+    {
+      "crossentropy": 2.71036159992218,
+      "epoch": 6.015,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003590265909830729,
+      "learning_rate": 0.0006326315789473685,
+      "loss": 54.2278,
+      "step": 399
+    },
+    {
+      "crossentropy": 2.782063364982605,
+      "epoch": 6.016,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003590265909830729,
+      "learning_rate": 0.0006315789473684211,
+      "loss": 55.5568,
+      "step": 400
+    },
+    {
+      "crossentropy": 2.783318877220154,
+      "epoch": 6.017,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003829320271809896,
+      "learning_rate": 0.0006305263157894737,
+      "loss": 56.6671,
+      "step": 401
+    },
+    {
+      "crossentropy": 2.9031968116760254,
+      "epoch": 6.018,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0004412333170572917,
+      "learning_rate": 0.0006294736842105263,
+      "loss": 55.2412,
+      "step": 402
+    },
+    {
+      "crossentropy": 2.8032177686691284,
+      "epoch": 6.019,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00045064290364583336,
+      "learning_rate": 0.000628421052631579,
+      "loss": 55.8731,
+      "step": 403
+    },
+    {
+      "crossentropy": 2.816411256790161,
+      "epoch": 6.02,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.00044193267822265623,
+      "learning_rate": 0.0006273684210526316,
+      "loss": 56.364,
+      "step": 404
+    },
+    {
+      "crossentropy": 2.7340190410614014,
+      "epoch": 6.021,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005151748657226563,
+      "learning_rate": 0.0006263157894736841,
+      "loss": 54.9243,
+      "step": 405
+    },
+    {
+      "crossentropy": 2.8077826499938965,
+      "epoch": 6.022,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004963556925455729,
+      "learning_rate": 0.0006252631578947368,
+      "loss": 56.1388,
+      "step": 406
+    },
+    {
+      "crossentropy": 2.8767653703689575,
+      "epoch": 6.023,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000574175516764323,
+      "learning_rate": 0.0006242105263157895,
+      "loss": 56.2231,
+      "step": 407
+    },
+    {
+      "crossentropy": 2.6605910062789917,
+      "epoch": 6.024,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0004887262980143229,
+      "learning_rate": 0.0006231578947368422,
+      "loss": 54.9768,
+      "step": 408
+    },
+    {
+      "crossentropy": 2.831836223602295,
+      "epoch": 6.025,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004900614420572917,
+      "learning_rate": 0.0006221052631578947,
+      "loss": 55.0988,
+      "step": 409
+    },
+    {
+      "crossentropy": 2.700801968574524,
+      "epoch": 6.026,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0004099527994791667,
+      "learning_rate": 0.0006210526315789474,
+      "loss": 54.9491,
+      "step": 410
+    },
+    {
+      "crossentropy": 2.933374762535095,
+      "epoch": 6.027,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0003191630045572917,
+      "learning_rate": 0.00062,
+      "loss": 56.9396,
+      "step": 411
+    },
+    {
+      "crossentropy": 2.74997615814209,
+      "epoch": 6.028,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0003903706868489583,
+      "learning_rate": 0.0006189473684210526,
+      "loss": 55.024,
+      "step": 412
+    },
+    {
+      "crossentropy": 2.916796088218689,
+      "epoch": 6.029,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00039310455322265624,
+      "learning_rate": 0.0006178947368421053,
+      "loss": 57.4697,
+      "step": 413
+    },
+    {
+      "crossentropy": 2.900514841079712,
+      "epoch": 6.03,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003916422526041667,
+      "learning_rate": 0.0006168421052631578,
+      "loss": 57.3729,
+      "step": 414
+    },
+    {
+      "crossentropy": 2.8999792337417603,
+      "epoch": 6.031,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0003986994425455729,
+      "learning_rate": 0.0006157894736842106,
+      "loss": 57.0101,
+      "step": 415
+    },
+    {
+      "crossentropy": 2.8940166234970093,
+      "epoch": 6.032,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004170099894205729,
+      "learning_rate": 0.0006147368421052632,
+      "loss": 59.3612,
+      "step": 416
+    },
+    {
+      "crossentropy": 2.7649847269058228,
+      "epoch": 6.033,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00042317708333333335,
+      "learning_rate": 0.0006136842105263159,
+      "loss": 55.2584,
+      "step": 417
+    },
+    {
+      "crossentropy": 2.8447442054748535,
+      "epoch": 6.034,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004200617472330729,
+      "learning_rate": 0.0006126315789473684,
+      "loss": 56.9598,
+      "step": 418
+    },
+    {
+      "crossentropy": 2.9044229984283447,
+      "epoch": 6.035,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004445393880208333,
+      "learning_rate": 0.000611578947368421,
+      "loss": 57.1472,
+      "step": 419
+    },
+    {
+      "crossentropy": 2.858897089958191,
+      "epoch": 6.036,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0004592259724934896,
+      "learning_rate": 0.0006105263157894737,
+      "loss": 58.3656,
+      "step": 420
+    },
+    {
+      "crossentropy": 2.781478762626648,
+      "epoch": 6.037,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004208882649739583,
+      "learning_rate": 0.0006094736842105263,
+      "loss": 55.7423,
+      "step": 421
+    },
+    {
+      "crossentropy": 2.7873969078063965,
+      "epoch": 6.038,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00041751861572265626,
+      "learning_rate": 0.0006084210526315789,
+      "loss": 56.0058,
+      "step": 422
+    },
+    {
+      "crossentropy": 2.7320199012756348,
+      "epoch": 6.039,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0004597981770833333,
+      "learning_rate": 0.0006073684210526316,
+      "loss": 55.8398,
+      "step": 423
+    },
+    {
+      "crossentropy": 2.859510898590088,
+      "epoch": 6.04,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0004856745402018229,
+      "learning_rate": 0.0006063157894736843,
+      "loss": 56.2963,
+      "step": 424
+    },
+    {
+      "crossentropy": 2.756445050239563,
+      "epoch": 6.041,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00048395792643229164,
+      "learning_rate": 0.0006052631578947369,
+      "loss": 55.111,
+      "step": 425
+    },
+    {
+      "crossentropy": 2.632467031478882,
+      "epoch": 6.042,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.0005929946899414062,
+      "learning_rate": 0.0006042105263157894,
+      "loss": 53.8061,
+      "step": 426
+    },
+    {
+      "crossentropy": 2.8806999921798706,
+      "epoch": 6.043,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005755106608072917,
+      "learning_rate": 0.0006031578947368421,
+      "loss": 57.9055,
+      "step": 427
+    },
+    {
+      "crossentropy": 2.818897008895874,
+      "epoch": 6.044,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00047092437744140626,
+      "learning_rate": 0.0006021052631578947,
+      "loss": 56.0326,
+      "step": 428
+    },
+    {
+      "crossentropy": 2.8834354877471924,
+      "epoch": 6.045,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004749933878580729,
+      "learning_rate": 0.0006010526315789474,
+      "loss": 57.5019,
+      "step": 429
+    },
+    {
+      "crossentropy": 2.856741189956665,
+      "epoch": 6.046,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0005070368448893229,
+      "learning_rate": 0.0006,
+      "loss": 56.8689,
+      "step": 430
+    },
+    {
+      "crossentropy": 2.8887990713119507,
+      "epoch": 6.047,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000510406494140625,
+      "learning_rate": 0.0005989473684210527,
+      "loss": 57.7479,
+      "step": 431
+    },
+    {
+      "crossentropy": 2.8145207166671753,
+      "epoch": 6.048,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004922866821289063,
+      "learning_rate": 0.0005978947368421053,
+      "loss": 56.8016,
+      "step": 432
+    },
+    {
+      "crossentropy": 2.841512441635132,
+      "epoch": 6.049,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.000485992431640625,
+      "learning_rate": 0.0005968421052631579,
+      "loss": 57.1147,
+      "step": 433
+    },
+    {
+      "crossentropy": 2.8247647285461426,
+      "epoch": 6.05,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000482177734375,
+      "learning_rate": 0.0005957894736842106,
+      "loss": 57.3273,
+      "step": 434
+    },
+    {
+      "crossentropy": 2.7687268257141113,
+      "epoch": 6.051,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.000565338134765625,
+      "learning_rate": 0.0005947368421052631,
+      "loss": 54.6459,
+      "step": 435
+    },
+    {
+      "crossentropy": 2.9791083335876465,
+      "epoch": 6.052,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0005961100260416667,
+      "learning_rate": 0.0005936842105263158,
+      "loss": 59.038,
+      "step": 436
+    },
+    {
+      "crossentropy": 2.80897319316864,
+      "epoch": 6.053,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005991617838541667,
+      "learning_rate": 0.0005926315789473684,
+      "loss": 56.0703,
+      "step": 437
+    },
+    {
+      "crossentropy": 2.8237274885177612,
+      "epoch": 6.054,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0006306330362955729,
+      "learning_rate": 0.0005915789473684211,
+      "loss": 56.6772,
+      "step": 438
+    },
+    {
+      "crossentropy": 2.7993067502975464,
+      "epoch": 6.055,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0005375544230143229,
+      "learning_rate": 0.0005905263157894736,
+      "loss": 56.7413,
+      "step": 439
+    },
+    {
+      "crossentropy": 2.8471277952194214,
+      "epoch": 6.056,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0004961649576822917,
+      "learning_rate": 0.0005894736842105263,
+      "loss": 56.4865,
+      "step": 440
+    },
+    {
+      "crossentropy": 2.8962905406951904,
+      "epoch": 6.057,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.000604248046875,
+      "learning_rate": 0.000588421052631579,
+      "loss": 56.8234,
+      "step": 441
+    },
+    {
+      "crossentropy": 2.8405044078826904,
+      "epoch": 6.058,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0004903157552083333,
+      "learning_rate": 0.0005873684210526316,
+      "loss": 57.3965,
+      "step": 442
+    },
+    {
+      "crossentropy": 2.89117431640625,
+      "epoch": 6.059,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004938125610351562,
+      "learning_rate": 0.0005863157894736842,
+      "loss": 57.63,
+      "step": 443
+    },
+    {
+      "crossentropy": 2.7827476263046265,
+      "epoch": 6.06,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00042292277018229164,
+      "learning_rate": 0.0005852631578947368,
+      "loss": 56.2097,
+      "step": 444
+    },
+    {
+      "crossentropy": 2.8849616050720215,
+      "epoch": 6.061,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0004900614420572917,
+      "learning_rate": 0.0005842105263157895,
+      "loss": 58.1,
+      "step": 445
+    },
+    {
+      "crossentropy": 2.994400978088379,
+      "epoch": 6.062,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00042012532552083336,
+      "learning_rate": 0.0005831578947368421,
+      "loss": 58.3295,
+      "step": 446
+    },
+    {
+      "crossentropy": 2.806332230567932,
+      "epoch": 6.063,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00044148763020833334,
+      "learning_rate": 0.0005821052631578948,
+      "loss": 56.9378,
+      "step": 447
+    },
+    {
+      "crossentropy": 2.8283108472824097,
+      "epoch": 6.064,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004353205362955729,
+      "learning_rate": 0.0005810526315789474,
+      "loss": 57.3855,
+      "step": 448
+    },
+    {
+      "crossentropy": 2.7301007509231567,
+      "epoch": 7.001,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0004729588826497396,
+      "learning_rate": 0.00058,
+      "loss": 54.0905,
+      "step": 449
+    },
+    {
+      "crossentropy": 2.741808295249939,
+      "epoch": 7.002,
+      "grad_norm": 0.81640625,
+      "grad_norm_var": 0.000577545166015625,
+      "learning_rate": 0.0005789473684210527,
+      "loss": 55.7839,
+      "step": 450
+    },
+    {
+      "crossentropy": 2.9567712545394897,
+      "epoch": 7.003,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.000531005859375,
+      "learning_rate": 0.0005778947368421053,
+      "loss": 58.1478,
+      "step": 451
+    },
+    {
+      "crossentropy": 2.740614175796509,
+      "epoch": 7.004,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000614166259765625,
+      "learning_rate": 0.0005768421052631579,
+      "loss": 54.8335,
+      "step": 452
+    },
+    {
+      "crossentropy": 2.791291117668152,
+      "epoch": 7.005,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006316502888997396,
+      "learning_rate": 0.0005757894736842105,
+      "loss": 57.0646,
+      "step": 453
+    },
+    {
+      "crossentropy": 2.830601930618286,
+      "epoch": 7.006,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0006243387858072916,
+      "learning_rate": 0.0005747368421052632,
+      "loss": 56.4297,
+      "step": 454
+    },
+    {
+      "crossentropy": 2.766074061393738,
+      "epoch": 7.007,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0006469090779622396,
+      "learning_rate": 0.0005736842105263158,
+      "loss": 55.315,
+      "step": 455
+    },
+    {
+      "crossentropy": 2.8447988033294678,
+      "epoch": 7.008,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0006357192993164063,
+      "learning_rate": 0.0005726315789473684,
+      "loss": 56.7864,
+      "step": 456
+    },
+    {
+      "crossentropy": 2.8546324968338013,
+      "epoch": 7.009,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0006388346354166667,
+      "learning_rate": 0.0005715789473684211,
+      "loss": 56.896,
+      "step": 457
+    },
+    {
+      "crossentropy": 2.8077392578125,
+      "epoch": 7.01,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0006528218587239583,
+      "learning_rate": 0.0005705263157894737,
+      "loss": 55.2438,
+      "step": 458
+    },
+    {
+      "crossentropy": 2.8590996265411377,
+      "epoch": 7.011,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0006296157836914063,
+      "learning_rate": 0.0005694736842105264,
+      "loss": 56.4089,
+      "step": 459
+    },
+    {
+      "crossentropy": 2.6845264434814453,
+      "epoch": 7.012,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00059814453125,
+      "learning_rate": 0.0005684210526315789,
+      "loss": 54.5441,
+      "step": 460
+    },
+    {
+      "crossentropy": 2.821642518043518,
+      "epoch": 7.013,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005889892578125,
+      "learning_rate": 0.0005673684210526316,
+      "loss": 57.5324,
+      "step": 461
+    },
+    {
+      "crossentropy": 2.8966156244277954,
+      "epoch": 7.014,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0006052652994791666,
+      "learning_rate": 0.0005663157894736842,
+      "loss": 57.1761,
+      "step": 462
+    },
+    {
+      "crossentropy": 2.9717679023742676,
+      "epoch": 7.015,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006062189737955729,
+      "learning_rate": 0.0005652631578947368,
+      "loss": 57.733,
+      "step": 463
+    },
+    {
+      "crossentropy": 2.8593705892562866,
+      "epoch": 7.016,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006060282389322917,
+      "learning_rate": 0.0005642105263157896,
+      "loss": 56.4377,
+      "step": 464
+    },
+    {
+      "crossentropy": 2.8285802602767944,
+      "epoch": 7.017,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005869547526041667,
+      "learning_rate": 0.0005631578947368421,
+      "loss": 56.9182,
+      "step": 465
+    },
+    {
+      "crossentropy": 2.80751633644104,
+      "epoch": 7.018,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0004215876261393229,
+      "learning_rate": 0.0005621052631578948,
+      "loss": 58.1339,
+      "step": 466
+    },
+    {
+      "crossentropy": 2.880888819694519,
+      "epoch": 7.019,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000418853759765625,
+      "learning_rate": 0.0005610526315789474,
+      "loss": 57.298,
+      "step": 467
+    },
+    {
+      "crossentropy": 2.818486213684082,
+      "epoch": 7.02,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0004012425740559896,
+      "learning_rate": 0.0005600000000000001,
+      "loss": 55.9215,
+      "step": 468
+    },
+    {
+      "crossentropy": 2.765039563179016,
+      "epoch": 7.021,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0003615697224934896,
+      "learning_rate": 0.0005589473684210526,
+      "loss": 56.3919,
+      "step": 469
+    },
+    {
+      "crossentropy": 2.765025496482849,
+      "epoch": 7.022,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0003956476847330729,
+      "learning_rate": 0.0005578947368421052,
+      "loss": 56.4707,
+      "step": 470
+    },
+    {
+      "crossentropy": 2.6866267919540405,
+      "epoch": 7.023,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0003679911295572917,
+      "learning_rate": 0.0005568421052631579,
+      "loss": 54.7156,
+      "step": 471
+    },
+    {
+      "crossentropy": 2.7979997396469116,
+      "epoch": 7.024,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.00043512980143229165,
+      "learning_rate": 0.0005557894736842106,
+      "loss": 56.7267,
+      "step": 472
+    },
+    {
+      "crossentropy": 2.851110816001892,
+      "epoch": 7.025,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00043074289957682293,
+      "learning_rate": 0.0005547368421052632,
+      "loss": 56.6983,
+      "step": 473
+    },
+    {
+      "crossentropy": 2.7913917303085327,
+      "epoch": 7.026,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0004107157389322917,
+      "learning_rate": 0.0005536842105263158,
+      "loss": 55.4107,
+      "step": 474
+    },
+    {
+      "crossentropy": 2.8196051120758057,
+      "epoch": 7.027,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0004679361979166667,
+      "learning_rate": 0.0005526315789473685,
+      "loss": 57.6855,
+      "step": 475
+    },
+    {
+      "crossentropy": 2.7812657356262207,
+      "epoch": 7.028,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0004709879557291667,
+      "learning_rate": 0.0005515789473684211,
+      "loss": 56.5311,
+      "step": 476
+    },
+    {
+      "crossentropy": 2.884629964828491,
+      "epoch": 7.029,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0005205790201822917,
+      "learning_rate": 0.0005505263157894736,
+      "loss": 56.0411,
+      "step": 477
+    },
+    {
+      "crossentropy": 2.770102024078369,
+      "epoch": 7.03,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005039850870768229,
+      "learning_rate": 0.0005494736842105263,
+      "loss": 55.6079,
+      "step": 478
+    },
+    {
+      "crossentropy": 2.7415486574172974,
+      "epoch": 7.031,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.000510406494140625,
+      "learning_rate": 0.0005484210526315789,
+      "loss": 56.3453,
+      "step": 479
+    },
+    {
+      "crossentropy": 2.850640296936035,
+      "epoch": 7.032,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005035400390625,
+      "learning_rate": 0.0005473684210526317,
+      "loss": 56.8195,
+      "step": 480
+    },
+    {
+      "crossentropy": 2.7248213291168213,
+      "epoch": 7.033,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000565338134765625,
+      "learning_rate": 0.0005463157894736843,
+      "loss": 54.4893,
+      "step": 481
+    },
+    {
+      "crossentropy": 2.8087538480758667,
+      "epoch": 7.034,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005940755208333333,
+      "learning_rate": 0.0005452631578947369,
+      "loss": 56.456,
+      "step": 482
+    },
+    {
+      "crossentropy": 2.7221765518188477,
+      "epoch": 7.035,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005904515584309896,
+      "learning_rate": 0.0005442105263157895,
+      "loss": 56.3871,
+      "step": 483
+    },
+    {
+      "crossentropy": 2.897437572479248,
+      "epoch": 7.036,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0005833307902018229,
+      "learning_rate": 0.0005431578947368421,
+      "loss": 56.8503,
+      "step": 484
+    },
+    {
+      "crossentropy": 2.9512122869491577,
+      "epoch": 7.037,
+      "grad_norm": 0.8359375,
+      "grad_norm_var": 0.0009389241536458333,
+      "learning_rate": 0.0005421052631578948,
+      "loss": 57.7135,
+      "step": 485
+    },
+    {
+      "crossentropy": 2.768813133239746,
+      "epoch": 7.038,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0009236653645833333,
+      "learning_rate": 0.0005410526315789473,
+      "loss": 55.8437,
+      "step": 486
+    },
+    {
+      "crossentropy": 2.994757056236267,
+      "epoch": 7.039,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.000919342041015625,
+      "learning_rate": 0.00054,
+      "loss": 57.5798,
+      "step": 487
+    },
+    {
+      "crossentropy": 2.804057478904724,
+      "epoch": 7.04,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0007410049438476562,
+      "learning_rate": 0.0005389473684210526,
+      "loss": 56.4264,
+      "step": 488
+    },
+    {
+      "crossentropy": 2.8012847900390625,
+      "epoch": 7.041,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0007410049438476562,
+      "learning_rate": 0.0005378947368421054,
+      "loss": 55.1591,
+      "step": 489
+    },
+    {
+      "crossentropy": 2.927749991416931,
+      "epoch": 7.042,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0007138570149739583,
+      "learning_rate": 0.0005368421052631579,
+      "loss": 56.8244,
+      "step": 490
+    },
+    {
+      "crossentropy": 2.840681791305542,
+      "epoch": 7.043,
+      "grad_norm": 0.8203125,
+      "grad_norm_var": 0.0008623758951822917,
+      "learning_rate": 0.0005357894736842105,
+      "loss": 56.2898,
+      "step": 491
+    },
+    {
+      "crossentropy": 2.743423581123352,
+      "epoch": 7.044,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00087890625,
+      "learning_rate": 0.0005347368421052632,
+      "loss": 56.1168,
+      "step": 492
+    },
+    {
+      "crossentropy": 2.800847887992859,
+      "epoch": 7.045,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0008481343587239584,
+      "learning_rate": 0.0005336842105263158,
+      "loss": 55.5752,
+      "step": 493
+    },
+    {
+      "crossentropy": 2.6926422119140625,
+      "epoch": 7.046,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0009749730428059896,
+      "learning_rate": 0.0005326315789473684,
+      "loss": 54.8747,
+      "step": 494
+    },
+    {
+      "crossentropy": 2.8151170015335083,
+      "epoch": 7.047,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.000980377197265625,
+      "learning_rate": 0.000531578947368421,
+      "loss": 55.6196,
+      "step": 495
+    },
+    {
+      "crossentropy": 2.9150657653808594,
+      "epoch": 7.048,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.001076189676920573,
+      "learning_rate": 0.0005305263157894737,
+      "loss": 57.0052,
+      "step": 496
+    },
+    {
+      "crossentropy": 2.840682625770569,
+      "epoch": 7.049,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0010121663411458333,
+      "learning_rate": 0.0005294736842105264,
+      "loss": 58.2104,
+      "step": 497
+    },
+    {
+      "crossentropy": 2.8163187503814697,
+      "epoch": 7.05,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0010121663411458333,
+      "learning_rate": 0.000528421052631579,
+      "loss": 56.675,
+      "step": 498
+    },
+    {
+      "crossentropy": 2.9374297857284546,
+      "epoch": 7.051,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0010660807291666666,
+      "learning_rate": 0.0005273684210526316,
+      "loss": 57.6299,
+      "step": 499
+    },
+    {
+      "crossentropy": 2.9131165742874146,
+      "epoch": 7.052,
+      "grad_norm": 0.828125,
+      "grad_norm_var": 0.0012598037719726562,
+      "learning_rate": 0.0005263157894736842,
+      "loss": 58.4242,
+      "step": 500
+    },
+    {
+      "crossentropy": 2.900420308113098,
+      "epoch": 7.053,
+      "grad_norm": 0.8203125,
+      "grad_norm_var": 0.0011463801066080729,
+      "learning_rate": 0.0005252631578947369,
+      "loss": 57.3699,
+      "step": 501
+    },
+    {
+      "crossentropy": 2.9151841402053833,
+      "epoch": 7.054,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0011463801066080729,
+      "learning_rate": 0.0005242105263157895,
+      "loss": 56.6537,
+      "step": 502
+    },
+    {
+      "crossentropy": 2.9279205799102783,
+      "epoch": 7.055,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.001144854227701823,
+      "learning_rate": 0.0005231578947368421,
+      "loss": 57.4459,
+      "step": 503
+    },
+    {
+      "crossentropy": 2.9333810806274414,
+      "epoch": 7.056,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.001114336649576823,
+      "learning_rate": 0.0005221052631578947,
+      "loss": 57.2862,
+      "step": 504
+    },
+    {
+      "crossentropy": 2.805972099304199,
+      "epoch": 7.057,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0011566162109375,
+      "learning_rate": 0.0005210526315789474,
+      "loss": 56.7537,
+      "step": 505
+    },
+    {
+      "crossentropy": 2.8755249977111816,
+      "epoch": 7.058,
+      "grad_norm": 0.82421875,
+      "grad_norm_var": 0.0013111750284830728,
+      "learning_rate": 0.0005200000000000001,
+      "loss": 58.3636,
+      "step": 506
+    },
+    {
+      "crossentropy": 2.9246153831481934,
+      "epoch": 7.059,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0012326558430989583,
+      "learning_rate": 0.0005189473684210526,
+      "loss": 57.2294,
+      "step": 507
+    },
+    {
+      "crossentropy": 2.8592110872268677,
+      "epoch": 7.06,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00120849609375,
+      "learning_rate": 0.0005178947368421053,
+      "loss": 57.1909,
+      "step": 508
+    },
+    {
+      "crossentropy": 2.755800485610962,
+      "epoch": 7.061,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0011682510375976562,
+      "learning_rate": 0.0005168421052631579,
+      "loss": 55.5823,
+      "step": 509
+    },
+    {
+      "crossentropy": 2.7610713243484497,
+      "epoch": 7.062,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0010129292805989584,
+      "learning_rate": 0.0005157894736842106,
+      "loss": 55.2065,
+      "step": 510
+    },
+    {
+      "crossentropy": 2.8712987899780273,
+      "epoch": 7.063,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0010129292805989584,
+      "learning_rate": 0.0005147368421052631,
+      "loss": 56.8722,
+      "step": 511
+    },
+    {
+      "crossentropy": 2.7960678339004517,
+      "epoch": 7.064,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0009702046712239584,
+      "learning_rate": 0.0005136842105263157,
+      "loss": 55.4123,
+      "step": 512
+    },
+    {
+      "crossentropy": 2.8373619318008423,
+      "epoch": 8.001,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0009103775024414063,
+      "learning_rate": 0.0005126315789473685,
+      "loss": 57.6566,
+      "step": 513
+    },
+    {
+      "crossentropy": 2.7967774868011475,
+      "epoch": 8.002,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0008508682250976563,
+      "learning_rate": 0.0005115789473684211,
+      "loss": 57.0669,
+      "step": 514
+    },
+    {
+      "crossentropy": 2.795915961265564,
+      "epoch": 8.003,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0006955464680989584,
+      "learning_rate": 0.0005105263157894738,
+      "loss": 54.5675,
+      "step": 515
+    },
+    {
+      "crossentropy": 2.754665493965149,
+      "epoch": 8.004,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005246480305989583,
+      "learning_rate": 0.0005094736842105263,
+      "loss": 56.3108,
+      "step": 516
+    },
+    {
+      "crossentropy": 2.8175487518310547,
+      "epoch": 8.005,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005121231079101562,
+      "learning_rate": 0.000508421052631579,
+      "loss": 56.5216,
+      "step": 517
+    },
+    {
+      "crossentropy": 2.884229898452759,
+      "epoch": 8.006,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.000531005859375,
+      "learning_rate": 0.0005073684210526316,
+      "loss": 56.2322,
+      "step": 518
+    },
+    {
+      "crossentropy": 2.8160749673843384,
+      "epoch": 8.007,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005578994750976562,
+      "learning_rate": 0.0005063157894736841,
+      "loss": 56.09,
+      "step": 519
+    },
+    {
+      "crossentropy": 2.8993260860443115,
+      "epoch": 8.008,
+      "grad_norm": 0.8125,
+      "grad_norm_var": 0.000665728251139323,
+      "learning_rate": 0.0005052631578947368,
+      "loss": 56.7618,
+      "step": 520
+    },
+    {
+      "crossentropy": 2.8044674396514893,
+      "epoch": 8.009,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00045878092447916666,
+      "learning_rate": 0.0005042105263157895,
+      "loss": 56.7056,
+      "step": 521
+    },
+    {
+      "crossentropy": 2.688468098640442,
+      "epoch": 8.01,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0006321589152018229,
+      "learning_rate": 0.0005031578947368422,
+      "loss": 54.6439,
+      "step": 522
+    },
+    {
+      "crossentropy": 2.77541720867157,
+      "epoch": 8.011,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006469090779622396,
+      "learning_rate": 0.0005021052631578948,
+      "loss": 55.0559,
+      "step": 523
+    },
+    {
+      "crossentropy": 2.9500958919525146,
+      "epoch": 8.012,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0006398518880208333,
+      "learning_rate": 0.0005010526315789474,
+      "loss": 58.3708,
+      "step": 524
+    },
+    {
+      "crossentropy": 2.7789478302001953,
+      "epoch": 8.013,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0008040746053059896,
+      "learning_rate": 0.0005,
+      "loss": 56.1477,
+      "step": 525
+    },
+    {
+      "crossentropy": 2.89370858669281,
+      "epoch": 8.014,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0008819580078125,
+      "learning_rate": 0.0004989473684210527,
+      "loss": 57.8208,
+      "step": 526
+    },
+    {
+      "crossentropy": 2.829719305038452,
+      "epoch": 8.015,
+      "grad_norm": 0.828125,
+      "grad_norm_var": 0.001163482666015625,
+      "learning_rate": 0.0004978947368421053,
+      "loss": 57.3503,
+      "step": 527
+    },
+    {
+      "crossentropy": 2.8827792406082153,
+      "epoch": 8.016,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.001163482666015625,
+      "learning_rate": 0.0004968421052631579,
+      "loss": 57.5831,
+      "step": 528
+    },
+    {
+      "crossentropy": 2.839862585067749,
+      "epoch": 8.017,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0011484146118164063,
+      "learning_rate": 0.0004957894736842105,
+      "loss": 55.7402,
+      "step": 529
+    },
+    {
+      "crossentropy": 2.7569580078125,
+      "epoch": 8.018,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0011270523071289062,
+      "learning_rate": 0.0004947368421052632,
+      "loss": 56.3813,
+      "step": 530
+    },
+    {
+      "crossentropy": 2.771231174468994,
+      "epoch": 8.019,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.001053619384765625,
+      "learning_rate": 0.0004936842105263158,
+      "loss": 57.1672,
+      "step": 531
+    },
+    {
+      "crossentropy": 2.787784457206726,
+      "epoch": 8.02,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0010528564453125,
+      "learning_rate": 0.0004926315789473684,
+      "loss": 56.1965,
+      "step": 532
+    },
+    {
+      "crossentropy": 2.716630458831787,
+      "epoch": 8.021,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0010461807250976562,
+      "learning_rate": 0.000491578947368421,
+      "loss": 55.2343,
+      "step": 533
+    },
+    {
+      "crossentropy": 2.8388242721557617,
+      "epoch": 8.022,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.001084136962890625,
+      "learning_rate": 0.0004905263157894737,
+      "loss": 56.8292,
+      "step": 534
+    },
+    {
+      "crossentropy": 2.8070307970046997,
+      "epoch": 8.023,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0010294596354166667,
+      "learning_rate": 0.0004894736842105264,
+      "loss": 55.36,
+      "step": 535
+    },
+    {
+      "crossentropy": 2.865221381187439,
+      "epoch": 8.024,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0009108861287434896,
+      "learning_rate": 0.000488421052631579,
+      "loss": 56.6106,
+      "step": 536
+    },
+    {
+      "crossentropy": 2.9020869731903076,
+      "epoch": 8.025,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0009165445963541667,
+      "learning_rate": 0.0004873684210526316,
+      "loss": 56.5227,
+      "step": 537
+    },
+    {
+      "crossentropy": 2.7119977474212646,
+      "epoch": 8.026,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0007786432902018229,
+      "learning_rate": 0.0004863157894736842,
+      "loss": 55.6506,
+      "step": 538
+    },
+    {
+      "crossentropy": 2.7730578184127808,
+      "epoch": 8.027,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0007781982421875,
+      "learning_rate": 0.00048526315789473683,
+      "loss": 56.5655,
+      "step": 539
+    },
+    {
+      "crossentropy": 2.736594557762146,
+      "epoch": 8.028,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0007769266764322916,
+      "learning_rate": 0.0004842105263157895,
+      "loss": 55.2709,
+      "step": 540
+    },
+    {
+      "crossentropy": 2.87870717048645,
+      "epoch": 8.029,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0006158828735351562,
+      "learning_rate": 0.00048315789473684213,
+      "loss": 56.1277,
+      "step": 541
+    },
+    {
+      "crossentropy": 2.8086968660354614,
+      "epoch": 8.03,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0006158828735351562,
+      "learning_rate": 0.00048210526315789476,
+      "loss": 56.0955,
+      "step": 542
+    },
+    {
+      "crossentropy": 2.807764172554016,
+      "epoch": 8.031,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00036869049072265627,
+      "learning_rate": 0.00048105263157894733,
+      "loss": 55.8439,
+      "step": 543
+    },
+    {
+      "crossentropy": 2.887509346008301,
+      "epoch": 8.032,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0005124409993489584,
+      "learning_rate": 0.00048,
+      "loss": 57.691,
+      "step": 544
+    },
+    {
+      "crossentropy": 2.8629679679870605,
+      "epoch": 8.033,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004961649576822917,
+      "learning_rate": 0.00047894736842105264,
+      "loss": 57.1656,
+      "step": 545
+    },
+    {
+      "crossentropy": 2.749383807182312,
+      "epoch": 8.034,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0004968643188476562,
+      "learning_rate": 0.00047789473684210526,
+      "loss": 54.5942,
+      "step": 546
+    },
+    {
+      "crossentropy": 2.771678328514099,
+      "epoch": 8.035,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00044091542561848957,
+      "learning_rate": 0.0004768421052631579,
+      "loss": 56.419,
+      "step": 547
+    },
+    {
+      "crossentropy": 2.6900097131729126,
+      "epoch": 8.036,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0004968643188476562,
+      "learning_rate": 0.00047578947368421057,
+      "loss": 55.0504,
+      "step": 548
+    },
+    {
+      "crossentropy": 2.769201636314392,
+      "epoch": 8.037,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000510406494140625,
+      "learning_rate": 0.0004747368421052632,
+      "loss": 55.9095,
+      "step": 549
+    },
+    {
+      "crossentropy": 2.9785958528518677,
+      "epoch": 8.038,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0006134033203125,
+      "learning_rate": 0.00047368421052631577,
+      "loss": 57.2934,
+      "step": 550
+    },
+    {
+      "crossentropy": 2.833102226257324,
+      "epoch": 8.039,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.000635210673014323,
+      "learning_rate": 0.0004726315789473684,
+      "loss": 57.6031,
+      "step": 551
+    },
+    {
+      "crossentropy": 2.8962067365646362,
+      "epoch": 8.04,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005619684855143229,
+      "learning_rate": 0.0004715789473684211,
+      "loss": 56.1431,
+      "step": 552
+    },
+    {
+      "crossentropy": 2.8563923835754395,
+      "epoch": 8.041,
+      "grad_norm": 0.82421875,
+      "grad_norm_var": 0.0008396784464518229,
+      "learning_rate": 0.0004705263157894737,
+      "loss": 58.5512,
+      "step": 553
+    },
+    {
+      "crossentropy": 2.8052828311920166,
+      "epoch": 8.042,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0008849461873372395,
+      "learning_rate": 0.0004694736842105263,
+      "loss": 54.951,
+      "step": 554
+    },
+    {
+      "crossentropy": 2.726549983024597,
+      "epoch": 8.043,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0009518941243489583,
+      "learning_rate": 0.00046842105263157895,
+      "loss": 55.8634,
+      "step": 555
+    },
+    {
+      "crossentropy": 2.8912360668182373,
+      "epoch": 8.044,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0009719212849934896,
+      "learning_rate": 0.00046736842105263163,
+      "loss": 56.6226,
+      "step": 556
+    },
+    {
+      "crossentropy": 2.8774843215942383,
+      "epoch": 8.045,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0009844462076822917,
+      "learning_rate": 0.0004663157894736842,
+      "loss": 57.6825,
+      "step": 557
+    },
+    {
+      "crossentropy": 2.7926796674728394,
+      "epoch": 8.046,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0009266535441080729,
+      "learning_rate": 0.00046526315789473683,
+      "loss": 56.6319,
+      "step": 558
+    },
+    {
+      "crossentropy": 2.8781834840774536,
+      "epoch": 8.047,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0008900960286458334,
+      "learning_rate": 0.00046421052631578946,
+      "loss": 57.2263,
+      "step": 559
+    },
+    {
+      "crossentropy": 2.768616557121277,
+      "epoch": 8.048,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0007138570149739583,
+      "learning_rate": 0.00046315789473684214,
+      "loss": 55.393,
+      "step": 560
+    },
+    {
+      "crossentropy": 2.8414195775985718,
+      "epoch": 8.049,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0007196426391601563,
+      "learning_rate": 0.00046210526315789476,
+      "loss": 56.6017,
+      "step": 561
+    },
+    {
+      "crossentropy": 2.886254072189331,
+      "epoch": 8.05,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0007079442342122396,
+      "learning_rate": 0.0004610526315789474,
+      "loss": 57.1002,
+      "step": 562
+    },
+    {
+      "crossentropy": 2.882540702819824,
+      "epoch": 8.051,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0007521947224934896,
+      "learning_rate": 0.00046,
+      "loss": 57.4033,
+      "step": 563
+    },
+    {
+      "crossentropy": 2.8176023960113525,
+      "epoch": 8.052,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0007369359334309896,
+      "learning_rate": 0.00045894736842105264,
+      "loss": 55.8245,
+      "step": 564
+    },
+    {
+      "crossentropy": 2.8578507900238037,
+      "epoch": 8.053,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0007994969685872396,
+      "learning_rate": 0.00045789473684210527,
+      "loss": 55.7208,
+      "step": 565
+    },
+    {
+      "crossentropy": 2.907991409301758,
+      "epoch": 8.054,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0006820042928059896,
+      "learning_rate": 0.0004568421052631579,
+      "loss": 57.6047,
+      "step": 566
+    },
+    {
+      "crossentropy": 2.9070833921432495,
+      "epoch": 8.055,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006647109985351562,
+      "learning_rate": 0.0004557894736842105,
+      "loss": 56.8514,
+      "step": 567
+    },
+    {
+      "crossentropy": 2.8488681316375732,
+      "epoch": 8.056,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.000757280985514323,
+      "learning_rate": 0.0004547368421052632,
+      "loss": 55.2764,
+      "step": 568
+    },
+    {
+      "crossentropy": 2.843410849571228,
+      "epoch": 8.057,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0003986994425455729,
+      "learning_rate": 0.0004536842105263158,
+      "loss": 57.3239,
+      "step": 569
+    },
+    {
+      "crossentropy": 2.7495408058166504,
+      "epoch": 8.058,
+      "grad_norm": 0.8203125,
+      "grad_norm_var": 0.000689697265625,
+      "learning_rate": 0.00045263157894736845,
+      "loss": 56.725,
+      "step": 570
+    },
+    {
+      "crossentropy": 2.8769084215164185,
+      "epoch": 8.059,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0006743748982747396,
+      "learning_rate": 0.0004515789473684211,
+      "loss": 58.1138,
+      "step": 571
+    },
+    {
+      "crossentropy": 2.865220785140991,
+      "epoch": 8.06,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0006632486979166667,
+      "learning_rate": 0.00045052631578947365,
+      "loss": 57.7083,
+      "step": 572
+    },
+    {
+      "crossentropy": 2.7678266763687134,
+      "epoch": 8.061,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006331761678059895,
+      "learning_rate": 0.00044947368421052633,
+      "loss": 56.7169,
+      "step": 573
+    },
+    {
+      "crossentropy": 2.716267228126526,
+      "epoch": 8.062,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006306330362955729,
+      "learning_rate": 0.00044842105263157895,
+      "loss": 55.4713,
+      "step": 574
+    },
+    {
+      "crossentropy": 2.746058225631714,
+      "epoch": 8.063,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000632476806640625,
+      "learning_rate": 0.0004473684210526316,
+      "loss": 56.1347,
+      "step": 575
+    },
+    {
+      "crossentropy": 2.8452965021133423,
+      "epoch": 8.064,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005889892578125,
+      "learning_rate": 0.0004463157894736842,
+      "loss": 57.0829,
+      "step": 576
+    },
+    {
+      "crossentropy": 2.7510355710983276,
+      "epoch": 9.001,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0006367365519205729,
+      "learning_rate": 0.0004452631578947369,
+      "loss": 56.112,
+      "step": 577
+    },
+    {
+      "crossentropy": 2.735834240913391,
+      "epoch": 9.002,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006235758463541667,
+      "learning_rate": 0.0004442105263157895,
+      "loss": 55.181,
+      "step": 578
+    },
+    {
+      "crossentropy": 2.731901526451111,
+      "epoch": 9.003,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000640869140625,
+      "learning_rate": 0.0004431578947368421,
+      "loss": 56.0573,
+      "step": 579
+    },
+    {
+      "crossentropy": 2.9925543069839478,
+      "epoch": 9.004,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0006418228149414062,
+      "learning_rate": 0.0004421052631578947,
+      "loss": 57.7087,
+      "step": 580
+    },
+    {
+      "crossentropy": 2.794908046722412,
+      "epoch": 9.005,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006575902303059896,
+      "learning_rate": 0.0004410526315789474,
+      "loss": 55.5605,
+      "step": 581
+    },
+    {
+      "crossentropy": 2.9317067861557007,
+      "epoch": 9.006,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0006458918253580729,
+      "learning_rate": 0.00044,
+      "loss": 58.2652,
+      "step": 582
+    },
+    {
+      "crossentropy": 2.9609873294830322,
+      "epoch": 9.007,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005431493123372396,
+      "learning_rate": 0.00043894736842105264,
+      "loss": 58.6218,
+      "step": 583
+    },
+    {
+      "crossentropy": 2.7990305423736572,
+      "epoch": 9.008,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005462010701497396,
+      "learning_rate": 0.00043789473684210527,
+      "loss": 57.2674,
+      "step": 584
+    },
+    {
+      "crossentropy": 2.7703919410705566,
+      "epoch": 9.009,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0002913792928059896,
+      "learning_rate": 0.00043684210526315795,
+      "loss": 55.587,
+      "step": 585
+    },
+    {
+      "crossentropy": 2.887115478515625,
+      "epoch": 9.01,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.00040486653645833336,
+      "learning_rate": 0.0004357894736842105,
+      "loss": 56.62,
+      "step": 586
+    },
+    {
+      "crossentropy": 2.9152616262435913,
+      "epoch": 9.011,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0004546483357747396,
+      "learning_rate": 0.00043473684210526315,
+      "loss": 56.9964,
+      "step": 587
+    },
+    {
+      "crossentropy": 2.8132872581481934,
+      "epoch": 9.012,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00042972564697265627,
+      "learning_rate": 0.00043368421052631577,
+      "loss": 55.0605,
+      "step": 588
+    },
+    {
+      "crossentropy": 2.7516273260116577,
+      "epoch": 9.013,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.000437164306640625,
+      "learning_rate": 0.00043263157894736845,
+      "loss": 56.6974,
+      "step": 589
+    },
+    {
+      "crossentropy": 2.7954412698745728,
+      "epoch": 9.014,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00043837229410807293,
+      "learning_rate": 0.0004315789473684211,
+      "loss": 56.952,
+      "step": 590
+    },
+    {
+      "crossentropy": 2.7435046434402466,
+      "epoch": 9.015,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004514058430989583,
+      "learning_rate": 0.0004305263157894737,
+      "loss": 56.8651,
+      "step": 591
+    },
+    {
+      "crossentropy": 2.85153329372406,
+      "epoch": 9.016,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0004892349243164062,
+      "learning_rate": 0.00042947368421052633,
+      "loss": 55.8497,
+      "step": 592
+    },
+    {
+      "crossentropy": 2.698570728302002,
+      "epoch": 9.017,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00039265950520833335,
+      "learning_rate": 0.00042842105263157896,
+      "loss": 55.0285,
+      "step": 593
+    },
+    {
+      "crossentropy": 2.8842196464538574,
+      "epoch": 9.018,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00041681925455729166,
+      "learning_rate": 0.0004273684210526316,
+      "loss": 56.7253,
+      "step": 594
+    },
+    {
+      "crossentropy": 2.703991651535034,
+      "epoch": 9.019,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000494384765625,
+      "learning_rate": 0.0004263157894736842,
+      "loss": 54.9482,
+      "step": 595
+    },
+    {
+      "crossentropy": 2.828360438346863,
+      "epoch": 9.02,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00047092437744140626,
+      "learning_rate": 0.00042526315789473683,
+      "loss": 56.6368,
+      "step": 596
+    },
+    {
+      "crossentropy": 2.7593008279800415,
+      "epoch": 9.021,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005039850870768229,
+      "learning_rate": 0.0004242105263157895,
+      "loss": 56.4671,
+      "step": 597
+    },
+    {
+      "crossentropy": 2.716967463493347,
+      "epoch": 9.022,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005449930826822917,
+      "learning_rate": 0.00042315789473684214,
+      "loss": 55.5724,
+      "step": 598
+    },
+    {
+      "crossentropy": 2.774570941925049,
+      "epoch": 9.023,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005816141764322917,
+      "learning_rate": 0.00042210526315789477,
+      "loss": 56.5112,
+      "step": 599
+    },
+    {
+      "crossentropy": 2.72983181476593,
+      "epoch": 9.024,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005899429321289063,
+      "learning_rate": 0.00042105263157894734,
+      "loss": 55.1346,
+      "step": 600
+    },
+    {
+      "crossentropy": 2.8420827388763428,
+      "epoch": 9.025,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005706787109375,
+      "learning_rate": 0.00042,
+      "loss": 56.3536,
+      "step": 601
+    },
+    {
+      "crossentropy": 2.880441188812256,
+      "epoch": 9.026,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00036722819010416666,
+      "learning_rate": 0.00041894736842105264,
+      "loss": 57.4732,
+      "step": 602
+    },
+    {
+      "crossentropy": 2.8829102516174316,
+      "epoch": 9.027,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0003374735514322917,
+      "learning_rate": 0.00041789473684210527,
+      "loss": 57.025,
+      "step": 603
+    },
+    {
+      "crossentropy": 2.7579070329666138,
+      "epoch": 9.028,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00033976236979166665,
+      "learning_rate": 0.0004168421052631579,
+      "loss": 54.3299,
+      "step": 604
+    },
+    {
+      "crossentropy": 2.79787278175354,
+      "epoch": 9.029,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003890355428059896,
+      "learning_rate": 0.0004157894736842106,
+      "loss": 56.4175,
+      "step": 605
+    },
+    {
+      "crossentropy": 2.814871907234192,
+      "epoch": 9.03,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0004109064737955729,
+      "learning_rate": 0.0004147368421052632,
+      "loss": 56.6412,
+      "step": 606
+    },
+    {
+      "crossentropy": 2.9164552688598633,
+      "epoch": 9.031,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00038553873697916665,
+      "learning_rate": 0.0004136842105263158,
+      "loss": 56.5248,
+      "step": 607
+    },
+    {
+      "crossentropy": 2.851742148399353,
+      "epoch": 9.032,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.000394439697265625,
+      "learning_rate": 0.0004126315789473684,
+      "loss": 57.026,
+      "step": 608
+    },
+    {
+      "crossentropy": 2.8921687602996826,
+      "epoch": 9.033,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.000424957275390625,
+      "learning_rate": 0.0004115789473684211,
+      "loss": 56.07,
+      "step": 609
+    },
+    {
+      "crossentropy": 2.748379707336426,
+      "epoch": 9.034,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.0004547119140625,
+      "learning_rate": 0.0004105263157894737,
+      "loss": 54.6257,
+      "step": 610
+    },
+    {
+      "crossentropy": 2.8249024152755737,
+      "epoch": 9.035,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004124323527018229,
+      "learning_rate": 0.00040947368421052633,
+      "loss": 55.8105,
+      "step": 611
+    },
+    {
+      "crossentropy": 2.793903946876526,
+      "epoch": 9.036,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0004124323527018229,
+      "learning_rate": 0.00040842105263157896,
+      "loss": 56.7126,
+      "step": 612
+    },
+    {
+      "crossentropy": 2.8926647901535034,
+      "epoch": 9.037,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0004018147786458333,
+      "learning_rate": 0.00040736842105263164,
+      "loss": 57.0243,
+      "step": 613
+    },
+    {
+      "crossentropy": 2.722157835960388,
+      "epoch": 9.038,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003448486328125,
+      "learning_rate": 0.0004063157894736842,
+      "loss": 56.15,
+      "step": 614
+    },
+    {
+      "crossentropy": 2.879178524017334,
+      "epoch": 9.039,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.000333404541015625,
+      "learning_rate": 0.00040526315789473684,
+      "loss": 56.5565,
+      "step": 615
+    },
+    {
+      "crossentropy": 2.870532989501953,
+      "epoch": 9.04,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00032450358072916665,
+      "learning_rate": 0.00040421052631578946,
+      "loss": 57.2944,
+      "step": 616
+    },
+    {
+      "crossentropy": 2.6587058305740356,
+      "epoch": 9.041,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003513971964518229,
+      "learning_rate": 0.0004031578947368421,
+      "loss": 55.9194,
+      "step": 617
+    },
+    {
+      "crossentropy": 2.8166728019714355,
+      "epoch": 9.042,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00035800933837890623,
+      "learning_rate": 0.00040210526315789477,
+      "loss": 56.4466,
+      "step": 618
+    },
+    {
+      "crossentropy": 2.7822681665420532,
+      "epoch": 9.043,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003753662109375,
+      "learning_rate": 0.0004010526315789474,
+      "loss": 55.9381,
+      "step": 619
+    },
+    {
+      "crossentropy": 2.7443701028823853,
+      "epoch": 9.044,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00037784576416015624,
+      "learning_rate": 0.0004,
+      "loss": 55.3256,
+      "step": 620
+    },
+    {
+      "crossentropy": 2.77031409740448,
+      "epoch": 9.045,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0003371556599934896,
+      "learning_rate": 0.0003989473684210526,
+      "loss": 55.9228,
+      "step": 621
+    },
+    {
+      "crossentropy": 2.7815150022506714,
+      "epoch": 9.046,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0003371556599934896,
+      "learning_rate": 0.00039789473684210527,
+      "loss": 55.1725,
+      "step": 622
+    },
+    {
+      "crossentropy": 2.8636655807495117,
+      "epoch": 9.047,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0003371556599934896,
+      "learning_rate": 0.0003968421052631579,
+      "loss": 56.2236,
+      "step": 623
+    },
+    {
+      "crossentropy": 2.8161659240722656,
+      "epoch": 9.048,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003356297810872396,
+      "learning_rate": 0.0003957894736842105,
+      "loss": 56.7137,
+      "step": 624
+    },
+    {
+      "crossentropy": 2.873541831970215,
+      "epoch": 9.049,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00027561187744140625,
+      "learning_rate": 0.00039473684210526315,
+      "loss": 58.0352,
+      "step": 625
+    },
+    {
+      "crossentropy": 2.830682158470154,
+      "epoch": 9.05,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00028934478759765623,
+      "learning_rate": 0.00039368421052631583,
+      "loss": 56.4589,
+      "step": 626
+    },
+    {
+      "crossentropy": 2.8676915168762207,
+      "epoch": 9.051,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0002979914347330729,
+      "learning_rate": 0.00039263157894736846,
+      "loss": 57.3925,
+      "step": 627
+    },
+    {
+      "crossentropy": 2.846311330795288,
+      "epoch": 9.052,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00031229654947916664,
+      "learning_rate": 0.000391578947368421,
+      "loss": 57.5314,
+      "step": 628
+    },
+    {
+      "crossentropy": 2.8266266584396362,
+      "epoch": 9.053,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00029474894205729166,
+      "learning_rate": 0.00039052631578947365,
+      "loss": 57.0562,
+      "step": 629
+    },
+    {
+      "crossentropy": 2.8677502870559692,
+      "epoch": 9.054,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00029474894205729166,
+      "learning_rate": 0.00038947368421052633,
+      "loss": 56.3468,
+      "step": 630
+    },
+    {
+      "crossentropy": 2.819862723350525,
+      "epoch": 9.055,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.00029474894205729166,
+      "learning_rate": 0.00038842105263157896,
+      "loss": 57.1646,
+      "step": 631
+    },
+    {
+      "crossentropy": 2.8373594284057617,
+      "epoch": 9.056,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0002873102823893229,
+      "learning_rate": 0.0003873684210526316,
+      "loss": 57.8049,
+      "step": 632
+    },
+    {
+      "crossentropy": 2.844488739967346,
+      "epoch": 9.057,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0002440770467122396,
+      "learning_rate": 0.0003863157894736842,
+      "loss": 57.1337,
+      "step": 633
+    },
+    {
+      "crossentropy": 2.8051549196243286,
+      "epoch": 9.058,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0002227783203125,
+      "learning_rate": 0.0003852631578947369,
+      "loss": 55.1455,
+      "step": 634
+    },
+    {
+      "crossentropy": 2.889759659767151,
+      "epoch": 9.059,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00022373199462890624,
+      "learning_rate": 0.00038421052631578946,
+      "loss": 57.317,
+      "step": 635
+    },
+    {
+      "crossentropy": 2.9444185495376587,
+      "epoch": 9.06,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0002135594685872396,
+      "learning_rate": 0.0003831578947368421,
+      "loss": 57.7695,
+      "step": 636
+    },
+    {
+      "crossentropy": 2.893734097480774,
+      "epoch": 9.061,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.00026543935139973957,
+      "learning_rate": 0.0003821052631578947,
+      "loss": 56.677,
+      "step": 637
+    },
+    {
+      "crossentropy": 2.8171656131744385,
+      "epoch": 9.062,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00028661092122395836,
+      "learning_rate": 0.0003810526315789474,
+      "loss": 55.8621,
+      "step": 638
+    },
+    {
+      "crossentropy": 2.9001771211624146,
+      "epoch": 9.063,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00028661092122395836,
+      "learning_rate": 0.00038,
+      "loss": 58.2953,
+      "step": 639
+    },
+    {
+      "crossentropy": 2.8124544620513916,
+      "epoch": 9.064,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.000241851806640625,
+      "learning_rate": 0.00037894736842105265,
+      "loss": 56.2274,
+      "step": 640
+    },
+    {
+      "crossentropy": 2.691709280014038,
+      "epoch": 10.001,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.00030492146809895835,
+      "learning_rate": 0.00037789473684210527,
+      "loss": 53.1371,
+      "step": 641
+    },
+    {
+      "crossentropy": 2.748944401741028,
+      "epoch": 10.002,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00030358632405598957,
+      "learning_rate": 0.0003768421052631579,
+      "loss": 55.4782,
+      "step": 642
+    },
+    {
+      "crossentropy": 2.8715380430221558,
+      "epoch": 10.003,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003127415974934896,
+      "learning_rate": 0.0003757894736842105,
+      "loss": 56.7533,
+      "step": 643
+    },
+    {
+      "crossentropy": 2.7285449504852295,
+      "epoch": 10.004,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0003110249837239583,
+      "learning_rate": 0.00037473684210526315,
+      "loss": 56.2589,
+      "step": 644
+    },
+    {
+      "crossentropy": 2.9220064878463745,
+      "epoch": 10.005,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003110249837239583,
+      "learning_rate": 0.0003736842105263158,
+      "loss": 55.9243,
+      "step": 645
+    },
+    {
+      "crossentropy": 2.886649250984192,
+      "epoch": 10.006,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0002898534138997396,
+      "learning_rate": 0.00037263157894736846,
+      "loss": 58.1926,
+      "step": 646
+    },
+    {
+      "crossentropy": 2.7266123294830322,
+      "epoch": 10.007,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0002898534138997396,
+      "learning_rate": 0.0003715789473684211,
+      "loss": 56.4874,
+      "step": 647
+    },
+    {
+      "crossentropy": 2.8823474645614624,
+      "epoch": 10.008,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0003193537394205729,
+      "learning_rate": 0.0003705263157894737,
+      "loss": 56.2834,
+      "step": 648
+    },
+    {
+      "crossentropy": 2.8225646018981934,
+      "epoch": 10.009,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0004058202107747396,
+      "learning_rate": 0.00036947368421052633,
+      "loss": 56.0732,
+      "step": 649
+    },
+    {
+      "crossentropy": 2.7276248931884766,
+      "epoch": 10.01,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00042622884114583334,
+      "learning_rate": 0.00036842105263157896,
+      "loss": 54.2119,
+      "step": 650
+    },
+    {
+      "crossentropy": 2.851339101791382,
+      "epoch": 10.011,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00046971638997395836,
+      "learning_rate": 0.0003673684210526316,
+      "loss": 57.2307,
+      "step": 651
+    },
+    {
+      "crossentropy": 2.81039559841156,
+      "epoch": 10.012,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003890355428059896,
+      "learning_rate": 0.0003663157894736842,
+      "loss": 55.5788,
+      "step": 652
+    },
+    {
+      "crossentropy": 2.917648434638977,
+      "epoch": 10.013,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.00048726399739583334,
+      "learning_rate": 0.00036526315789473684,
+      "loss": 58.2029,
+      "step": 653
+    },
+    {
+      "crossentropy": 2.8261080980300903,
+      "epoch": 10.014,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005116780598958333,
+      "learning_rate": 0.0003642105263157895,
+      "loss": 56.476,
+      "step": 654
+    },
+    {
+      "crossentropy": 2.7618919610977173,
+      "epoch": 10.015,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005080540974934896,
+      "learning_rate": 0.00036315789473684214,
+      "loss": 55.7106,
+      "step": 655
+    },
+    {
+      "crossentropy": 2.804520845413208,
+      "epoch": 10.016,
+      "grad_norm": 0.7890625,
+      "grad_norm_var": 0.0005889256795247396,
+      "learning_rate": 0.00036210526315789477,
+      "loss": 56.374,
+      "step": 656
+    },
+    {
+      "crossentropy": 2.751194953918457,
+      "epoch": 10.017,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0005141576131184896,
+      "learning_rate": 0.00036105263157894734,
+      "loss": 55.439,
+      "step": 657
+    },
+    {
+      "crossentropy": 2.893984794616699,
+      "epoch": 10.018,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005156834920247396,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 59.3323,
+      "step": 658
+    },
+    {
+      "crossentropy": 2.833603858947754,
+      "epoch": 10.019,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0005395889282226562,
+      "learning_rate": 0.00035894736842105265,
+      "loss": 56.9289,
+      "step": 659
+    },
+    {
+      "crossentropy": 2.7936803102493286,
+      "epoch": 10.02,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005368550618489583,
+      "learning_rate": 0.0003578947368421053,
+      "loss": 55.5988,
+      "step": 660
+    },
+    {
+      "crossentropy": 2.797438144683838,
+      "epoch": 10.021,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000537109375,
+      "learning_rate": 0.0003568421052631579,
+      "loss": 56.618,
+      "step": 661
+    },
+    {
+      "crossentropy": 2.6816130876541138,
+      "epoch": 10.022,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0006326675415039062,
+      "learning_rate": 0.0003557894736842105,
+      "loss": 55.4491,
+      "step": 662
+    },
+    {
+      "crossentropy": 2.775187849998474,
+      "epoch": 10.023,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0007029215494791667,
+      "learning_rate": 0.0003547368421052632,
+      "loss": 54.9814,
+      "step": 663
+    },
+    {
+      "crossentropy": 2.9283132553100586,
+      "epoch": 10.024,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006931940714518229,
+      "learning_rate": 0.0003536842105263158,
+      "loss": 57.377,
+      "step": 664
+    },
+    {
+      "crossentropy": 2.754921555519104,
+      "epoch": 10.025,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000608062744140625,
+      "learning_rate": 0.0003526315789473684,
+      "loss": 55.0399,
+      "step": 665
+    },
+    {
+      "crossentropy": 2.87182080745697,
+      "epoch": 10.026,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005950927734375,
+      "learning_rate": 0.00035157894736842103,
+      "loss": 56.8635,
+      "step": 666
+    },
+    {
+      "crossentropy": 2.787114143371582,
+      "epoch": 10.027,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0006545384724934896,
+      "learning_rate": 0.0003505263157894737,
+      "loss": 56.3069,
+      "step": 667
+    },
+    {
+      "crossentropy": 2.770491600036621,
+      "epoch": 10.028,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0006581624348958333,
+      "learning_rate": 0.00034947368421052634,
+      "loss": 56.8387,
+      "step": 668
+    },
+    {
+      "crossentropy": 2.8516522645950317,
+      "epoch": 10.029,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005614598592122396,
+      "learning_rate": 0.00034842105263157896,
+      "loss": 56.9971,
+      "step": 669
+    },
+    {
+      "crossentropy": 2.750314712524414,
+      "epoch": 10.03,
+      "grad_norm": 0.69921875,
+      "grad_norm_var": 0.0007109959920247396,
+      "learning_rate": 0.0003473684210526316,
+      "loss": 54.2891,
+      "step": 670
+    },
+    {
+      "crossentropy": 2.7711139917373657,
+      "epoch": 10.031,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0007138570149739583,
+      "learning_rate": 0.0003463157894736842,
+      "loss": 55.5093,
+      "step": 671
+    },
+    {
+      "crossentropy": 2.7986541986465454,
+      "epoch": 10.032,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006138483683268229,
+      "learning_rate": 0.00034526315789473684,
+      "loss": 56.5041,
+      "step": 672
+    },
+    {
+      "crossentropy": 2.8304941654205322,
+      "epoch": 10.033,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0005960464477539062,
+      "learning_rate": 0.00034421052631578947,
+      "loss": 56.5678,
+      "step": 673
+    },
+    {
+      "crossentropy": 2.748772144317627,
+      "epoch": 10.034,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0006336847941080729,
+      "learning_rate": 0.0003431578947368421,
+      "loss": 56.2156,
+      "step": 674
+    },
+    {
+      "crossentropy": 2.911113739013672,
+      "epoch": 10.035,
+      "grad_norm": 0.80859375,
+      "grad_norm_var": 0.0008005777994791667,
+      "learning_rate": 0.00034210526315789477,
+      "loss": 57.6437,
+      "step": 675
+    },
+    {
+      "crossentropy": 2.8851195573806763,
+      "epoch": 10.036,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0008117039998372396,
+      "learning_rate": 0.0003410526315789474,
+      "loss": 57.8075,
+      "step": 676
+    },
+    {
+      "crossentropy": 2.8386086225509644,
+      "epoch": 10.037,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.000852203369140625,
+      "learning_rate": 0.00034,
+      "loss": 56.9642,
+      "step": 677
+    },
+    {
+      "crossentropy": 2.837552309036255,
+      "epoch": 10.038,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0007862726847330729,
+      "learning_rate": 0.0003389473684210526,
+      "loss": 56.0464,
+      "step": 678
+    },
+    {
+      "crossentropy": 2.7876830101013184,
+      "epoch": 10.039,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0007181803385416667,
+      "learning_rate": 0.0003378947368421053,
+      "loss": 57.0013,
+      "step": 679
+    },
+    {
+      "crossentropy": 2.6957716941833496,
+      "epoch": 10.04,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0007008234659830729,
+      "learning_rate": 0.0003368421052631579,
+      "loss": 54.8812,
+      "step": 680
+    },
+    {
+      "crossentropy": 2.858887195587158,
+      "epoch": 10.041,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0007476806640625,
+      "learning_rate": 0.00033578947368421053,
+      "loss": 57.4649,
+      "step": 681
+    },
+    {
+      "crossentropy": 2.835731267929077,
+      "epoch": 10.042,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0007766087849934896,
+      "learning_rate": 0.00033473684210526315,
+      "loss": 57.3096,
+      "step": 682
+    },
+    {
+      "crossentropy": 2.7914698123931885,
+      "epoch": 10.043,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0006528218587239583,
+      "learning_rate": 0.00033368421052631583,
+      "loss": 54.5885,
+      "step": 683
+    },
+    {
+      "crossentropy": 2.7969202995300293,
+      "epoch": 10.044,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0006947835286458333,
+      "learning_rate": 0.00033263157894736846,
+      "loss": 55.199,
+      "step": 684
+    },
+    {
+      "crossentropy": 2.7273718118667603,
+      "epoch": 10.045,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0006947835286458333,
+      "learning_rate": 0.00033157894736842103,
+      "loss": 54.9177,
+      "step": 685
+    },
+    {
+      "crossentropy": 2.9325153827667236,
+      "epoch": 10.046,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0005777994791666667,
+      "learning_rate": 0.00033052631578947366,
+      "loss": 57.8565,
+      "step": 686
+    },
+    {
+      "crossentropy": 2.8253042697906494,
+      "epoch": 10.047,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0005955378214518229,
+      "learning_rate": 0.00032947368421052634,
+      "loss": 56.367,
+      "step": 687
+    },
+    {
+      "crossentropy": 2.9794070720672607,
+      "epoch": 10.048,
+      "grad_norm": 0.80078125,
+      "grad_norm_var": 0.0007196426391601563,
+      "learning_rate": 0.00032842105263157896,
+      "loss": 58.8961,
+      "step": 688
+    },
+    {
+      "crossentropy": 2.8630117177963257,
+      "epoch": 10.049,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.000693511962890625,
+      "learning_rate": 0.0003273684210526316,
+      "loss": 55.8875,
+      "step": 689
+    },
+    {
+      "crossentropy": 2.885907292366028,
+      "epoch": 10.05,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0007049560546875,
+      "learning_rate": 0.0003263157894736842,
+      "loss": 57.5656,
+      "step": 690
+    },
+    {
+      "crossentropy": 2.809995174407959,
+      "epoch": 10.051,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005050023396809896,
+      "learning_rate": 0.0003252631578947369,
+      "loss": 57.0467,
+      "step": 691
+    },
+    {
+      "crossentropy": 2.9418861865997314,
+      "epoch": 10.052,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00048421223958333335,
+      "learning_rate": 0.00032421052631578947,
+      "loss": 57.9401,
+      "step": 692
+    },
+    {
+      "crossentropy": 2.795152425765991,
+      "epoch": 10.053,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004597345987955729,
+      "learning_rate": 0.0003231578947368421,
+      "loss": 56.1651,
+      "step": 693
+    },
+    {
+      "crossentropy": 2.81339955329895,
+      "epoch": 10.054,
+      "grad_norm": 0.8515625,
+      "grad_norm_var": 0.0010171890258789062,
+      "learning_rate": 0.0003221052631578947,
+      "loss": 56.595,
+      "step": 694
+    },
+    {
+      "crossentropy": 2.9207112789154053,
+      "epoch": 10.055,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.001004473368326823,
+      "learning_rate": 0.0003210526315789474,
+      "loss": 57.6239,
+      "step": 695
+    },
+    {
+      "crossentropy": 2.7555843591690063,
+      "epoch": 10.056,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0009958902994791667,
+      "learning_rate": 0.00032,
+      "loss": 55.6938,
+      "step": 696
+    },
+    {
+      "crossentropy": 2.7774757146835327,
+      "epoch": 10.057,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0009724299112955729,
+      "learning_rate": 0.00031894736842105265,
+      "loss": 56.9648,
+      "step": 697
+    },
+    {
+      "crossentropy": 2.934939742088318,
+      "epoch": 10.058,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0009144465128580729,
+      "learning_rate": 0.0003178947368421053,
+      "loss": 57.7034,
+      "step": 698
+    },
+    {
+      "crossentropy": 2.8009008169174194,
+      "epoch": 10.059,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0009213765462239583,
+      "learning_rate": 0.00031684210526315785,
+      "loss": 56.4452,
+      "step": 699
+    },
+    {
+      "crossentropy": 2.9469012022018433,
+      "epoch": 10.06,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0008447647094726562,
+      "learning_rate": 0.00031578947368421053,
+      "loss": 58.8681,
+      "step": 700
+    },
+    {
+      "crossentropy": 2.7601804733276367,
+      "epoch": 10.061,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0008562723795572917,
+      "learning_rate": 0.00031473684210526316,
+      "loss": 56.4526,
+      "step": 701
+    },
+    {
+      "crossentropy": 2.8851720094680786,
+      "epoch": 10.062,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0008406956990559896,
+      "learning_rate": 0.0003136842105263158,
+      "loss": 56.0664,
+      "step": 702
+    },
+    {
+      "crossentropy": 2.7828023433685303,
+      "epoch": 10.063,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0008483250935872396,
+      "learning_rate": 0.0003126315789473684,
+      "loss": 55.3705,
+      "step": 703
+    },
+    {
+      "crossentropy": 2.9105297327041626,
+      "epoch": 10.064,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0007705052693684895,
+      "learning_rate": 0.0003115789473684211,
+      "loss": 57.2773,
+      "step": 704
+    },
+    {
+      "crossentropy": 2.706776738166809,
+      "epoch": 11.001,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0008045832316080729,
+      "learning_rate": 0.0003105263157894737,
+      "loss": 54.4294,
+      "step": 705
+    },
+    {
+      "crossentropy": 2.785434126853943,
+      "epoch": 11.002,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000811767578125,
+      "learning_rate": 0.0003094736842105263,
+      "loss": 56.077,
+      "step": 706
+    },
+    {
+      "crossentropy": 2.708212375640869,
+      "epoch": 11.003,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0008254369099934896,
+      "learning_rate": 0.0003084210526315789,
+      "loss": 53.1839,
+      "step": 707
+    },
+    {
+      "crossentropy": 2.7735451459884644,
+      "epoch": 11.004,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0008300145467122396,
+      "learning_rate": 0.0003073684210526316,
+      "loss": 56.1457,
+      "step": 708
+    },
+    {
+      "crossentropy": 2.8523166179656982,
+      "epoch": 11.005,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0001713434855143229,
+      "learning_rate": 0.0003063157894736842,
+      "loss": 57.1347,
+      "step": 709
+    },
+    {
+      "crossentropy": 2.914563536643982,
+      "epoch": 11.006,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0002522150675455729,
+      "learning_rate": 0.00030526315789473684,
+      "loss": 57.723,
+      "step": 710
+    },
+    {
+      "crossentropy": 2.7213577032089233,
+      "epoch": 11.007,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00027033487955729164,
+      "learning_rate": 0.00030421052631578947,
+      "loss": 55.6901,
+      "step": 711
+    },
+    {
+      "crossentropy": 2.912691831588745,
+      "epoch": 11.008,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002608617146809896,
+      "learning_rate": 0.00030315789473684215,
+      "loss": 57.1725,
+      "step": 712
+    },
+    {
+      "crossentropy": 2.8348140716552734,
+      "epoch": 11.009,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0002766927083333333,
+      "learning_rate": 0.0003021052631578947,
+      "loss": 56.1903,
+      "step": 713
+    },
+    {
+      "crossentropy": 2.8859379291534424,
+      "epoch": 11.01,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00032145182291666666,
+      "learning_rate": 0.00030105263157894735,
+      "loss": 58.3572,
+      "step": 714
+    },
+    {
+      "crossentropy": 2.778756260871887,
+      "epoch": 11.011,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00032526652018229166,
+      "learning_rate": 0.0003,
+      "loss": 56.3487,
+      "step": 715
+    },
+    {
+      "crossentropy": 2.982430934906006,
+      "epoch": 11.012,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0003539403279622396,
+      "learning_rate": 0.00029894736842105265,
+      "loss": 57.9075,
+      "step": 716
+    },
+    {
+      "crossentropy": 2.8630356788635254,
+      "epoch": 11.013,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.000394439697265625,
+      "learning_rate": 0.0002978947368421053,
+      "loss": 56.721,
+      "step": 717
+    },
+    {
+      "crossentropy": 2.6543571949005127,
+      "epoch": 11.014,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.000394439697265625,
+      "learning_rate": 0.0002968421052631579,
+      "loss": 54.8384,
+      "step": 718
+    },
+    {
+      "crossentropy": 2.869687557220459,
+      "epoch": 11.015,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.000382232666015625,
+      "learning_rate": 0.00029578947368421053,
+      "loss": 57.8614,
+      "step": 719
+    },
+    {
+      "crossentropy": 2.846762180328369,
+      "epoch": 11.016,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00034885406494140626,
+      "learning_rate": 0.00029473684210526316,
+      "loss": 57.1553,
+      "step": 720
+    },
+    {
+      "crossentropy": 2.7584152221679688,
+      "epoch": 11.017,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003280003865559896,
+      "learning_rate": 0.0002936842105263158,
+      "loss": 55.2654,
+      "step": 721
+    },
+    {
+      "crossentropy": 2.8962432146072388,
+      "epoch": 11.018,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003682454427083333,
+      "learning_rate": 0.0002926315789473684,
+      "loss": 56.6584,
+      "step": 722
+    },
+    {
+      "crossentropy": 2.7644320726394653,
+      "epoch": 11.019,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003590265909830729,
+      "learning_rate": 0.00029157894736842104,
+      "loss": 55.4874,
+      "step": 723
+    },
+    {
+      "crossentropy": 2.6588387489318848,
+      "epoch": 11.02,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0004241943359375,
+      "learning_rate": 0.0002905263157894737,
+      "loss": 54.2554,
+      "step": 724
+    },
+    {
+      "crossentropy": 2.8226871490478516,
+      "epoch": 11.021,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0004330952962239583,
+      "learning_rate": 0.00028947368421052634,
+      "loss": 56.599,
+      "step": 725
+    },
+    {
+      "crossentropy": 2.8901766538619995,
+      "epoch": 11.022,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003859837849934896,
+      "learning_rate": 0.00028842105263157897,
+      "loss": 56.2554,
+      "step": 726
+    },
+    {
+      "crossentropy": 2.772639751434326,
+      "epoch": 11.023,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003712336222330729,
+      "learning_rate": 0.0002873684210526316,
+      "loss": 56.7091,
+      "step": 727
+    },
+    {
+      "crossentropy": 2.8242876529693604,
+      "epoch": 11.024,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00035502115885416665,
+      "learning_rate": 0.0002863157894736842,
+      "loss": 55.5862,
+      "step": 728
+    },
+    {
+      "crossentropy": 2.9208909273147583,
+      "epoch": 11.025,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00034122467041015626,
+      "learning_rate": 0.00028526315789473685,
+      "loss": 56.8342,
+      "step": 729
+    },
+    {
+      "crossentropy": 2.6857054233551025,
+      "epoch": 11.026,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.000321197509765625,
+      "learning_rate": 0.00028421052631578947,
+      "loss": 54.6538,
+      "step": 730
+    },
+    {
+      "crossentropy": 2.8337063789367676,
+      "epoch": 11.027,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0003021240234375,
+      "learning_rate": 0.0002831578947368421,
+      "loss": 57.5415,
+      "step": 731
+    },
+    {
+      "crossentropy": 2.786600351333618,
+      "epoch": 11.028,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00020084381103515624,
+      "learning_rate": 0.0002821052631578948,
+      "loss": 55.5913,
+      "step": 732
+    },
+    {
+      "crossentropy": 2.690516948699951,
+      "epoch": 11.029,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00027256011962890626,
+      "learning_rate": 0.0002810526315789474,
+      "loss": 53.9815,
+      "step": 733
+    },
+    {
+      "crossentropy": 2.7604899406433105,
+      "epoch": 11.03,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00028324127197265625,
+      "learning_rate": 0.00028000000000000003,
+      "loss": 54.824,
+      "step": 734
+    },
+    {
+      "crossentropy": 2.7863725423812866,
+      "epoch": 11.031,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.00026950836181640627,
+      "learning_rate": 0.0002789473684210526,
+      "loss": 55.8541,
+      "step": 735
+    },
+    {
+      "crossentropy": 2.8232542276382446,
+      "epoch": 11.032,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.000302886962890625,
+      "learning_rate": 0.0002778947368421053,
+      "loss": 56.7185,
+      "step": 736
+    },
+    {
+      "crossentropy": 2.7417391538619995,
+      "epoch": 11.033,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0003330866495768229,
+      "learning_rate": 0.0002768421052631579,
+      "loss": 57.2269,
+      "step": 737
+    },
+    {
+      "crossentropy": 2.819975972175598,
+      "epoch": 11.034,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0003265380859375,
+      "learning_rate": 0.00027578947368421053,
+      "loss": 56.6551,
+      "step": 738
+    },
+    {
+      "crossentropy": 2.9095921516418457,
+      "epoch": 11.035,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00040836334228515624,
+      "learning_rate": 0.00027473684210526316,
+      "loss": 56.8075,
+      "step": 739
+    },
+    {
+      "crossentropy": 2.91646945476532,
+      "epoch": 11.036,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0003956476847330729,
+      "learning_rate": 0.00027368421052631584,
+      "loss": 57.2162,
+      "step": 740
+    },
+    {
+      "crossentropy": 2.9138376712799072,
+      "epoch": 11.037,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0004088719685872396,
+      "learning_rate": 0.00027263157894736847,
+      "loss": 56.7284,
+      "step": 741
+    },
+    {
+      "crossentropy": 2.84540057182312,
+      "epoch": 11.038,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0005045572916666667,
+      "learning_rate": 0.00027157894736842104,
+      "loss": 55.8447,
+      "step": 742
+    },
+    {
+      "crossentropy": 2.7151050567626953,
+      "epoch": 11.039,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0005116780598958333,
+      "learning_rate": 0.00027052631578947366,
+      "loss": 55.9625,
+      "step": 743
+    },
+    {
+      "crossentropy": 2.9206697940826416,
+      "epoch": 11.04,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005619684855143229,
+      "learning_rate": 0.0002694736842105263,
+      "loss": 57.6432,
+      "step": 744
+    },
+    {
+      "crossentropy": 2.8370519876480103,
+      "epoch": 11.041,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.000577545166015625,
+      "learning_rate": 0.00026842105263157897,
+      "loss": 55.4367,
+      "step": 745
+    },
+    {
+      "crossentropy": 2.8395227193832397,
+      "epoch": 11.042,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005156834920247396,
+      "learning_rate": 0.0002673684210526316,
+      "loss": 56.171,
+      "step": 746
+    },
+    {
+      "crossentropy": 2.786525845527649,
+      "epoch": 11.043,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005467096964518229,
+      "learning_rate": 0.0002663157894736842,
+      "loss": 57.2315,
+      "step": 747
+    },
+    {
+      "crossentropy": 2.9187395572662354,
+      "epoch": 11.044,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005472819010416667,
+      "learning_rate": 0.00026526315789473685,
+      "loss": 57.2511,
+      "step": 748
+    },
+    {
+      "crossentropy": 2.7996715307235718,
+      "epoch": 11.045,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0006182352701822917,
+      "learning_rate": 0.0002642105263157895,
+      "loss": 57.045,
+      "step": 749
+    },
+    {
+      "crossentropy": 2.8842201232910156,
+      "epoch": 11.046,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0006001154581705729,
+      "learning_rate": 0.0002631578947368421,
+      "loss": 57.6562,
+      "step": 750
+    },
+    {
+      "crossentropy": 2.805160880088806,
+      "epoch": 11.047,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004795710245768229,
+      "learning_rate": 0.0002621052631578947,
+      "loss": 56.6321,
+      "step": 751
+    },
+    {
+      "crossentropy": 2.9018155336380005,
+      "epoch": 11.048,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00048802693684895835,
+      "learning_rate": 0.00026105263157894735,
+      "loss": 57.7864,
+      "step": 752
+    },
+    {
+      "crossentropy": 2.8916486501693726,
+      "epoch": 11.049,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0005156834920247396,
+      "learning_rate": 0.00026000000000000003,
+      "loss": 56.4045,
+      "step": 753
+    },
+    {
+      "crossentropy": 2.8715128898620605,
+      "epoch": 11.05,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0005429585774739584,
+      "learning_rate": 0.00025894736842105266,
+      "loss": 57.9977,
+      "step": 754
+    },
+    {
+      "crossentropy": 2.798703193664551,
+      "epoch": 11.051,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005083719889322917,
+      "learning_rate": 0.0002578947368421053,
+      "loss": 57.3152,
+      "step": 755
+    },
+    {
+      "crossentropy": 2.9445585012435913,
+      "epoch": 11.052,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0005024592081705729,
+      "learning_rate": 0.00025684210526315786,
+      "loss": 57.4762,
+      "step": 756
+    },
+    {
+      "crossentropy": 2.8702229261398315,
+      "epoch": 11.053,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0004900614420572917,
+      "learning_rate": 0.00025578947368421054,
+      "loss": 57.5399,
+      "step": 757
+    },
+    {
+      "crossentropy": 2.8421502113342285,
+      "epoch": 11.054,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003636042277018229,
+      "learning_rate": 0.00025473684210526316,
+      "loss": 58.3388,
+      "step": 758
+    },
+    {
+      "crossentropy": 2.9485079050064087,
+      "epoch": 11.055,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003483454386393229,
+      "learning_rate": 0.0002536842105263158,
+      "loss": 58.1595,
+      "step": 759
+    },
+    {
+      "crossentropy": 2.7640316486358643,
+      "epoch": 11.056,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003193537394205729,
+      "learning_rate": 0.0002526315789473684,
+      "loss": 55.4479,
+      "step": 760
+    },
+    {
+      "crossentropy": 2.9783310890197754,
+      "epoch": 11.057,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.00032755533854166664,
+      "learning_rate": 0.0002515789473684211,
+      "loss": 58.1017,
+      "step": 761
+    },
+    {
+      "crossentropy": 2.8926788568496704,
+      "epoch": 11.058,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.00034357706705729165,
+      "learning_rate": 0.0002505263157894737,
+      "loss": 56.6529,
+      "step": 762
+    },
+    {
+      "crossentropy": 2.82997465133667,
+      "epoch": 11.059,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00034357706705729165,
+      "learning_rate": 0.00024947368421052635,
+      "loss": 56.8071,
+      "step": 763
+    },
+    {
+      "crossentropy": 2.813700556755066,
+      "epoch": 11.06,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0003742853800455729,
+      "learning_rate": 0.00024842105263157897,
+      "loss": 55.8355,
+      "step": 764
+    },
+    {
+      "crossentropy": 2.7938244342803955,
+      "epoch": 11.061,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0002491633097330729,
+      "learning_rate": 0.0002473684210526316,
+      "loss": 55.8545,
+      "step": 765
+    },
+    {
+      "crossentropy": 2.8322267532348633,
+      "epoch": 11.062,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00024102528889973957,
+      "learning_rate": 0.0002463157894736842,
+      "loss": 57.2321,
+      "step": 766
+    },
+    {
+      "crossentropy": 2.7156633138656616,
+      "epoch": 11.063,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00024102528889973957,
+      "learning_rate": 0.00024526315789473685,
+      "loss": 54.8035,
+      "step": 767
+    },
+    {
+      "crossentropy": 2.634229063987732,
+      "epoch": 11.064,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00023167928059895834,
+      "learning_rate": 0.0002442105263157895,
+      "loss": 54.4477,
+      "step": 768
+    },
+    {
+      "crossentropy": 2.7296793460845947,
+      "epoch": 12.001,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0002196629842122396,
+      "learning_rate": 0.0002431578947368421,
+      "loss": 56.2054,
+      "step": 769
+    },
+    {
+      "crossentropy": 2.8033626079559326,
+      "epoch": 12.002,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0002196629842122396,
+      "learning_rate": 0.00024210526315789475,
+      "loss": 56.5435,
+      "step": 770
+    },
+    {
+      "crossentropy": 2.8827661275863647,
+      "epoch": 12.003,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0002608617146809896,
+      "learning_rate": 0.00024105263157894738,
+      "loss": 56.553,
+      "step": 771
+    },
+    {
+      "crossentropy": 2.798128128051758,
+      "epoch": 12.004,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0002598444620768229,
+      "learning_rate": 0.00024,
+      "loss": 56.1054,
+      "step": 772
+    },
+    {
+      "crossentropy": 2.79728901386261,
+      "epoch": 12.005,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003072102864583333,
+      "learning_rate": 0.00023894736842105263,
+      "loss": 57.2369,
+      "step": 773
+    },
+    {
+      "crossentropy": 2.758781909942627,
+      "epoch": 12.006,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003191630045572917,
+      "learning_rate": 0.00023789473684210529,
+      "loss": 54.4973,
+      "step": 774
+    },
+    {
+      "crossentropy": 2.9363932609558105,
+      "epoch": 12.007,
+      "grad_norm": 0.8046875,
+      "grad_norm_var": 0.0005096435546875,
+      "learning_rate": 0.00023684210526315788,
+      "loss": 57.2391,
+      "step": 775
+    },
+    {
+      "crossentropy": 2.8779611587524414,
+      "epoch": 12.008,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0004961649576822917,
+      "learning_rate": 0.00023578947368421054,
+      "loss": 56.8895,
+      "step": 776
+    },
+    {
+      "crossentropy": 2.824180841445923,
+      "epoch": 12.009,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00047550201416015627,
+      "learning_rate": 0.00023473684210526316,
+      "loss": 57.5237,
+      "step": 777
+    },
+    {
+      "crossentropy": 2.8232827186584473,
+      "epoch": 12.01,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00047397613525390625,
+      "learning_rate": 0.00023368421052631582,
+      "loss": 56.4012,
+      "step": 778
+    },
+    {
+      "crossentropy": 2.817031145095825,
+      "epoch": 12.011,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0004521052042643229,
+      "learning_rate": 0.00023263157894736841,
+      "loss": 55.2773,
+      "step": 779
+    },
+    {
+      "crossentropy": 2.731096386909485,
+      "epoch": 12.012,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00044733683268229166,
+      "learning_rate": 0.00023157894736842107,
+      "loss": 55.6655,
+      "step": 780
+    },
+    {
+      "crossentropy": 2.7005906105041504,
+      "epoch": 12.013,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00045166015625,
+      "learning_rate": 0.0002305263157894737,
+      "loss": 55.448,
+      "step": 781
+    },
+    {
+      "crossentropy": 2.7934017181396484,
+      "epoch": 12.014,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.0005266825358072917,
+      "learning_rate": 0.00022947368421052632,
+      "loss": 54.7745,
+      "step": 782
+    },
+    {
+      "crossentropy": 2.8664556741714478,
+      "epoch": 12.015,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005228042602539063,
+      "learning_rate": 0.00022842105263157895,
+      "loss": 57.5525,
+      "step": 783
+    },
+    {
+      "crossentropy": 2.797169327735901,
+      "epoch": 12.016,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.000516510009765625,
+      "learning_rate": 0.0002273684210526316,
+      "loss": 56.7851,
+      "step": 784
+    },
+    {
+      "crossentropy": 2.8436896800994873,
+      "epoch": 12.017,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00045617421468098957,
+      "learning_rate": 0.00022631578947368422,
+      "loss": 56.0802,
+      "step": 785
+    },
+    {
+      "crossentropy": 2.664197087287903,
+      "epoch": 12.018,
+      "grad_norm": 0.79296875,
+      "grad_norm_var": 0.0005853652954101562,
+      "learning_rate": 0.00022526315789473682,
+      "loss": 54.7135,
+      "step": 786
+    },
+    {
+      "crossentropy": 2.6850061416625977,
+      "epoch": 12.019,
+      "grad_norm": 0.78125,
+      "grad_norm_var": 0.0006209691365559896,
+      "learning_rate": 0.00022421052631578948,
+      "loss": 55.2802,
+      "step": 787
+    },
+    {
+      "crossentropy": 2.7967450618743896,
+      "epoch": 12.02,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0006245930989583333,
+      "learning_rate": 0.0002231578947368421,
+      "loss": 57.1233,
+      "step": 788
+    },
+    {
+      "crossentropy": 2.872409701347351,
+      "epoch": 12.021,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0006245930989583333,
+      "learning_rate": 0.00022210526315789476,
+      "loss": 56.7216,
+      "step": 789
+    },
+    {
+      "crossentropy": 2.8131096363067627,
+      "epoch": 12.022,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0006245930989583333,
+      "learning_rate": 0.00022105263157894735,
+      "loss": 55.5856,
+      "step": 790
+    },
+    {
+      "crossentropy": 2.847323417663574,
+      "epoch": 12.023,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00041300455729166666,
+      "learning_rate": 0.00022,
+      "loss": 56.5617,
+      "step": 791
+    },
+    {
+      "crossentropy": 2.8835920095443726,
+      "epoch": 12.024,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00039615631103515623,
+      "learning_rate": 0.00021894736842105263,
+      "loss": 57.2592,
+      "step": 792
+    },
+    {
+      "crossentropy": 2.830420136451721,
+      "epoch": 12.025,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00040531158447265625,
+      "learning_rate": 0.00021789473684210526,
+      "loss": 56.2414,
+      "step": 793
+    },
+    {
+      "crossentropy": 2.755424976348877,
+      "epoch": 12.026,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003895441691080729,
+      "learning_rate": 0.00021684210526315789,
+      "loss": 55.1508,
+      "step": 794
+    },
+    {
+      "crossentropy": 2.737109422683716,
+      "epoch": 12.027,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003903706868489583,
+      "learning_rate": 0.00021578947368421054,
+      "loss": 55.5294,
+      "step": 795
+    },
+    {
+      "crossentropy": 2.8102558851242065,
+      "epoch": 12.028,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0004221598307291667,
+      "learning_rate": 0.00021473684210526316,
+      "loss": 55.6168,
+      "step": 796
+    },
+    {
+      "crossentropy": 2.8057838678359985,
+      "epoch": 12.029,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.00044854482014973957,
+      "learning_rate": 0.0002136842105263158,
+      "loss": 56.8656,
+      "step": 797
+    },
+    {
+      "crossentropy": 2.7969436645507812,
+      "epoch": 12.03,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004514058430989583,
+      "learning_rate": 0.00021263157894736842,
+      "loss": 56.5856,
+      "step": 798
+    },
+    {
+      "crossentropy": 2.854472517967224,
+      "epoch": 12.031,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004595438639322917,
+      "learning_rate": 0.00021157894736842107,
+      "loss": 57.0911,
+      "step": 799
+    },
+    {
+      "crossentropy": 2.705453634262085,
+      "epoch": 12.032,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0004622777303059896,
+      "learning_rate": 0.00021052631578947367,
+      "loss": 53.9657,
+      "step": 800
+    },
+    {
+      "crossentropy": 2.823037266731262,
+      "epoch": 12.033,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0004622777303059896,
+      "learning_rate": 0.00020947368421052632,
+      "loss": 56.4739,
+      "step": 801
+    },
+    {
+      "crossentropy": 2.826178789138794,
+      "epoch": 12.034,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003407796223958333,
+      "learning_rate": 0.00020842105263157895,
+      "loss": 57.7884,
+      "step": 802
+    },
+    {
+      "crossentropy": 2.868850827217102,
+      "epoch": 12.035,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0003110249837239583,
+      "learning_rate": 0.0002073684210526316,
+      "loss": 56.8438,
+      "step": 803
+    },
+    {
+      "crossentropy": 2.870843529701233,
+      "epoch": 12.036,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0003218968709309896,
+      "learning_rate": 0.0002063157894736842,
+      "loss": 57.301,
+      "step": 804
+    },
+    {
+      "crossentropy": 2.8651864528656006,
+      "epoch": 12.037,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00031305948893229165,
+      "learning_rate": 0.00020526315789473685,
+      "loss": 57.0191,
+      "step": 805
+    },
+    {
+      "crossentropy": 2.9826983213424683,
+      "epoch": 12.038,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0003559748331705729,
+      "learning_rate": 0.00020421052631578948,
+      "loss": 57.7126,
+      "step": 806
+    },
+    {
+      "crossentropy": 2.7978984117507935,
+      "epoch": 12.039,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0003956476847330729,
+      "learning_rate": 0.0002031578947368421,
+      "loss": 55.453,
+      "step": 807
+    },
+    {
+      "crossentropy": 2.668187379837036,
+      "epoch": 12.04,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.000443267822265625,
+      "learning_rate": 0.00020210526315789473,
+      "loss": 53.316,
+      "step": 808
+    },
+    {
+      "crossentropy": 2.7966721057891846,
+      "epoch": 12.041,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00042292277018229164,
+      "learning_rate": 0.00020105263157894738,
+      "loss": 55.6218,
+      "step": 809
+    },
+    {
+      "crossentropy": 2.739244222640991,
+      "epoch": 12.042,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0004404703776041667,
+      "learning_rate": 0.0002,
+      "loss": 56.3849,
+      "step": 810
+    },
+    {
+      "crossentropy": 2.783705234527588,
+      "epoch": 12.043,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00045369466145833335,
+      "learning_rate": 0.00019894736842105264,
+      "loss": 57.1809,
+      "step": 811
+    },
+    {
+      "crossentropy": 2.8895143270492554,
+      "epoch": 12.044,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000437164306640625,
+      "learning_rate": 0.00019789473684210526,
+      "loss": 58.2889,
+      "step": 812
+    },
+    {
+      "crossentropy": 2.8347127437591553,
+      "epoch": 12.045,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.00041478474934895834,
+      "learning_rate": 0.00019684210526315791,
+      "loss": 56.1132,
+      "step": 813
+    },
+    {
+      "crossentropy": 2.8544065952301025,
+      "epoch": 12.046,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0003651301066080729,
+      "learning_rate": 0.0001957894736842105,
+      "loss": 56.9511,
+      "step": 814
+    },
+    {
+      "crossentropy": 2.7046180963516235,
+      "epoch": 12.047,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003600438435872396,
+      "learning_rate": 0.00019473684210526317,
+      "loss": 54.0523,
+      "step": 815
+    },
+    {
+      "crossentropy": 2.912235379219055,
+      "epoch": 12.048,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003560384114583333,
+      "learning_rate": 0.0001936842105263158,
+      "loss": 57.7615,
+      "step": 816
+    },
+    {
+      "crossentropy": 2.8122063875198364,
+      "epoch": 12.049,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00035953521728515625,
+      "learning_rate": 0.00019263157894736845,
+      "loss": 57.801,
+      "step": 817
+    },
+    {
+      "crossentropy": 2.74889874458313,
+      "epoch": 12.05,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00034688313802083335,
+      "learning_rate": 0.00019157894736842104,
+      "loss": 57.198,
+      "step": 818
+    },
+    {
+      "crossentropy": 2.9050692319869995,
+      "epoch": 12.051,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00029697418212890623,
+      "learning_rate": 0.0001905263157894737,
+      "loss": 56.6941,
+      "step": 819
+    },
+    {
+      "crossentropy": 2.7928906679153442,
+      "epoch": 12.052,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00028584798177083335,
+      "learning_rate": 0.00018947368421052632,
+      "loss": 56.1308,
+      "step": 820
+    },
+    {
+      "crossentropy": 2.7914161682128906,
+      "epoch": 12.053,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00034383138020833336,
+      "learning_rate": 0.00018842105263157895,
+      "loss": 56.491,
+      "step": 821
+    },
+    {
+      "crossentropy": 2.7516348361968994,
+      "epoch": 12.054,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0002929051717122396,
+      "learning_rate": 0.00018736842105263158,
+      "loss": 56.2496,
+      "step": 822
+    },
+    {
+      "crossentropy": 2.7576178312301636,
+      "epoch": 12.055,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0002929051717122396,
+      "learning_rate": 0.00018631578947368423,
+      "loss": 56.6343,
+      "step": 823
+    },
+    {
+      "crossentropy": 2.8195626735687256,
+      "epoch": 12.056,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0002486546834309896,
+      "learning_rate": 0.00018526315789473685,
+      "loss": 57.3709,
+      "step": 824
+    },
+    {
+      "crossentropy": 2.7876213788986206,
+      "epoch": 12.057,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0002486546834309896,
+      "learning_rate": 0.00018421052631578948,
+      "loss": 55.6639,
+      "step": 825
+    },
+    {
+      "crossentropy": 2.88731849193573,
+      "epoch": 12.058,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00024153391520182293,
+      "learning_rate": 0.0001831578947368421,
+      "loss": 57.4164,
+      "step": 826
+    },
+    {
+      "crossentropy": 2.9272228479385376,
+      "epoch": 12.059,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.00028279622395833336,
+      "learning_rate": 0.00018210526315789476,
+      "loss": 57.3789,
+      "step": 827
+    },
+    {
+      "crossentropy": 2.833387613296509,
+      "epoch": 12.06,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0003864924112955729,
+      "learning_rate": 0.00018105263157894739,
+      "loss": 57.4141,
+      "step": 828
+    },
+    {
+      "crossentropy": 2.896293044090271,
+      "epoch": 12.061,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0003895441691080729,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 58.0117,
+      "step": 829
+    },
+    {
+      "crossentropy": 2.831289052963257,
+      "epoch": 12.062,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00039265950520833335,
+      "learning_rate": 0.00017894736842105264,
+      "loss": 56.8003,
+      "step": 830
+    },
+    {
+      "crossentropy": 2.8674780130386353,
+      "epoch": 12.063,
+      "grad_norm": 0.69140625,
+      "grad_norm_var": 0.0005523045857747396,
+      "learning_rate": 0.00017789473684210526,
+      "loss": 56.9051,
+      "step": 831
+    },
+    {
+      "crossentropy": 2.9149069786071777,
+      "epoch": 12.064,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0006467183430989583,
+      "learning_rate": 0.0001768421052631579,
+      "loss": 57.5528,
+      "step": 832
+    },
+    {
+      "crossentropy": 2.824967622756958,
+      "epoch": 13.001,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0007888158162434896,
+      "learning_rate": 0.00017578947368421052,
+      "loss": 56.2444,
+      "step": 833
+    },
+    {
+      "crossentropy": 2.784726858139038,
+      "epoch": 13.002,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0007802327473958333,
+      "learning_rate": 0.00017473684210526317,
+      "loss": 54.6207,
+      "step": 834
+    },
+    {
+      "crossentropy": 2.924685478210449,
+      "epoch": 13.003,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0007664362589518229,
+      "learning_rate": 0.0001736842105263158,
+      "loss": 57.5746,
+      "step": 835
+    },
+    {
+      "crossentropy": 2.897099494934082,
+      "epoch": 13.004,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.0008188883463541666,
+      "learning_rate": 0.00017263157894736842,
+      "loss": 56.227,
+      "step": 836
+    },
+    {
+      "crossentropy": 2.7194520235061646,
+      "epoch": 13.005,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0008188883463541666,
+      "learning_rate": 0.00017157894736842105,
+      "loss": 54.1649,
+      "step": 837
+    },
+    {
+      "crossentropy": 2.7639235258102417,
+      "epoch": 13.006,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000791168212890625,
+      "learning_rate": 0.0001705263157894737,
+      "loss": 55.093,
+      "step": 838
+    },
+    {
+      "crossentropy": 2.7386432886123657,
+      "epoch": 13.007,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0008066813151041667,
+      "learning_rate": 0.0001694736842105263,
+      "loss": 54.6862,
+      "step": 839
+    },
+    {
+      "crossentropy": 2.814002513885498,
+      "epoch": 13.008,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0008066813151041667,
+      "learning_rate": 0.00016842105263157895,
+      "loss": 56.0865,
+      "step": 840
+    },
+    {
+      "crossentropy": 2.8583909273147583,
+      "epoch": 13.009,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0007933934529622395,
+      "learning_rate": 0.00016736842105263158,
+      "loss": 56.4426,
+      "step": 841
+    },
+    {
+      "crossentropy": 2.79672372341156,
+      "epoch": 13.01,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.000775146484375,
+      "learning_rate": 0.00016631578947368423,
+      "loss": 55.4208,
+      "step": 842
+    },
+    {
+      "crossentropy": 2.8741607666015625,
+      "epoch": 13.011,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0007715225219726562,
+      "learning_rate": 0.00016526315789473683,
+      "loss": 56.4795,
+      "step": 843
+    },
+    {
+      "crossentropy": 2.7828328609466553,
+      "epoch": 13.012,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.000803375244140625,
+      "learning_rate": 0.00016421052631578948,
+      "loss": 54.505,
+      "step": 844
+    },
+    {
+      "crossentropy": 2.7500873804092407,
+      "epoch": 13.013,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0007974624633789063,
+      "learning_rate": 0.0001631578947368421,
+      "loss": 54.9745,
+      "step": 845
+    },
+    {
+      "crossentropy": 2.7167465686798096,
+      "epoch": 13.014,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0006510416666666666,
+      "learning_rate": 0.00016210526315789473,
+      "loss": 54.637,
+      "step": 846
+    },
+    {
+      "crossentropy": 2.838550090789795,
+      "epoch": 13.015,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.000553131103515625,
+      "learning_rate": 0.00016105263157894736,
+      "loss": 56.9125,
+      "step": 847
+    },
+    {
+      "crossentropy": 2.728639602661133,
+      "epoch": 13.016,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00042667388916015623,
+      "learning_rate": 0.00016,
+      "loss": 55.2942,
+      "step": 848
+    },
+    {
+      "crossentropy": 2.8884626626968384,
+      "epoch": 13.017,
+      "grad_norm": 0.81640625,
+      "grad_norm_var": 0.000719134012858073,
+      "learning_rate": 0.00015894736842105264,
+      "loss": 58.9905,
+      "step": 849
+    },
+    {
+      "crossentropy": 2.6712993383407593,
+      "epoch": 13.018,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.000681304931640625,
+      "learning_rate": 0.00015789473684210527,
+      "loss": 54.4483,
+      "step": 850
+    },
+    {
+      "crossentropy": 2.8165026903152466,
+      "epoch": 13.019,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.000687408447265625,
+      "learning_rate": 0.0001568421052631579,
+      "loss": 56.7351,
+      "step": 851
+    },
+    {
+      "crossentropy": 2.9343361854553223,
+      "epoch": 13.02,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0006713231404622395,
+      "learning_rate": 0.00015578947368421054,
+      "loss": 58.3815,
+      "step": 852
+    },
+    {
+      "crossentropy": 2.931985378265381,
+      "epoch": 13.021,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0007403055826822917,
+      "learning_rate": 0.00015473684210526314,
+      "loss": 57.2322,
+      "step": 853
+    },
+    {
+      "crossentropy": 2.8127466440200806,
+      "epoch": 13.022,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0007314046223958333,
+      "learning_rate": 0.0001536842105263158,
+      "loss": 55.7789,
+      "step": 854
+    },
+    {
+      "crossentropy": 2.8048046827316284,
+      "epoch": 13.023,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0007932027180989583,
+      "learning_rate": 0.00015263157894736842,
+      "loss": 56.2506,
+      "step": 855
+    },
+    {
+      "crossentropy": 2.787923812866211,
+      "epoch": 13.024,
+      "grad_norm": 0.703125,
+      "grad_norm_var": 0.0008656819661458333,
+      "learning_rate": 0.00015157894736842108,
+      "loss": 54.8036,
+      "step": 856
+    },
+    {
+      "crossentropy": 2.829706907272339,
+      "epoch": 13.025,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0007893880208333333,
+      "learning_rate": 0.00015052631578947367,
+      "loss": 56.4524,
+      "step": 857
+    },
+    {
+      "crossentropy": 2.810854196548462,
+      "epoch": 13.026,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0007893880208333333,
+      "learning_rate": 0.00014947368421052633,
+      "loss": 56.9006,
+      "step": 858
+    },
+    {
+      "crossentropy": 2.7314698696136475,
+      "epoch": 13.027,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0007609049479166667,
+      "learning_rate": 0.00014842105263157895,
+      "loss": 55.7139,
+      "step": 859
+    },
+    {
+      "crossentropy": 2.747408866882324,
+      "epoch": 13.028,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0007440567016601563,
+      "learning_rate": 0.00014736842105263158,
+      "loss": 54.023,
+      "step": 860
+    },
+    {
+      "crossentropy": 2.8353370428085327,
+      "epoch": 13.029,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0007287979125976563,
+      "learning_rate": 0.0001463157894736842,
+      "loss": 55.9548,
+      "step": 861
+    },
+    {
+      "crossentropy": 2.757072687149048,
+      "epoch": 13.03,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0007481257120768229,
+      "learning_rate": 0.00014526315789473686,
+      "loss": 54.7874,
+      "step": 862
+    },
+    {
+      "crossentropy": 2.8938580751419067,
+      "epoch": 13.031,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.0008686701456705729,
+      "learning_rate": 0.00014421052631578948,
+      "loss": 58.1519,
+      "step": 863
+    },
+    {
+      "crossentropy": 2.7633159160614014,
+      "epoch": 13.032,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0008707046508789062,
+      "learning_rate": 0.0001431578947368421,
+      "loss": 56.3047,
+      "step": 864
+    },
+    {
+      "crossentropy": 2.810257911682129,
+      "epoch": 13.033,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.000528717041015625,
+      "learning_rate": 0.00014210526315789474,
+      "loss": 57.6018,
+      "step": 865
+    },
+    {
+      "crossentropy": 2.753029227256775,
+      "epoch": 13.034,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000520769755045573,
+      "learning_rate": 0.0001410526315789474,
+      "loss": 56.7448,
+      "step": 866
+    },
+    {
+      "crossentropy": 2.7661619186401367,
+      "epoch": 13.035,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0005124409993489584,
+      "learning_rate": 0.00014000000000000001,
+      "loss": 56.7924,
+      "step": 867
+    },
+    {
+      "crossentropy": 2.915842056274414,
+      "epoch": 13.036,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00047143300374348957,
+      "learning_rate": 0.00013894736842105264,
+      "loss": 57.8143,
+      "step": 868
+    },
+    {
+      "crossentropy": 2.838146448135376,
+      "epoch": 13.037,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.000445556640625,
+      "learning_rate": 0.00013789473684210527,
+      "loss": 57.094,
+      "step": 869
+    },
+    {
+      "crossentropy": 2.917192220687866,
+      "epoch": 13.038,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0004608154296875,
+      "learning_rate": 0.00013684210526315792,
+      "loss": 58.5578,
+      "step": 870
+    },
+    {
+      "crossentropy": 2.8409940004348755,
+      "epoch": 13.039,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00040384928385416664,
+      "learning_rate": 0.00013578947368421052,
+      "loss": 55.9825,
+      "step": 871
+    },
+    {
+      "crossentropy": 2.70103657245636,
+      "epoch": 13.04,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0003208796183268229,
+      "learning_rate": 0.00013473684210526314,
+      "loss": 55.2032,
+      "step": 872
+    },
+    {
+      "crossentropy": 2.7406742572784424,
+      "epoch": 13.041,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003031412760416667,
+      "learning_rate": 0.0001336842105263158,
+      "loss": 56.4148,
+      "step": 873
+    },
+    {
+      "crossentropy": 2.919757843017578,
+      "epoch": 13.042,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.00034357706705729165,
+      "learning_rate": 0.00013263157894736842,
+      "loss": 57.4301,
+      "step": 874
+    },
+    {
+      "crossentropy": 2.7461795806884766,
+      "epoch": 13.043,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0004185358683268229,
+      "learning_rate": 0.00013157894736842105,
+      "loss": 54.4688,
+      "step": 875
+    },
+    {
+      "crossentropy": 2.7317864894866943,
+      "epoch": 13.044,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0004252115885416667,
+      "learning_rate": 0.00013052631578947368,
+      "loss": 55.3466,
+      "step": 876
+    },
+    {
+      "crossentropy": 2.761776566505432,
+      "epoch": 13.045,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0004353205362955729,
+      "learning_rate": 0.00012947368421052633,
+      "loss": 56.4126,
+      "step": 877
+    },
+    {
+      "crossentropy": 2.8709728717803955,
+      "epoch": 13.046,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003788630167643229,
+      "learning_rate": 0.00012842105263157893,
+      "loss": 55.527,
+      "step": 878
+    },
+    {
+      "crossentropy": 2.9433149099349976,
+      "epoch": 13.047,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0002878824869791667,
+      "learning_rate": 0.00012736842105263158,
+      "loss": 56.7329,
+      "step": 879
+    },
+    {
+      "crossentropy": 2.7954342365264893,
+      "epoch": 13.048,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0003611246744791667,
+      "learning_rate": 0.0001263157894736842,
+      "loss": 56.6293,
+      "step": 880
+    },
+    {
+      "crossentropy": 2.9298503398895264,
+      "epoch": 13.049,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00029703776041666664,
+      "learning_rate": 0.00012526315789473686,
+      "loss": 57.9576,
+      "step": 881
+    },
+    {
+      "crossentropy": 2.910145401954651,
+      "epoch": 13.05,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.00032393137613932293,
+      "learning_rate": 0.00012421052631578949,
+      "loss": 57.0496,
+      "step": 882
+    },
+    {
+      "crossentropy": 2.80023729801178,
+      "epoch": 13.051,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.00032138824462890625,
+      "learning_rate": 0.0001231578947368421,
+      "loss": 56.4184,
+      "step": 883
+    },
+    {
+      "crossentropy": 2.8765305280685425,
+      "epoch": 13.052,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003326416015625,
+      "learning_rate": 0.00012210526315789474,
+      "loss": 57.1571,
+      "step": 884
+    },
+    {
+      "crossentropy": 2.877085328102112,
+      "epoch": 13.053,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003402074178059896,
+      "learning_rate": 0.00012105263157894738,
+      "loss": 56.3713,
+      "step": 885
+    },
+    {
+      "crossentropy": 2.825315833091736,
+      "epoch": 13.054,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00033162434895833335,
+      "learning_rate": 0.00012,
+      "loss": 58.7512,
+      "step": 886
+    },
+    {
+      "crossentropy": 2.9117591381073,
+      "epoch": 13.055,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003371556599934896,
+      "learning_rate": 0.00011894736842105264,
+      "loss": 57.4971,
+      "step": 887
+    },
+    {
+      "crossentropy": 2.95816707611084,
+      "epoch": 13.056,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0003371556599934896,
+      "learning_rate": 0.00011789473684210527,
+      "loss": 58.199,
+      "step": 888
+    },
+    {
+      "crossentropy": 2.877577304840088,
+      "epoch": 13.057,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.00033969879150390624,
+      "learning_rate": 0.00011684210526315791,
+      "loss": 57.5653,
+      "step": 889
+    },
+    {
+      "crossentropy": 2.839280843734741,
+      "epoch": 13.058,
+      "grad_norm": 0.796875,
+      "grad_norm_var": 0.0005238215128580729,
+      "learning_rate": 0.00011578947368421053,
+      "loss": 55.8712,
+      "step": 890
+    },
+    {
+      "crossentropy": 2.9066461324691772,
+      "epoch": 13.059,
+      "grad_norm": 0.78515625,
+      "grad_norm_var": 0.0005823135375976563,
+      "learning_rate": 0.00011473684210526316,
+      "loss": 57.5233,
+      "step": 891
+    },
+    {
+      "crossentropy": 2.963346004486084,
+      "epoch": 13.06,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005877176920572917,
+      "learning_rate": 0.0001136842105263158,
+      "loss": 58.7213,
+      "step": 892
+    },
+    {
+      "crossentropy": 2.887979745864868,
+      "epoch": 13.061,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0005551656087239583,
+      "learning_rate": 0.00011263157894736841,
+      "loss": 57.1343,
+      "step": 893
+    },
+    {
+      "crossentropy": 2.8576492071151733,
+      "epoch": 13.062,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005816141764322917,
+      "learning_rate": 0.00011157894736842105,
+      "loss": 57.0907,
+      "step": 894
+    },
+    {
+      "crossentropy": 2.950958728790283,
+      "epoch": 13.063,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.0006372451782226563,
+      "learning_rate": 0.00011052631578947368,
+      "loss": 56.6607,
+      "step": 895
+    },
+    {
+      "crossentropy": 2.8504726886749268,
+      "epoch": 13.064,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.000571441650390625,
+      "learning_rate": 0.00010947368421052632,
+      "loss": 57.7151,
+      "step": 896
+    },
+    {
+      "crossentropy": 2.878578782081604,
+      "epoch": 14.001,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005177179972330729,
+      "learning_rate": 0.00010842105263157894,
+      "loss": 55.9718,
+      "step": 897
+    },
+    {
+      "crossentropy": 2.782250165939331,
+      "epoch": 14.002,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005083719889322917,
+      "learning_rate": 0.00010736842105263158,
+      "loss": 57.0449,
+      "step": 898
+    },
+    {
+      "crossentropy": 2.8027374744415283,
+      "epoch": 14.003,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0004974365234375,
+      "learning_rate": 0.00010631578947368421,
+      "loss": 55.2197,
+      "step": 899
+    },
+    {
+      "crossentropy": 2.7636749744415283,
+      "epoch": 14.004,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0004819234212239583,
+      "learning_rate": 0.00010526315789473683,
+      "loss": 55.1626,
+      "step": 900
+    },
+    {
+      "crossentropy": 2.7231706380844116,
+      "epoch": 14.005,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00047702789306640624,
+      "learning_rate": 0.00010421052631578947,
+      "loss": 54.001,
+      "step": 901
+    },
+    {
+      "crossentropy": 2.883240580558777,
+      "epoch": 14.006,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.000469970703125,
+      "learning_rate": 0.0001031578947368421,
+      "loss": 58.3301,
+      "step": 902
+    },
+    {
+      "crossentropy": 2.6990073919296265,
+      "epoch": 14.007,
+      "grad_norm": 0.703125,
+      "grad_norm_var": 0.0005655288696289062,
+      "learning_rate": 0.00010210526315789474,
+      "loss": 55.7926,
+      "step": 903
+    },
+    {
+      "crossentropy": 2.786812424659729,
+      "epoch": 14.008,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005706787109375,
+      "learning_rate": 0.00010105263157894737,
+      "loss": 56.0858,
+      "step": 904
+    },
+    {
+      "crossentropy": 2.8873369693756104,
+      "epoch": 14.009,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.000347900390625,
+      "learning_rate": 0.0001,
+      "loss": 57.1081,
+      "step": 905
+    },
+    {
+      "crossentropy": 2.8812583684921265,
+      "epoch": 14.01,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.00028018951416015626,
+      "learning_rate": 9.894736842105263e-05,
+      "loss": 56.194,
+      "step": 906
+    },
+    {
+      "crossentropy": 2.73581063747406,
+      "epoch": 14.011,
+      "grad_norm": 0.77734375,
+      "grad_norm_var": 0.00038547515869140624,
+      "learning_rate": 9.789473684210526e-05,
+      "loss": 56.1636,
+      "step": 907
+    },
+    {
+      "crossentropy": 2.7859939336776733,
+      "epoch": 14.012,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00038859049479166664,
+      "learning_rate": 9.68421052631579e-05,
+      "loss": 56.4249,
+      "step": 908
+    },
+    {
+      "crossentropy": 2.8880062103271484,
+      "epoch": 14.013,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00034122467041015626,
+      "learning_rate": 9.578947368421052e-05,
+      "loss": 58.172,
+      "step": 909
+    },
+    {
+      "crossentropy": 2.772728443145752,
+      "epoch": 14.014,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003218968709309896,
+      "learning_rate": 9.473684210526316e-05,
+      "loss": 55.1069,
+      "step": 910
+    },
+    {
+      "crossentropy": 2.77231502532959,
+      "epoch": 14.015,
+      "grad_norm": 0.7734375,
+      "grad_norm_var": 0.0004025777180989583,
+      "learning_rate": 9.368421052631579e-05,
+      "loss": 54.3101,
+      "step": 911
+    },
+    {
+      "crossentropy": 2.7247848510742188,
+      "epoch": 14.016,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00040791829427083335,
+      "learning_rate": 9.263157894736843e-05,
+      "loss": 55.4261,
+      "step": 912
+    },
+    {
+      "crossentropy": 2.907187581062317,
+      "epoch": 14.017,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00040486653645833336,
+      "learning_rate": 9.157894736842105e-05,
+      "loss": 57.3701,
+      "step": 913
+    },
+    {
+      "crossentropy": 2.816227078437805,
+      "epoch": 14.018,
+      "grad_norm": 0.69921875,
+      "grad_norm_var": 0.000506591796875,
+      "learning_rate": 9.052631578947369e-05,
+      "loss": 55.6567,
+      "step": 914
+    },
+    {
+      "crossentropy": 2.8363900184631348,
+      "epoch": 14.019,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0005202611287434896,
+      "learning_rate": 8.947368421052632e-05,
+      "loss": 57.0809,
+      "step": 915
+    },
+    {
+      "crossentropy": 2.856706380844116,
+      "epoch": 14.02,
+      "grad_norm": 0.69921875,
+      "grad_norm_var": 0.0006189346313476562,
+      "learning_rate": 8.842105263157894e-05,
+      "loss": 57.1048,
+      "step": 916
+    },
+    {
+      "crossentropy": 2.8060814142227173,
+      "epoch": 14.021,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0006182352701822917,
+      "learning_rate": 8.736842105263158e-05,
+      "loss": 55.9918,
+      "step": 917
+    },
+    {
+      "crossentropy": 2.956759810447693,
+      "epoch": 14.022,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0006382624308268229,
+      "learning_rate": 8.631578947368421e-05,
+      "loss": 57.2669,
+      "step": 918
+    },
+    {
+      "crossentropy": 2.754731297492981,
+      "epoch": 14.023,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0006067276000976563,
+      "learning_rate": 8.526315789473685e-05,
+      "loss": 53.8387,
+      "step": 919
+    },
+    {
+      "crossentropy": 2.7637789249420166,
+      "epoch": 14.024,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0006652196248372396,
+      "learning_rate": 8.421052631578948e-05,
+      "loss": 55.6373,
+      "step": 920
+    },
+    {
+      "crossentropy": 2.721121907234192,
+      "epoch": 14.025,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0006586074829101563,
+      "learning_rate": 8.315789473684212e-05,
+      "loss": 54.8625,
+      "step": 921
+    },
+    {
+      "crossentropy": 2.8451952934265137,
+      "epoch": 14.026,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0005782445271809896,
+      "learning_rate": 8.210526315789474e-05,
+      "loss": 57.2868,
+      "step": 922
+    },
+    {
+      "crossentropy": 2.796187996864319,
+      "epoch": 14.027,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0004668553670247396,
+      "learning_rate": 8.105263157894737e-05,
+      "loss": 56.2234,
+      "step": 923
+    },
+    {
+      "crossentropy": 2.824811339378357,
+      "epoch": 14.028,
+      "grad_norm": 0.703125,
+      "grad_norm_var": 0.000528717041015625,
+      "learning_rate": 8e-05,
+      "loss": 56.7024,
+      "step": 924
+    },
+    {
+      "crossentropy": 2.7085684537887573,
+      "epoch": 14.029,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0005868911743164062,
+      "learning_rate": 7.894736842105263e-05,
+      "loss": 56.1883,
+      "step": 925
+    },
+    {
+      "crossentropy": 2.7962204217910767,
+      "epoch": 14.03,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0005782445271809896,
+      "learning_rate": 7.789473684210527e-05,
+      "loss": 57.1829,
+      "step": 926
+    },
+    {
+      "crossentropy": 2.796944737434387,
+      "epoch": 14.031,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.0005136489868164063,
+      "learning_rate": 7.68421052631579e-05,
+      "loss": 56.7551,
+      "step": 927
+    },
+    {
+      "crossentropy": 2.7813810110092163,
+      "epoch": 14.032,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0005212783813476563,
+      "learning_rate": 7.578947368421054e-05,
+      "loss": 56.2743,
+      "step": 928
+    },
+    {
+      "crossentropy": 2.8674248456954956,
+      "epoch": 14.033,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0005212783813476563,
+      "learning_rate": 7.473684210526316e-05,
+      "loss": 58.4157,
+      "step": 929
+    },
+    {
+      "crossentropy": 2.7428762912750244,
+      "epoch": 14.034,
+      "grad_norm": 0.69140625,
+      "grad_norm_var": 0.0005609512329101563,
+      "learning_rate": 7.368421052631579e-05,
+      "loss": 55.4523,
+      "step": 930
+    },
+    {
+      "crossentropy": 2.7889604568481445,
+      "epoch": 14.035,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.0005304336547851563,
+      "learning_rate": 7.263157894736843e-05,
+      "loss": 56.8295,
+      "step": 931
+    },
+    {
+      "crossentropy": 2.78811252117157,
+      "epoch": 14.036,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.0005039850870768229,
+      "learning_rate": 7.157894736842105e-05,
+      "loss": 56.2246,
+      "step": 932
+    },
+    {
+      "crossentropy": 2.797974109649658,
+      "epoch": 14.037,
+      "grad_norm": 0.69921875,
+      "grad_norm_var": 0.0005853652954101562,
+      "learning_rate": 7.05263157894737e-05,
+      "loss": 56.1583,
+      "step": 933
+    },
+    {
+      "crossentropy": 2.9307310581207275,
+      "epoch": 14.038,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0005853652954101562,
+      "learning_rate": 6.947368421052632e-05,
+      "loss": 57.1375,
+      "step": 934
+    },
+    {
+      "crossentropy": 2.8903194665908813,
+      "epoch": 14.039,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.000559234619140625,
+      "learning_rate": 6.842105263157896e-05,
+      "loss": 57.0463,
+      "step": 935
+    },
+    {
+      "crossentropy": 2.865442395210266,
+      "epoch": 14.04,
+      "grad_norm": 0.69921875,
+      "grad_norm_var": 0.0005340576171875,
+      "learning_rate": 6.736842105263157e-05,
+      "loss": 56.6052,
+      "step": 936
+    },
+    {
+      "crossentropy": 2.9218584299087524,
+      "epoch": 14.041,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0005228678385416667,
+      "learning_rate": 6.631578947368421e-05,
+      "loss": 57.2276,
+      "step": 937
+    },
+    {
+      "crossentropy": 2.819886326789856,
+      "epoch": 14.042,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0005167007446289062,
+      "learning_rate": 6.526315789473684e-05,
+      "loss": 57.4423,
+      "step": 938
+    },
+    {
+      "crossentropy": 2.6768345832824707,
+      "epoch": 14.043,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0005228042602539063,
+      "learning_rate": 6.421052631578946e-05,
+      "loss": 54.1839,
+      "step": 939
+    },
+    {
+      "crossentropy": 2.7807631492614746,
+      "epoch": 14.044,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00047607421875,
+      "learning_rate": 6.31578947368421e-05,
+      "loss": 55.5346,
+      "step": 940
+    },
+    {
+      "crossentropy": 2.8685619831085205,
+      "epoch": 14.045,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.000400543212890625,
+      "learning_rate": 6.210526315789474e-05,
+      "loss": 57.8767,
+      "step": 941
+    },
+    {
+      "crossentropy": 2.8203389644622803,
+      "epoch": 14.046,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.00039647420247395835,
+      "learning_rate": 6.105263157894737e-05,
+      "loss": 56.1573,
+      "step": 942
+    },
+    {
+      "crossentropy": 2.8836749792099,
+      "epoch": 14.047,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.00034122467041015626,
+      "learning_rate": 6e-05,
+      "loss": 56.838,
+      "step": 943
+    },
+    {
+      "crossentropy": 2.9158384799957275,
+      "epoch": 14.048,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000382232666015625,
+      "learning_rate": 5.8947368421052634e-05,
+      "loss": 58.3756,
+      "step": 944
+    },
+    {
+      "crossentropy": 2.7543821334838867,
+      "epoch": 14.049,
+      "grad_norm": 0.6953125,
+      "grad_norm_var": 0.0004414240519205729,
+      "learning_rate": 5.789473684210527e-05,
+      "loss": 55.6096,
+      "step": 945
+    },
+    {
+      "crossentropy": 2.918057680130005,
+      "epoch": 14.05,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003985087076822917,
+      "learning_rate": 5.68421052631579e-05,
+      "loss": 56.9547,
+      "step": 946
+    },
+    {
+      "crossentropy": 2.8211581707000732,
+      "epoch": 14.051,
+      "grad_norm": 0.71484375,
+      "grad_norm_var": 0.00040868123372395836,
+      "learning_rate": 5.5789473684210526e-05,
+      "loss": 55.7815,
+      "step": 947
+    },
+    {
+      "crossentropy": 2.7923535108566284,
+      "epoch": 14.052,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.000323486328125,
+      "learning_rate": 5.473684210526316e-05,
+      "loss": 55.9897,
+      "step": 948
+    },
+    {
+      "crossentropy": 2.8318454027175903,
+      "epoch": 14.053,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.00029239654541015627,
+      "learning_rate": 5.368421052631579e-05,
+      "loss": 57.443,
+      "step": 949
+    },
+    {
+      "crossentropy": 2.7717995643615723,
+      "epoch": 14.054,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.000290679931640625,
+      "learning_rate": 5.263157894736842e-05,
+      "loss": 55.1007,
+      "step": 950
+    },
+    {
+      "crossentropy": 2.7858974933624268,
+      "epoch": 14.055,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.00029341379801432293,
+      "learning_rate": 5.157894736842105e-05,
+      "loss": 57.4654,
+      "step": 951
+    },
+    {
+      "crossentropy": 2.9458006620407104,
+      "epoch": 14.056,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00025018056233723957,
+      "learning_rate": 5.052631578947368e-05,
+      "loss": 57.8048,
+      "step": 952
+    },
+    {
+      "crossentropy": 2.7548274993896484,
+      "epoch": 14.057,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.00025202433268229165,
+      "learning_rate": 4.9473684210526315e-05,
+      "loss": 56.2672,
+      "step": 953
+    },
+    {
+      "crossentropy": 2.8330026865005493,
+      "epoch": 14.058,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00024763743082682293,
+      "learning_rate": 4.842105263157895e-05,
+      "loss": 56.3621,
+      "step": 954
+    },
+    {
+      "crossentropy": 2.775303602218628,
+      "epoch": 14.059,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00024051666259765624,
+      "learning_rate": 4.736842105263158e-05,
+      "loss": 56.4681,
+      "step": 955
+    },
+    {
+      "crossentropy": 2.84080708026886,
+      "epoch": 14.06,
+      "grad_norm": 0.76953125,
+      "grad_norm_var": 0.0003524144490559896,
+      "learning_rate": 4.6315789473684214e-05,
+      "loss": 57.448,
+      "step": 956
+    },
+    {
+      "crossentropy": 2.891412377357483,
+      "epoch": 14.061,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003524144490559896,
+      "learning_rate": 4.5263157894736846e-05,
+      "loss": 57.3745,
+      "step": 957
+    },
+    {
+      "crossentropy": 2.778434157371521,
+      "epoch": 14.062,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00034885406494140626,
+      "learning_rate": 4.421052631578947e-05,
+      "loss": 55.2243,
+      "step": 958
+    },
+    {
+      "crossentropy": 2.949416756629944,
+      "epoch": 14.063,
+      "grad_norm": 0.7578125,
+      "grad_norm_var": 0.000376129150390625,
+      "learning_rate": 4.3157894736842105e-05,
+      "loss": 59.2026,
+      "step": 959
+    },
+    {
+      "crossentropy": 2.902771472930908,
+      "epoch": 14.064,
+      "grad_norm": 0.76171875,
+      "grad_norm_var": 0.00039005279541015625,
+      "learning_rate": 4.210526315789474e-05,
+      "loss": 56.5269,
+      "step": 960
+    },
+    {
+      "crossentropy": 2.698053002357483,
+      "epoch": 15.001,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003310521443684896,
+      "learning_rate": 4.105263157894737e-05,
+      "loss": 54.2453,
+      "step": 961
+    },
+    {
+      "crossentropy": 2.75066876411438,
+      "epoch": 15.002,
+      "grad_norm": 0.7421875,
+      "grad_norm_var": 0.0003021240234375,
+      "learning_rate": 4e-05,
+      "loss": 56.7785,
+      "step": 962
+    },
+    {
+      "crossentropy": 2.881886601448059,
+      "epoch": 15.003,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00028781890869140626,
+      "learning_rate": 3.8947368421052636e-05,
+      "loss": 56.9223,
+      "step": 963
+    },
+    {
+      "crossentropy": 2.8759313821792603,
+      "epoch": 15.004,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.000254058837890625,
+      "learning_rate": 3.789473684210527e-05,
+      "loss": 56.867,
+      "step": 964
+    },
+    {
+      "crossentropy": 2.783510446548462,
+      "epoch": 15.005,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00025202433268229165,
+      "learning_rate": 3.6842105263157895e-05,
+      "loss": 56.8626,
+      "step": 965
+    },
+    {
+      "crossentropy": 2.778250217437744,
+      "epoch": 15.006,
+      "grad_norm": 0.6953125,
+      "grad_norm_var": 0.00035196940104166666,
+      "learning_rate": 3.578947368421053e-05,
+      "loss": 54.6814,
+      "step": 966
+    },
+    {
+      "crossentropy": 2.755368709564209,
+      "epoch": 15.007,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00035196940104166666,
+      "learning_rate": 3.473684210526316e-05,
+      "loss": 56.5905,
+      "step": 967
+    },
+    {
+      "crossentropy": 2.8220062255859375,
+      "epoch": 15.008,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.00035196940104166666,
+      "learning_rate": 3.3684210526315786e-05,
+      "loss": 56.3069,
+      "step": 968
+    },
+    {
+      "crossentropy": 2.8596099615097046,
+      "epoch": 15.009,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.0003598531087239583,
+      "learning_rate": 3.263157894736842e-05,
+      "loss": 56.5936,
+      "step": 969
+    },
+    {
+      "crossentropy": 2.797579288482666,
+      "epoch": 15.01,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.00037988026936848957,
+      "learning_rate": 3.157894736842105e-05,
+      "loss": 55.846,
+      "step": 970
+    },
+    {
+      "crossentropy": 2.6855785846710205,
+      "epoch": 15.011,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003142674763997396,
+      "learning_rate": 3.0526315789473684e-05,
+      "loss": 55.2178,
+      "step": 971
+    },
+    {
+      "crossentropy": 2.7928699254989624,
+      "epoch": 15.012,
+      "grad_norm": 0.6953125,
+      "grad_norm_var": 0.0004012425740559896,
+      "learning_rate": 2.9473684210526317e-05,
+      "loss": 54.3003,
+      "step": 972
+    },
+    {
+      "crossentropy": 2.7457377910614014,
+      "epoch": 15.013,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0004042943318684896,
+      "learning_rate": 2.842105263157895e-05,
+      "loss": 54.456,
+      "step": 973
+    },
+    {
+      "crossentropy": 2.825121760368347,
+      "epoch": 15.014,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.0003916422526041667,
+      "learning_rate": 2.736842105263158e-05,
+      "loss": 56.7055,
+      "step": 974
+    },
+    {
+      "crossentropy": 2.6830825805664062,
+      "epoch": 15.015,
+      "grad_norm": 0.75390625,
+      "grad_norm_var": 0.000363922119140625,
+      "learning_rate": 2.631578947368421e-05,
+      "loss": 55.219,
+      "step": 975
+    },
+    {
+      "crossentropy": 2.712595582008362,
+      "epoch": 15.016,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.00029850006103515625,
+      "learning_rate": 2.526315789473684e-05,
+      "loss": 55.8602,
+      "step": 976
+    },
+    {
+      "crossentropy": 2.9039396047592163,
+      "epoch": 15.017,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.000296783447265625,
+      "learning_rate": 2.4210526315789474e-05,
+      "loss": 56.9117,
+      "step": 977
+    },
+    {
+      "crossentropy": 2.8268353939056396,
+      "epoch": 15.018,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003692626953125,
+      "learning_rate": 2.3157894736842107e-05,
+      "loss": 55.0809,
+      "step": 978
+    },
+    {
+      "crossentropy": 2.846464157104492,
+      "epoch": 15.019,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.00037174224853515626,
+      "learning_rate": 2.2105263157894736e-05,
+      "loss": 57.271,
+      "step": 979
+    },
+    {
+      "crossentropy": 2.920137643814087,
+      "epoch": 15.02,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.00037174224853515626,
+      "learning_rate": 2.105263157894737e-05,
+      "loss": 58.0873,
+      "step": 980
+    },
+    {
+      "crossentropy": 2.5910059213638306,
+      "epoch": 15.021,
+      "grad_norm": 0.703125,
+      "grad_norm_var": 0.00040461222330729165,
+      "learning_rate": 2e-05,
+      "loss": 54.132,
+      "step": 981
+    },
+    {
+      "crossentropy": 2.8294392824172974,
+      "epoch": 15.022,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.0003458658854166667,
+      "learning_rate": 1.8947368421052634e-05,
+      "loss": 57.1192,
+      "step": 982
+    },
+    {
+      "crossentropy": 2.8623571395874023,
+      "epoch": 15.023,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.00035578409830729166,
+      "learning_rate": 1.7894736842105264e-05,
+      "loss": 56.702,
+      "step": 983
+    },
+    {
+      "crossentropy": 2.8070725202560425,
+      "epoch": 15.024,
+      "grad_norm": 0.69921875,
+      "grad_norm_var": 0.0004221598307291667,
+      "learning_rate": 1.6842105263157893e-05,
+      "loss": 55.9888,
+      "step": 984
+    },
+    {
+      "crossentropy": 2.7253717184066772,
+      "epoch": 15.025,
+      "grad_norm": 0.73046875,
+      "grad_norm_var": 0.00042057037353515625,
+      "learning_rate": 1.5789473684210526e-05,
+      "loss": 56.3073,
+      "step": 985
+    },
+    {
+      "crossentropy": 2.8384543657302856,
+      "epoch": 15.026,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.00042057037353515625,
+      "learning_rate": 1.4736842105263159e-05,
+      "loss": 56.2086,
+      "step": 986
+    },
+    {
+      "crossentropy": 2.8448028564453125,
+      "epoch": 15.027,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.00041681925455729166,
+      "learning_rate": 1.368421052631579e-05,
+      "loss": 57.0379,
+      "step": 987
+    },
+    {
+      "crossentropy": 2.7504059076309204,
+      "epoch": 15.028,
+      "grad_norm": 0.75,
+      "grad_norm_var": 0.00033671061197916666,
+      "learning_rate": 1.263157894736842e-05,
+      "loss": 56.9316,
+      "step": 988
+    },
+    {
+      "crossentropy": 2.8394349813461304,
+      "epoch": 15.029,
+      "grad_norm": 0.765625,
+      "grad_norm_var": 0.0003941218058268229,
+      "learning_rate": 1.1578947368421053e-05,
+      "loss": 57.3295,
+      "step": 989
+    },
+    {
+      "crossentropy": 2.8349748849868774,
+      "epoch": 15.03,
+      "grad_norm": 0.70703125,
+      "grad_norm_var": 0.0004261652628580729,
+      "learning_rate": 1.0526315789473684e-05,
+      "loss": 56.5169,
+      "step": 990
+    },
+    {
+      "crossentropy": 2.8274471759796143,
+      "epoch": 15.031,
+      "grad_norm": 0.74609375,
+      "grad_norm_var": 0.0004093805948893229,
+      "learning_rate": 9.473684210526317e-06,
+      "loss": 56.015,
+      "step": 991
+    },
+    {
+      "crossentropy": 2.8977288007736206,
+      "epoch": 15.032,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.00041681925455729166,
+      "learning_rate": 8.421052631578947e-06,
+      "loss": 56.5352,
+      "step": 992
+    },
+    {
+      "crossentropy": 2.9102202653884888,
+      "epoch": 15.033,
+      "grad_norm": 0.71875,
+      "grad_norm_var": 0.0004290262858072917,
+      "learning_rate": 7.368421052631579e-06,
+      "loss": 57.9547,
+      "step": 993
+    },
+    {
+      "crossentropy": 2.9221819639205933,
+      "epoch": 15.034,
+      "grad_norm": 0.734375,
+      "grad_norm_var": 0.0003496805826822917,
+      "learning_rate": 6.31578947368421e-06,
+      "loss": 56.7362,
+      "step": 994
+    },
+    {
+      "crossentropy": 2.800140857696533,
+      "epoch": 15.035,
+      "grad_norm": 0.73828125,
+      "grad_norm_var": 0.0003496805826822917,
+      "learning_rate": 5.263157894736842e-06,
+      "loss": 56.3905,
+      "step": 995
+    },
+    {
+      "crossentropy": 2.6751104593276978,
+      "epoch": 15.036,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.00036970774332682293,
+      "learning_rate": 4.210526315789473e-06,
+      "loss": 55.695,
+      "step": 996
+    },
+    {
+      "crossentropy": 2.7358882427215576,
+      "epoch": 15.037,
+      "grad_norm": 0.72265625,
+      "grad_norm_var": 0.0003255208333333333,
+      "learning_rate": 3.157894736842105e-06,
+      "loss": 57.0907,
+      "step": 997
+    },
+    {
+      "crossentropy": 2.7598263025283813,
+      "epoch": 15.038,
+      "grad_norm": 0.7109375,
+      "grad_norm_var": 0.0003191630045572917,
+      "learning_rate": 2.1052631578947366e-06,
+      "loss": 55.385,
+      "step": 998
+    },
+    {
+      "crossentropy": 2.8410521745681763,
+      "epoch": 15.039,
+      "grad_norm": 0.6953125,
+      "grad_norm_var": 0.00035800933837890623,
+      "learning_rate": 1.0526315789473683e-06,
+      "loss": 57.0506,
+      "step": 999
+    },
+    {
+      "crossentropy": 2.7283369302749634,
+      "epoch": 15.04,
+      "grad_norm": 0.7265625,
+      "grad_norm_var": 0.000311279296875,
+      "learning_rate": 0.0,
+      "loss": 57.0943,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 250,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.033762523578368e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}