diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,36033 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.25,
+  "eval_steps": 2000,
+  "global_step": 30000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 8.333333333333333e-05,
+      "grad_norm": 33.25,
+      "learning_rate": 0.0001,
+      "loss": 7.5723,
+      "loss/crossentropy": 2.127359576523304,
+      "loss/hidden": 3.5390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22891336753964425,
+      "step": 10
+    },
+    {
+      "epoch": 0.00016666666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.769791666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3973,
+      "loss/crossentropy": 1.9497304677963256,
+      "loss/hidden": 3.56875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20572521891444923,
+      "step": 20
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.0494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3789,
+      "loss/crossentropy": 1.873328334838152,
+      "loss/hidden": 3.451953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17487502712756395,
+      "step": 30
+    },
+    {
+      "epoch": 0.0003333333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.137434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.4058,
+      "loss/crossentropy": 1.9326039656996727,
+      "loss/hidden": 3.460546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20465592797845603,
+      "step": 40
+    },
+    {
+      "epoch": 0.0004166666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.910872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.4246,
+      "loss/crossentropy": 1.91383947879076,
+      "loss/hidden": 3.4359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20186774488538503,
+      "step": 50
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 40.75,
+      "grad_norm_var": 5.63515625,
+      "learning_rate": 0.0001,
+      "loss": 7.5044,
+      "loss/crossentropy": 1.92522851228714,
+      "loss/hidden": 3.497265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19990789387375116,
+      "step": 60
+    },
+    {
+      "epoch": 0.0005833333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 9.306266253378193e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.357,
+      "loss/crossentropy": 1.9717040583491325,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909018535166979,
+      "step": 70
+    },
+    {
+      "epoch": 0.0006666666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.306266259608481e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3788,
+      "loss/crossentropy": 1.9090200901031493,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17854449450969695,
+      "step": 80
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.45625,
+      "learning_rate": 0.0001,
+      "loss": 7.4741,
+      "loss/crossentropy": 1.8572393998503685,
+      "loss/hidden": 3.571875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19387567192316055,
+      "step": 90
+    },
+    {
+      "epoch": 0.0008333333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.949934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.4307,
+      "loss/crossentropy": 2.076287841796875,
+      "loss/hidden": 3.423046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21235600393265486,
+      "step": 100
+    },
+    {
+      "epoch": 0.0009166666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.5143229166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.4063,
+      "loss/crossentropy": 2.0186501413583757,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20398833509534597,
+      "step": 110
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.6372395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.4431,
+      "loss/crossentropy": 1.9501568615436553,
+      "loss/hidden": 3.541015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1999204397201538,
+      "step": 120
+    },
+    {
+      "epoch": 0.0010833333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 4.7837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.3751,
+      "loss/crossentropy": 2.0366190388798713,
+      "loss/hidden": 3.4203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18896052613854408,
+      "step": 130
+    },
+    {
+      "epoch": 0.0011666666666666668,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.43125,
+      "learning_rate": 0.0001,
+      "loss": 7.4741,
+      "loss/crossentropy": 1.8939224407076836,
+      "loss/hidden": 3.4390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19272235538810492,
+      "step": 140
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 33.0,
+      "grad_norm_var": 27.179622395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.5153,
+      "loss/crossentropy": 2.0108851701021195,
+      "loss/hidden": 3.4078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20005665346980095,
+      "step": 150
+    },
+    {
+      "epoch": 0.0013333333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 58.23951822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.4085,
+      "loss/crossentropy": 1.9532957911491393,
+      "loss/hidden": 3.505078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2099555365741253,
+      "step": 160
+    },
+    {
+      "epoch": 0.0014166666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 39.82858072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.4047,
+      "loss/crossentropy": 1.9121289029717445,
+      "loss/hidden": 3.439453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977582860738039,
+      "step": 170
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 41.25,
+      "grad_norm_var": 2.3565788464023603e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3435,
+      "loss/crossentropy": 1.9281877517700194,
+      "loss/hidden": 3.50546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19753940496593714,
+      "step": 180
+    },
+    {
+      "epoch": 0.0015833333333333333,
+      "grad_norm": 35.25,
+      "grad_norm_var": 2.3565788455580467e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.471,
+      "loss/crossentropy": 2.032211276888847,
+      "loss/hidden": 3.503515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19554367773234843,
+      "step": 190
+    },
+    {
+      "epoch": 0.0016666666666666668,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.676497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2708,
+      "loss/crossentropy": 2.005619902908802,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833203172311187,
+      "step": 200
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 30.75,
+      "grad_norm_var": 8.633268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.4096,
+      "loss/crossentropy": 2.043836605548859,
+      "loss/hidden": 3.467578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20938539411872625,
+      "step": 210
+    },
+    {
+      "epoch": 0.0018333333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 23.039322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3709,
+      "loss/crossentropy": 2.054931232333183,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20825725365430117,
+      "step": 220
+    },
+    {
+      "epoch": 0.0019166666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.0732816604826787e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3838,
+      "loss/crossentropy": 1.9404870495200157,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20280153322964906,
+      "step": 230
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 30.375,
+      "grad_norm_var": 12.29765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3121,
+      "loss/crossentropy": 1.8668742343783378,
+      "loss/hidden": 3.47578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21291877124458553,
+      "step": 240
+    },
+    {
+      "epoch": 0.0020833333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 56.5400390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3905,
+      "loss/crossentropy": 1.9617563866078853,
+      "loss/hidden": 3.51640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919050351716578,
+      "step": 250
+    },
+    {
+      "epoch": 0.0021666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 47.96640625,
+      "learning_rate": 0.0001,
+      "loss": 7.3528,
+      "loss/crossentropy": 1.8532208397984504,
+      "loss/hidden": 3.551171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19277248289436102,
+      "step": 260
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 32.0,
+      "grad_norm_var": 18.443684895833332,
+      "learning_rate": 0.0001,
+      "loss": 7.3309,
+      "loss/crossentropy": 2.0671744480729104,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19330325871706008,
+      "step": 270
+    },
+    {
+      "epoch": 0.0023333333333333335,
+      "grad_norm": 41.0,
+      "grad_norm_var": 23.93125,
+      "learning_rate": 0.0001,
+      "loss": 7.2214,
+      "loss/crossentropy": 1.9675503507256509,
+      "loss/hidden": 3.451953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18025756757706404,
+      "step": 280
+    },
+    {
+      "epoch": 0.002416666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 17.74765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2949,
+      "loss/crossentropy": 1.9656010001897812,
+      "loss/hidden": 3.479296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20177473574876786,
+      "step": 290
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 43.5,
+      "grad_norm_var": 4.011194233137573e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4422,
+      "loss/crossentropy": 1.9836331263184548,
+      "loss/hidden": 3.30546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909252552315593,
+      "step": 300
+    },
+    {
+      "epoch": 0.0025833333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.0111942329792384e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4124,
+      "loss/crossentropy": 2.082610194385052,
+      "loss/hidden": 3.462890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21536452919244767,
+      "step": 310
+    },
+    {
+      "epoch": 0.0026666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.5056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.3504,
+      "loss/crossentropy": 1.9580433249473572,
+      "loss/hidden": 3.429296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865969082340598,
+      "step": 320
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.421809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2673,
+      "loss/crossentropy": 2.0938573241233827,
+      "loss/hidden": 3.360546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21014094576239586,
+      "step": 330
+    },
+    {
+      "epoch": 0.0028333333333333335,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.9494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3136,
+      "loss/crossentropy": 1.8406853549182416,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1884180039167404,
+      "step": 340
+    },
+    {
+      "epoch": 0.002916666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 14.85625,
+      "learning_rate": 0.0001,
+      "loss": 7.3683,
+      "loss/crossentropy": 1.972689888626337,
+      "loss/hidden": 3.431640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20450354916974903,
+      "step": 350
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 35.0,
+      "grad_norm_var": 17.483072916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.3037,
+      "loss/crossentropy": 1.995607402920723,
+      "loss/hidden": 3.412890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20541023295372723,
+      "step": 360
+    },
+    {
+      "epoch": 0.0030833333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.827351345607127e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4571,
+      "loss/crossentropy": 1.9991235852241516,
+      "loss/hidden": 3.4359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20315398182719946,
+      "step": 370
+    },
+    {
+      "epoch": 0.0031666666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.971875,
+      "learning_rate": 0.0001,
+      "loss": 7.3758,
+      "loss/crossentropy": 2.0666207425296306,
+      "loss/hidden": 3.555859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20830214060842991,
+      "step": 380
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 34.25,
+      "grad_norm_var": 109.1759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.4163,
+      "loss/crossentropy": 1.942155936360359,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19059792384505272,
+      "step": 390
+    },
+    {
+      "epoch": 0.0033333333333333335,
+      "grad_norm": 29.0,
+      "grad_norm_var": 10.253125,
+      "learning_rate": 0.0001,
+      "loss": 7.24,
+      "loss/crossentropy": 1.8737600088119506,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19837535824626684,
+      "step": 400
+    },
+    {
+      "epoch": 0.003416666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.715284576287676e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3633,
+      "loss/crossentropy": 2.0632002711296082,
+      "loss/hidden": 3.73828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22990943398326635,
+      "step": 410
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.715284568977354e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3609,
+      "loss/crossentropy": 2.0630046755075453,
+      "loss/hidden": 3.31484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19341134056448936,
+      "step": 420
+    },
+    {
+      "epoch": 0.0035833333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.5770833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3068,
+      "loss/crossentropy": 1.9697899878025056,
+      "loss/hidden": 3.48515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20131503157317637,
+      "step": 430
+    },
+    {
+      "epoch": 0.0036666666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.013997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.4332,
+      "loss/crossentropy": 1.7330221965909005,
+      "loss/hidden": 3.5515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2054268393665552,
+      "step": 440
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 34.0,
+      "grad_norm_var": 6.847330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3632,
+      "loss/crossentropy": 1.9284149013459682,
+      "loss/hidden": 3.41640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18791294917464257,
+      "step": 450
+    },
+    {
+      "epoch": 0.003833333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.338541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.4044,
+      "loss/crossentropy": 1.9736344426870347,
+      "loss/hidden": 3.544921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20589777827262878,
+      "step": 460
+    },
+    {
+      "epoch": 0.003916666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.7884765625,
+      "learning_rate": 0.0001,
+      "loss": 7.437,
+      "loss/crossentropy": 1.9907344430685043,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21705251093953848,
+      "step": 470
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 29.75,
+      "grad_norm_var": 20.412434895833332,
+      "learning_rate": 0.0001,
+      "loss": 7.2693,
+      "loss/crossentropy": 1.9349011391401292,
+      "loss/hidden": 3.591015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21220431197434664,
+      "step": 480
+    },
+    {
+      "epoch": 0.004083333333333333,
+      "grad_norm": 43.0,
+      "grad_norm_var": 27.826822916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.3501,
+      "loss/crossentropy": 1.980695901811123,
+      "loss/hidden": 3.42265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19862887468189,
+      "step": 490
+    },
+    {
+      "epoch": 0.004166666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 18.60390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3467,
+      "loss/crossentropy": 2.033171671628952,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20085344444960357,
+      "step": 500
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 32.0,
+      "grad_norm_var": 21.980208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2657,
+      "loss/crossentropy": 1.915445376932621,
+      "loss/hidden": 3.414453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20700605846941472,
+      "step": 510
+    },
+    {
+      "epoch": 0.004333333333333333,
+      "grad_norm": 42.5,
+      "grad_norm_var": 22.976822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.413,
+      "loss/crossentropy": 1.9141934983432294,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19435560163110494,
+      "step": 520
+    },
+    {
+      "epoch": 0.004416666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 24.456184895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.2444,
+      "loss/crossentropy": 2.0416490510106087,
+      "loss/hidden": 3.4328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19158751983195543,
+      "step": 530
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 33.25,
+      "grad_norm_var": 18.768489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3152,
+      "loss/crossentropy": 2.004468746483326,
+      "loss/hidden": 3.362890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1849455814808607,
+      "step": 540
+    },
+    {
+      "epoch": 0.004583333333333333,
+      "grad_norm": 38.75,
+      "grad_norm_var": 22.685416666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.363,
+      "loss/crossentropy": 2.100373297929764,
+      "loss/hidden": 3.348828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20310039576143027,
+      "step": 550
+    },
+    {
+      "epoch": 0.004666666666666667,
+      "grad_norm": 65.5,
+      "grad_norm_var": 87.36354166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.4567,
+      "loss/crossentropy": 1.9387385457754136,
+      "loss/hidden": 3.570703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21101371981203557,
+      "step": 560
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 35.0,
+      "grad_norm_var": 68.48899739583334,
+      "learning_rate": 0.0001,
+      "loss": 7.3065,
+      "loss/crossentropy": 1.8911512687802314,
+      "loss/hidden": 3.446484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19550031386315822,
+      "step": 570
+    },
+    {
+      "epoch": 0.004833333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 10.197916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3044,
+      "loss/crossentropy": 2.0782967150211333,
+      "loss/hidden": 3.488671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20857007205486297,
+      "step": 580
+    },
+    {
+      "epoch": 0.004916666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.111393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3021,
+      "loss/crossentropy": 1.9965147748589516,
+      "loss/hidden": 3.33125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19327867813408375,
+      "step": 590
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 32.0,
+      "grad_norm_var": 12.423893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3111,
+      "loss/crossentropy": 1.9416659727692605,
+      "loss/hidden": 3.4671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19690530523657798,
+      "step": 600
+    },
+    {
+      "epoch": 0.005083333333333333,
+      "grad_norm": 35.25,
+      "grad_norm_var": 17.788541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2461,
+      "loss/crossentropy": 1.8854174569249154,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19389428701251746,
+      "step": 610
+    },
+    {
+      "epoch": 0.005166666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 19.061458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3305,
+      "loss/crossentropy": 1.9388531804084779,
+      "loss/hidden": 3.534765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1920976996421814,
+      "step": 620
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 33.25,
+      "grad_norm_var": 15.886393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.3302,
+      "loss/crossentropy": 2.0138145625591277,
+      "loss/hidden": 3.365234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1938948707655072,
+      "step": 630
+    },
+    {
+      "epoch": 0.005333333333333333,
+      "grad_norm": 46.75,
+      "grad_norm_var": 27.868489583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2544,
+      "loss/crossentropy": 2.0624490082263947,
+      "loss/hidden": 3.35703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968511760234833,
+      "step": 640
+    },
+    {
+      "epoch": 0.005416666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 29.6291015625,
+      "learning_rate": 0.0001,
+      "loss": 7.3249,
+      "loss/crossentropy": 1.910299976170063,
+      "loss/hidden": 3.54765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20261993119493127,
+      "step": 650
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 51.5,
+      "grad_norm_var": 43.927083333333336,
+      "learning_rate": 0.0001,
+      "loss": 7.3305,
+      "loss/crossentropy": 1.8762669041752815,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17283952739089728,
+      "step": 660
+    },
+    {
+      "epoch": 0.005583333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 43.02233072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.238,
+      "loss/crossentropy": 1.886153695732355,
+      "loss/hidden": 3.46328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1988078447058797,
+      "step": 670
+    },
+    {
+      "epoch": 0.005666666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 49.3494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3053,
+      "loss/crossentropy": 2.0484144926071166,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20717981327325105,
+      "step": 680
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 39.25,
+      "grad_norm_var": 39.08639322916667,
+      "learning_rate": 0.0001,
+      "loss": 7.299,
+      "loss/crossentropy": 1.8476220905780791,
+      "loss/hidden": 3.475390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946946369484067,
+      "step": 690
+    },
+    {
+      "epoch": 0.005833333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 6.162239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3848,
+      "loss/crossentropy": 1.9601008757948875,
+      "loss/hidden": 3.41328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1941233342513442,
+      "step": 700
+    },
+    {
+      "epoch": 0.005916666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 109.7134765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2428,
+      "loss/crossentropy": 1.945550049096346,
+      "loss/hidden": 3.311328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20260647609829902,
+      "step": 710
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 42.75,
+      "grad_norm_var": 68.07291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3668,
+      "loss/crossentropy": 1.9872938007116319,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2058260628953576,
+      "step": 720
+    },
+    {
+      "epoch": 0.006083333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 35.67858072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2796,
+      "loss/crossentropy": 2.049102196097374,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19005358330905436,
+      "step": 730
+    },
+    {
+      "epoch": 0.006166666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 13.388541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3165,
+      "loss/crossentropy": 2.000891861319542,
+      "loss/hidden": 3.34609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18923768866807222,
+      "step": 740
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.78515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2249,
+      "loss/crossentropy": 1.9180716082453728,
+      "loss/hidden": 3.405078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19066319558769465,
+      "step": 750
+    },
+    {
+      "epoch": 0.006333333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.3947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2432,
+      "loss/crossentropy": 1.9366105154156685,
+      "loss/hidden": 3.309375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816173456609249,
+      "step": 760
+    },
+    {
+      "epoch": 0.006416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.420833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.375,
+      "loss/crossentropy": 1.9630885288119315,
+      "loss/hidden": 3.5234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20541108381003143,
+      "step": 770
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.01875,
+      "learning_rate": 0.0001,
+      "loss": 7.2215,
+      "loss/crossentropy": 1.8655893057584763,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17294289451092482,
+      "step": 780
+    },
+    {
+      "epoch": 0.006583333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 31.75390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3612,
+      "loss/crossentropy": 2.009425923228264,
+      "loss/hidden": 3.426953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19349737111479043,
+      "step": 790
+    },
+    {
+      "epoch": 0.006666666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 256.80358072916664,
+      "learning_rate": 0.0001,
+      "loss": 7.2728,
+      "loss/crossentropy": 1.9748301401734352,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1777269348502159,
+      "step": 800
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.159309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3129,
+      "loss/crossentropy": 2.023061712086201,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18830966539680957,
+      "step": 810
+    },
+    {
+      "epoch": 0.006833333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 13.510872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1579,
+      "loss/crossentropy": 2.0136666923761366,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22291016448289155,
+      "step": 820
+    },
+    {
+      "epoch": 0.0069166666666666664,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.578125,
+      "learning_rate": 0.0001,
+      "loss": 7.3571,
+      "loss/crossentropy": 1.9204814687371254,
+      "loss/hidden": 3.419921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20071878507733346,
+      "step": 830
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 30.0,
+      "grad_norm_var": 12.862239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3315,
+      "loss/crossentropy": 1.894421537220478,
+      "loss/hidden": 3.394140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19227557741105555,
+      "step": 840
+    },
+    {
+      "epoch": 0.007083333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 9.105143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1733,
+      "loss/crossentropy": 1.917093026638031,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18238474596291782,
+      "step": 850
+    },
+    {
+      "epoch": 0.007166666666666667,
+      "grad_norm": 38.5,
+      "grad_norm_var": 9.84375,
+      "learning_rate": 0.0001,
+      "loss": 7.2598,
+      "loss/crossentropy": 1.971535401046276,
+      "loss/hidden": 3.456640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19669556841254235,
+      "step": 860
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.320572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2687,
+      "loss/crossentropy": 2.0588334918022158,
+      "loss/hidden": 3.4078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20278499722480775,
+      "step": 870
+    },
+    {
+      "epoch": 0.007333333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 9.912239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2785,
+      "loss/crossentropy": 2.085460352897644,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2105460435152054,
+      "step": 880
+    },
+    {
+      "epoch": 0.007416666666666667,
+      "grad_norm": 37.75,
+      "grad_norm_var": 22.637434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.3492,
+      "loss/crossentropy": 1.9137649670243264,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21051024934276938,
+      "step": 890
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.320572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2658,
+      "loss/crossentropy": 1.9578298330307007,
+      "loss/hidden": 3.453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18825461007654667,
+      "step": 900
+    },
+    {
+      "epoch": 0.007583333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.7322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.283,
+      "loss/crossentropy": 1.92996806204319,
+      "loss/hidden": 3.50078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107054302468896,
+      "step": 910
+    },
+    {
+      "epoch": 0.007666666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.770572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3248,
+      "loss/crossentropy": 2.0229760840535165,
+      "loss/hidden": 3.50859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2188296400010586,
+      "step": 920
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.871875,
+      "learning_rate": 0.0001,
+      "loss": 7.2833,
+      "loss/crossentropy": 1.9864106953144074,
+      "loss/hidden": 3.520703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2264138963073492,
+      "step": 930
+    },
+    {
+      "epoch": 0.007833333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 6.97265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2909,
+      "loss/crossentropy": 1.871564555168152,
+      "loss/hidden": 3.64296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20268587870523333,
+      "step": 940
+    },
+    {
+      "epoch": 0.007916666666666667,
+      "grad_norm": 36.25,
+      "grad_norm_var": 6.064518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2552,
+      "loss/crossentropy": 1.8941823691129684,
+      "loss/hidden": 3.434375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19341870285570623,
+      "step": 950
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.970768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2977,
+      "loss/crossentropy": 1.9714406847953796,
+      "loss/hidden": 3.262109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801678713411093,
+      "step": 960
+    },
+    {
+      "epoch": 0.008083333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 49.87083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.5097,
+      "loss/crossentropy": 1.8991910368204117,
+      "loss/hidden": 3.495703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19257726836949587,
+      "step": 970
+    },
+    {
+      "epoch": 0.008166666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.0388009758452352e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2699,
+      "loss/crossentropy": 1.9593139976263045,
+      "loss/hidden": 3.4109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18533678650856017,
+      "step": 980
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.038800976104286e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.233,
+      "loss/crossentropy": 1.9886707931756973,
+      "loss/hidden": 3.484765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2002451216802001,
+      "step": 990
+    },
+    {
+      "epoch": 0.008333333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.783072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2139,
+      "loss/crossentropy": 1.982450208067894,
+      "loss/hidden": 3.299609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19080037772655487,
+      "step": 1000
+    },
+    {
+      "epoch": 0.008416666666666666,
+      "grad_norm": 38.75,
+      "grad_norm_var": 7.674739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2816,
+      "loss/crossentropy": 1.8640219122171402,
+      "loss/hidden": 3.4703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18669026363641023,
+      "step": 1010
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 36.5,
+      "grad_norm_var": 6.824934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2353,
+      "loss/crossentropy": 1.8924013078212738,
+      "loss/hidden": 3.405859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1948640437796712,
+      "step": 1020
+    },
+    {
+      "epoch": 0.008583333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.4916666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.4108,
+      "loss/crossentropy": 1.9207781434059144,
+      "loss/hidden": 3.463671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21109477896243334,
+      "step": 1030
+    },
+    {
+      "epoch": 0.008666666666666666,
+      "grad_norm": 39.75,
+      "grad_norm_var": 7.131705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.378,
+      "loss/crossentropy": 1.9148639217019081,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19126542042940856,
+      "step": 1040
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.465625,
+      "learning_rate": 0.0001,
+      "loss": 7.3071,
+      "loss/crossentropy": 1.9637104764580726,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1933687962591648,
+      "step": 1050
+    },
+    {
+      "epoch": 0.008833333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.6837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2891,
+      "loss/crossentropy": 1.898938211798668,
+      "loss/hidden": 3.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20170777514576912,
+      "step": 1060
+    },
+    {
+      "epoch": 0.008916666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 3.20625,
+      "learning_rate": 0.0001,
+      "loss": 7.3074,
+      "loss/crossentropy": 1.875780703127384,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1741238312795758,
+      "step": 1070
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.985872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2715,
+      "loss/crossentropy": 1.8494450084865093,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1831109117716551,
+      "step": 1080
+    },
+    {
+      "epoch": 0.009083333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 29.108072916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.3007,
+      "loss/crossentropy": 1.9538781769573688,
+      "loss/hidden": 3.47890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19717481937259435,
+      "step": 1090
+    },
+    {
+      "epoch": 0.009166666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.83515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2203,
+      "loss/crossentropy": 1.9142641603946686,
+      "loss/hidden": 3.3875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19166908524930476,
+      "step": 1100
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 37.5,
+      "grad_norm_var": 6.734830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.326,
+      "loss/crossentropy": 2.016929093003273,
+      "loss/hidden": 3.394921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18157481551170349,
+      "step": 1110
+    },
+    {
+      "epoch": 0.009333333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.9652184708252966e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4337,
+      "loss/crossentropy": 1.982716566324234,
+      "loss/hidden": 3.396875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18657856304198503,
+      "step": 1120
+    },
+    {
+      "epoch": 0.009416666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.965218471471219e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1922,
+      "loss/crossentropy": 1.8523512050509452,
+      "loss/hidden": 3.466015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19456181898713112,
+      "step": 1130
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.073281661622271e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2354,
+      "loss/crossentropy": 1.9637323051691056,
+      "loss/hidden": 3.360546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18722216449677945,
+      "step": 1140
+    },
+    {
+      "epoch": 0.009583333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.786393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2889,
+      "loss/crossentropy": 1.9170468106865883,
+      "loss/hidden": 3.496875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19212216939777135,
+      "step": 1150
+    },
+    {
+      "epoch": 0.009666666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.0218098958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2555,
+      "loss/crossentropy": 2.1359424561262133,
+      "loss/hidden": 3.3421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815421387553215,
+      "step": 1160
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.94765625,
+      "learning_rate": 0.0001,
+      "loss": 7.277,
+      "loss/crossentropy": 1.827228417992592,
+      "loss/hidden": 3.475390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859055045992136,
+      "step": 1170
+    },
+    {
+      "epoch": 0.009833333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.2119140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2983,
+      "loss/crossentropy": 1.9235713258385658,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20665957890450953,
+      "step": 1180
+    },
+    {
+      "epoch": 0.009916666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.3067057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2995,
+      "loss/crossentropy": 2.06373308300972,
+      "loss/hidden": 3.301953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894452802836895,
+      "step": 1190
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 31.875,
+      "grad_norm_var": 9.71528458302303e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.366,
+      "loss/crossentropy": 1.9264583349227906,
+      "loss/hidden": 3.404296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899071641266346,
+      "step": 1200
+    },
+    {
+      "epoch": 0.010083333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 9.71528457661623e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3836,
+      "loss/crossentropy": 1.7987317427992822,
+      "loss/hidden": 3.514453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20049565955996512,
+      "step": 1210
+    },
+    {
+      "epoch": 0.010166666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2874,
+      "loss/crossentropy": 1.7559522911906242,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18231481197290123,
+      "step": 1220
+    },
+    {
+      "epoch": 0.01025,
+      "grad_norm": 37.75,
+      "grad_norm_var": 4.79765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2641,
+      "loss/crossentropy": 1.9591181308031083,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18080000337213278,
+      "step": 1230
+    },
+    {
+      "epoch": 0.010333333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.617122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2299,
+      "loss/crossentropy": 1.996937209367752,
+      "loss/hidden": 3.291796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17810146119445563,
+      "step": 1240
+    },
+    {
+      "epoch": 0.010416666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.272916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.31,
+      "loss/crossentropy": 1.8335382163524627,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968407742679119,
+      "step": 1250
+    },
+    {
+      "epoch": 0.0105,
+      "grad_norm": 34.25,
+      "grad_norm_var": 17.586393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2393,
+      "loss/crossentropy": 2.059291285276413,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19821931160986422,
+      "step": 1260
+    },
+    {
+      "epoch": 0.010583333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.60390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2797,
+      "loss/crossentropy": 1.9161242499947548,
+      "loss/hidden": 3.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1930157072842121,
+      "step": 1270
+    },
+    {
+      "epoch": 0.010666666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.16015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2758,
+      "loss/crossentropy": 1.946441213786602,
+      "loss/hidden": 3.425,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005026711151004,
+      "step": 1280
+    },
+    {
+      "epoch": 0.01075,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.9330729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2913,
+      "loss/crossentropy": 1.981220117211342,
+      "loss/hidden": 3.437109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20416617318987845,
+      "step": 1290
+    },
+    {
+      "epoch": 0.010833333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.75390625,
+      "learning_rate": 0.0001,
+      "loss": 7.284,
+      "loss/crossentropy": 2.0280979484319688,
+      "loss/hidden": 3.434375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17846004143357277,
+      "step": 1300
+    },
+    {
+      "epoch": 0.010916666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.831705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2516,
+      "loss/crossentropy": 1.9297500923275948,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18526345416903495,
+      "step": 1310
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.120247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.278,
+      "loss/crossentropy": 1.9518141075968742,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2055202412419021,
+      "step": 1320
+    },
+    {
+      "epoch": 0.011083333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 10.774739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2781,
+      "loss/crossentropy": 2.0175456278026105,
+      "loss/hidden": 3.473828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20037419367581605,
+      "step": 1330
+    },
+    {
+      "epoch": 0.011166666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 7.321164001851736e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4105,
+      "loss/crossentropy": 1.9731741398572922,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19264885168522597,
+      "step": 1340
+    },
+    {
+      "epoch": 0.01125,
+      "grad_norm": 34.5,
+      "grad_norm_var": 7.32116400149522e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2418,
+      "loss/crossentropy": 2.0094937115907667,
+      "loss/hidden": 3.330859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1830705888569355,
+      "step": 1350
+    },
+    {
+      "epoch": 0.011333333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 6.511393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2534,
+      "loss/crossentropy": 1.8466748759150504,
+      "loss/hidden": 3.255859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17569661382585763,
+      "step": 1360
+    },
+    {
+      "epoch": 0.011416666666666667,
+      "grad_norm": 27.75,
+      "grad_norm_var": 6.679622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1384,
+      "loss/crossentropy": 1.933292143791914,
+      "loss/hidden": 3.423828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18499917536973953,
+      "step": 1370
+    },
+    {
+      "epoch": 0.0115,
+      "grad_norm": 33.0,
+      "grad_norm_var": 17.3541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.3339,
+      "loss/crossentropy": 1.9056864947080612,
+      "loss/hidden": 3.523828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22226835750043392,
+      "step": 1380
+    },
+    {
+      "epoch": 0.011583333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 12.066666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3259,
+      "loss/crossentropy": 1.971571609377861,
+      "loss/hidden": 3.337109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19764023572206496,
+      "step": 1390
+    },
+    {
+      "epoch": 0.011666666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.26015625,
+      "learning_rate": 0.0001,
+      "loss": 7.3098,
+      "loss/crossentropy": 1.8674658387899399,
+      "loss/hidden": 3.403515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18480249140411614,
+      "step": 1400
+    },
+    {
+      "epoch": 0.01175,
+      "grad_norm": 42.0,
+      "grad_norm_var": 1.3401023452742157e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3637,
+      "loss/crossentropy": 2.1348548874258997,
+      "loss/hidden": 3.63984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23163463342934848,
+      "step": 1410
+    },
+    {
+      "epoch": 0.011833333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.3401023450764544e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2733,
+      "loss/crossentropy": 2.0169719740748406,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19758520759642123,
+      "step": 1420
+    },
+    {
+      "epoch": 0.011916666666666667,
+      "grad_norm": 35.75,
+      "grad_norm_var": 2.9155598958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.256,
+      "loss/crossentropy": 2.0090753018856047,
+      "loss/hidden": 3.27578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18631072752177716,
+      "step": 1430
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.63515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3997,
+      "loss/crossentropy": 2.062183755636215,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17819378785789014,
+      "step": 1440
+    },
+    {
+      "epoch": 0.012083333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.536458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1829,
+      "loss/crossentropy": 1.9344430059194564,
+      "loss/hidden": 3.5171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17964769918471574,
+      "step": 1450
+    },
+    {
+      "epoch": 0.012166666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 18.2009765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3129,
+      "loss/crossentropy": 1.9684132590889931,
+      "loss/hidden": 3.333203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17724689934402704,
+      "step": 1460
+    },
+    {
+      "epoch": 0.01225,
+      "grad_norm": 30.875,
+      "grad_norm_var": 57.805989583333336,
+      "learning_rate": 0.0001,
+      "loss": 7.2262,
+      "loss/crossentropy": 1.9477881044149399,
+      "loss/hidden": 3.336328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18544685766100882,
+      "step": 1470
+    },
+    {
+      "epoch": 0.012333333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.303059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2255,
+      "loss/crossentropy": 2.0414924576878546,
+      "loss/hidden": 3.444140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19923711474984884,
+      "step": 1480
+    },
+    {
+      "epoch": 0.012416666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 37.46555989583333,
+      "learning_rate": 0.0001,
+      "loss": 7.2586,
+      "loss/crossentropy": 1.9072564639151097,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19603695515543224,
+      "step": 1490
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.642708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3078,
+      "loss/crossentropy": 1.9207553781569005,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19159257756546139,
+      "step": 1500
+    },
+    {
+      "epoch": 0.012583333333333334,
+      "grad_norm": 45.0,
+      "grad_norm_var": 12.5,
+      "learning_rate": 0.0001,
+      "loss": 7.2107,
+      "loss/crossentropy": 1.908722074329853,
+      "loss/hidden": 3.325390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19376476481556892,
+      "step": 1510
+    },
+    {
+      "epoch": 0.012666666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 65.01295572916666,
+      "learning_rate": 0.0001,
+      "loss": 7.3764,
+      "loss/crossentropy": 1.9983658343553543,
+      "loss/hidden": 3.370703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18891664631664754,
+      "step": 1520
+    },
+    {
+      "epoch": 0.01275,
+      "grad_norm": 33.5,
+      "grad_norm_var": 45.13645833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2604,
+      "loss/crossentropy": 1.9507597625255584,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20500100627541543,
+      "step": 1530
+    },
+    {
+      "epoch": 0.012833333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 12.14140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3091,
+      "loss/crossentropy": 2.0078801713883876,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18963296040892602,
+      "step": 1540
+    },
+    {
+      "epoch": 0.012916666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.196809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2892,
+      "loss/crossentropy": 1.9844240590929985,
+      "loss/hidden": 3.436328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20112392045557498,
+      "step": 1550
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.860416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2492,
+      "loss/crossentropy": 1.9080413848161697,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1872928399592638,
+      "step": 1560
+    },
+    {
+      "epoch": 0.013083333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 17.445768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.4383,
+      "loss/crossentropy": 2.0557202830910684,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1941663671284914,
+      "step": 1570
+    },
+    {
+      "epoch": 0.013166666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 25.912239583333335,
+      "learning_rate": 0.0001,
+      "loss": 7.3577,
+      "loss/crossentropy": 2.019923450052738,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19055009353905916,
+      "step": 1580
+    },
+    {
+      "epoch": 0.01325,
+      "grad_norm": 32.75,
+      "grad_norm_var": 7.486393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3418,
+      "loss/crossentropy": 1.9912660360336303,
+      "loss/hidden": 3.425,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19967138823121786,
+      "step": 1590
+    },
+    {
+      "epoch": 0.013333333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 13.739322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.329,
+      "loss/crossentropy": 1.976361881196499,
+      "loss/hidden": 3.4578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1988378567621112,
+      "step": 1600
+    },
+    {
+      "epoch": 0.013416666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 8.120768229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.3197,
+      "loss/crossentropy": 1.9507541894912719,
+      "loss/hidden": 3.448046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1952985545620322,
+      "step": 1610
+    },
+    {
+      "epoch": 0.0135,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.485183333490755e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3644,
+      "loss/crossentropy": 2.0464282095432282,
+      "loss/hidden": 3.36953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19348717443645,
+      "step": 1620
+    },
+    {
+      "epoch": 0.013583333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.680143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.233,
+      "loss/crossentropy": 1.9385505408048629,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1836165890097618,
+      "step": 1630
+    },
+    {
+      "epoch": 0.013666666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.2791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1957,
+      "loss/crossentropy": 1.939904248714447,
+      "loss/hidden": 3.254296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178390958532691,
+      "step": 1640
+    },
+    {
+      "epoch": 0.01375,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.909375,
+      "learning_rate": 0.0001,
+      "loss": 7.1277,
+      "loss/crossentropy": 1.8794662043452264,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18380587548017502,
+      "step": 1650
+    },
+    {
+      "epoch": 0.013833333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.2900390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2484,
+      "loss/crossentropy": 1.9310500830411912,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18547237031161784,
+      "step": 1660
+    },
+    {
+      "epoch": 0.013916666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.414322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2795,
+      "loss/crossentropy": 1.9204590037465095,
+      "loss/hidden": 3.41171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19697155160829424,
+      "step": 1670
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.1650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2973,
+      "loss/crossentropy": 2.122091555595398,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18852671552449465,
+      "step": 1680
+    },
+    {
+      "epoch": 0.014083333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.1994140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3361,
+      "loss/crossentropy": 2.156261298060417,
+      "loss/hidden": 3.516015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20998501256108285,
+      "step": 1690
+    },
+    {
+      "epoch": 0.014166666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.687955729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2512,
+      "loss/crossentropy": 1.9670209422707559,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20212535988539457,
+      "step": 1700
+    },
+    {
+      "epoch": 0.01425,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.2947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.3058,
+      "loss/crossentropy": 2.0846331417560577,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20701619572937488,
+      "step": 1710
+    },
+    {
+      "epoch": 0.014333333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.5455729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.3898,
+      "loss/crossentropy": 2.029135060310364,
+      "loss/hidden": 3.43125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19423509035259484,
+      "step": 1720
+    },
+    {
+      "epoch": 0.014416666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 25.8306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2523,
+      "loss/crossentropy": 1.8986969135701657,
+      "loss/hidden": 3.3953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18308270815759897,
+      "step": 1730
+    },
+    {
+      "epoch": 0.0145,
+      "grad_norm": 35.5,
+      "grad_norm_var": 8.847330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.385,
+      "loss/crossentropy": 1.9802353143692017,
+      "loss/hidden": 3.454296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20486843958497047,
+      "step": 1740
+    },
+    {
+      "epoch": 0.014583333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.5166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2707,
+      "loss/crossentropy": 1.8174491569399833,
+      "loss/hidden": 3.43125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18351891580969096,
+      "step": 1750
+    },
+    {
+      "epoch": 0.014666666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.098893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.279,
+      "loss/crossentropy": 2.161423406004906,
+      "loss/hidden": 3.280859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18401132225990297,
+      "step": 1760
+    },
+    {
+      "epoch": 0.01475,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.0384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2269,
+      "loss/crossentropy": 1.8804115176200866,
+      "loss/hidden": 3.401171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18950734920799733,
+      "step": 1770
+    },
+    {
+      "epoch": 0.014833333333333334,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.3917132166325434e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.321,
+      "loss/crossentropy": 1.9431498855352403,
+      "loss/hidden": 3.61484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20938508119434118,
+      "step": 1780
+    },
+    {
+      "epoch": 0.014916666666666667,
+      "grad_norm": 48.0,
+      "grad_norm_var": 4.3917132079346054e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2738,
+      "loss/crossentropy": 1.991785955429077,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19057564698159696,
+      "step": 1790
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 30.875,
+      "grad_norm_var": 32.218684895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.3331,
+      "loss/crossentropy": 1.9663019806146622,
+      "loss/hidden": 3.361328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18586504831910133,
+      "step": 1800
+    },
+    {
+      "epoch": 0.015083333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.289322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2361,
+      "loss/crossentropy": 1.9916753873229027,
+      "loss/hidden": 3.4328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20035373903810977,
+      "step": 1810
+    },
+    {
+      "epoch": 0.015166666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.5205729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1769,
+      "loss/crossentropy": 2.0380251288414,
+      "loss/hidden": 3.321875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18559953831136228,
+      "step": 1820
+    },
+    {
+      "epoch": 0.01525,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.2384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.318,
+      "loss/crossentropy": 1.9261852458119393,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18253942150622607,
+      "step": 1830
+    },
+    {
+      "epoch": 0.015333333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.781705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2674,
+      "loss/crossentropy": 1.8436972185969354,
+      "loss/hidden": 3.3375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18778917007148266,
+      "step": 1840
+    },
+    {
+      "epoch": 0.015416666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.9494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3212,
+      "loss/crossentropy": 2.020959322154522,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19789048228412867,
+      "step": 1850
+    },
+    {
+      "epoch": 0.0155,
+      "grad_norm": 37.25,
+      "grad_norm_var": 31.046809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2136,
+      "loss/crossentropy": 1.8541840478777885,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17573605831712485,
+      "step": 1860
+    },
+    {
+      "epoch": 0.015583333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 32.603580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2691,
+      "loss/crossentropy": 1.9017258927226066,
+      "loss/hidden": 3.344140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17979437448084354,
+      "step": 1870
+    },
+    {
+      "epoch": 0.015666666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.289518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2647,
+      "loss/crossentropy": 1.9352346003055572,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19231825936585664,
+      "step": 1880
+    },
+    {
+      "epoch": 0.01575,
+      "grad_norm": 35.75,
+      "grad_norm_var": 1.6906090573603364e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.4439,
+      "loss/crossentropy": 1.943103864789009,
+      "loss/hidden": 3.44765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879008635878563,
+      "step": 1890
+    },
+    {
+      "epoch": 0.015833333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.6906090571002895e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3468,
+      "loss/crossentropy": 1.9841052293777466,
+      "loss/hidden": 3.466796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19862290378659964,
+      "step": 1900
+    },
+    {
+      "epoch": 0.015916666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 295.4098307291667,
+      "learning_rate": 0.0001,
+      "loss": 7.4047,
+      "loss/crossentropy": 1.9083608210086822,
+      "loss/hidden": 3.30625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18380118031054735,
+      "step": 1910
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 30.25,
+      "grad_norm_var": 342.01458333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.3901,
+      "loss/crossentropy": 2.0147466957569122,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1957575250416994,
+      "step": 1920
+    },
+    {
+      "epoch": 0.016083333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.322916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3132,
+      "loss/crossentropy": 1.8240422800183296,
+      "loss/hidden": 3.43515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18423110404983162,
+      "step": 1930
+    },
+    {
+      "epoch": 0.016166666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 21.651822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2571,
+      "loss/crossentropy": 1.9640603423118592,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19020091257989408,
+      "step": 1940
+    },
+    {
+      "epoch": 0.01625,
+      "grad_norm": 38.75,
+      "grad_norm_var": 23.867122395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2404,
+      "loss/crossentropy": 1.7818415738642215,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17666226997971535,
+      "step": 1950
+    },
+    {
+      "epoch": 0.01633333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.885416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.34,
+      "loss/crossentropy": 1.9440691128373147,
+      "loss/hidden": 3.605859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21677861418575048,
+      "step": 1960
+    },
+    {
+      "epoch": 0.016416666666666666,
+      "grad_norm": 36.75,
+      "grad_norm_var": 5.0841416521866726e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3199,
+      "loss/crossentropy": 1.976449340581894,
+      "loss/hidden": 3.5921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2107225891202688,
+      "step": 1970
+    },
+    {
+      "epoch": 0.0165,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.084141640391941e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.389,
+      "loss/crossentropy": 2.0012796193361284,
+      "loss/hidden": 3.39140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19476055447012186,
+      "step": 1980
+    },
+    {
+      "epoch": 0.016583333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 69.82805989583333,
+      "learning_rate": 0.0001,
+      "loss": 7.2807,
+      "loss/crossentropy": 2.057659167051315,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19637791439890862,
+      "step": 1990
+    },
+    {
+      "epoch": 0.016666666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.4427083333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2848,
+      "loss/crossentropy": 2.1016288846731186,
+      "loss/hidden": 3.500390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19531550351530313,
+      "step": 2000
+    },
+    {
+      "epoch": 0.01675,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.6884765625,
+      "learning_rate": 0.0001,
+      "loss": 7.153,
+      "loss/crossentropy": 1.9537018239498138,
+      "loss/hidden": 3.43203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18602741174399853,
+      "step": 2010
+    },
+    {
+      "epoch": 0.016833333333333332,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.06015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2727,
+      "loss/crossentropy": 1.9659764975309373,
+      "loss/hidden": 3.3953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1949648916721344,
+      "step": 2020
+    },
+    {
+      "epoch": 0.016916666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.5697265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2785,
+      "loss/crossentropy": 1.8973619878292083,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18199929129332304,
+      "step": 2030
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 44.25,
+      "grad_norm_var": 14.552018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2118,
+      "loss/crossentropy": 1.9548767402768135,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19267449919134377,
+      "step": 2040
+    },
+    {
+      "epoch": 0.017083333333333332,
+      "grad_norm": 30.625,
+      "grad_norm_var": 12.8181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.3086,
+      "loss/crossentropy": 1.9166896522045136,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18487014006823302,
+      "step": 2050
+    },
+    {
+      "epoch": 0.017166666666666667,
+      "grad_norm": 37.5,
+      "grad_norm_var": 7.750455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2843,
+      "loss/crossentropy": 1.8707634150981902,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18515090448781849,
+      "step": 2060
+    },
+    {
+      "epoch": 0.01725,
+      "grad_norm": 38.75,
+      "grad_norm_var": 9.956184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2942,
+      "loss/crossentropy": 1.8519902683794498,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1862996906042099,
+      "step": 2070
+    },
+    {
+      "epoch": 0.017333333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.755989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.246,
+      "loss/crossentropy": 1.939876636862755,
+      "loss/hidden": 3.439453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1866559438407421,
+      "step": 2080
+    },
+    {
+      "epoch": 0.017416666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 11.2634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.4043,
+      "loss/crossentropy": 1.95171078145504,
+      "loss/hidden": 3.436328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19473332334309817,
+      "step": 2090
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 33.25,
+      "grad_norm_var": 8.5150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1663,
+      "loss/crossentropy": 1.997160328924656,
+      "loss/hidden": 3.2921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17636931519955396,
+      "step": 2100
+    },
+    {
+      "epoch": 0.017583333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.155989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2116,
+      "loss/crossentropy": 1.9414543956518173,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18860510028898717,
+      "step": 2110
+    },
+    {
+      "epoch": 0.017666666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.614322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.196,
+      "loss/crossentropy": 2.0807690426707266,
+      "loss/hidden": 3.3625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18818104062229396,
+      "step": 2120
+    },
+    {
+      "epoch": 0.01775,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.62265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2894,
+      "loss/crossentropy": 1.9585611373186111,
+      "loss/hidden": 3.430078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19511018712073563,
+      "step": 2130
+    },
+    {
+      "epoch": 0.017833333333333333,
+      "grad_norm": 38.5,
+      "grad_norm_var": 11.9634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.276,
+      "loss/crossentropy": 1.942204162478447,
+      "loss/hidden": 3.32265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18475140817463398,
+      "step": 2140
+    },
+    {
+      "epoch": 0.017916666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.11875,
+      "learning_rate": 0.0001,
+      "loss": 7.2854,
+      "loss/crossentropy": 1.9810814321041108,
+      "loss/hidden": 3.429296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18507389035075902,
+      "step": 2150
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.000455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2508,
+      "loss/crossentropy": 2.141017886996269,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19872815869748592,
+      "step": 2160
+    },
+    {
+      "epoch": 0.018083333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.131184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3377,
+      "loss/crossentropy": 1.9983078092336655,
+      "loss/hidden": 3.45390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18931595589965583,
+      "step": 2170
+    },
+    {
+      "epoch": 0.018166666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.385416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3045,
+      "loss/crossentropy": 1.9222343221306801,
+      "loss/hidden": 3.433203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17565734032541513,
+      "step": 2180
+    },
+    {
+      "epoch": 0.01825,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.27890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2599,
+      "loss/crossentropy": 2.0131379783153536,
+      "loss/hidden": 3.276953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18543861098587514,
+      "step": 2190
+    },
+    {
+      "epoch": 0.018333333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.69765625,
+      "learning_rate": 0.0001,
+      "loss": 7.185,
+      "loss/crossentropy": 1.8784870684146882,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19702841751277447,
+      "step": 2200
+    },
+    {
+      "epoch": 0.018416666666666668,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.620995007756109e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3567,
+      "loss/crossentropy": 1.9045794017612934,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889135385863483,
+      "step": 2210
+    },
+    {
+      "epoch": 0.0185,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.620995005009888e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2957,
+      "loss/crossentropy": 1.7774473994970321,
+      "loss/hidden": 3.346484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16910541616380215,
+      "step": 2220
+    },
+    {
+      "epoch": 0.018583333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 13.5150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3112,
+      "loss/crossentropy": 2.0893754065036774,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19306899029761554,
+      "step": 2230
+    },
+    {
+      "epoch": 0.018666666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.10390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3108,
+      "loss/crossentropy": 1.9501435413956643,
+      "loss/hidden": 3.2921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1907532515004277,
+      "step": 2240
+    },
+    {
+      "epoch": 0.01875,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.5739081064856954e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3127,
+      "loss/crossentropy": 1.936707255989313,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19168335273861886,
+      "step": 2250
+    },
+    {
+      "epoch": 0.018833333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.310872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2447,
+      "loss/crossentropy": 2.0652841836214066,
+      "loss/hidden": 3.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22945739645510912,
+      "step": 2260
+    },
+    {
+      "epoch": 0.018916666666666665,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.591666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2527,
+      "loss/crossentropy": 1.8847783967852592,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17899461332708597,
+      "step": 2270
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.779622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2593,
+      "loss/crossentropy": 1.8318727254867553,
+      "loss/hidden": 3.388671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841424321755767,
+      "step": 2280
+    },
+    {
+      "epoch": 0.019083333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.3056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2545,
+      "loss/crossentropy": 1.9013791278004646,
+      "loss/hidden": 3.4734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20086494665592908,
+      "step": 2290
+    },
+    {
+      "epoch": 0.019166666666666665,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.9077473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3744,
+      "loss/crossentropy": 1.7958017572760583,
+      "loss/hidden": 3.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2134784586727619,
+      "step": 2300
+    },
+    {
+      "epoch": 0.01925,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.1228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3052,
+      "loss/crossentropy": 2.032643774151802,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19537682719528676,
+      "step": 2310
+    },
+    {
+      "epoch": 0.019333333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.6864583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2865,
+      "loss/crossentropy": 1.9596822246909142,
+      "loss/hidden": 3.372265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18629487175494433,
+      "step": 2320
+    },
+    {
+      "epoch": 0.019416666666666665,
+      "grad_norm": 36.25,
+      "grad_norm_var": 4.873372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3174,
+      "loss/crossentropy": 2.0085081800818445,
+      "loss/hidden": 3.308203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19248725334182382,
+      "step": 2330
+    },
+    {
+      "epoch": 0.0195,
+      "grad_norm": 36.75,
+      "grad_norm_var": 6.5744140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3592,
+      "loss/crossentropy": 1.9844809271395207,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18588697481900454,
+      "step": 2340
+    },
+    {
+      "epoch": 0.019583333333333335,
+      "grad_norm": 38.5,
+      "grad_norm_var": 1.2824703428670262e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3201,
+      "loss/crossentropy": 1.9626279659569263,
+      "loss/hidden": 3.3796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18211031127721072,
+      "step": 2350
+    },
+    {
+      "epoch": 0.019666666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.282470342291358e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3481,
+      "loss/crossentropy": 2.0168631345033647,
+      "loss/hidden": 3.423046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20021111145615578,
+      "step": 2360
+    },
+    {
+      "epoch": 0.01975,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.151811406201619e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2642,
+      "loss/crossentropy": 2.0320428051054478,
+      "loss/hidden": 3.651171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17399079585447907,
+      "step": 2370
+    },
+    {
+      "epoch": 0.019833333333333335,
+      "grad_norm": 28.125,
+      "grad_norm_var": 7.537955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1327,
+      "loss/crossentropy": 2.011017915606499,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19383390806615353,
+      "step": 2380
+    },
+    {
+      "epoch": 0.019916666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 8.124934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3095,
+      "loss/crossentropy": 1.9657955452799798,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18745789155364037,
+      "step": 2390
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 34.75,
+      "grad_norm_var": 6.0822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2852,
+      "loss/crossentropy": 1.9604591697454452,
+      "loss/hidden": 3.470703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20235047750175,
+      "step": 2400
+    },
+    {
+      "epoch": 0.020083333333333335,
+      "grad_norm": 32.75,
+      "grad_norm_var": 28.183072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2632,
+      "loss/crossentropy": 1.944566234201193,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18141100630164148,
+      "step": 2410
+    },
+    {
+      "epoch": 0.020166666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 30.720572916666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2207,
+      "loss/crossentropy": 1.992261415719986,
+      "loss/hidden": 3.365234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20662249848246575,
+      "step": 2420
+    },
+    {
+      "epoch": 0.02025,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.2684895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3458,
+      "loss/crossentropy": 1.891785178333521,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19321308471262455,
+      "step": 2430
+    },
+    {
+      "epoch": 0.02033333333333333,
+      "grad_norm": 4664066048.0,
+      "grad_norm_var": 1.3595944877801428e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1966,
+      "loss/crossentropy": 1.934097670018673,
+      "loss/hidden": 3.3375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1767245288938284,
+      "step": 2440
+    },
+    {
+      "epoch": 0.020416666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.3595944861768701e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3028,
+      "loss/crossentropy": 1.9819417625665665,
+      "loss/hidden": 3.3421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1829627426341176,
+      "step": 2450
+    },
+    {
+      "epoch": 0.0205,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.872330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1355,
+      "loss/crossentropy": 1.9402824997901917,
+      "loss/hidden": 3.478125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19555577524006368,
+      "step": 2460
+    },
+    {
+      "epoch": 0.020583333333333332,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.3962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1806,
+      "loss/crossentropy": 1.8212218895554542,
+      "loss/hidden": 3.303125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18379593417048454,
+      "step": 2470
+    },
+    {
+      "epoch": 0.020666666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.39140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3467,
+      "loss/crossentropy": 1.9318528160452844,
+      "loss/hidden": 3.475390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22001873292028903,
+      "step": 2480
+    },
+    {
+      "epoch": 0.02075,
+      "grad_norm": 48.75,
+      "grad_norm_var": 23.338541666666668,
+      "learning_rate": 0.0001,
+      "loss": 7.26,
+      "loss/crossentropy": 1.7162697166204453,
+      "loss/hidden": 3.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2028669610619545,
+      "step": 2490
+    },
+    {
+      "epoch": 0.020833333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 25.046875,
+      "learning_rate": 0.0001,
+      "loss": 7.0735,
+      "loss/crossentropy": 1.9614305034279824,
+      "loss/hidden": 3.397265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1880598498508334,
+      "step": 2500
+    },
+    {
+      "epoch": 0.020916666666666667,
+      "grad_norm": 48.25,
+      "grad_norm_var": 7.393116035913265e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3996,
+      "loss/crossentropy": 2.056662857532501,
+      "loss/hidden": 3.4109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19303689561784268,
+      "step": 2510
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.393116032760547e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3348,
+      "loss/crossentropy": 2.052212342619896,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18852751776576043,
+      "step": 2520
+    },
+    {
+      "epoch": 0.021083333333333332,
+      "grad_norm": 30.0,
+      "grad_norm_var": 9.699739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2244,
+      "loss/crossentropy": 1.9503659322857856,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18111072592437266,
+      "step": 2530
+    },
+    {
+      "epoch": 0.021166666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.4875,
+      "learning_rate": 0.0001,
+      "loss": 7.1369,
+      "loss/crossentropy": 1.96889967918396,
+      "loss/hidden": 3.255859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18848092965781688,
+      "step": 2540
+    },
+    {
+      "epoch": 0.02125,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.251497395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2827,
+      "loss/crossentropy": 1.9892879366874694,
+      "loss/hidden": 3.458203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1890705717727542,
+      "step": 2550
+    },
+    {
+      "epoch": 0.021333333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.1483723958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3539,
+      "loss/crossentropy": 1.9124898374080659,
+      "loss/hidden": 3.475390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17518300358206035,
+      "step": 2560
+    },
+    {
+      "epoch": 0.021416666666666667,
+      "grad_norm": 36.25,
+      "grad_norm_var": 2.5650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2363,
+      "loss/crossentropy": 1.976254241168499,
+      "loss/hidden": 3.612109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2144654095172882,
+      "step": 2570
+    },
+    {
+      "epoch": 0.0215,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.50625,
+      "learning_rate": 0.0001,
+      "loss": 7.3128,
+      "loss/crossentropy": 1.9184046685695648,
+      "loss/hidden": 3.469140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21248191315680742,
+      "step": 2580
+    },
+    {
+      "epoch": 0.021583333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3331.639583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.298,
+      "loss/crossentropy": 1.9172766536474228,
+      "loss/hidden": 3.43828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916635401546955,
+      "step": 2590
+    },
+    {
+      "epoch": 0.021666666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3321.0608723958335,
+      "learning_rate": 0.0001,
+      "loss": 7.2846,
+      "loss/crossentropy": 1.9668544963002206,
+      "loss/hidden": 3.433984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1907737607136369,
+      "step": 2600
+    },
+    {
+      "epoch": 0.02175,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.981398732307433e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4946,
+      "loss/crossentropy": 1.9127167880535125,
+      "loss/hidden": 3.434375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960796787403524,
+      "step": 2610
+    },
+    {
+      "epoch": 0.021833333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 7.98139873078123e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3146,
+      "loss/crossentropy": 1.996176090836525,
+      "loss/hidden": 3.433984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1958073414862156,
+      "step": 2620
+    },
+    {
+      "epoch": 0.021916666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.320572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3888,
+      "loss/crossentropy": 2.0682441249489782,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959614047780633,
+      "step": 2630
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.614322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3234,
+      "loss/crossentropy": 2.0070124730467795,
+      "loss/hidden": 3.4625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18341080471873283,
+      "step": 2640
+    },
+    {
+      "epoch": 0.022083333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.0249348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2761,
+      "loss/crossentropy": 1.967797513306141,
+      "loss/hidden": 3.372265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19041062686592342,
+      "step": 2650
+    },
+    {
+      "epoch": 0.022166666666666668,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.9171223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2174,
+      "loss/crossentropy": 1.974215391278267,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17478577867150308,
+      "step": 2660
+    },
+    {
+      "epoch": 0.02225,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.5660807291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2954,
+      "loss/crossentropy": 1.7671419337391854,
+      "loss/hidden": 3.617578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18350028842687607,
+      "step": 2670
+    },
+    {
+      "epoch": 0.022333333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 3.29765625,
+      "learning_rate": 0.0001,
+      "loss": 7.352,
+      "loss/crossentropy": 1.9952654451131822,
+      "loss/hidden": 3.455859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19615703728049994,
+      "step": 2680
+    },
+    {
+      "epoch": 0.022416666666666668,
+      "grad_norm": 36.25,
+      "grad_norm_var": 6.985416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.277,
+      "loss/crossentropy": 2.0219733342528343,
+      "loss/hidden": 3.354296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18562499135732652,
+      "step": 2690
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 41.0,
+      "grad_norm_var": 9.537239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3328,
+      "loss/crossentropy": 2.0624974846839903,
+      "loss/hidden": 3.502734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.209426661580801,
+      "step": 2700
+    },
+    {
+      "epoch": 0.022583333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.608268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2719,
+      "loss/crossentropy": 1.9747545182704926,
+      "loss/hidden": 3.336328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18756564743816853,
+      "step": 2710
+    },
+    {
+      "epoch": 0.02266666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.6083333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3397,
+      "loss/crossentropy": 2.012811505794525,
+      "loss/hidden": 3.483203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20681646689772606,
+      "step": 2720
+    },
+    {
+      "epoch": 0.02275,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.201041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2686,
+      "loss/crossentropy": 1.9331883117556572,
+      "loss/hidden": 3.463671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20859357807785273,
+      "step": 2730
+    },
+    {
+      "epoch": 0.022833333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.1176432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2957,
+      "loss/crossentropy": 1.926119977235794,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20198929514735936,
+      "step": 2740
+    },
+    {
+      "epoch": 0.022916666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.3317057291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2201,
+      "loss/crossentropy": 2.0820239409804344,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20356852617114782,
+      "step": 2750
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.29375,
+      "learning_rate": 0.0001,
+      "loss": 7.1686,
+      "loss/crossentropy": 1.9975199311971665,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18412500955164432,
+      "step": 2760
+    },
+    {
+      "epoch": 0.023083333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.133072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2266,
+      "loss/crossentropy": 2.1272772580385206,
+      "loss/hidden": 3.491015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22087238356471062,
+      "step": 2770
+    },
+    {
+      "epoch": 0.023166666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.6455729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2456,
+      "loss/crossentropy": 1.8811777606606483,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19566608127206564,
+      "step": 2780
+    },
+    {
+      "epoch": 0.02325,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.508072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2141,
+      "loss/crossentropy": 1.9168774321675301,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17958773095160724,
+      "step": 2790
+    },
+    {
+      "epoch": 0.023333333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.005143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2308,
+      "loss/crossentropy": 1.9714482218027114,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1880871295928955,
+      "step": 2800
+    },
+    {
+      "epoch": 0.023416666666666665,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.439583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1113,
+      "loss/crossentropy": 1.987887953221798,
+      "loss/hidden": 3.38203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19080262612551452,
+      "step": 2810
+    },
+    {
+      "epoch": 0.0235,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.260416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.345,
+      "loss/crossentropy": 1.948936927318573,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18754115160554646,
+      "step": 2820
+    },
+    {
+      "epoch": 0.023583333333333335,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.3934895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2216,
+      "loss/crossentropy": 1.8182113721966744,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18257561102509498,
+      "step": 2830
+    },
+    {
+      "epoch": 0.023666666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.7332682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3073,
+      "loss/crossentropy": 2.11132807135582,
+      "loss/hidden": 3.462109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1939231114462018,
+      "step": 2840
+    },
+    {
+      "epoch": 0.02375,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.3650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2723,
+      "loss/crossentropy": 1.9388402462005616,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19834725223481656,
+      "step": 2850
+    },
+    {
+      "epoch": 0.023833333333333335,
+      "grad_norm": 39.75,
+      "grad_norm_var": 6.1634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3178,
+      "loss/crossentropy": 1.9689893007278443,
+      "loss/hidden": 3.360546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19809619728475808,
+      "step": 2860
+    },
+    {
+      "epoch": 0.023916666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 7.628059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3359,
+      "loss/crossentropy": 2.027157208323479,
+      "loss/hidden": 3.39765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937933772802353,
+      "step": 2870
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 64.0,
+      "grad_norm_var": 70.29583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2869,
+      "loss/crossentropy": 1.9325517199933528,
+      "loss/hidden": 3.315234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18406210988759994,
+      "step": 2880
+    },
+    {
+      "epoch": 0.024083333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 66.80390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3758,
+      "loss/crossentropy": 1.9735887706279756,
+      "loss/hidden": 3.495703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1949346676468849,
+      "step": 2890
+    },
+    {
+      "epoch": 0.024166666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.49375,
+      "learning_rate": 0.0001,
+      "loss": 7.3159,
+      "loss/crossentropy": 2.036432480812073,
+      "loss/hidden": 3.4328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2019389744848013,
+      "step": 2900
+    },
+    {
+      "epoch": 0.02425,
+      "grad_norm": 44.5,
+      "grad_norm_var": 19.467122395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1925,
+      "loss/crossentropy": 2.054004454612732,
+      "loss/hidden": 3.24765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1765751089900732,
+      "step": 2910
+    },
+    {
+      "epoch": 0.024333333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 16.184830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2403,
+      "loss/crossentropy": 1.9157620191574096,
+      "loss/hidden": 3.461328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18368239123374225,
+      "step": 2920
+    },
+    {
+      "epoch": 0.024416666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 32.4791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2583,
+      "loss/crossentropy": 2.1513744235038756,
+      "loss/hidden": 3.2515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182504341006279,
+      "step": 2930
+    },
+    {
+      "epoch": 0.0245,
+      "grad_norm": 31.875,
+      "grad_norm_var": 27.1947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2578,
+      "loss/crossentropy": 1.9272947192192078,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19014938455075026,
+      "step": 2940
+    },
+    {
+      "epoch": 0.024583333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 19.096809895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.2335,
+      "loss/crossentropy": 2.084392386674881,
+      "loss/hidden": 3.483984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20528450179845095,
+      "step": 2950
+    },
+    {
+      "epoch": 0.024666666666666667,
+      "grad_norm": 35.5,
+      "grad_norm_var": 20.470833333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2383,
+      "loss/crossentropy": 1.9980163618922233,
+      "loss/hidden": 3.508203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19989981167018414,
+      "step": 2960
+    },
+    {
+      "epoch": 0.02475,
+      "grad_norm": 40.0,
+      "grad_norm_var": 17.977018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3249,
+      "loss/crossentropy": 1.9498015969991684,
+      "loss/hidden": 3.4703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932325528934598,
+      "step": 2970
+    },
+    {
+      "epoch": 0.024833333333333332,
+      "grad_norm": 4764729344.0,
+      "grad_norm_var": 1.4189153377602678e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3108,
+      "loss/crossentropy": 1.9112057030200957,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198336130566895,
+      "step": 2980
+    },
+    {
+      "epoch": 0.024916666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.4189153376361864e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2281,
+      "loss/crossentropy": 1.911822970956564,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17767341239377857,
+      "step": 2990
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 36.0,
+      "grad_norm_var": 7.55,
+      "learning_rate": 0.0001,
+      "loss": 7.1727,
+      "loss/crossentropy": 2.0090511620044706,
+      "loss/hidden": 3.30390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18732633236795665,
+      "step": 3000
+    },
+    {
+      "epoch": 0.025083333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.983268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2384,
+      "loss/crossentropy": 1.8529448568820954,
+      "loss/hidden": 3.457421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17940086945891381,
+      "step": 3010
+    },
+    {
+      "epoch": 0.025166666666666667,
+      "grad_norm": 36.5,
+      "grad_norm_var": 8.026822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.191,
+      "loss/crossentropy": 1.8407732352614403,
+      "loss/hidden": 3.431640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18361600376665593,
+      "step": 3020
+    },
+    {
+      "epoch": 0.02525,
+      "grad_norm": 30.625,
+      "grad_norm_var": 6.8150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2616,
+      "loss/crossentropy": 1.9298807948827743,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18290845435112715,
+      "step": 3030
+    },
+    {
+      "epoch": 0.025333333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.673372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3267,
+      "loss/crossentropy": 1.8523324035108089,
+      "loss/hidden": 3.45234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20208420231938362,
+      "step": 3040
+    },
+    {
+      "epoch": 0.025416666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.653125,
+      "learning_rate": 0.0001,
+      "loss": 7.1421,
+      "loss/crossentropy": 1.8332019917666913,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813520684838295,
+      "step": 3050
+    },
+    {
+      "epoch": 0.0255,
+      "grad_norm": 29.375,
+      "grad_norm_var": 8.258072916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3024,
+      "loss/crossentropy": 2.017191821336746,
+      "loss/hidden": 3.47265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20145763643085957,
+      "step": 3060
+    },
+    {
+      "epoch": 0.025583333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.99140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2919,
+      "loss/crossentropy": 2.0802741676568983,
+      "loss/hidden": 3.2875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20051774512976409,
+      "step": 3070
+    },
+    {
+      "epoch": 0.025666666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 6.896840476364637e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3932,
+      "loss/crossentropy": 1.917404356598854,
+      "loss/hidden": 3.442578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19901793487370015,
+      "step": 3080
+    },
+    {
+      "epoch": 0.02575,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.896840478406214e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2225,
+      "loss/crossentropy": 1.8461115274578332,
+      "loss/hidden": 3.4578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18547605257481337,
+      "step": 3090
+    },
+    {
+      "epoch": 0.025833333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 16.2197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1225,
+      "loss/crossentropy": 1.929587610065937,
+      "loss/hidden": 3.309765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1925740350037813,
+      "step": 3100
+    },
+    {
+      "epoch": 0.025916666666666668,
+      "grad_norm": 37.25,
+      "grad_norm_var": 17.442122395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1508,
+      "loss/crossentropy": 1.9709638714790345,
+      "loss/hidden": 3.34140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18460453823208808,
+      "step": 3110
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.7931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1906,
+      "loss/crossentropy": 2.0181252285838127,
+      "loss/hidden": 3.4359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18887849394232034,
+      "step": 3120
+    },
+    {
+      "epoch": 0.026083333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 12.138541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3697,
+      "loss/crossentropy": 2.027159887552261,
+      "loss/hidden": 3.438671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20400238893926143,
+      "step": 3130
+    },
+    {
+      "epoch": 0.026166666666666668,
+      "grad_norm": 29.375,
+      "grad_norm_var": 10.220768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1989,
+      "loss/crossentropy": 2.016098701953888,
+      "loss/hidden": 3.330859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17855217922478914,
+      "step": 3140
+    },
+    {
+      "epoch": 0.02625,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.47890625,
+      "learning_rate": 0.0001,
+      "loss": 7.3656,
+      "loss/crossentropy": 1.9397494405508042,
+      "loss/hidden": 3.625390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2223890544846654,
+      "step": 3150
+    },
+    {
+      "epoch": 0.026333333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3563.3494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3418,
+      "loss/crossentropy": 1.8907858923077583,
+      "loss/hidden": 3.4125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18646195530891418,
+      "step": 3160
+    },
+    {
+      "epoch": 0.026416666666666668,
+      "grad_norm": 34.0,
+      "grad_norm_var": 7.910872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2217,
+      "loss/crossentropy": 1.9405520528554916,
+      "loss/hidden": 3.324609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.179235702380538,
+      "step": 3170
+    },
+    {
+      "epoch": 0.0265,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.532291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2206,
+      "loss/crossentropy": 2.064801651239395,
+      "loss/hidden": 3.234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18341027162969112,
+      "step": 3180
+    },
+    {
+      "epoch": 0.026583333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.162434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2847,
+      "loss/crossentropy": 2.023468551039696,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19835583344101906,
+      "step": 3190
+    },
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.3574,
+      "loss/crossentropy": 1.9950867831707,
+      "loss/hidden": 3.32578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19693203810602428,
+      "step": 3200
+    },
+    {
+      "epoch": 0.02675,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.21015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1773,
+      "loss/crossentropy": 1.859079574048519,
+      "loss/hidden": 3.353515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18082537315785885,
+      "step": 3210
+    },
+    {
+      "epoch": 0.026833333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.4770833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2378,
+      "loss/crossentropy": 2.074732157588005,
+      "loss/hidden": 3.315234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19291758965700864,
+      "step": 3220
+    },
+    {
+      "epoch": 0.026916666666666665,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.3343098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1845,
+      "loss/crossentropy": 1.9518444910645485,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21211670562624932,
+      "step": 3230
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9811848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2353,
+      "loss/crossentropy": 1.908726019412279,
+      "loss/hidden": 3.40546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893859247211367,
+      "step": 3240
+    },
+    {
+      "epoch": 0.027083333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.86640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2767,
+      "loss/crossentropy": 1.83207483664155,
+      "loss/hidden": 3.50703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1776171642355621,
+      "step": 3250
+    },
+    {
+      "epoch": 0.027166666666666665,
+      "grad_norm": 29.25,
+      "grad_norm_var": 15.318489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2133,
+      "loss/crossentropy": 2.0540854394435883,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19790230877697468,
+      "step": 3260
+    },
+    {
+      "epoch": 0.02725,
+      "grad_norm": 29.75,
+      "grad_norm_var": 13.468684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2909,
+      "loss/crossentropy": 2.1424532026052474,
+      "loss/hidden": 3.409765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20493081752210857,
+      "step": 3270
+    },
+    {
+      "epoch": 0.027333333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 8.204166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2174,
+      "loss/crossentropy": 1.8585446879267693,
+      "loss/hidden": 3.417578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19328337758779526,
+      "step": 3280
+    },
+    {
+      "epoch": 0.027416666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.145121829761844e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2046,
+      "loss/crossentropy": 1.935442891716957,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17625004909932612,
+      "step": 3290
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.680208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2284,
+      "loss/crossentropy": 1.925745696015656,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18133268980309367,
+      "step": 3300
+    },
+    {
+      "epoch": 0.027583333333333335,
+      "grad_norm": 36.0,
+      "grad_norm_var": 33.514322916666664,
+      "learning_rate": 0.0001,
+      "loss": 7.1611,
+      "loss/crossentropy": 1.934358787536621,
+      "loss/hidden": 3.308203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927736137062311,
+      "step": 3310
+    },
+    {
+      "epoch": 0.027666666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 164.93326822916666,
+      "learning_rate": 0.0001,
+      "loss": 7.2316,
+      "loss/crossentropy": 1.8626260861754418,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18418960385024546,
+      "step": 3320
+    },
+    {
+      "epoch": 0.02775,
+      "grad_norm": 31.875,
+      "grad_norm_var": 162.27545572916668,
+      "learning_rate": 0.0001,
+      "loss": 7.2925,
+      "loss/crossentropy": 1.9802043229341506,
+      "loss/hidden": 3.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877164736390114,
+      "step": 3330
+    },
+    {
+      "epoch": 0.027833333333333335,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.8229166666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1663,
+      "loss/crossentropy": 1.9671151712536812,
+      "loss/hidden": 3.40859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19295375384390354,
+      "step": 3340
+    },
+    {
+      "epoch": 0.027916666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.1892317587907108e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2766,
+      "loss/crossentropy": 2.1149969041347503,
+      "loss/hidden": 3.4578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20657425597310067,
+      "step": 3350
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 33.0,
+      "grad_norm_var": 30.4509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2297,
+      "loss/crossentropy": 2.0748997569084167,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20927353482693434,
+      "step": 3360
+    },
+    {
+      "epoch": 0.02808333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.559309895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1992,
+      "loss/crossentropy": 1.7197825767099857,
+      "loss/hidden": 3.430859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1713511989451945,
+      "step": 3370
+    },
+    {
+      "epoch": 0.028166666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 8.859309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2538,
+      "loss/crossentropy": 1.9892003759741783,
+      "loss/hidden": 3.37265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924882870167494,
+      "step": 3380
+    },
+    {
+      "epoch": 0.02825,
+      "grad_norm": 35.5,
+      "grad_norm_var": 20.360416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3701,
+      "loss/crossentropy": 2.0029900878667832,
+      "loss/hidden": 3.4203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1986021962016821,
+      "step": 3390
+    },
+    {
+      "epoch": 0.028333333333333332,
+      "grad_norm": 38.0,
+      "grad_norm_var": 5.1509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3267,
+      "loss/crossentropy": 1.9108583837747575,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18161017037928104,
+      "step": 3400
+    },
+    {
+      "epoch": 0.028416666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 24.479166666666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2388,
+      "loss/crossentropy": 1.936232401430607,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21553935762494802,
+      "step": 3410
+    },
+    {
+      "epoch": 0.0285,
+      "grad_norm": 37.0,
+      "grad_norm_var": 14.5244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1867,
+      "loss/crossentropy": 1.9755698680877685,
+      "loss/hidden": 3.409375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19823556169867515,
+      "step": 3420
+    },
+    {
+      "epoch": 0.028583333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.248372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2332,
+      "loss/crossentropy": 1.8893498435616494,
+      "loss/hidden": 3.371484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18988429009914398,
+      "step": 3430
+    },
+    {
+      "epoch": 0.028666666666666667,
+      "grad_norm": 38.25,
+      "grad_norm_var": 4.315559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3593,
+      "loss/crossentropy": 1.9460684165358544,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18759000655263663,
+      "step": 3440
+    },
+    {
+      "epoch": 0.02875,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.5150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3355,
+      "loss/crossentropy": 1.843932069838047,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18119751177728177,
+      "step": 3450
+    },
+    {
+      "epoch": 0.028833333333333332,
+      "grad_norm": 37.0,
+      "grad_norm_var": 8.888997395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.4385,
+      "loss/crossentropy": 1.9830153197050095,
+      "loss/hidden": 3.5125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973841156810522,
+      "step": 3460
+    },
+    {
+      "epoch": 0.028916666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.643489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3091,
+      "loss/crossentropy": 1.9742940947413445,
+      "loss/hidden": 3.35234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943078614771366,
+      "step": 3470
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.71875,
+      "learning_rate": 0.0001,
+      "loss": 7.4103,
+      "loss/crossentropy": 2.0165591590106486,
+      "loss/hidden": 3.31875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18567599412053823,
+      "step": 3480
+    },
+    {
+      "epoch": 0.029083333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.623893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1689,
+      "loss/crossentropy": 1.8976772978901864,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17701343428343536,
+      "step": 3490
+    },
+    {
+      "epoch": 0.029166666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.560070433854806e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4131,
+      "loss/crossentropy": 1.8664681144058703,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18947559054940938,
+      "step": 3500
+    },
+    {
+      "epoch": 0.02925,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.5600704344738157e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2189,
+      "loss/crossentropy": 1.9561820104718208,
+      "loss/hidden": 3.42109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2091151824221015,
+      "step": 3510
+    },
+    {
+      "epoch": 0.029333333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.15390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2372,
+      "loss/crossentropy": 1.9695313319563865,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19255149811506272,
+      "step": 3520
+    },
+    {
+      "epoch": 0.029416666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 22.8572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2377,
+      "loss/crossentropy": 2.091267944872379,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2002470575273037,
+      "step": 3530
+    },
+    {
+      "epoch": 0.0295,
+      "grad_norm": 31.0,
+      "grad_norm_var": 12.553059895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.3079,
+      "loss/crossentropy": 2.050358733534813,
+      "loss/hidden": 3.278515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18941335808485746,
+      "step": 3540
+    },
+    {
+      "epoch": 0.029583333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.202018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2156,
+      "loss/crossentropy": 1.9859870471060277,
+      "loss/hidden": 3.35,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879461133852601,
+      "step": 3550
+    },
+    {
+      "epoch": 0.029666666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.253125,
+      "learning_rate": 0.0001,
+      "loss": 7.2648,
+      "loss/crossentropy": 1.9784325510263443,
+      "loss/hidden": 3.460546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18699357379227877,
+      "step": 3560
+    },
+    {
+      "epoch": 0.02975,
+      "grad_norm": 35.0,
+      "grad_norm_var": 19.438541666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1386,
+      "loss/crossentropy": 1.868053089082241,
+      "loss/hidden": 3.45625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19436977664008737,
+      "step": 3570
+    },
+    {
+      "epoch": 0.029833333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 21.012239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2663,
+      "loss/crossentropy": 2.0065255418419836,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18793275263160467,
+      "step": 3580
+    },
+    {
+      "epoch": 0.029916666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 12.87890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2809,
+      "loss/crossentropy": 1.8890562951564789,
+      "loss/hidden": 3.338671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1753269799053669,
+      "step": 3590
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 30.25,
+      "grad_norm_var": 10.074934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.9849582225084306,
+      "loss/hidden": 3.344140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19147213082760572,
+      "step": 3600
+    },
+    {
+      "epoch": 0.030083333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.9020182291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1947,
+      "loss/crossentropy": 1.927957746386528,
+      "loss/hidden": 3.256640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846942937001586,
+      "step": 3610
+    },
+    {
+      "epoch": 0.030166666666666668,
+      "grad_norm": 38.5,
+      "grad_norm_var": 5.361393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.169,
+      "loss/crossentropy": 1.9098347902297974,
+      "loss/hidden": 3.480859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18104900792241096,
+      "step": 3620
+    },
+    {
+      "epoch": 0.03025,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.13125,
+      "learning_rate": 0.0001,
+      "loss": 7.144,
+      "loss/crossentropy": 1.7323764339089394,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18193134814500808,
+      "step": 3630
+    },
+    {
+      "epoch": 0.030333333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.798143776342934e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.289,
+      "loss/crossentropy": 1.9735290616750718,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1941153584048152,
+      "step": 3640
+    },
+    {
+      "epoch": 0.030416666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 10.925,
+      "learning_rate": 0.0001,
+      "loss": 7.2602,
+      "loss/crossentropy": 1.8394082367420197,
+      "loss/hidden": 3.36875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17473349459469317,
+      "step": 3650
+    },
+    {
+      "epoch": 0.0305,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.091080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2306,
+      "loss/crossentropy": 1.8136562265455722,
+      "loss/hidden": 3.41171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17720981892198323,
+      "step": 3660
+    },
+    {
+      "epoch": 0.030583333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.7747395833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2363,
+      "loss/crossentropy": 1.947695901989937,
+      "loss/hidden": 3.412890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18550261501222848,
+      "step": 3670
+    },
+    {
+      "epoch": 0.030666666666666665,
+      "grad_norm": 36.5,
+      "grad_norm_var": 7.684442640909425e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3297,
+      "loss/crossentropy": 1.8268452778458595,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1864478962495923,
+      "step": 3680
+    },
+    {
+      "epoch": 0.03075,
+      "grad_norm": 29.25,
+      "grad_norm_var": 7.684442642078237e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3815,
+      "loss/crossentropy": 2.044998970627785,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19564838595688344,
+      "step": 3690
+    },
+    {
+      "epoch": 0.030833333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.412239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2426,
+      "loss/crossentropy": 1.9021501779556274,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1848083844408393,
+      "step": 3700
+    },
+    {
+      "epoch": 0.030916666666666665,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.105208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3239,
+      "loss/crossentropy": 2.0004999369382857,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1975437816232443,
+      "step": 3710
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.511458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.3091,
+      "loss/crossentropy": 1.9500782281160354,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1928380098193884,
+      "step": 3720
+    },
+    {
+      "epoch": 0.031083333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.1056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2011,
+      "loss/crossentropy": 2.0728418275713922,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1800950923934579,
+      "step": 3730
+    },
+    {
+      "epoch": 0.031166666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.7327473958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2847,
+      "loss/crossentropy": 2.007042866945267,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19376245383173227,
+      "step": 3740
+    },
+    {
+      "epoch": 0.03125,
+      "grad_norm": 39.0,
+      "grad_norm_var": 45.46139322916667,
+      "learning_rate": 0.0001,
+      "loss": 7.0946,
+      "loss/crossentropy": 1.9019963197410106,
+      "loss/hidden": 3.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17656781915575265,
+      "step": 3750
+    },
+    {
+      "epoch": 0.03133333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.8014953780616896e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2806,
+      "loss/crossentropy": 1.9263739496469499,
+      "loss/hidden": 3.438671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1908752067014575,
+      "step": 3760
+    },
+    {
+      "epoch": 0.03141666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.8014953838162746e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2388,
+      "loss/crossentropy": 1.965820837020874,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19951398996636271,
+      "step": 3770
+    },
+    {
+      "epoch": 0.0315,
+      "grad_norm": 29.875,
+      "grad_norm_var": 70.99576822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.346,
+      "loss/crossentropy": 1.9226266965270042,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18729710709303618,
+      "step": 3780
+    },
+    {
+      "epoch": 0.03158333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 66.08125,
+      "learning_rate": 0.0001,
+      "loss": 7.2438,
+      "loss/crossentropy": 1.8775743916630745,
+      "loss/hidden": 3.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19226148668676615,
+      "step": 3790
+    },
+    {
+      "epoch": 0.03166666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.207291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2464,
+      "loss/crossentropy": 1.960534116625786,
+      "loss/hidden": 3.50703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20045664943754674,
+      "step": 3800
+    },
+    {
+      "epoch": 0.03175,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.6436848958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1574,
+      "loss/crossentropy": 1.7869322314858436,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1767852820456028,
+      "step": 3810
+    },
+    {
+      "epoch": 0.03183333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.9650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2089,
+      "loss/crossentropy": 2.004594850540161,
+      "loss/hidden": 3.271875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17339777592569589,
+      "step": 3820
+    },
+    {
+      "epoch": 0.03191666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.370572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1548,
+      "loss/crossentropy": 1.9374451048672199,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855450887233019,
+      "step": 3830
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.9514973958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2219,
+      "loss/crossentropy": 1.9749352604150772,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18502496387809514,
+      "step": 3840
+    },
+    {
+      "epoch": 0.03208333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.7186848958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1275,
+      "loss/crossentropy": 2.0351619154214857,
+      "loss/hidden": 3.33515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17815996166318654,
+      "step": 3850
+    },
+    {
+      "epoch": 0.03216666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 10.888997395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2478,
+      "loss/crossentropy": 1.976029747724533,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18774439115077257,
+      "step": 3860
+    },
+    {
+      "epoch": 0.03225,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.245572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2188,
+      "loss/crossentropy": 1.9812984496355057,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19507236313074827,
+      "step": 3870
+    },
+    {
+      "epoch": 0.03233333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 152.28229166666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2458,
+      "loss/crossentropy": 1.984493115544319,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19963345900177956,
+      "step": 3880
+    },
+    {
+      "epoch": 0.03241666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.1625,
+      "learning_rate": 0.0001,
+      "loss": 7.389,
+      "loss/crossentropy": 1.9200308933854102,
+      "loss/hidden": 3.479296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1876375088468194,
+      "step": 3890
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.81015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2937,
+      "loss/crossentropy": 1.983275644481182,
+      "loss/hidden": 3.360546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18901068139821292,
+      "step": 3900
+    },
+    {
+      "epoch": 0.03258333333333333,
+      "grad_norm": 42.75,
+      "grad_norm_var": 13.860416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2993,
+      "loss/crossentropy": 2.002154844999313,
+      "loss/hidden": 3.51796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924261325970292,
+      "step": 3910
+    },
+    {
+      "epoch": 0.03266666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 25.407747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2245,
+      "loss/crossentropy": 1.9173000499606132,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19295201431959869,
+      "step": 3920
+    },
+    {
+      "epoch": 0.03275,
+      "grad_norm": 29.625,
+      "grad_norm_var": 18.592643229166665,
+      "learning_rate": 0.0001,
+      "loss": 7.2637,
+      "loss/crossentropy": 2.0171883307397365,
+      "loss/hidden": 3.290234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17894033640623092,
+      "step": 3930
+    },
+    {
+      "epoch": 0.03283333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.7561848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2559,
+      "loss/crossentropy": 1.973395998775959,
+      "loss/hidden": 3.53671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20406727930530905,
+      "step": 3940
+    },
+    {
+      "epoch": 0.032916666666666664,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.5947916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2626,
+      "loss/crossentropy": 1.95164203196764,
+      "loss/hidden": 3.408203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1913003684952855,
+      "step": 3950
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.794205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2546,
+      "loss/crossentropy": 2.006377327442169,
+      "loss/hidden": 3.27578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18781272880733013,
+      "step": 3960
+    },
+    {
+      "epoch": 0.03308333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.484830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1955,
+      "loss/crossentropy": 1.8874030619859696,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17777375876903534,
+      "step": 3970
+    },
+    {
+      "epoch": 0.033166666666666664,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.337239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3097,
+      "loss/crossentropy": 1.9324250996112824,
+      "loss/hidden": 3.219140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1771338377147913,
+      "step": 3980
+    },
+    {
+      "epoch": 0.03325,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.3025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.8945629693567754,
+      "loss/hidden": 3.38203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17384705552831292,
+      "step": 3990
+    },
+    {
+      "epoch": 0.03333333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 12.415625,
+      "learning_rate": 0.0001,
+      "loss": 7.2273,
+      "loss/crossentropy": 1.954460334777832,
+      "loss/hidden": 3.33046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20084381606429816,
+      "step": 4000
+    },
+    {
+      "epoch": 0.033416666666666664,
+      "grad_norm": 34.25,
+      "grad_norm_var": 8.487434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.262,
+      "loss/crossentropy": 1.8653246648609638,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17524506486952304,
+      "step": 4010
+    },
+    {
+      "epoch": 0.0335,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.2333333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2548,
+      "loss/crossentropy": 1.9703700937330724,
+      "loss/hidden": 3.245703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18107774439267815,
+      "step": 4020
+    },
+    {
+      "epoch": 0.03358333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.416666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2275,
+      "loss/crossentropy": 2.0328017130494116,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19874867778271438,
+      "step": 4030
+    },
+    {
+      "epoch": 0.033666666666666664,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.8785807291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1966,
+      "loss/crossentropy": 2.034155286103487,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1778541506268084,
+      "step": 4040
+    },
+    {
+      "epoch": 0.03375,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.233333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.391,
+      "loss/crossentropy": 1.956402349472046,
+      "loss/hidden": 3.274609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18872720897197723,
+      "step": 4050
+    },
+    {
+      "epoch": 0.03383333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.89144705786341e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1737,
+      "loss/crossentropy": 1.983364275097847,
+      "loss/hidden": 3.54609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20503977742046117,
+      "step": 4060
+    },
+    {
+      "epoch": 0.033916666666666664,
+      "grad_norm": 47.5,
+      "grad_norm_var": 23.84140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2168,
+      "loss/crossentropy": 1.9249817907810212,
+      "loss/hidden": 3.3609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1917848229408264,
+      "step": 4070
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 32.25,
+      "grad_norm_var": 20.892708333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.3556,
+      "loss/crossentropy": 1.9946053527295589,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20452868696302176,
+      "step": 4080
+    },
+    {
+      "epoch": 0.034083333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.035872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2173,
+      "loss/crossentropy": 1.9789577759802341,
+      "loss/hidden": 3.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17965719159692525,
+      "step": 4090
+    },
+    {
+      "epoch": 0.034166666666666665,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.5254557291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1648,
+      "loss/crossentropy": 1.95723911896348,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18108865497633814,
+      "step": 4100
+    },
+    {
+      "epoch": 0.03425,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.8067057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1109,
+      "loss/crossentropy": 1.962564404308796,
+      "loss/hidden": 3.441796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19903685674071311,
+      "step": 4110
+    },
+    {
+      "epoch": 0.034333333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.3650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2889,
+      "loss/crossentropy": 1.9317152202129364,
+      "loss/hidden": 3.3890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894014548510313,
+      "step": 4120
+    },
+    {
+      "epoch": 0.034416666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.6869140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2726,
+      "loss/crossentropy": 1.888142066448927,
+      "loss/hidden": 3.511328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19417026322335004,
+      "step": 4130
+    },
+    {
+      "epoch": 0.0345,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.608333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2027,
+      "loss/crossentropy": 1.9418023459613323,
+      "loss/hidden": 3.26484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18381806444376708,
+      "step": 4140
+    },
+    {
+      "epoch": 0.034583333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.323958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2148,
+      "loss/crossentropy": 1.9318063765764237,
+      "loss/hidden": 3.480078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20140955168753863,
+      "step": 4150
+    },
+    {
+      "epoch": 0.034666666666666665,
+      "grad_norm": 175.0,
+      "grad_norm_var": 1278.2921223958333,
+      "learning_rate": 0.0001,
+      "loss": 7.1926,
+      "loss/crossentropy": 1.8987142145633698,
+      "loss/hidden": 3.428515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20260288771241902,
+      "step": 4160
+    },
+    {
+      "epoch": 0.03475,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1286.3910807291666,
+      "learning_rate": 0.0001,
+      "loss": 7.1593,
+      "loss/crossentropy": 2.034174790978432,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18820980936288834,
+      "step": 4170
+    },
+    {
+      "epoch": 0.034833333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 15.941666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2882,
+      "loss/crossentropy": 1.9306144535541534,
+      "loss/hidden": 3.449609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19824343249201776,
+      "step": 4180
+    },
+    {
+      "epoch": 0.034916666666666665,
+      "grad_norm": 31.125,
+      "grad_norm_var": 85.56770833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1008,
+      "loss/crossentropy": 1.951157060265541,
+      "loss/hidden": 3.349609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17776616513729096,
+      "step": 4190
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 38.25,
+      "grad_norm_var": 11.1625,
+      "learning_rate": 0.0001,
+      "loss": 7.2101,
+      "loss/crossentropy": 1.9059597700834274,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1789044144563377,
+      "step": 4200
+    },
+    {
+      "epoch": 0.035083333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.891447057895392e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4328,
+      "loss/crossentropy": 1.9277472093701362,
+      "loss/hidden": 3.615234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19346159268170596,
+      "step": 4210
+    },
+    {
+      "epoch": 0.035166666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.06875,
+      "learning_rate": 0.0001,
+      "loss": 7.137,
+      "loss/crossentropy": 1.976280900835991,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18772026468068362,
+      "step": 4220
+    },
+    {
+      "epoch": 0.03525,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.734830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.265,
+      "loss/crossentropy": 1.9235099971294403,
+      "loss/hidden": 3.493359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18395463675260543,
+      "step": 4230
+    },
+    {
+      "epoch": 0.035333333333333335,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.07890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1643,
+      "loss/crossentropy": 1.8247085146605968,
+      "loss/hidden": 3.302734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17993250191211702,
+      "step": 4240
+    },
+    {
+      "epoch": 0.035416666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 233.09557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2326,
+      "loss/crossentropy": 1.900027759373188,
+      "loss/hidden": 3.49140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891545556485653,
+      "step": 4250
+    },
+    {
+      "epoch": 0.0355,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.8525390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2668,
+      "loss/crossentropy": 2.0083697855472566,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19048811830580234,
+      "step": 4260
+    },
+    {
+      "epoch": 0.035583333333333335,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.295247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2289,
+      "loss/crossentropy": 1.8813670575618744,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20953154526650905,
+      "step": 4270
+    },
+    {
+      "epoch": 0.035666666666666666,
+      "grad_norm": 4009754624.0,
+      "grad_norm_var": 1.0048832424736108e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2215,
+      "loss/crossentropy": 2.036851942539215,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18767790962010622,
+      "step": 4280
+    },
+    {
+      "epoch": 0.03575,
+      "grad_norm": 35.5,
+      "grad_norm_var": 1.0048832419306232e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1731,
+      "loss/crossentropy": 1.8813318252563476,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17605223394930364,
+      "step": 4290
+    },
+    {
+      "epoch": 0.035833333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 8.367708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2143,
+      "loss/crossentropy": 1.7793952539563178,
+      "loss/hidden": 3.450390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17674990901723503,
+      "step": 4300
+    },
+    {
+      "epoch": 0.035916666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.8125,
+      "learning_rate": 0.0001,
+      "loss": 7.1788,
+      "loss/crossentropy": 2.114986628293991,
+      "loss/hidden": 3.28203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17624464612454177,
+      "step": 4310
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 29.375,
+      "grad_norm_var": 9.32265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1168,
+      "loss/crossentropy": 1.965844801068306,
+      "loss/hidden": 3.227734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18552668560296298,
+      "step": 4320
+    },
+    {
+      "epoch": 0.036083333333333335,
+      "grad_norm": 33.5,
+      "grad_norm_var": 12.855143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2536,
+      "loss/crossentropy": 2.029114265739918,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20089028123766184,
+      "step": 4330
+    },
+    {
+      "epoch": 0.036166666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.6889973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1575,
+      "loss/crossentropy": 1.9205705314874648,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19222395457327365,
+      "step": 4340
+    },
+    {
+      "epoch": 0.03625,
+      "grad_norm": 29.875,
+      "grad_norm_var": 18.380143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2364,
+      "loss/crossentropy": 2.080160215497017,
+      "loss/hidden": 3.32265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2045929528772831,
+      "step": 4350
+    },
+    {
+      "epoch": 0.036333333333333336,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.9692057291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1983,
+      "loss/crossentropy": 1.9375086218118667,
+      "loss/hidden": 3.4015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18691876921802758,
+      "step": 4360
+    },
+    {
+      "epoch": 0.03641666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.981398734615348e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3271,
+      "loss/crossentropy": 2.005251409113407,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18281115759164096,
+      "step": 4370
+    },
+    {
+      "epoch": 0.0365,
+      "grad_norm": 32.5,
+      "grad_norm_var": 52.04420572916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2931,
+      "loss/crossentropy": 1.9374827235937118,
+      "loss/hidden": 3.3859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19405684284865857,
+      "step": 4380
+    },
+    {
+      "epoch": 0.036583333333333336,
+      "grad_norm": 34.75,
+      "grad_norm_var": 20.840559895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2968,
+      "loss/crossentropy": 1.9063488006591798,
+      "loss/hidden": 3.42421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20247304160147905,
+      "step": 4390
+    },
+    {
+      "epoch": 0.03666666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 23.69140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2249,
+      "loss/crossentropy": 1.9918562591075897,
+      "loss/hidden": 3.3015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18467331416904925,
+      "step": 4400
+    },
+    {
+      "epoch": 0.03675,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.3025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.117,
+      "loss/crossentropy": 1.9270878240466118,
+      "loss/hidden": 3.48828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20000088401138783,
+      "step": 4410
+    },
+    {
+      "epoch": 0.036833333333333336,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.7552083333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.3101,
+      "loss/crossentropy": 2.071250331401825,
+      "loss/hidden": 3.58828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19236841816455125,
+      "step": 4420
+    },
+    {
+      "epoch": 0.03691666666666667,
+      "grad_norm": 62.75,
+      "grad_norm_var": 65.09837239583334,
+      "learning_rate": 0.0001,
+      "loss": 7.1969,
+      "loss/crossentropy": 1.966628861427307,
+      "loss/hidden": 3.471484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20192189700901508,
+      "step": 4430
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 27.75,
+      "grad_norm_var": 66.3728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1672,
+      "loss/crossentropy": 1.8696994706988335,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916260415688157,
+      "step": 4440
+    },
+    {
+      "epoch": 0.037083333333333336,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.7775390625,
+      "learning_rate": 0.0001,
+      "loss": 7.278,
+      "loss/crossentropy": 1.9153100848197937,
+      "loss/hidden": 3.465625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20059756319969893,
+      "step": 4450
+    },
+    {
+      "epoch": 0.03716666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.753125,
+      "learning_rate": 0.0001,
+      "loss": 7.2516,
+      "loss/crossentropy": 2.06074358522892,
+      "loss/hidden": 3.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1892743492498994,
+      "step": 4460
+    },
+    {
+      "epoch": 0.03725,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.079166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1622,
+      "loss/crossentropy": 1.954483150690794,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1837238807231188,
+      "step": 4470
+    },
+    {
+      "epoch": 0.037333333333333336,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.5337890625,
+      "learning_rate": 0.0001,
+      "loss": 7.241,
+      "loss/crossentropy": 1.9478939175605774,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20669157840311528,
+      "step": 4480
+    },
+    {
+      "epoch": 0.03741666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.60390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2972,
+      "loss/crossentropy": 1.890691903233528,
+      "loss/hidden": 3.432421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19282174929976464,
+      "step": 4490
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.24140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2593,
+      "loss/crossentropy": 1.8015254758298398,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17511156313121318,
+      "step": 4500
+    },
+    {
+      "epoch": 0.03758333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.465625,
+      "learning_rate": 0.0001,
+      "loss": 7.188,
+      "loss/crossentropy": 1.984817424416542,
+      "loss/hidden": 3.3078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17105563301593066,
+      "step": 4510
+    },
+    {
+      "epoch": 0.03766666666666667,
+      "grad_norm": 40.75,
+      "grad_norm_var": 21.156705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2147,
+      "loss/crossentropy": 1.9874697402119637,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19757193494588138,
+      "step": 4520
+    },
+    {
+      "epoch": 0.03775,
+      "grad_norm": 37.5,
+      "grad_norm_var": 342.69166666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2505,
+      "loss/crossentropy": 1.845185561478138,
+      "loss/hidden": 3.376953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192365712672472,
+      "step": 4530
+    },
+    {
+      "epoch": 0.03783333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 330.7806640625,
+      "learning_rate": 0.0001,
+      "loss": 7.203,
+      "loss/crossentropy": 2.0427382573485375,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17846857775002717,
+      "step": 4540
+    },
+    {
+      "epoch": 0.03791666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.216080729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1229,
+      "loss/crossentropy": 1.8890719719231128,
+      "loss/hidden": 3.436328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911027053371072,
+      "step": 4550
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.1801432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2172,
+      "loss/crossentropy": 2.0080476641654967,
+      "loss/hidden": 3.361328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192045708745718,
+      "step": 4560
+    },
+    {
+      "epoch": 0.03808333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.338997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1599,
+      "loss/crossentropy": 1.9886827737092971,
+      "loss/hidden": 3.30546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17904456499964,
+      "step": 4570
+    },
+    {
+      "epoch": 0.03816666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.708333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.289,
+      "loss/crossentropy": 2.07514126971364,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21552205746993422,
+      "step": 4580
+    },
+    {
+      "epoch": 0.03825,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.268684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3541,
+      "loss/crossentropy": 1.8997907042503357,
+      "loss/hidden": 3.496484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20105953328311443,
+      "step": 4590
+    },
+    {
+      "epoch": 0.03833333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.0416015625,
+      "learning_rate": 0.0001,
+      "loss": 7.3417,
+      "loss/crossentropy": 1.9448037043213844,
+      "loss/hidden": 3.4546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19288307912647723,
+      "step": 4600
+    },
+    {
+      "epoch": 0.03841666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.5434895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1478,
+      "loss/crossentropy": 1.9898255258798598,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18608010280877352,
+      "step": 4610
+    },
+    {
+      "epoch": 0.0385,
+      "grad_norm": 32.75,
+      "grad_norm_var": 11.652018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.3242,
+      "loss/crossentropy": 2.009296101331711,
+      "loss/hidden": 3.334765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843553153797984,
+      "step": 4620
+    },
+    {
+      "epoch": 0.03858333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 13.0119140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1898,
+      "loss/crossentropy": 1.9398988649249076,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18743706177920103,
+      "step": 4630
+    },
+    {
+      "epoch": 0.03866666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.059830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1591,
+      "loss/crossentropy": 1.853212970495224,
+      "loss/hidden": 3.433203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17872758507728576,
+      "step": 4640
+    },
+    {
+      "epoch": 0.03875,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.08515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1896,
+      "loss/crossentropy": 2.0782950162887572,
+      "loss/hidden": 3.225390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17811987847089766,
+      "step": 4650
+    },
+    {
+      "epoch": 0.03883333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 2.575671875150546e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3045,
+      "loss/crossentropy": 2.0085189670324324,
+      "loss/hidden": 3.325390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20201543867588043,
+      "step": 4660
+    },
+    {
+      "epoch": 0.03891666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.5756718751928384e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0863,
+      "loss/crossentropy": 1.9253064393997192,
+      "loss/hidden": 3.444140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21200950276106595,
+      "step": 4670
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.4166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2533,
+      "loss/crossentropy": 1.9820420145988464,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19638330191373826,
+      "step": 4680
+    },
+    {
+      "epoch": 0.03908333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.9374348958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1681,
+      "loss/crossentropy": 1.8506903916597366,
+      "loss/hidden": 3.420703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865494478493929,
+      "step": 4690
+    },
+    {
+      "epoch": 0.03916666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.941666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2402,
+      "loss/crossentropy": 2.0131797283887862,
+      "loss/hidden": 3.546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20369560234248638,
+      "step": 4700
+    },
+    {
+      "epoch": 0.03925,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.776497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2278,
+      "loss/crossentropy": 1.7893060639500618,
+      "loss/hidden": 3.353515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18423357047140598,
+      "step": 4710
+    },
+    {
+      "epoch": 0.03933333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.2072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2598,
+      "loss/crossentropy": 1.9981599599123,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18187536485493183,
+      "step": 4720
+    },
+    {
+      "epoch": 0.03941666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.0259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2328,
+      "loss/crossentropy": 1.969702386856079,
+      "loss/hidden": 3.4296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20060717556625604,
+      "step": 4730
+    },
+    {
+      "epoch": 0.0395,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.226822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.8951549768447875,
+      "loss/hidden": 3.269921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19447669275105,
+      "step": 4740
+    },
+    {
+      "epoch": 0.03958333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.737434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2429,
+      "loss/crossentropy": 2.080011251568794,
+      "loss/hidden": 3.290625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19000240378081798,
+      "step": 4750
+    },
+    {
+      "epoch": 0.03966666666666667,
+      "grad_norm": 28.125,
+      "grad_norm_var": 3.9301432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1889,
+      "loss/crossentropy": 1.9444591283798218,
+      "loss/hidden": 3.34921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841985611245036,
+      "step": 4760
+    },
+    {
+      "epoch": 0.03975,
+      "grad_norm": 29.125,
+      "grad_norm_var": 6.077083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2539,
+      "loss/crossentropy": 1.9875277951359749,
+      "loss/hidden": 3.476171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19982990454882382,
+      "step": 4770
+    },
+    {
+      "epoch": 0.03983333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.3684895833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2338,
+      "loss/crossentropy": 2.018096478283405,
+      "loss/hidden": 3.358984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1817854383960366,
+      "step": 4780
+    },
+    {
+      "epoch": 0.03991666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.211458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2624,
+      "loss/crossentropy": 1.9615140050649642,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18594696801155805,
+      "step": 4790
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.9509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3972,
+      "loss/crossentropy": 1.9175285905599595,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19690106008201838,
+      "step": 4800
+    },
+    {
+      "epoch": 0.04008333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 61.534375,
+      "learning_rate": 0.0001,
+      "loss": 7.2732,
+      "loss/crossentropy": 1.8362210482358932,
+      "loss/hidden": 3.422265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1917069649323821,
+      "step": 4810
+    },
+    {
+      "epoch": 0.04016666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 57.2712890625,
+      "learning_rate": 0.0001,
+      "loss": 7.3767,
+      "loss/crossentropy": 2.133534160256386,
+      "loss/hidden": 3.51015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21782590821385384,
+      "step": 4820
+    },
+    {
+      "epoch": 0.04025,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.082291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2232,
+      "loss/crossentropy": 1.951946584880352,
+      "loss/hidden": 3.4390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19556211698800324,
+      "step": 4830
+    },
+    {
+      "epoch": 0.04033333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.76015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2368,
+      "loss/crossentropy": 1.9654059708118439,
+      "loss/hidden": 3.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19498455952852964,
+      "step": 4840
+    },
+    {
+      "epoch": 0.04041666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 6.287239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2589,
+      "loss/crossentropy": 1.9640298388898372,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852282472886145,
+      "step": 4850
+    },
+    {
+      "epoch": 0.0405,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.3489583333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2713,
+      "loss/crossentropy": 1.896032267808914,
+      "loss/hidden": 3.442578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20271046720445157,
+      "step": 4860
+    },
+    {
+      "epoch": 0.04058333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 54.316080729166664,
+      "learning_rate": 0.0001,
+      "loss": 7.1648,
+      "loss/crossentropy": 1.9707864001393318,
+      "loss/hidden": 3.50234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.192245776951313,
+      "step": 4870
+    },
+    {
+      "epoch": 0.04066666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.3556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2257,
+      "loss/crossentropy": 1.7333999305963517,
+      "loss/hidden": 3.49765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17861571693792938,
+      "step": 4880
+    },
+    {
+      "epoch": 0.04075,
+      "grad_norm": 31.25,
+      "grad_norm_var": 952.7306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2747,
+      "loss/crossentropy": 1.9761729918420314,
+      "loss/hidden": 3.381640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18439243715256454,
+      "step": 4890
+    },
+    {
+      "epoch": 0.04083333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 951.7489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1721,
+      "loss/crossentropy": 1.976510974764824,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18401159718632698,
+      "step": 4900
+    },
+    {
+      "epoch": 0.040916666666666664,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.0353515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3696,
+      "loss/crossentropy": 1.8498829215765,
+      "loss/hidden": 3.497265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19358056131750345,
+      "step": 4910
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.943684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3129,
+      "loss/crossentropy": 1.993993727862835,
+      "loss/hidden": 3.331640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032605566084385,
+      "step": 4920
+    },
+    {
+      "epoch": 0.04108333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.502018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.161,
+      "loss/crossentropy": 1.8945767790079118,
+      "loss/hidden": 3.301953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1947425028309226,
+      "step": 4930
+    },
+    {
+      "epoch": 0.041166666666666664,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.505989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3919,
+      "loss/crossentropy": 1.9350941553711891,
+      "loss/hidden": 3.70625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20500725619494914,
+      "step": 4940
+    },
+    {
+      "epoch": 0.04125,
+      "grad_norm": 29.875,
+      "grad_norm_var": 18.84140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2145,
+      "loss/crossentropy": 2.050392009317875,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18438252825289964,
+      "step": 4950
+    },
+    {
+      "epoch": 0.04133333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 21.251041666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.148,
+      "loss/crossentropy": 1.8834857508540153,
+      "loss/hidden": 3.46640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20490446221083403,
+      "step": 4960
+    },
+    {
+      "epoch": 0.041416666666666664,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.446809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1951,
+      "loss/crossentropy": 1.9395118832588196,
+      "loss/hidden": 3.240625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18307873839512467,
+      "step": 4970
+    },
+    {
+      "epoch": 0.0415,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.9369140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3024,
+      "loss/crossentropy": 2.0352481067180634,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20577008351683618,
+      "step": 4980
+    },
+    {
+      "epoch": 0.04158333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 7.832217005980997e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2377,
+      "loss/crossentropy": 1.969672729074955,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20676071532070636,
+      "step": 4990
+    },
+    {
+      "epoch": 0.041666666666666664,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.487955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1396,
+      "loss/crossentropy": 1.9443906158208848,
+      "loss/hidden": 3.26953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18803976774215697,
+      "step": 5000
+    },
+    {
+      "epoch": 0.04175,
+      "grad_norm": 36.25,
+      "grad_norm_var": 1.848278976250949e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3254,
+      "loss/crossentropy": 1.8845553174614906,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17818044014275075,
+      "step": 5010
+    },
+    {
+      "epoch": 0.041833333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.848278975319464e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2489,
+      "loss/crossentropy": 1.882246272265911,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1847611464560032,
+      "step": 5020
+    },
+    {
+      "epoch": 0.041916666666666665,
+      "grad_norm": 2785017856.0,
+      "grad_norm_var": 4.847702676187994e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2388,
+      "loss/crossentropy": 2.051472933590412,
+      "loss/hidden": 3.255859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18801066745072603,
+      "step": 5030
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.847702674592411e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1795,
+      "loss/crossentropy": 1.8703616850078106,
+      "loss/hidden": 3.474609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20192000661045312,
+      "step": 5040
+    },
+    {
+      "epoch": 0.042083333333333334,
+      "grad_norm": 49.75,
+      "grad_norm_var": 22.523372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1898,
+      "loss/crossentropy": 1.9267778247594833,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18499485440552235,
+      "step": 5050
+    },
+    {
+      "epoch": 0.042166666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 25.1697265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1415,
+      "loss/crossentropy": 1.8913537606596946,
+      "loss/hidden": 3.242578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1697727482765913,
+      "step": 5060
+    },
+    {
+      "epoch": 0.04225,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.05,
+      "learning_rate": 0.0001,
+      "loss": 7.1967,
+      "loss/crossentropy": 2.001948466897011,
+      "loss/hidden": 3.284765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18900279961526395,
+      "step": 5070
+    },
+    {
+      "epoch": 0.042333333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.2384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 2.0092270568013193,
+      "loss/hidden": 3.31796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18744736202061177,
+      "step": 5080
+    },
+    {
+      "epoch": 0.042416666666666665,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.459830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3271,
+      "loss/crossentropy": 1.9950493648648262,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18363405521959067,
+      "step": 5090
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.466666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2988,
+      "loss/crossentropy": 1.96678287088871,
+      "loss/hidden": 3.278515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19190824180841445,
+      "step": 5100
+    },
+    {
+      "epoch": 0.042583333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.0796223958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1883,
+      "loss/crossentropy": 2.0521468251943586,
+      "loss/hidden": 3.342578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20394890960305928,
+      "step": 5110
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.670768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1932,
+      "loss/crossentropy": 1.894241599738598,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18954524621367455,
+      "step": 5120
+    },
+    {
+      "epoch": 0.04275,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.25390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.9609125055372716,
+      "loss/hidden": 3.294921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18426925949752332,
+      "step": 5130
+    },
+    {
+      "epoch": 0.042833333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.32265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1633,
+      "loss/crossentropy": 1.9538515985012055,
+      "loss/hidden": 3.438671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18203786462545396,
+      "step": 5140
+    },
+    {
+      "epoch": 0.042916666666666665,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.286393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1574,
+      "loss/crossentropy": 1.8362636983394622,
+      "loss/hidden": 3.331640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1781992632895708,
+      "step": 5150
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.076497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2141,
+      "loss/crossentropy": 2.0191104084253313,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19397008065134286,
+      "step": 5160
+    },
+    {
+      "epoch": 0.043083333333333335,
+      "grad_norm": 29.25,
+      "grad_norm_var": 11.00390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2431,
+      "loss/crossentropy": 1.8258389428257942,
+      "loss/hidden": 3.272265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18870493108406663,
+      "step": 5170
+    },
+    {
+      "epoch": 0.043166666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 11.309309895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2124,
+      "loss/crossentropy": 1.9334841251373291,
+      "loss/hidden": 3.464453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005193768069148,
+      "step": 5180
+    },
+    {
+      "epoch": 0.04325,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.317122395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1607,
+      "loss/crossentropy": 1.9018930494785309,
+      "loss/hidden": 3.324609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909161601215601,
+      "step": 5190
+    },
+    {
+      "epoch": 0.043333333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 5.542708333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 1.9978152081370353,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20025275563821196,
+      "step": 5200
+    },
+    {
+      "epoch": 0.043416666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 24.744791666666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2824,
+      "loss/crossentropy": 1.9716323763132095,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18187592513859271,
+      "step": 5210
+    },
+    {
+      "epoch": 0.0435,
+      "grad_norm": 30.0,
+      "grad_norm_var": 83.13307291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1963,
+      "loss/crossentropy": 2.118371768295765,
+      "loss/hidden": 3.206640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18061571158468723,
+      "step": 5220
+    },
+    {
+      "epoch": 0.043583333333333335,
+      "grad_norm": 31.375,
+      "grad_norm_var": 83.16354166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1726,
+      "loss/crossentropy": 1.953398549556732,
+      "loss/hidden": 3.40546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18984109461307525,
+      "step": 5230
+    },
+    {
+      "epoch": 0.043666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.545572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1202,
+      "loss/crossentropy": 1.9805527925491333,
+      "loss/hidden": 3.428125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19125975519418717,
+      "step": 5240
+    },
+    {
+      "epoch": 0.04375,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.461393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2709,
+      "loss/crossentropy": 1.9394566029310227,
+      "loss/hidden": 3.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17967610489577054,
+      "step": 5250
+    },
+    {
+      "epoch": 0.043833333333333335,
+      "grad_norm": 30.125,
+      "grad_norm_var": 12.862239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1176,
+      "loss/crossentropy": 1.933772175014019,
+      "loss/hidden": 3.349609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18062447737902404,
+      "step": 5260
+    },
+    {
+      "epoch": 0.043916666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 14.672330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2236,
+      "loss/crossentropy": 2.000246451795101,
+      "loss/hidden": 3.305859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18543695881962777,
+      "step": 5270
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.2853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2063,
+      "loss/crossentropy": 1.8413677945733071,
+      "loss/hidden": 3.430859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18291340861469507,
+      "step": 5280
+    },
+    {
+      "epoch": 0.044083333333333335,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.5697265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2327,
+      "loss/crossentropy": 1.9906944945454597,
+      "loss/hidden": 3.312109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18375325668603182,
+      "step": 5290
+    },
+    {
+      "epoch": 0.04416666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 6.108333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3108,
+      "loss/crossentropy": 1.8987505346536637,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1902456210926175,
+      "step": 5300
+    },
+    {
+      "epoch": 0.04425,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.784309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1342,
+      "loss/crossentropy": 2.0367724671959877,
+      "loss/hidden": 3.45234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810287894681096,
+      "step": 5310
+    },
+    {
+      "epoch": 0.044333333333333336,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.567122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1938,
+      "loss/crossentropy": 1.8333687089383601,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1711188904941082,
+      "step": 5320
+    },
+    {
+      "epoch": 0.04441666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.0830729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1841,
+      "loss/crossentropy": 2.008009234815836,
+      "loss/hidden": 3.408203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1856338769197464,
+      "step": 5330
+    },
+    {
+      "epoch": 0.0445,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.0052083333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.312,
+      "loss/crossentropy": 2.0298721551895142,
+      "loss/hidden": 3.43828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19467000029981135,
+      "step": 5340
+    },
+    {
+      "epoch": 0.044583333333333336,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.8518229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2527,
+      "loss/crossentropy": 1.946864341199398,
+      "loss/hidden": 3.4203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18655043169856073,
+      "step": 5350
+    },
+    {
+      "epoch": 0.04466666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 8.37890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2448,
+      "loss/crossentropy": 1.986108300089836,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911590563133359,
+      "step": 5360
+    },
+    {
+      "epoch": 0.04475,
+      "grad_norm": 31.75,
+      "grad_norm_var": 8.673893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2376,
+      "loss/crossentropy": 1.9386815264821053,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18868138398975134,
+      "step": 5370
+    },
+    {
+      "epoch": 0.044833333333333336,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.776822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2728,
+      "loss/crossentropy": 1.9182999551296234,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1911846835166216,
+      "step": 5380
+    },
+    {
+      "epoch": 0.04491666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.9879557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2084,
+      "loss/crossentropy": 2.096855753660202,
+      "loss/hidden": 3.420703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1850230830721557,
+      "step": 5390
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.78515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2046,
+      "loss/crossentropy": 2.1250849097967146,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18073338493704796,
+      "step": 5400
+    },
+    {
+      "epoch": 0.045083333333333336,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.0245041383900454e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4045,
+      "loss/crossentropy": 1.8851775869727134,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1752549959346652,
+      "step": 5410
+    },
+    {
+      "epoch": 0.04516666666666667,
+      "grad_norm": 35.5,
+      "grad_norm_var": 5.024504136293068e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3242,
+      "loss/crossentropy": 2.0536662988364696,
+      "loss/hidden": 3.443359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19238244714215397,
+      "step": 5420
+    },
+    {
+      "epoch": 0.04525,
+      "grad_norm": 31.0,
+      "grad_norm_var": 37.017643229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2895,
+      "loss/crossentropy": 1.911766418814659,
+      "loss/hidden": 3.533203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18743917755782605,
+      "step": 5430
+    },
+    {
+      "epoch": 0.04533333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 17.709375,
+      "learning_rate": 0.0001,
+      "loss": 7.2069,
+      "loss/crossentropy": 2.0367508992552756,
+      "loss/hidden": 3.52109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2101318620145321,
+      "step": 5440
+    },
+    {
+      "epoch": 0.04541666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.678580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2162,
+      "loss/crossentropy": 1.9954749569296837,
+      "loss/hidden": 3.311328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17785828560590744,
+      "step": 5450
+    },
+    {
+      "epoch": 0.0455,
+      "grad_norm": 30.0,
+      "grad_norm_var": 9.764102533822118e+16,
+      "learning_rate": 0.0001,
+      "loss": 7.2735,
+      "loss/crossentropy": 2.088523915410042,
+      "loss/hidden": 3.2859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178264039196074,
+      "step": 5460
+    },
+    {
+      "epoch": 0.04558333333333333,
+      "grad_norm": 38.75,
+      "grad_norm_var": 204.03326822916668,
+      "learning_rate": 0.0001,
+      "loss": 7.1777,
+      "loss/crossentropy": 1.9395600646734237,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719443194568157,
+      "step": 5470
+    },
+    {
+      "epoch": 0.04566666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 197.82890625,
+      "learning_rate": 0.0001,
+      "loss": 7.3162,
+      "loss/crossentropy": 1.8322873935103416,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17720702290534973,
+      "step": 5480
+    },
+    {
+      "epoch": 0.04575,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.2583333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.188,
+      "loss/crossentropy": 1.9261434845626355,
+      "loss/hidden": 3.338671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18068330511450767,
+      "step": 5490
+    },
+    {
+      "epoch": 0.04583333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.4999801304737034e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3305,
+      "loss/crossentropy": 1.8927360117435454,
+      "loss/hidden": 3.36953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1788069920614362,
+      "step": 5500
+    },
+    {
+      "epoch": 0.04591666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 7.321164001103054e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3202,
+      "loss/crossentropy": 1.9236746475100517,
+      "loss/hidden": 3.2640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17970658484846352,
+      "step": 5510
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.2517332698401866e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2784,
+      "loss/crossentropy": 1.7933879487216473,
+      "loss/hidden": 3.293359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16929342793300747,
+      "step": 5520
+    },
+    {
+      "epoch": 0.04608333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.5488932291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0643,
+      "loss/crossentropy": 1.9319801151752471,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735060378909111,
+      "step": 5530
+    },
+    {
+      "epoch": 0.04616666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.5353515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2155,
+      "loss/crossentropy": 2.0292537212371826,
+      "loss/hidden": 3.48203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2000424973666668,
+      "step": 5540
+    },
+    {
+      "epoch": 0.04625,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.726497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3183,
+      "loss/crossentropy": 2.046224331855774,
+      "loss/hidden": 3.470703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20262836031615733,
+      "step": 5550
+    },
+    {
+      "epoch": 0.04633333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.4666015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1892,
+      "loss/crossentropy": 1.9176844008266927,
+      "loss/hidden": 3.408984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835050239227712,
+      "step": 5560
+    },
+    {
+      "epoch": 0.04641666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.359594487167984e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.25,
+      "loss/crossentropy": 1.9457813799381256,
+      "loss/hidden": 3.419921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19265421126037835,
+      "step": 5570
+    },
+    {
+      "epoch": 0.0465,
+      "grad_norm": 29.25,
+      "grad_norm_var": 9.5369140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0833,
+      "loss/crossentropy": 1.8953604146838188,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19760203436017038,
+      "step": 5580
+    },
+    {
+      "epoch": 0.04658333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.1322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2633,
+      "loss/crossentropy": 2.192387434840202,
+      "loss/hidden": 3.40078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20997173935174943,
+      "step": 5590
+    },
+    {
+      "epoch": 0.04666666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.9613932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1848,
+      "loss/crossentropy": 1.9025974899530411,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18724445197731257,
+      "step": 5600
+    },
+    {
+      "epoch": 0.04675,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.4145833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2077,
+      "loss/crossentropy": 1.8980001240968705,
+      "loss/hidden": 3.425,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18669269513338804,
+      "step": 5610
+    },
+    {
+      "epoch": 0.04683333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 157.37682291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3404,
+      "loss/crossentropy": 2.0061571091413497,
+      "loss/hidden": 3.483984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119721319526434,
+      "step": 5620
+    },
+    {
+      "epoch": 0.04691666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.900455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.307,
+      "loss/crossentropy": 1.9688873529434203,
+      "loss/hidden": 3.325,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1799240939319134,
+      "step": 5630
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.467708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1158,
+      "loss/crossentropy": 1.9035341456532477,
+      "loss/hidden": 3.463671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894345859065652,
+      "step": 5640
+    },
+    {
+      "epoch": 0.04708333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.3541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2714,
+      "loss/crossentropy": 1.98752211779356,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18840861935168504,
+      "step": 5650
+    },
+    {
+      "epoch": 0.04716666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.378580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.124,
+      "loss/crossentropy": 1.8570011980831622,
+      "loss/hidden": 3.327734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17682648468762635,
+      "step": 5660
+    },
+    {
+      "epoch": 0.04725,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.8770182291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1694,
+      "loss/crossentropy": 1.944897037744522,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17851272821426392,
+      "step": 5670
+    },
+    {
+      "epoch": 0.04733333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 6.12890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2989,
+      "loss/crossentropy": 2.0170719534158708,
+      "loss/hidden": 3.38203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19368235394358635,
+      "step": 5680
+    },
+    {
+      "epoch": 0.04741666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.084141647551967e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2023,
+      "loss/crossentropy": 1.8989535093307495,
+      "loss/hidden": 3.429296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889783263206482,
+      "step": 5690
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.012955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2679,
+      "loss/crossentropy": 1.937063890695572,
+      "loss/hidden": 3.41640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2034616844728589,
+      "step": 5700
+    },
+    {
+      "epoch": 0.04758333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.137955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1541,
+      "loss/crossentropy": 1.8730784460902214,
+      "loss/hidden": 3.303125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18296674843877553,
+      "step": 5710
+    },
+    {
+      "epoch": 0.04766666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 50.691080729166664,
+      "learning_rate": 0.0001,
+      "loss": 7.2326,
+      "loss/crossentropy": 1.9757542133331298,
+      "loss/hidden": 3.299609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815430372953415,
+      "step": 5720
+    },
+    {
+      "epoch": 0.04775,
+      "grad_norm": 27.5,
+      "grad_norm_var": 76.8822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2095,
+      "loss/crossentropy": 1.8371678739786148,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1928836090490222,
+      "step": 5730
+    },
+    {
+      "epoch": 0.04783333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 7.036874271525002e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.315,
+      "loss/crossentropy": 1.9386666312813758,
+      "loss/hidden": 3.57734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1786883294582367,
+      "step": 5740
+    },
+    {
+      "epoch": 0.04791666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.572330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.112,
+      "loss/crossentropy": 1.8475066609680653,
+      "loss/hidden": 3.22890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1685674911364913,
+      "step": 5750
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.036393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2146,
+      "loss/crossentropy": 1.8362929932773113,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18263157112523914,
+      "step": 5760
+    },
+    {
+      "epoch": 0.04808333333333333,
+      "grad_norm": 41.0,
+      "grad_norm_var": 5.0841416496613523e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2376,
+      "loss/crossentropy": 1.992903184890747,
+      "loss/hidden": 3.325,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18581640850752593,
+      "step": 5770
+    },
+    {
+      "epoch": 0.04816666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.0841416490077395e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1021,
+      "loss/crossentropy": 1.8959848627448082,
+      "loss/hidden": 3.447265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1710287369787693,
+      "step": 5780
+    },
+    {
+      "epoch": 0.04825,
+      "grad_norm": 35.0,
+      "grad_norm_var": 289.79993489583336,
+      "learning_rate": 0.0001,
+      "loss": 7.2732,
+      "loss/crossentropy": 1.9366097539663314,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18170985132455825,
+      "step": 5790
+    },
+    {
+      "epoch": 0.04833333333333333,
+      "grad_norm": 54.5,
+      "grad_norm_var": 34.83639322916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2101,
+      "loss/crossentropy": 2.073981301486492,
+      "loss/hidden": 3.446875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20776797477155923,
+      "step": 5800
+    },
+    {
+      "epoch": 0.04841666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 35.93515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2147,
+      "loss/crossentropy": 1.8804841123521328,
+      "loss/hidden": 3.371484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16808632388710976,
+      "step": 5810
+    },
+    {
+      "epoch": 0.0485,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.8875,
+      "learning_rate": 0.0001,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.9499920338392258,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18369101714342834,
+      "step": 5820
+    },
+    {
+      "epoch": 0.04858333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.4247395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2225,
+      "loss/crossentropy": 1.8653924077749253,
+      "loss/hidden": 3.526171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19181002490222454,
+      "step": 5830
+    },
+    {
+      "epoch": 0.048666666666666664,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.1439318802114807e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3217,
+      "loss/crossentropy": 1.969824068248272,
+      "loss/hidden": 3.58828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20675076451152563,
+      "step": 5840
+    },
+    {
+      "epoch": 0.04875,
+      "grad_norm": 38.25,
+      "grad_norm_var": 5.338541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.141,
+      "loss/crossentropy": 1.9277636885643006,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18821494542062284,
+      "step": 5850
+    },
+    {
+      "epoch": 0.04883333333333333,
+      "grad_norm": 3019898880.0,
+      "grad_norm_var": 5.6998681567140256e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2917,
+      "loss/crossentropy": 1.8802243813872337,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18886747639626264,
+      "step": 5860
+    },
+    {
+      "epoch": 0.048916666666666664,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.6998681553613626e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2099,
+      "loss/crossentropy": 2.0238874942064284,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17390586640685796,
+      "step": 5870
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 37.5,
+      "grad_norm_var": 9.881354820405691e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2301,
+      "loss/crossentropy": 2.0223625183105467,
+      "loss/hidden": 3.409765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19623048808425664,
+      "step": 5880
+    },
+    {
+      "epoch": 0.04908333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 9.881354818210497e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1501,
+      "loss/crossentropy": 1.8737989097833634,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16763302199542524,
+      "step": 5890
+    },
+    {
+      "epoch": 0.049166666666666664,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.7712890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2475,
+      "loss/crossentropy": 1.8810926377773285,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1831731954589486,
+      "step": 5900
+    },
+    {
+      "epoch": 0.04925,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.748893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2608,
+      "loss/crossentropy": 2.062958613038063,
+      "loss/hidden": 3.55390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21889359913766385,
+      "step": 5910
+    },
+    {
+      "epoch": 0.04933333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 1021.2436848958333,
+      "learning_rate": 0.0001,
+      "loss": 7.2497,
+      "loss/crossentropy": 1.898280094563961,
+      "loss/hidden": 3.454296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18923573587089776,
+      "step": 5920
+    },
+    {
+      "epoch": 0.049416666666666664,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1012.7455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.283,
+      "loss/crossentropy": 2.062563756108284,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18664450161159038,
+      "step": 5930
+    },
+    {
+      "epoch": 0.0495,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.758214050691854e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.377,
+      "loss/crossentropy": 1.9490419641137122,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910331157967448,
+      "step": 5940
+    },
+    {
+      "epoch": 0.04958333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.758214053740413e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.263,
+      "loss/crossentropy": 1.9519288882613182,
+      "loss/hidden": 3.376953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19702470256015658,
+      "step": 5950
+    },
+    {
+      "epoch": 0.049666666666666665,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.3080729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3382,
+      "loss/crossentropy": 2.08356169462204,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919506683945656,
+      "step": 5960
+    },
+    {
+      "epoch": 0.04975,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.986393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2119,
+      "loss/crossentropy": 1.8876198440790177,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18486291356384754,
+      "step": 5970
+    },
+    {
+      "epoch": 0.049833333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.50390625,
+      "learning_rate": 0.0001,
+      "loss": 7.162,
+      "loss/crossentropy": 1.9698369055986404,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18667118791490794,
+      "step": 5980
+    },
+    {
+      "epoch": 0.049916666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 53.614322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3357,
+      "loss/crossentropy": 2.067339900135994,
+      "loss/hidden": 3.45859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2117473976686597,
+      "step": 5990
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 31.0,
+      "grad_norm_var": 45.57890625,
+      "learning_rate": 0.0001,
+      "loss": 7.195,
+      "loss/crossentropy": 1.9594978839159012,
+      "loss/hidden": 3.250390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1724390959367156,
+      "step": 6000
+    },
+    {
+      "epoch": 0.050083333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.4427083333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1326,
+      "loss/crossentropy": 1.8880977675318718,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18454405162483453,
+      "step": 6010
+    },
+    {
+      "epoch": 0.050166666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.503599518374517e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3398,
+      "loss/crossentropy": 2.040569248795509,
+      "loss/hidden": 3.351171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19138383716344834,
+      "step": 6020
+    },
+    {
+      "epoch": 0.05025,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.85390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1766,
+      "loss/crossentropy": 2.0133946374058724,
+      "loss/hidden": 3.38984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18018098138272762,
+      "step": 6030
+    },
+    {
+      "epoch": 0.050333333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.374739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.4233,
+      "loss/crossentropy": 2.0365489065647124,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943318447098136,
+      "step": 6040
+    },
+    {
+      "epoch": 0.050416666666666665,
+      "grad_norm": 38.0,
+      "grad_norm_var": 6.753580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1707,
+      "loss/crossentropy": 2.019296956062317,
+      "loss/hidden": 3.463671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1895071506500244,
+      "step": 6050
+    },
+    {
+      "epoch": 0.0505,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.6509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1773,
+      "loss/crossentropy": 1.9771725565195084,
+      "loss/hidden": 3.310546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18213878907263278,
+      "step": 6060
+    },
+    {
+      "epoch": 0.050583333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.2442057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0858,
+      "loss/crossentropy": 1.9022766441106795,
+      "loss/hidden": 3.548828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20166466273367406,
+      "step": 6070
+    },
+    {
+      "epoch": 0.050666666666666665,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.9160807291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1795,
+      "loss/crossentropy": 1.961769598722458,
+      "loss/hidden": 3.4546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1797276422381401,
+      "step": 6080
+    },
+    {
+      "epoch": 0.05075,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.016666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1952,
+      "loss/crossentropy": 1.9573363393545151,
+      "loss/hidden": 3.24765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16141145564615728,
+      "step": 6090
+    },
+    {
+      "epoch": 0.050833333333333335,
+      "grad_norm": 35.75,
+      "grad_norm_var": 140.92545572916666,
+      "learning_rate": 0.0001,
+      "loss": 7.2033,
+      "loss/crossentropy": 1.8590539634227752,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18147912044078113,
+      "step": 6100
+    },
+    {
+      "epoch": 0.050916666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.3018772934252755e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3202,
+      "loss/crossentropy": 2.0075626581907273,
+      "loss/hidden": 3.6578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2295171281322837,
+      "step": 6110
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 44.0,
+      "grad_norm_var": 18.712434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1691,
+      "loss/crossentropy": 1.8493891522288322,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17705651726573707,
+      "step": 6120
+    },
+    {
+      "epoch": 0.051083333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 19.46015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1577,
+      "loss/crossentropy": 2.1470450803637506,
+      "loss/hidden": 3.2515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17646269779652357,
+      "step": 6130
+    },
+    {
+      "epoch": 0.051166666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.367708333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2079,
+      "loss/crossentropy": 1.9435604512691498,
+      "loss/hidden": 3.409375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852132784202695,
+      "step": 6140
+    },
+    {
+      "epoch": 0.05125,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.164322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2506,
+      "loss/crossentropy": 1.8571129024028779,
+      "loss/hidden": 3.533984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20155182387679815,
+      "step": 6150
+    },
+    {
+      "epoch": 0.051333333333333335,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.8613932291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1611,
+      "loss/crossentropy": 1.930855180323124,
+      "loss/hidden": 3.4859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17727227360010148,
+      "step": 6160
+    },
+    {
+      "epoch": 0.051416666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 4.572916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2118,
+      "loss/crossentropy": 1.973878675699234,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19820086359977723,
+      "step": 6170
+    },
+    {
+      "epoch": 0.0515,
+      "grad_norm": 29.875,
+      "grad_norm_var": 7.744205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3277,
+      "loss/crossentropy": 2.0238889157772064,
+      "loss/hidden": 3.530078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2186591826379299,
+      "step": 6180
+    },
+    {
+      "epoch": 0.051583333333333335,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.8692057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1675,
+      "loss/crossentropy": 1.9423449069261551,
+      "loss/hidden": 3.43515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19517161659896373,
+      "step": 6190
+    },
+    {
+      "epoch": 0.051666666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 11.745572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3246,
+      "loss/crossentropy": 2.0157795667648317,
+      "loss/hidden": 3.300390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19794415701180695,
+      "step": 6200
+    },
+    {
+      "epoch": 0.05175,
+      "grad_norm": 31.0,
+      "grad_norm_var": 73.48723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1391,
+      "loss/crossentropy": 1.8031481310725213,
+      "loss/hidden": 3.5703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18407038692384958,
+      "step": 6210
+    },
+    {
+      "epoch": 0.051833333333333335,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.1051432291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1885,
+      "loss/crossentropy": 2.1037486374378203,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19484200291335582,
+      "step": 6220
+    },
+    {
+      "epoch": 0.051916666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.786393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2227,
+      "loss/crossentropy": 1.926634880900383,
+      "loss/hidden": 3.423828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19628361649811268,
+      "step": 6230
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2018229166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1722,
+      "loss/crossentropy": 2.1934867203235626,
+      "loss/hidden": 3.428125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19669417291879654,
+      "step": 6240
+    },
+    {
+      "epoch": 0.052083333333333336,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.0205729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2297,
+      "loss/crossentropy": 1.886684399843216,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18135682679712772,
+      "step": 6250
+    },
+    {
+      "epoch": 0.05216666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.359375,
+      "learning_rate": 0.0001,
+      "loss": 7.2224,
+      "loss/crossentropy": 1.9629553347826003,
+      "loss/hidden": 3.3953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19749300181865692,
+      "step": 6260
+    },
+    {
+      "epoch": 0.05225,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.217782107599537e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3326,
+      "loss/crossentropy": 2.1625852316617964,
+      "loss/hidden": 3.262890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18893173113465309,
+      "step": 6270
+    },
+    {
+      "epoch": 0.052333333333333336,
+      "grad_norm": 29.75,
+      "grad_norm_var": 6.217781961524684e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2523,
+      "loss/crossentropy": 1.9883932605385781,
+      "loss/hidden": 3.46796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21924990750849246,
+      "step": 6280
+    },
+    {
+      "epoch": 0.05241666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 23.372916666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1854,
+      "loss/crossentropy": 2.079185092449188,
+      "loss/hidden": 3.28671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18457219656556845,
+      "step": 6290
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 33.0,
+      "grad_norm_var": 19.95390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2171,
+      "loss/crossentropy": 1.9342011585831642,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18812895119190215,
+      "step": 6300
+    },
+    {
+      "epoch": 0.052583333333333336,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.0369140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1543,
+      "loss/crossentropy": 1.9667259231209755,
+      "loss/hidden": 3.38359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891764525324106,
+      "step": 6310
+    },
+    {
+      "epoch": 0.05266666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.3833333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1719,
+      "loss/crossentropy": 1.883527809381485,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18346853349357845,
+      "step": 6320
+    },
+    {
+      "epoch": 0.05275,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.1529214870959075e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.277,
+      "loss/crossentropy": 1.9553183883428573,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17686478048563004,
+      "step": 6330
+    },
+    {
+      "epoch": 0.052833333333333336,
+      "grad_norm": 38.75,
+      "grad_norm_var": 1.1529214856642518e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3041,
+      "loss/crossentropy": 1.9471174091100694,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17402494940906763,
+      "step": 6340
+    },
+    {
+      "epoch": 0.05291666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 25.320572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.161,
+      "loss/crossentropy": 1.9604827284812927,
+      "loss/hidden": 3.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19619520641863347,
+      "step": 6350
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 29.375,
+      "grad_norm_var": 1.1171223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1766,
+      "loss/crossentropy": 2.016821599006653,
+      "loss/hidden": 3.36875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20081317238509655,
+      "step": 6360
+    },
+    {
+      "epoch": 0.05308333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.622330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3007,
+      "loss/crossentropy": 2.0598457247018813,
+      "loss/hidden": 3.478515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20236392691731453,
+      "step": 6370
+    },
+    {
+      "epoch": 0.05316666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.2853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2402,
+      "loss/crossentropy": 1.9487939864397048,
+      "loss/hidden": 3.283984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17032433431595564,
+      "step": 6380
+    },
+    {
+      "epoch": 0.05325,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.0697265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2519,
+      "loss/crossentropy": 1.79762829169631,
+      "loss/hidden": 3.334765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17330624535679817,
+      "step": 6390
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.148958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2327,
+      "loss/crossentropy": 1.8139575704932214,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19837790299206973,
+      "step": 6400
+    },
+    {
+      "epoch": 0.05341666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.787955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1422,
+      "loss/crossentropy": 1.8317793741822244,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19687960278242828,
+      "step": 6410
+    },
+    {
+      "epoch": 0.0535,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.9747395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2303,
+      "loss/crossentropy": 1.8772493988275527,
+      "loss/hidden": 3.4328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18726407140493392,
+      "step": 6420
+    },
+    {
+      "epoch": 0.05358333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.876822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2184,
+      "loss/crossentropy": 1.924004825949669,
+      "loss/hidden": 3.4578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18612550031393765,
+      "step": 6430
+    },
+    {
+      "epoch": 0.05366666666666667,
+      "grad_norm": 55.0,
+      "grad_norm_var": 38.6572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2838,
+      "loss/crossentropy": 1.971806037425995,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19059366807341577,
+      "step": 6440
+    },
+    {
+      "epoch": 0.05375,
+      "grad_norm": 33.25,
+      "grad_norm_var": 31.5244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2897,
+      "loss/crossentropy": 2.0371891140937803,
+      "loss/hidden": 3.419921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19689792413264512,
+      "step": 6450
+    },
+    {
+      "epoch": 0.05383333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 7.057291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1589,
+      "loss/crossentropy": 1.815991611778736,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18021370992064475,
+      "step": 6460
+    },
+    {
+      "epoch": 0.05391666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.139518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2613,
+      "loss/crossentropy": 1.9856169998645783,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19363545887172223,
+      "step": 6470
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 75.5,
+      "grad_norm_var": 8467.976041666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3998,
+      "loss/crossentropy": 2.0012280732393264,
+      "loss/hidden": 3.42421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21550402007997035,
+      "step": 6480
+    },
+    {
+      "epoch": 0.05408333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 8393.768489583334,
+      "learning_rate": 0.0001,
+      "loss": 7.1939,
+      "loss/crossentropy": 1.9625369042158127,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19765603579580784,
+      "step": 6490
+    },
+    {
+      "epoch": 0.05416666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.258333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2261,
+      "loss/crossentropy": 1.9609766840934753,
+      "loss/hidden": 3.34609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17874824199825526,
+      "step": 6500
+    },
+    {
+      "epoch": 0.05425,
+      "grad_norm": 40.75,
+      "grad_norm_var": 39.4853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1973,
+      "loss/crossentropy": 2.050165778398514,
+      "loss/hidden": 3.35390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17521476447582246,
+      "step": 6510
+    },
+    {
+      "epoch": 0.05433333333333333,
+      "grad_norm": 29.125,
+      "grad_norm_var": 59.44583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.194,
+      "loss/crossentropy": 1.8978053271770476,
+      "loss/hidden": 3.405078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1794705156236887,
+      "step": 6520
+    },
+    {
+      "epoch": 0.05441666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.263541147958615e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3198,
+      "loss/crossentropy": 2.025116375088692,
+      "loss/hidden": 3.44375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18641012702137233,
+      "step": 6530
+    },
+    {
+      "epoch": 0.0545,
+      "grad_norm": 70.5,
+      "grad_norm_var": 1.2635411437105544e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1999,
+      "loss/crossentropy": 1.8800811529159547,
+      "loss/hidden": 3.541015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20000641271471978,
+      "step": 6540
+    },
+    {
+      "epoch": 0.05458333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 366.4457682291667,
+      "learning_rate": 0.0001,
+      "loss": 7.2098,
+      "loss/crossentropy": 1.88163965344429,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18751533031463624,
+      "step": 6550
+    },
+    {
+      "epoch": 0.05466666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 13.5025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1116,
+      "loss/crossentropy": 1.8538421645760537,
+      "loss/hidden": 3.262890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17072358094155787,
+      "step": 6560
+    },
+    {
+      "epoch": 0.05475,
+      "grad_norm": 36.25,
+      "grad_norm_var": 16.593684895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1972,
+      "loss/crossentropy": 1.9755418807268144,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18982569519430398,
+      "step": 6570
+    },
+    {
+      "epoch": 0.05483333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 15.738997395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1361,
+      "loss/crossentropy": 1.706727545708418,
+      "loss/hidden": 3.405859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18362588211894035,
+      "step": 6580
+    },
+    {
+      "epoch": 0.05491666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.137955729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.247,
+      "loss/crossentropy": 1.8340010598301888,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1724576648324728,
+      "step": 6590
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 40.25,
+      "grad_norm_var": 13.128580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.3337,
+      "loss/crossentropy": 1.93518505692482,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17210706733167172,
+      "step": 6600
+    },
+    {
+      "epoch": 0.05508333333333333,
+      "grad_norm": 41.5,
+      "grad_norm_var": 1.3990009827900785e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.276,
+      "loss/crossentropy": 1.979572981595993,
+      "loss/hidden": 3.3390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19375349935144187,
+      "step": 6610
+    },
+    {
+      "epoch": 0.05516666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.3990009825535199e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1396,
+      "loss/crossentropy": 1.8551251679658889,
+      "loss/hidden": 3.466015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1853062679991126,
+      "step": 6620
+    },
+    {
+      "epoch": 0.05525,
+      "grad_norm": 38.0,
+      "grad_norm_var": 27.79765625,
+      "learning_rate": 0.0001,
+      "loss": 7.4334,
+      "loss/crossentropy": 1.956180915236473,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2130817821249366,
+      "step": 6630
+    },
+    {
+      "epoch": 0.05533333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 35.608333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1717,
+      "loss/crossentropy": 1.8848136879503727,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17505101822316646,
+      "step": 6640
+    },
+    {
+      "epoch": 0.05541666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 24.198958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2696,
+      "loss/crossentropy": 2.1142563939094545,
+      "loss/hidden": 3.303125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016379315406084,
+      "step": 6650
+    },
+    {
+      "epoch": 0.0555,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.18923175838631e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2787,
+      "loss/crossentropy": 1.877379097044468,
+      "loss/hidden": 3.28671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17101834136992694,
+      "step": 6660
+    },
+    {
+      "epoch": 0.05558333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 31.5,
+      "learning_rate": 0.0001,
+      "loss": 7.1617,
+      "loss/crossentropy": 1.967728652060032,
+      "loss/hidden": 3.2625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16908176708966494,
+      "step": 6670
+    },
+    {
+      "epoch": 0.05566666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 10.295833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1966,
+      "loss/crossentropy": 1.9864210113883018,
+      "loss/hidden": 3.312890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17814714480191468,
+      "step": 6680
+    },
+    {
+      "epoch": 0.05575,
+      "grad_norm": 30.25,
+      "grad_norm_var": 8.192643229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2298,
+      "loss/crossentropy": 1.9852227136492728,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18118114713579417,
+      "step": 6690
+    },
+    {
+      "epoch": 0.05583333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.3362978299559725e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2515,
+      "loss/crossentropy": 1.9546830110251903,
+      "loss/hidden": 3.516796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20408000089228154,
+      "step": 6700
+    },
+    {
+      "epoch": 0.05591666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 1.0178463309959793e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.311,
+      "loss/crossentropy": 1.8227561213076116,
+      "loss/hidden": 3.466015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1906059012748301,
+      "step": 6710
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.212434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1884,
+      "loss/crossentropy": 1.8481850996613503,
+      "loss/hidden": 3.460546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1807496551424265,
+      "step": 6720
+    },
+    {
+      "epoch": 0.05608333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.244752697754334e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3472,
+      "loss/crossentropy": 2.0029425621032715,
+      "loss/hidden": 3.3171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18047370873391627,
+      "step": 6730
+    },
+    {
+      "epoch": 0.05616666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.2447526969454625e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2428,
+      "loss/crossentropy": 2.070193290710449,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19471360072493554,
+      "step": 6740
+    },
+    {
+      "epoch": 0.05625,
+      "grad_norm": 34.75,
+      "grad_norm_var": 10.155143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2488,
+      "loss/crossentropy": 1.9138814441859722,
+      "loss/hidden": 3.458984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1864965006709099,
+      "step": 6750
+    },
+    {
+      "epoch": 0.05633333333333333,
+      "grad_norm": 57.25,
+      "grad_norm_var": 65.1634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2888,
+      "loss/crossentropy": 1.9506281644105912,
+      "loss/hidden": 3.5109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21938262470066547,
+      "step": 6760
+    },
+    {
+      "epoch": 0.056416666666666664,
+      "grad_norm": 31.75,
+      "grad_norm_var": 72.72493489583333,
+      "learning_rate": 0.0001,
+      "loss": 7.1287,
+      "loss/crossentropy": 1.8292874485254287,
+      "loss/hidden": 3.433984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17849782444536685,
+      "step": 6770
+    },
+    {
+      "epoch": 0.0565,
+      "grad_norm": 35.25,
+      "grad_norm_var": 1.7566853074176182e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2384,
+      "loss/crossentropy": 1.8679364308714868,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17026210222393273,
+      "step": 6780
+    },
+    {
+      "epoch": 0.05658333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.7566852899057772e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2511,
+      "loss/crossentropy": 1.9325649484992027,
+      "loss/hidden": 3.3375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18963379934430122,
+      "step": 6790
+    },
+    {
+      "epoch": 0.056666666666666664,
+      "grad_norm": 31.125,
+      "grad_norm_var": 7281.924934895834,
+      "learning_rate": 0.0001,
+      "loss": 7.2677,
+      "loss/crossentropy": 1.9077018201351166,
+      "loss/hidden": 3.416015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19200860038399697,
+      "step": 6800
+    },
+    {
+      "epoch": 0.05675,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.996875,
+      "learning_rate": 0.0001,
+      "loss": 7.1384,
+      "loss/crossentropy": 1.9394373342394828,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18492530584335326,
+      "step": 6810
+    },
+    {
+      "epoch": 0.05683333333333333,
+      "grad_norm": 27.75,
+      "grad_norm_var": 5.395572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2212,
+      "loss/crossentropy": 1.9307297945022583,
+      "loss/hidden": 3.49609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20020147040486336,
+      "step": 6820
+    },
+    {
+      "epoch": 0.056916666666666664,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.728125,
+      "learning_rate": 0.0001,
+      "loss": 7.1808,
+      "loss/crossentropy": 1.8511271178722382,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18137127477675677,
+      "step": 6830
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 28.5,
+      "grad_norm_var": 21.429166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2759,
+      "loss/crossentropy": 2.0388415962457658,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18986355643719435,
+      "step": 6840
+    },
+    {
+      "epoch": 0.05708333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.610872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2423,
+      "loss/crossentropy": 2.036486195027828,
+      "loss/hidden": 3.365625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19451405573636293,
+      "step": 6850
+    },
+    {
+      "epoch": 0.057166666666666664,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.421875,
+      "learning_rate": 0.0001,
+      "loss": 7.3234,
+      "loss/crossentropy": 1.9453188359737397,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18416350986808538,
+      "step": 6860
+    },
+    {
+      "epoch": 0.05725,
+      "grad_norm": 30.875,
+      "grad_norm_var": 38.827083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2471,
+      "loss/crossentropy": 2.118215595185757,
+      "loss/hidden": 3.2765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18926584757864476,
+      "step": 6870
+    },
+    {
+      "epoch": 0.05733333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 42.50462239583333,
+      "learning_rate": 0.0001,
+      "loss": 7.1489,
+      "loss/crossentropy": 2.0976534694433213,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18597040716558694,
+      "step": 6880
+    },
+    {
+      "epoch": 0.057416666666666664,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.66640625,
+      "learning_rate": 0.0001,
+      "loss": 7.229,
+      "loss/crossentropy": 2.0895703792572022,
+      "loss/hidden": 3.451953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18702018074691296,
+      "step": 6890
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.1205729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1657,
+      "loss/crossentropy": 2.014855782687664,
+      "loss/hidden": 3.44296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19445941504091024,
+      "step": 6900
+    },
+    {
+      "epoch": 0.057583333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.93125,
+      "learning_rate": 0.0001,
+      "loss": 7.2919,
+      "loss/crossentropy": 1.94414019882679,
+      "loss/hidden": 3.447265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919761449098587,
+      "step": 6910
+    },
+    {
+      "epoch": 0.057666666666666665,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.228580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3015,
+      "loss/crossentropy": 1.9129973486065865,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791930379346013,
+      "step": 6920
+    },
+    {
+      "epoch": 0.05775,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.5739081030638426e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2774,
+      "loss/crossentropy": 2.124286249279976,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18690722845494748,
+      "step": 6930
+    },
+    {
+      "epoch": 0.057833333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.784375,
+      "learning_rate": 0.0001,
+      "loss": 7.1534,
+      "loss/crossentropy": 1.817179811000824,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18679010756313802,
+      "step": 6940
+    },
+    {
+      "epoch": 0.057916666666666665,
+      "grad_norm": 30.125,
+      "grad_norm_var": 238.24557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2343,
+      "loss/crossentropy": 2.0582338005304335,
+      "loss/hidden": 3.34140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17898097299039364,
+      "step": 6950
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 30.5,
+      "grad_norm_var": 11.303580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2738,
+      "loss/crossentropy": 2.0119086511433126,
+      "loss/hidden": 3.342578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978511307388544,
+      "step": 6960
+    },
+    {
+      "epoch": 0.058083333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.4184895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2317,
+      "loss/crossentropy": 1.9403888657689095,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19264563638716936,
+      "step": 6970
+    },
+    {
+      "epoch": 0.058166666666666665,
+      "grad_norm": 35.75,
+      "grad_norm_var": 2.44140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2749,
+      "loss/crossentropy": 1.9770619094371795,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889339266344905,
+      "step": 6980
+    },
+    {
+      "epoch": 0.05825,
+      "grad_norm": 28.875,
+      "grad_norm_var": 14.030989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2932,
+      "loss/crossentropy": 1.8545608639717102,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1864480933174491,
+      "step": 6990
+    },
+    {
+      "epoch": 0.058333333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.9218098958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0992,
+      "loss/crossentropy": 1.830371941626072,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17968840897083282,
+      "step": 7000
+    },
+    {
+      "epoch": 0.058416666666666665,
+      "grad_norm": 28.75,
+      "grad_norm_var": 11.968684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2045,
+      "loss/crossentropy": 1.9251724004745483,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17747993357479572,
+      "step": 7010
+    },
+    {
+      "epoch": 0.0585,
+      "grad_norm": 30.125,
+      "grad_norm_var": 12.1087890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1432,
+      "loss/crossentropy": 1.8945713669061661,
+      "loss/hidden": 3.40546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1888335794210434,
+      "step": 7020
+    },
+    {
+      "epoch": 0.058583333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 8.276822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1716,
+      "loss/crossentropy": 1.9128420159220696,
+      "loss/hidden": 3.19453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1727126305922866,
+      "step": 7030
+    },
+    {
+      "epoch": 0.058666666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 17.006705729166665,
+      "learning_rate": 0.0001,
+      "loss": 7.1884,
+      "loss/crossentropy": 1.991888478398323,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19327655341476202,
+      "step": 7040
+    },
+    {
+      "epoch": 0.05875,
+      "grad_norm": 42.25,
+      "grad_norm_var": 12.85390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2039,
+      "loss/crossentropy": 1.845263669639826,
+      "loss/hidden": 3.456640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19932820042595267,
+      "step": 7050
+    },
+    {
+      "epoch": 0.058833333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5221.237434895834,
+      "learning_rate": 0.0001,
+      "loss": 7.3444,
+      "loss/crossentropy": 1.9406698502600193,
+      "loss/hidden": 3.4875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18545874748378993,
+      "step": 7060
+    },
+    {
+      "epoch": 0.058916666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 21.25390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3752,
+      "loss/crossentropy": 2.0439330741763113,
+      "loss/hidden": 3.38984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20598556995391845,
+      "step": 7070
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 31.25,
+      "grad_norm_var": 13.151497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1997,
+      "loss/crossentropy": 2.000889018177986,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18128069918602704,
+      "step": 7080
+    },
+    {
+      "epoch": 0.059083333333333335,
+      "grad_norm": 29.25,
+      "grad_norm_var": 13.904166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3386,
+      "loss/crossentropy": 1.9829541355371476,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19588420819491148,
+      "step": 7090
+    },
+    {
+      "epoch": 0.059166666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.495572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2047,
+      "loss/crossentropy": 1.88180633187294,
+      "loss/hidden": 3.605078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20174703951925038,
+      "step": 7100
+    },
+    {
+      "epoch": 0.05925,
+      "grad_norm": 32.5,
+      "grad_norm_var": 61.45390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2763,
+      "loss/crossentropy": 1.9542135357856751,
+      "loss/hidden": 3.489453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21102935615926982,
+      "step": 7110
+    },
+    {
+      "epoch": 0.059333333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 71.41458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1667,
+      "loss/crossentropy": 1.926981085538864,
+      "loss/hidden": 3.461328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17952323369681836,
+      "step": 7120
+    },
+    {
+      "epoch": 0.059416666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 38.1884765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3224,
+      "loss/crossentropy": 1.9933403447270392,
+      "loss/hidden": 3.466796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21018762681633235,
+      "step": 7130
+    },
+    {
+      "epoch": 0.0595,
+      "grad_norm": 34.25,
+      "grad_norm_var": 23.698893229166668,
+      "learning_rate": 0.0001,
+      "loss": 7.2229,
+      "loss/crossentropy": 1.8195350259542464,
+      "loss/hidden": 3.424609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1856760375201702,
+      "step": 7140
+    },
+    {
+      "epoch": 0.059583333333333335,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.039583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1395,
+      "loss/crossentropy": 2.0024550408124924,
+      "loss/hidden": 3.40390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1889307852834463,
+      "step": 7150
+    },
+    {
+      "epoch": 0.059666666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 7.2103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2664,
+      "loss/crossentropy": 2.0668145686388018,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19147251602262258,
+      "step": 7160
+    },
+    {
+      "epoch": 0.05975,
+      "grad_norm": 29.75,
+      "grad_norm_var": 18.46015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1747,
+      "loss/crossentropy": 2.0236642003059386,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18423411902040243,
+      "step": 7170
+    },
+    {
+      "epoch": 0.059833333333333336,
+      "grad_norm": 31.625,
+      "grad_norm_var": 15.653580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1739,
+      "loss/crossentropy": 2.0101078763604163,
+      "loss/hidden": 3.32109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18118085842579604,
+      "step": 7180
+    },
+    {
+      "epoch": 0.05991666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 25.6759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2501,
+      "loss/crossentropy": 2.0987291634082794,
+      "loss/hidden": 3.35859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19870839845389127,
+      "step": 7190
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 33.0,
+      "grad_norm_var": 26.830143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.232,
+      "loss/crossentropy": 2.0165331818163397,
+      "loss/hidden": 3.388671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20182917425408958,
+      "step": 7200
+    },
+    {
+      "epoch": 0.060083333333333336,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.4629557291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2123,
+      "loss/crossentropy": 1.9919808238744736,
+      "loss/hidden": 3.303515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857657887041569,
+      "step": 7210
+    },
+    {
+      "epoch": 0.06016666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.876822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1908,
+      "loss/crossentropy": 2.048959070444107,
+      "loss/hidden": 3.290234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18348812833428382,
+      "step": 7220
+    },
+    {
+      "epoch": 0.06025,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.8577473958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2264,
+      "loss/crossentropy": 1.9573696821928024,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977718524634838,
+      "step": 7230
+    },
+    {
+      "epoch": 0.060333333333333336,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.6363932291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1928,
+      "loss/crossentropy": 1.8912595480680465,
+      "loss/hidden": 3.4921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922957869246602,
+      "step": 7240
+    },
+    {
+      "epoch": 0.06041666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.995833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2126,
+      "loss/crossentropy": 1.852707712352276,
+      "loss/hidden": 3.3578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17833833750337363,
+      "step": 7250
+    },
+    {
+      "epoch": 0.0605,
+      "grad_norm": 27.375,
+      "grad_norm_var": 36.7791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2043,
+      "loss/crossentropy": 2.023664565384388,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.195465612411499,
+      "step": 7260
+    },
+    {
+      "epoch": 0.060583333333333336,
+      "grad_norm": 30.375,
+      "grad_norm_var": 42.77708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2333,
+      "loss/crossentropy": 1.9334395371377469,
+      "loss/hidden": 3.3234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18526536021381618,
+      "step": 7270
+    },
+    {
+      "epoch": 0.06066666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 652.9705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3917,
+      "loss/crossentropy": 2.0037595987319947,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1814060116186738,
+      "step": 7280
+    },
+    {
+      "epoch": 0.06075,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.891447032789868e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2725,
+      "loss/crossentropy": 2.023529142141342,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1940064739435911,
+      "step": 7290
+    },
+    {
+      "epoch": 0.060833333333333336,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.2143229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2438,
+      "loss/crossentropy": 1.9430765897035598,
+      "loss/hidden": 3.51640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1979274285957217,
+      "step": 7300
+    },
+    {
+      "epoch": 0.06091666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.655989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1017,
+      "loss/crossentropy": 1.8648948967456818,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17332021631300448,
+      "step": 7310
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.9671223958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1943,
+      "loss/crossentropy": 1.953578931093216,
+      "loss/hidden": 3.526953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22345543652772903,
+      "step": 7320
+    },
+    {
+      "epoch": 0.06108333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.653059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2887,
+      "loss/crossentropy": 1.9954859048128128,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19549989588558675,
+      "step": 7330
+    },
+    {
+      "epoch": 0.06116666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.7104166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.325,
+      "loss/crossentropy": 2.1271569937467576,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951669655740261,
+      "step": 7340
+    },
+    {
+      "epoch": 0.06125,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.9317057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2158,
+      "loss/crossentropy": 1.8807506158947944,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18400093652307986,
+      "step": 7350
+    },
+    {
+      "epoch": 0.06133333333333333,
+      "grad_norm": 35.75,
+      "grad_norm_var": 10.7962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.3313,
+      "loss/crossentropy": 1.98387304097414,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802327126264572,
+      "step": 7360
+    },
+    {
+      "epoch": 0.06141666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.9666015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2317,
+      "loss/crossentropy": 1.8400464594364165,
+      "loss/hidden": 3.471875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2012029906734824,
+      "step": 7370
+    },
+    {
+      "epoch": 0.0615,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.478125,
+      "learning_rate": 0.0001,
+      "loss": 7.336,
+      "loss/crossentropy": 1.9901346012949943,
+      "loss/hidden": 3.492578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21220925878733396,
+      "step": 7380
+    },
+    {
+      "epoch": 0.06158333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.945833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2569,
+      "loss/crossentropy": 2.025828644633293,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852064810693264,
+      "step": 7390
+    },
+    {
+      "epoch": 0.06166666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 11.0212890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1639,
+      "loss/crossentropy": 1.931120578944683,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19699512235820293,
+      "step": 7400
+    },
+    {
+      "epoch": 0.06175,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2259,
+      "loss/crossentropy": 1.9323313042521477,
+      "loss/hidden": 3.54140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18886004630476236,
+      "step": 7410
+    },
+    {
+      "epoch": 0.06183333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 6.25390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2594,
+      "loss/crossentropy": 1.9984138324856757,
+      "loss/hidden": 3.42734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.207680026255548,
+      "step": 7420
+    },
+    {
+      "epoch": 0.06191666666666667,
+      "grad_norm": 39.0,
+      "grad_norm_var": 7.705989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1694,
+      "loss/crossentropy": 1.9141117088496684,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18680279217660428,
+      "step": 7430
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 32.25,
+      "grad_norm_var": 31.816666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.227,
+      "loss/crossentropy": 1.915043169260025,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18126301970332862,
+      "step": 7440
+    },
+    {
+      "epoch": 0.06208333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 25.651041666666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2287,
+      "loss/crossentropy": 1.9978440970182418,
+      "loss/hidden": 3.35703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19587596897035836,
+      "step": 7450
+    },
+    {
+      "epoch": 0.06216666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.303580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1764,
+      "loss/crossentropy": 2.01900063380599,
+      "loss/hidden": 3.327734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18093273509293795,
+      "step": 7460
+    },
+    {
+      "epoch": 0.06225,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.61640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1635,
+      "loss/crossentropy": 1.9081674709916114,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18367233984172343,
+      "step": 7470
+    },
+    {
+      "epoch": 0.06233333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.0056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1685,
+      "loss/crossentropy": 1.9261738657951355,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18073464091867208,
+      "step": 7480
+    },
+    {
+      "epoch": 0.06241666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 99.3087890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1488,
+      "loss/crossentropy": 1.9166977405548096,
+      "loss/hidden": 3.420703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18761946000158786,
+      "step": 7490
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.8020833333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2591,
+      "loss/crossentropy": 1.902839483320713,
+      "loss/hidden": 3.453515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19028478916734456,
+      "step": 7500
+    },
+    {
+      "epoch": 0.06258333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.9697916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1728,
+      "loss/crossentropy": 1.6974174082279205,
+      "loss/hidden": 3.4421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1784145675599575,
+      "step": 7510
+    },
+    {
+      "epoch": 0.06266666666666666,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.8655598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1342,
+      "loss/crossentropy": 1.9428329408168792,
+      "loss/hidden": 3.394140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18656713459640742,
+      "step": 7520
+    },
+    {
+      "epoch": 0.06275,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.423958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2,
+      "loss/crossentropy": 1.8696777828037738,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18415667656809093,
+      "step": 7530
+    },
+    {
+      "epoch": 0.06283333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.8379557291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2881,
+      "loss/crossentropy": 1.9623609468340875,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18007479943335056,
+      "step": 7540
+    },
+    {
+      "epoch": 0.06291666666666666,
+      "grad_norm": 2801795072.0,
+      "grad_norm_var": 4.90628464974024e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2655,
+      "loss/crossentropy": 1.9910573005676269,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20073548033833505,
+      "step": 7550
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.9062846466465914e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2488,
+      "loss/crossentropy": 1.8616838872432708,
+      "loss/hidden": 3.50078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19058813713490963,
+      "step": 7560
+    },
+    {
+      "epoch": 0.06308333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.57265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2364,
+      "loss/crossentropy": 1.9160969465970994,
+      "loss/hidden": 3.48203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20963120143860578,
+      "step": 7570
+    },
+    {
+      "epoch": 0.06316666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.820572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.4052,
+      "loss/crossentropy": 1.90859245210886,
+      "loss/hidden": 3.458984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20211333967745304,
+      "step": 7580
+    },
+    {
+      "epoch": 0.06325,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.809309895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2456,
+      "loss/crossentropy": 1.828519704192877,
+      "loss/hidden": 3.40078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19046486187726258,
+      "step": 7590
+    },
+    {
+      "epoch": 0.06333333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.226822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1943,
+      "loss/crossentropy": 2.030736616253853,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21096080131828784,
+      "step": 7600
+    },
+    {
+      "epoch": 0.06341666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 4.0228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2782,
+      "loss/crossentropy": 2.016854166984558,
+      "loss/hidden": 3.348828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18759209662675858,
+      "step": 7610
+    },
+    {
+      "epoch": 0.0635,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.439322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1895,
+      "loss/crossentropy": 1.918825089931488,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19764372929930688,
+      "step": 7620
+    },
+    {
+      "epoch": 0.06358333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 5.9072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2783,
+      "loss/crossentropy": 1.9146866127848625,
+      "loss/hidden": 3.46640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974598715081811,
+      "step": 7630
+    },
+    {
+      "epoch": 0.06366666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.93125,
+      "learning_rate": 0.0001,
+      "loss": 7.2713,
+      "loss/crossentropy": 1.9532035797834397,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2067047307267785,
+      "step": 7640
+    },
+    {
+      "epoch": 0.06375,
+      "grad_norm": 32.5,
+      "grad_norm_var": 13.1259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1361,
+      "loss/crossentropy": 1.824920044839382,
+      "loss/hidden": 3.440234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19241070449352266,
+      "step": 7650
+    },
+    {
+      "epoch": 0.06383333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 10.99375,
+      "learning_rate": 0.0001,
+      "loss": 7.2802,
+      "loss/crossentropy": 1.7821342311799526,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18938454277813435,
+      "step": 7660
+    },
+    {
+      "epoch": 0.06391666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 52.74557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3821,
+      "loss/crossentropy": 2.130902296304703,
+      "loss/hidden": 3.3890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2177656836807728,
+      "step": 7670
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 32.0,
+      "grad_norm_var": 54.44895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1582,
+      "loss/crossentropy": 1.9051862180233001,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17711994536221026,
+      "step": 7680
+    },
+    {
+      "epoch": 0.06408333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 26.953580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3007,
+      "loss/crossentropy": 1.997354769706726,
+      "loss/hidden": 3.52265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18910952042788268,
+      "step": 7690
+    },
+    {
+      "epoch": 0.06416666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 9.301041666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3028,
+      "loss/crossentropy": 1.832273904979229,
+      "loss/hidden": 3.504296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21006521843373777,
+      "step": 7700
+    },
+    {
+      "epoch": 0.06425,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.3291666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3769,
+      "loss/crossentropy": 2.0843783155083657,
+      "loss/hidden": 3.465234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20692231114953757,
+      "step": 7710
+    },
+    {
+      "epoch": 0.06433333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.6322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2327,
+      "loss/crossentropy": 2.079830232262611,
+      "loss/hidden": 3.43828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18693204522132872,
+      "step": 7720
+    },
+    {
+      "epoch": 0.06441666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.6650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2392,
+      "loss/crossentropy": 1.9599878072738648,
+      "loss/hidden": 3.264453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17181740291416644,
+      "step": 7730
+    },
+    {
+      "epoch": 0.0645,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.8177083333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2603,
+      "loss/crossentropy": 1.9764328211545945,
+      "loss/hidden": 3.511328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19167867433279753,
+      "step": 7740
+    },
+    {
+      "epoch": 0.06458333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.98125,
+      "learning_rate": 0.0001,
+      "loss": 7.2676,
+      "loss/crossentropy": 1.9611605733633042,
+      "loss/hidden": 3.50546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19947379268705845,
+      "step": 7750
+    },
+    {
+      "epoch": 0.06466666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4520833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2259,
+      "loss/crossentropy": 2.044002892076969,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19694972168654204,
+      "step": 7760
+    },
+    {
+      "epoch": 0.06475,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.122330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1708,
+      "loss/crossentropy": 2.0414538815617562,
+      "loss/hidden": 3.237109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17013669721782207,
+      "step": 7770
+    },
+    {
+      "epoch": 0.06483333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.2395833333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.3133,
+      "loss/crossentropy": 1.910868799686432,
+      "loss/hidden": 3.420703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19090145360678434,
+      "step": 7780
+    },
+    {
+      "epoch": 0.06491666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 0.9747395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2692,
+      "loss/crossentropy": 2.0562462225556373,
+      "loss/hidden": 3.28203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18266476448625327,
+      "step": 7790
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.957291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3863,
+      "loss/crossentropy": 1.9470960587263106,
+      "loss/hidden": 3.483203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22827607877552508,
+      "step": 7800
+    },
+    {
+      "epoch": 0.06508333333333334,
+      "grad_norm": 38.5,
+      "grad_norm_var": 13.29765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2335,
+      "loss/crossentropy": 1.9175234675407409,
+      "loss/hidden": 3.426953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19259056225419044,
+      "step": 7810
+    },
+    {
+      "epoch": 0.06516666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 12.7603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2009,
+      "loss/crossentropy": 1.9584806606173515,
+      "loss/hidden": 3.4421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21137978304177524,
+      "step": 7820
+    },
+    {
+      "epoch": 0.06525,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.91875,
+      "learning_rate": 0.0001,
+      "loss": 7.2006,
+      "loss/crossentropy": 1.9486504584550857,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18704792261123657,
+      "step": 7830
+    },
+    {
+      "epoch": 0.06533333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.18125,
+      "learning_rate": 0.0001,
+      "loss": 7.2527,
+      "loss/crossentropy": 1.9827014431357384,
+      "loss/hidden": 3.440625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188728235848248,
+      "step": 7840
+    },
+    {
+      "epoch": 0.06541666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 86.52682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.39,
+      "loss/crossentropy": 2.0009460031986235,
+      "loss/hidden": 3.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1901438269764185,
+      "step": 7850
+    },
+    {
+      "epoch": 0.0655,
+      "grad_norm": 33.5,
+      "grad_norm_var": 82.5634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.282,
+      "loss/crossentropy": 2.0347262173891068,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961182001978159,
+      "step": 7860
+    },
+    {
+      "epoch": 0.06558333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.896875,
+      "learning_rate": 0.0001,
+      "loss": 7.1761,
+      "loss/crossentropy": 1.9007416464388371,
+      "loss/hidden": 3.43203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19023145977407693,
+      "step": 7870
+    },
+    {
+      "epoch": 0.06566666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.138541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2198,
+      "loss/crossentropy": 2.1901776075363157,
+      "loss/hidden": 3.4015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21192686930298804,
+      "step": 7880
+    },
+    {
+      "epoch": 0.06575,
+      "grad_norm": 31.875,
+      "grad_norm_var": 5.09375,
+      "learning_rate": 0.0001,
+      "loss": 7.1888,
+      "loss/crossentropy": 1.9310210451483727,
+      "loss/hidden": 3.37265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.176263870857656,
+      "step": 7890
+    },
+    {
+      "epoch": 0.06583333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.770833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2021,
+      "loss/crossentropy": 1.9472015351057053,
+      "loss/hidden": 3.2953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1718252243474126,
+      "step": 7900
+    },
+    {
+      "epoch": 0.06591666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 11.4072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1642,
+      "loss/crossentropy": 2.060355693101883,
+      "loss/hidden": 3.2875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17545824572443963,
+      "step": 7910
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.0473683717269925e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1745,
+      "loss/crossentropy": 1.7840346470475197,
+      "loss/hidden": 3.740625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19948329012840987,
+      "step": 7920
+    },
+    {
+      "epoch": 0.06608333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.0473683711683812e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3275,
+      "loss/crossentropy": 1.9387343034148217,
+      "loss/hidden": 3.476953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1954814039170742,
+      "step": 7930
+    },
+    {
+      "epoch": 0.06616666666666667,
+      "grad_norm": 36.25,
+      "grad_norm_var": 9.870572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1739,
+      "loss/crossentropy": 1.8542084500193596,
+      "loss/hidden": 3.39140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18380195926874876,
+      "step": 7940
+    },
+    {
+      "epoch": 0.06625,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.309309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2471,
+      "loss/crossentropy": 1.891501422226429,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21512426622211933,
+      "step": 7950
+    },
+    {
+      "epoch": 0.06633333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.7405598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1933,
+      "loss/crossentropy": 1.8226936116814614,
+      "loss/hidden": 3.441796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1818669240921736,
+      "step": 7960
+    },
+    {
+      "epoch": 0.06641666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 199.25625,
+      "learning_rate": 0.0001,
+      "loss": 7.2723,
+      "loss/crossentropy": 1.8292644947767258,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19608462285250425,
+      "step": 7970
+    },
+    {
+      "epoch": 0.0665,
+      "grad_norm": 32.75,
+      "grad_norm_var": 204.10149739583332,
+      "learning_rate": 0.0001,
+      "loss": 7.2325,
+      "loss/crossentropy": 1.9156912505626678,
+      "loss/hidden": 3.544140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19286852981895208,
+      "step": 7980
+    },
+    {
+      "epoch": 0.06658333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 28.8447265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1774,
+      "loss/crossentropy": 1.8666133657097816,
+      "loss/hidden": 3.385546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17876000609248877,
+      "step": 7990
+    },
+    {
+      "epoch": 0.06666666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.275,
+      "learning_rate": 0.0001,
+      "loss": 7.2216,
+      "loss/crossentropy": 1.841892847418785,
+      "loss/hidden": 3.421484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1745407897979021,
+      "step": 8000
+    },
+    {
+      "epoch": 0.06675,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.4702473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2601,
+      "loss/crossentropy": 1.8827014505863189,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20147160943597556,
+      "step": 8010
+    },
+    {
+      "epoch": 0.06683333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.549934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3408,
+      "loss/crossentropy": 1.9311093546450138,
+      "loss/hidden": 3.568359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22386483792215586,
+      "step": 8020
+    },
+    {
+      "epoch": 0.06691666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.2541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1497,
+      "loss/crossentropy": 1.8826589100062847,
+      "loss/hidden": 3.29921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17312360629439355,
+      "step": 8030
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.233268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2066,
+      "loss/crossentropy": 1.9970761120319367,
+      "loss/hidden": 3.319140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19054881893098355,
+      "step": 8040
+    },
+    {
+      "epoch": 0.06708333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 8652.215559895832,
+      "learning_rate": 0.0001,
+      "loss": 7.2576,
+      "loss/crossentropy": 1.9152855359017849,
+      "loss/hidden": 3.455078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18708332600072025,
+      "step": 8050
+    },
+    {
+      "epoch": 0.06716666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 5.430208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2357,
+      "loss/crossentropy": 2.031227025389671,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193635768070817,
+      "step": 8060
+    },
+    {
+      "epoch": 0.06725,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.4931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1251,
+      "loss/crossentropy": 1.8489420399069787,
+      "loss/hidden": 3.458984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17814766112715005,
+      "step": 8070
+    },
+    {
+      "epoch": 0.06733333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 27.78515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2487,
+      "loss/crossentropy": 1.884847517311573,
+      "loss/hidden": 3.404296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2042497692629695,
+      "step": 8080
+    },
+    {
+      "epoch": 0.06741666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 4.5035995046731245e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1985,
+      "loss/crossentropy": 1.8161820538342,
+      "loss/hidden": 3.344140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17485685739666224,
+      "step": 8090
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 42.5,
+      "grad_norm_var": 19.139583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1604,
+      "loss/crossentropy": 1.9721700698137283,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1950763339176774,
+      "step": 8100
+    },
+    {
+      "epoch": 0.06758333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 14.6416015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2636,
+      "loss/crossentropy": 2.0180334128439426,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20730137974023818,
+      "step": 8110
+    },
+    {
+      "epoch": 0.06766666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 22.159309895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.2743,
+      "loss/crossentropy": 2.1068934723734856,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977040534839034,
+      "step": 8120
+    },
+    {
+      "epoch": 0.06775,
+      "grad_norm": 30.875,
+      "grad_norm_var": 445.7035807291667,
+      "learning_rate": 0.0001,
+      "loss": 7.2164,
+      "loss/crossentropy": 1.9630880862474442,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19089617524296046,
+      "step": 8130
+    },
+    {
+      "epoch": 0.06783333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.3145182291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2163,
+      "loss/crossentropy": 1.8661000967025756,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19853502176702023,
+      "step": 8140
+    },
+    {
+      "epoch": 0.06791666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.31015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1223,
+      "loss/crossentropy": 1.9666982188820838,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1820853654295206,
+      "step": 8150
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 35.0,
+      "grad_norm_var": 29.7681640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2284,
+      "loss/crossentropy": 2.037832336127758,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19564964435994625,
+      "step": 8160
+    },
+    {
+      "epoch": 0.06808333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 27.930208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1623,
+      "loss/crossentropy": 1.8319560870528222,
+      "loss/hidden": 3.411328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18993047419935466,
+      "step": 8170
+    },
+    {
+      "epoch": 0.06816666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.486458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1666,
+      "loss/crossentropy": 1.9121002182364464,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18671723157167436,
+      "step": 8180
+    },
+    {
+      "epoch": 0.06825,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.5837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1998,
+      "loss/crossentropy": 1.9544257938861846,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18110576160252095,
+      "step": 8190
+    },
+    {
+      "epoch": 0.06833333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.4129557291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1602,
+      "loss/crossentropy": 1.8966913163661956,
+      "loss/hidden": 3.325390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17959749344736337,
+      "step": 8200
+    },
+    {
+      "epoch": 0.06841666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.2067057291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1619,
+      "loss/crossentropy": 1.9131664127111434,
+      "loss/hidden": 3.354296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17485106661915778,
+      "step": 8210
+    },
+    {
+      "epoch": 0.0685,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.4936848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2113,
+      "loss/crossentropy": 2.0453411146998404,
+      "loss/hidden": 3.2796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18515327256172895,
+      "step": 8220
+    },
+    {
+      "epoch": 0.06858333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.59375,
+      "learning_rate": 0.0001,
+      "loss": 7.2128,
+      "loss/crossentropy": 1.9506754860281945,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18393214661628007,
+      "step": 8230
+    },
+    {
+      "epoch": 0.06866666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.8,
+      "learning_rate": 0.0001,
+      "loss": 7.2449,
+      "loss/crossentropy": 1.9423611342906952,
+      "loss/hidden": 3.25078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17854795306921006,
+      "step": 8240
+    },
+    {
+      "epoch": 0.06875,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.6035807291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2422,
+      "loss/crossentropy": 1.9720477700233459,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188589053414762,
+      "step": 8250
+    },
+    {
+      "epoch": 0.06883333333333333,
+      "grad_norm": 123.0,
+      "grad_norm_var": 535.6372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1691,
+      "loss/crossentropy": 1.99368005245924,
+      "loss/hidden": 3.354296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1896060870960355,
+      "step": 8260
+    },
+    {
+      "epoch": 0.06891666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 521.7205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2283,
+      "loss/crossentropy": 1.896087047457695,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775512419641018,
+      "step": 8270
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 29.125,
+      "grad_norm_var": 78.86399739583334,
+      "learning_rate": 0.0001,
+      "loss": 7.2136,
+      "loss/crossentropy": 1.7727789789438249,
+      "loss/hidden": 3.475390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19561070688068866,
+      "step": 8280
+    },
+    {
+      "epoch": 0.06908333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 35.99583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.32,
+      "loss/crossentropy": 2.039402049779892,
+      "loss/hidden": 3.409765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993569979444146,
+      "step": 8290
+    },
+    {
+      "epoch": 0.06916666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 19.953125,
+      "learning_rate": 0.0001,
+      "loss": 7.1865,
+      "loss/crossentropy": 1.8603923156857491,
+      "loss/hidden": 3.55234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19390070009976626,
+      "step": 8300
+    },
+    {
+      "epoch": 0.06925,
+      "grad_norm": 31.625,
+      "grad_norm_var": 15.540625,
+      "learning_rate": 0.0001,
+      "loss": 7.1888,
+      "loss/crossentropy": 1.828030915558338,
+      "loss/hidden": 3.348828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19069764837622644,
+      "step": 8310
+    },
+    {
+      "epoch": 0.06933333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.298893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2229,
+      "loss/crossentropy": 1.8442706629633903,
+      "loss/hidden": 3.43125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816394252702594,
+      "step": 8320
+    },
+    {
+      "epoch": 0.06941666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.35625,
+      "learning_rate": 0.0001,
+      "loss": 7.2461,
+      "loss/crossentropy": 1.895699442923069,
+      "loss/hidden": 3.4546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19321332015097142,
+      "step": 8330
+    },
+    {
+      "epoch": 0.0695,
+      "grad_norm": 40.0,
+      "grad_norm_var": 36.78723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1403,
+      "loss/crossentropy": 1.9688566237688065,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19061133153736592,
+      "step": 8340
+    },
+    {
+      "epoch": 0.06958333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 39.82473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1012,
+      "loss/crossentropy": 1.9205253809690475,
+      "loss/hidden": 3.401171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1715885553508997,
+      "step": 8350
+    },
+    {
+      "epoch": 0.06966666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.15625,
+      "learning_rate": 0.0001,
+      "loss": 7.2205,
+      "loss/crossentropy": 1.949224701523781,
+      "loss/hidden": 3.41328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1786526907235384,
+      "step": 8360
+    },
+    {
+      "epoch": 0.06975,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.7025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1994,
+      "loss/crossentropy": 1.9869706839323045,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1956912238150835,
+      "step": 8370
+    },
+    {
+      "epoch": 0.06983333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 19.805143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2692,
+      "loss/crossentropy": 1.9487977296113967,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18634006716310977,
+      "step": 8380
+    },
+    {
+      "epoch": 0.06991666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.615625,
+      "learning_rate": 0.0001,
+      "loss": 7.3005,
+      "loss/crossentropy": 1.9617695778608322,
+      "loss/hidden": 3.437109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1997200610116124,
+      "step": 8390
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.7393229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3869,
+      "loss/crossentropy": 1.9588475212454797,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17612226950004697,
+      "step": 8400
+    },
+    {
+      "epoch": 0.07008333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 5.204472194874912e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2926,
+      "loss/crossentropy": 2.057014559209347,
+      "loss/hidden": 3.5015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19315509274601936,
+      "step": 8410
+    },
+    {
+      "epoch": 0.07016666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.2044721999248544e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2189,
+      "loss/crossentropy": 1.9825481489300727,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18724329564720393,
+      "step": 8420
+    },
+    {
+      "epoch": 0.07025,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.370833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1289,
+      "loss/crossentropy": 1.8243992432951928,
+      "loss/hidden": 3.409375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827236920595169,
+      "step": 8430
+    },
+    {
+      "epoch": 0.07033333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.078125,
+      "learning_rate": 0.0001,
+      "loss": 7.1837,
+      "loss/crossentropy": 2.044503265619278,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17393332067877054,
+      "step": 8440
+    },
+    {
+      "epoch": 0.07041666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.126822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.915854911506176,
+      "loss/hidden": 3.50703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19963504755869507,
+      "step": 8450
+    },
+    {
+      "epoch": 0.0705,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.7983723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2567,
+      "loss/crossentropy": 1.9769734531641006,
+      "loss/hidden": 3.295703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.181718045566231,
+      "step": 8460
+    },
+    {
+      "epoch": 0.07058333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 20.070572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1527,
+      "loss/crossentropy": 1.8116442456841468,
+      "loss/hidden": 3.45546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18748198114335538,
+      "step": 8470
+    },
+    {
+      "epoch": 0.07066666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 28.205143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2199,
+      "loss/crossentropy": 1.9818651676177979,
+      "loss/hidden": 3.523828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894064551219344,
+      "step": 8480
+    },
+    {
+      "epoch": 0.07075,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.1083252943282854e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2738,
+      "loss/crossentropy": 1.9650639399886132,
+      "loss/hidden": 3.30390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17558318749070168,
+      "step": 8490
+    },
+    {
+      "epoch": 0.07083333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 9.544205729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2025,
+      "loss/crossentropy": 1.9761282935738564,
+      "loss/hidden": 3.425390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19507634602487087,
+      "step": 8500
+    },
+    {
+      "epoch": 0.07091666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.1723307291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1301,
+      "loss/crossentropy": 1.9544002994894982,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756620042026043,
+      "step": 8510
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.7962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0789,
+      "loss/crossentropy": 1.9353080570697785,
+      "loss/hidden": 3.38359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18456000369042158,
+      "step": 8520
+    },
+    {
+      "epoch": 0.07108333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.6405598958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1855,
+      "loss/crossentropy": 2.0061560861766337,
+      "loss/hidden": 3.321875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18838093373924494,
+      "step": 8530
+    },
+    {
+      "epoch": 0.07116666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 12.155989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2221,
+      "loss/crossentropy": 1.9317010313272476,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18269629161804915,
+      "step": 8540
+    },
+    {
+      "epoch": 0.07125,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.755208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1957,
+      "loss/crossentropy": 1.9866686046123505,
+      "loss/hidden": 3.3171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18688506465405225,
+      "step": 8550
+    },
+    {
+      "epoch": 0.07133333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.524739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2597,
+      "loss/crossentropy": 1.8756410598754882,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17877673637121916,
+      "step": 8560
+    },
+    {
+      "epoch": 0.07141666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.362955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2339,
+      "loss/crossentropy": 1.9563673168420792,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17372122146189212,
+      "step": 8570
+    },
+    {
+      "epoch": 0.0715,
+      "grad_norm": 35.0,
+      "grad_norm_var": 3.120247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2685,
+      "loss/crossentropy": 1.8956820949912072,
+      "loss/hidden": 3.33046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17256692852824926,
+      "step": 8580
+    },
+    {
+      "epoch": 0.07158333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 12.751497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1566,
+      "loss/crossentropy": 1.9644134506583213,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18599690701812505,
+      "step": 8590
+    },
+    {
+      "epoch": 0.07166666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 12.847330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.152,
+      "loss/crossentropy": 1.9605856448411942,
+      "loss/hidden": 3.37265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17473262287676333,
+      "step": 8600
+    },
+    {
+      "epoch": 0.07175,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.9955729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2577,
+      "loss/crossentropy": 2.0416051924228666,
+      "loss/hidden": 3.46953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19386923424899577,
+      "step": 8610
+    },
+    {
+      "epoch": 0.07183333333333333,
+      "grad_norm": 46.25,
+      "grad_norm_var": 16312.484309895834,
+      "learning_rate": 0.0001,
+      "loss": 7.2449,
+      "loss/crossentropy": 1.8780235946178436,
+      "loss/hidden": 3.42421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18178340941667556,
+      "step": 8620
+    },
+    {
+      "epoch": 0.07191666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 16171.3572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2881,
+      "loss/crossentropy": 2.017537733912468,
+      "loss/hidden": 3.430859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20194268729537726,
+      "step": 8630
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.9639973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0493,
+      "loss/crossentropy": 1.7677518472075462,
+      "loss/hidden": 3.414453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18404914420098067,
+      "step": 8640
+    },
+    {
+      "epoch": 0.07208333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.115625,
+      "learning_rate": 0.0001,
+      "loss": 7.2341,
+      "loss/crossentropy": 1.81800634264946,
+      "loss/hidden": 3.346484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17132537085562943,
+      "step": 8650
+    },
+    {
+      "epoch": 0.07216666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 8.748418045451962e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2491,
+      "loss/crossentropy": 1.9027793630957603,
+      "loss/hidden": 3.291015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17584120333194733,
+      "step": 8660
+    },
+    {
+      "epoch": 0.07225,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.729166666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2757,
+      "loss/crossentropy": 1.8777679741382598,
+      "loss/hidden": 3.534375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19321729261428117,
+      "step": 8670
+    },
+    {
+      "epoch": 0.07233333333333333,
+      "grad_norm": 39.75,
+      "grad_norm_var": 6.401822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1494,
+      "loss/crossentropy": 1.9891476452350616,
+      "loss/hidden": 3.312890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188967870734632,
+      "step": 8680
+    },
+    {
+      "epoch": 0.07241666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.230208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1312,
+      "loss/crossentropy": 1.9339799106121063,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178991923853755,
+      "step": 8690
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.9478515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3105,
+      "loss/crossentropy": 2.0435035467147826,
+      "loss/hidden": 3.5171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2305173959583044,
+      "step": 8700
+    },
+    {
+      "epoch": 0.07258333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.911393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.3734,
+      "loss/crossentropy": 2.069814404845238,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19525352716445923,
+      "step": 8710
+    },
+    {
+      "epoch": 0.07266666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.888997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2247,
+      "loss/crossentropy": 1.8153776928782464,
+      "loss/hidden": 3.489453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20906406342983247,
+      "step": 8720
+    },
+    {
+      "epoch": 0.07275,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.466666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2225,
+      "loss/crossentropy": 1.986364483833313,
+      "loss/hidden": 3.32109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18280372805893422,
+      "step": 8730
+    },
+    {
+      "epoch": 0.07283333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.3343098958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1908,
+      "loss/crossentropy": 1.9479957491159439,
+      "loss/hidden": 3.314453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17259119898080827,
+      "step": 8740
+    },
+    {
+      "epoch": 0.07291666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.1603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1168,
+      "loss/crossentropy": 1.984895572811365,
+      "loss/hidden": 3.251171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18064502384513617,
+      "step": 8750
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.8889973958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2268,
+      "loss/crossentropy": 1.8017809867858887,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17566005773842336,
+      "step": 8760
+    },
+    {
+      "epoch": 0.07308333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.761458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1696,
+      "loss/crossentropy": 1.900542539358139,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17970035411417484,
+      "step": 8770
+    },
+    {
+      "epoch": 0.07316666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.3166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2412,
+      "loss/crossentropy": 2.041232463717461,
+      "loss/hidden": 3.43671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20939856991171837,
+      "step": 8780
+    },
+    {
+      "epoch": 0.07325,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.689322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2721,
+      "loss/crossentropy": 2.005451811850071,
+      "loss/hidden": 3.348828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18856565970927477,
+      "step": 8790
+    },
+    {
+      "epoch": 0.07333333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.9098307291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.27,
+      "loss/crossentropy": 2.00799767524004,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1781530775129795,
+      "step": 8800
+    },
+    {
+      "epoch": 0.07341666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 2.780143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2018,
+      "loss/crossentropy": 2.035827913880348,
+      "loss/hidden": 3.473828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20112563883885742,
+      "step": 8810
+    },
+    {
+      "epoch": 0.0735,
+      "grad_norm": 29.875,
+      "grad_norm_var": 4.736393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0714,
+      "loss/crossentropy": 1.9053474813699722,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1742972746491432,
+      "step": 8820
+    },
+    {
+      "epoch": 0.07358333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.2541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2318,
+      "loss/crossentropy": 1.8243301972746848,
+      "loss/hidden": 3.406640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931868800893426,
+      "step": 8830
+    },
+    {
+      "epoch": 0.07366666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.14140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2375,
+      "loss/crossentropy": 2.120525896549225,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1751797879114747,
+      "step": 8840
+    },
+    {
+      "epoch": 0.07375,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.7525390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1501,
+      "loss/crossentropy": 1.9116109699010848,
+      "loss/hidden": 3.499609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20058850292116404,
+      "step": 8850
+    },
+    {
+      "epoch": 0.07383333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.690625,
+      "learning_rate": 0.0001,
+      "loss": 7.2157,
+      "loss/crossentropy": 1.918264901638031,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18105561528354883,
+      "step": 8860
+    },
+    {
+      "epoch": 0.07391666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.4238932291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1731,
+      "loss/crossentropy": 1.997542105615139,
+      "loss/hidden": 3.196875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17583039812743664,
+      "step": 8870
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.2075980013739968e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2201,
+      "loss/crossentropy": 2.035869190096855,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19501792285591363,
+      "step": 8880
+    },
+    {
+      "epoch": 0.07408333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.207598000929855e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1496,
+      "loss/crossentropy": 1.7965922892093658,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16304808165878057,
+      "step": 8890
+    },
+    {
+      "epoch": 0.07416666666666667,
+      "grad_norm": 3909091328.0,
+      "grad_norm_var": 9.550621728791244e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2495,
+      "loss/crossentropy": 2.001436561346054,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20285465717315673,
+      "step": 8900
+    },
+    {
+      "epoch": 0.07425,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.550621713969274e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2073,
+      "loss/crossentropy": 2.0041428416967393,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20391638875007628,
+      "step": 8910
+    },
+    {
+      "epoch": 0.07433333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 11.8072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0817,
+      "loss/crossentropy": 1.960794059932232,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17380897589027883,
+      "step": 8920
+    },
+    {
+      "epoch": 0.07441666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.9770833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2189,
+      "loss/crossentropy": 1.9762230925261974,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17565663568675519,
+      "step": 8930
+    },
+    {
+      "epoch": 0.0745,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.035872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.171,
+      "loss/crossentropy": 1.8802208930253983,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18836744036525488,
+      "step": 8940
+    },
+    {
+      "epoch": 0.07458333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.203580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1634,
+      "loss/crossentropy": 2.038408327102661,
+      "loss/hidden": 3.265234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17698741499334575,
+      "step": 8950
+    },
+    {
+      "epoch": 0.07466666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.43515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2006,
+      "loss/crossentropy": 1.9461330994963646,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18866215460002422,
+      "step": 8960
+    },
+    {
+      "epoch": 0.07475,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.9285807291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1538,
+      "loss/crossentropy": 1.8740751318633557,
+      "loss/hidden": 3.4328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18583609201014042,
+      "step": 8970
+    },
+    {
+      "epoch": 0.07483333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.3988932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1565,
+      "loss/crossentropy": 1.8552331678569316,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.174785933829844,
+      "step": 8980
+    },
+    {
+      "epoch": 0.07491666666666667,
+      "grad_norm": 40.75,
+      "grad_norm_var": 7.595572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1589,
+      "loss/crossentropy": 2.018748515844345,
+      "loss/hidden": 3.3578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19232797529548407,
+      "step": 8990
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 42.25,
+      "grad_norm_var": 37.480143229166664,
+      "learning_rate": 0.0001,
+      "loss": 7.1619,
+      "loss/crossentropy": 1.9784274086356164,
+      "loss/hidden": 3.54453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19451803993433714,
+      "step": 9000
+    },
+    {
+      "epoch": 0.07508333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 36.496875,
+      "learning_rate": 0.0001,
+      "loss": 7.1749,
+      "loss/crossentropy": 2.0763546764850616,
+      "loss/hidden": 3.287109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18347182385623456,
+      "step": 9010
+    },
+    {
+      "epoch": 0.07516666666666667,
+      "grad_norm": 47.5,
+      "grad_norm_var": 20.393489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1598,
+      "loss/crossentropy": 1.9717566907405852,
+      "loss/hidden": 3.255859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17117710653692483,
+      "step": 9020
+    },
+    {
+      "epoch": 0.07525,
+      "grad_norm": 30.375,
+      "grad_norm_var": 19.7056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.234,
+      "loss/crossentropy": 1.9948984555900098,
+      "loss/hidden": 3.47734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19654347184114157,
+      "step": 9030
+    },
+    {
+      "epoch": 0.07533333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.1129557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2049,
+      "loss/crossentropy": 1.9566779106855392,
+      "loss/hidden": 3.43515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19929395206272602,
+      "step": 9040
+    },
+    {
+      "epoch": 0.07541666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9082682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1134,
+      "loss/crossentropy": 1.8074573144316672,
+      "loss/hidden": 3.337109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17522313240915538,
+      "step": 9050
+    },
+    {
+      "epoch": 0.0755,
+      "grad_norm": 28.25,
+      "grad_norm_var": 5.104166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0296,
+      "loss/crossentropy": 1.8022950239479543,
+      "loss/hidden": 3.423046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18761953189969063,
+      "step": 9060
+    },
+    {
+      "epoch": 0.07558333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.562239583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2742,
+      "loss/crossentropy": 2.078881235420704,
+      "loss/hidden": 3.393359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19961594343185424,
+      "step": 9070
+    },
+    {
+      "epoch": 0.07566666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 23.4962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0822,
+      "loss/crossentropy": 1.8851124197244644,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1875220352783799,
+      "step": 9080
+    },
+    {
+      "epoch": 0.07575,
+      "grad_norm": 32.25,
+      "grad_norm_var": 23.117708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2021,
+      "loss/crossentropy": 1.8932191953063011,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18913868740200995,
+      "step": 9090
+    },
+    {
+      "epoch": 0.07583333333333334,
+      "grad_norm": 27.5,
+      "grad_norm_var": 12.273893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1001,
+      "loss/crossentropy": 2.0837344110012053,
+      "loss/hidden": 3.291796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18248570300638675,
+      "step": 9100
+    },
+    {
+      "epoch": 0.07591666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.482291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2,
+      "loss/crossentropy": 1.8633555322885513,
+      "loss/hidden": 3.509375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894706480205059,
+      "step": 9110
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.141666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.251,
+      "loss/crossentropy": 1.9421419739723205,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.187481820397079,
+      "step": 9120
+    },
+    {
+      "epoch": 0.07608333333333334,
+      "grad_norm": 39.75,
+      "grad_norm_var": 9.668684895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2181,
+      "loss/crossentropy": 2.014544752240181,
+      "loss/hidden": 3.397265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18573549184948207,
+      "step": 9130
+    },
+    {
+      "epoch": 0.07616666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 8.687434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1915,
+      "loss/crossentropy": 1.9612121179699897,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18028553631156682,
+      "step": 9140
+    },
+    {
+      "epoch": 0.07625,
+      "grad_norm": 34.0,
+      "grad_norm_var": 5.188997395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0426,
+      "loss/crossentropy": 1.8344813786447047,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17982687037438155,
+      "step": 9150
+    },
+    {
+      "epoch": 0.07633333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 6.635872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.242,
+      "loss/crossentropy": 1.9214183151721955,
+      "loss/hidden": 3.284765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19156268667429685,
+      "step": 9160
+    },
+    {
+      "epoch": 0.07641666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.1572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2402,
+      "loss/crossentropy": 2.055627739429474,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2214739166200161,
+      "step": 9170
+    },
+    {
+      "epoch": 0.0765,
+      "grad_norm": 49.75,
+      "grad_norm_var": 416.1322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2374,
+      "loss/crossentropy": 1.877545380592346,
+      "loss/hidden": 3.628515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19467846769839525,
+      "step": 9180
+    },
+    {
+      "epoch": 0.07658333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 454.45774739583334,
+      "learning_rate": 0.0001,
+      "loss": 7.2325,
+      "loss/crossentropy": 2.07303985953331,
+      "loss/hidden": 3.411328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1902431882917881,
+      "step": 9190
+    },
+    {
+      "epoch": 0.07666666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.3917132086249184e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1667,
+      "loss/crossentropy": 1.9947309635579586,
+      "loss/hidden": 3.2984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17162247989326715,
+      "step": 9200
+    },
+    {
+      "epoch": 0.07675,
+      "grad_norm": 31.875,
+      "grad_norm_var": 17.014583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2712,
+      "loss/crossentropy": 1.9334057569503784,
+      "loss/hidden": 3.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19680443294346334,
+      "step": 9210
+    },
+    {
+      "epoch": 0.07683333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 13.737239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2037,
+      "loss/crossentropy": 1.9268360614776612,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18807857297360897,
+      "step": 9220
+    },
+    {
+      "epoch": 0.07691666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 8.670132813815612e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3321,
+      "loss/crossentropy": 1.991458609700203,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18901226110756397,
+      "step": 9230
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 32.5,
+      "grad_norm_var": 45.81848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2186,
+      "loss/crossentropy": 1.8337467461824417,
+      "loss/hidden": 3.41171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18715285304933788,
+      "step": 9240
+    },
+    {
+      "epoch": 0.07708333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 54.86015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1042,
+      "loss/crossentropy": 1.899548515677452,
+      "loss/hidden": 3.496875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1904792718589306,
+      "step": 9250
+    },
+    {
+      "epoch": 0.07716666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.133072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1672,
+      "loss/crossentropy": 2.0225892096757887,
+      "loss/hidden": 3.199609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1757547115907073,
+      "step": 9260
+    },
+    {
+      "epoch": 0.07725,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.4634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2112,
+      "loss/crossentropy": 1.8952244639396667,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19042186178267,
+      "step": 9270
+    },
+    {
+      "epoch": 0.07733333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.151811409763807e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1947,
+      "loss/crossentropy": 1.92966673001647,
+      "loss/hidden": 3.309375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17150711864233018,
+      "step": 9280
+    },
+    {
+      "epoch": 0.07741666666666666,
+      "grad_norm": 35.0,
+      "grad_norm_var": 1.571886422806041e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2936,
+      "loss/crossentropy": 2.11180402636528,
+      "loss/hidden": 3.611328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21517626009881496,
+      "step": 9290
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.135872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1262,
+      "loss/crossentropy": 1.8919033020734788,
+      "loss/hidden": 3.291796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16641295626759528,
+      "step": 9300
+    },
+    {
+      "epoch": 0.07758333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.58515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1427,
+      "loss/crossentropy": 1.9885196268558503,
+      "loss/hidden": 3.467578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20802160874009132,
+      "step": 9310
+    },
+    {
+      "epoch": 0.07766666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.1364583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2223,
+      "loss/crossentropy": 2.008994552493095,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19474735893309117,
+      "step": 9320
+    },
+    {
+      "epoch": 0.07775,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.6853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0724,
+      "loss/crossentropy": 1.9341641962528229,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20453856308013202,
+      "step": 9330
+    },
+    {
+      "epoch": 0.07783333333333334,
+      "grad_norm": 37.0,
+      "grad_norm_var": 3.6552083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2828,
+      "loss/crossentropy": 2.112407311797142,
+      "loss/hidden": 3.248046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18990872390568256,
+      "step": 9340
+    },
+    {
+      "epoch": 0.07791666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.970833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2668,
+      "loss/crossentropy": 1.995858334004879,
+      "loss/hidden": 3.3890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229170560836792,
+      "step": 9350
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.68515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2618,
+      "loss/crossentropy": 1.9759315863251685,
+      "loss/hidden": 3.38359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.184008826687932,
+      "step": 9360
+    },
+    {
+      "epoch": 0.07808333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.64765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2351,
+      "loss/crossentropy": 1.964530172944069,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19536323696374894,
+      "step": 9370
+    },
+    {
+      "epoch": 0.07816666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.1129557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2376,
+      "loss/crossentropy": 1.9365324467420577,
+      "loss/hidden": 3.473828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193184126727283,
+      "step": 9380
+    },
+    {
+      "epoch": 0.07825,
+      "grad_norm": 31.5,
+      "grad_norm_var": 18.044205729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1192,
+      "loss/crossentropy": 2.03958273679018,
+      "loss/hidden": 3.300390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18845226969569923,
+      "step": 9390
+    },
+    {
+      "epoch": 0.07833333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 20.661458333333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2336,
+      "loss/crossentropy": 1.9258702397346497,
+      "loss/hidden": 3.346875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20382192749530076,
+      "step": 9400
+    },
+    {
+      "epoch": 0.07841666666666666,
+      "grad_norm": 39.5,
+      "grad_norm_var": 8.745833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0934,
+      "loss/crossentropy": 1.9922194987535478,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19633182231336832,
+      "step": 9410
+    },
+    {
+      "epoch": 0.0785,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.108072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1712,
+      "loss/crossentropy": 1.831391155719757,
+      "loss/hidden": 3.465234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18680636398494244,
+      "step": 9420
+    },
+    {
+      "epoch": 0.07858333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 0.5747395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2869,
+      "loss/crossentropy": 2.0193733513355254,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822111329063773,
+      "step": 9430
+    },
+    {
+      "epoch": 0.07866666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 10.3822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2302,
+      "loss/crossentropy": 1.8986568033695221,
+      "loss/hidden": 3.293359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18191856034100057,
+      "step": 9440
+    },
+    {
+      "epoch": 0.07875,
+      "grad_norm": 30.25,
+      "grad_norm_var": 17.498372395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1577,
+      "loss/crossentropy": 1.8924290344119072,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18720396794378757,
+      "step": 9450
+    },
+    {
+      "epoch": 0.07883333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.7625,
+      "learning_rate": 0.0001,
+      "loss": 7.176,
+      "loss/crossentropy": 1.8431319534778594,
+      "loss/hidden": 3.437109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1925075102597475,
+      "step": 9460
+    },
+    {
+      "epoch": 0.07891666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.309309895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1943,
+      "loss/crossentropy": 1.9800647720694542,
+      "loss/hidden": 3.319921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18900530487298967,
+      "step": 9470
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.1056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1064,
+      "loss/crossentropy": 1.9323790550231934,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18270213007926941,
+      "step": 9480
+    },
+    {
+      "epoch": 0.07908333333333334,
+      "grad_norm": 36.25,
+      "grad_norm_var": 2.927515591570359e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2507,
+      "loss/crossentropy": 1.8533887356519698,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833003519102931,
+      "step": 9490
+    },
+    {
+      "epoch": 0.07916666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.92751559080385e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2466,
+      "loss/crossentropy": 2.010724890232086,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1888010736554861,
+      "step": 9500
+    },
+    {
+      "epoch": 0.07925,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.25390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2362,
+      "loss/crossentropy": 1.9017160810530185,
+      "loss/hidden": 3.451953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20483599258586765,
+      "step": 9510
+    },
+    {
+      "epoch": 0.07933333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 13.957747395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.3128,
+      "loss/crossentropy": 1.917439240217209,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16879426389932634,
+      "step": 9520
+    },
+    {
+      "epoch": 0.07941666666666666,
+      "grad_norm": 38.75,
+      "grad_norm_var": 30.634375,
+      "learning_rate": 0.0001,
+      "loss": 7.2802,
+      "loss/crossentropy": 2.0289409935474394,
+      "loss/hidden": 3.43515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19048050362616778,
+      "step": 9530
+    },
+    {
+      "epoch": 0.0795,
+      "grad_norm": 28.0,
+      "grad_norm_var": 25.167122395833335,
+      "learning_rate": 0.0001,
+      "loss": 7.1468,
+      "loss/crossentropy": 1.7849997401237487,
+      "loss/hidden": 3.480078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18105144426226616,
+      "step": 9540
+    },
+    {
+      "epoch": 0.07958333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.5832967226589184e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2816,
+      "loss/crossentropy": 1.9449206456542014,
+      "loss/hidden": 3.60234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1812568910419941,
+      "step": 9550
+    },
+    {
+      "epoch": 0.07966666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.5832967232775782e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1754,
+      "loss/crossentropy": 1.7890570968389512,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17575022913515567,
+      "step": 9560
+    },
+    {
+      "epoch": 0.07975,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.748418048102062e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2619,
+      "loss/crossentropy": 1.9128870755434035,
+      "loss/hidden": 3.663671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20060238279402257,
+      "step": 9570
+    },
+    {
+      "epoch": 0.07983333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.4955729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2103,
+      "loss/crossentropy": 1.8678260400891304,
+      "loss/hidden": 3.4265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20040124338120222,
+      "step": 9580
+    },
+    {
+      "epoch": 0.07991666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 16.593489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1735,
+      "loss/crossentropy": 1.7469847694039344,
+      "loss/hidden": 3.29453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16893960312008857,
+      "step": 9590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 31.0,
+      "grad_norm_var": 14.876497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2358,
+      "loss/crossentropy": 2.0145472913980482,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21668562162667512,
+      "step": 9600
+    },
+    {
+      "epoch": 0.08008333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.614583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2138,
+      "loss/crossentropy": 1.8460545286536216,
+      "loss/hidden": 3.547265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19596499875187873,
+      "step": 9610
+    },
+    {
+      "epoch": 0.08016666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.2824703444335985e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2641,
+      "loss/crossentropy": 1.9291812598705291,
+      "loss/hidden": 3.374609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1993147611618042,
+      "step": 9620
+    },
+    {
+      "epoch": 0.08025,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.253059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0381,
+      "loss/crossentropy": 1.9004499420523644,
+      "loss/hidden": 3.28046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17315188869833947,
+      "step": 9630
+    },
+    {
+      "epoch": 0.08033333333333334,
+      "grad_norm": 49.75,
+      "grad_norm_var": 24.8875,
+      "learning_rate": 0.0001,
+      "loss": 7.2074,
+      "loss/crossentropy": 2.0676659286022185,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19399872440844773,
+      "step": 9640
+    },
+    {
+      "epoch": 0.08041666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 21.844205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 1.8846231751143931,
+      "loss/hidden": 3.569921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813927460461855,
+      "step": 9650
+    },
+    {
+      "epoch": 0.0805,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.473372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2297,
+      "loss/crossentropy": 1.95650455057621,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18862175196409225,
+      "step": 9660
+    },
+    {
+      "epoch": 0.08058333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.873893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2405,
+      "loss/crossentropy": 2.0532629549503327,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1972910648211837,
+      "step": 9670
+    },
+    {
+      "epoch": 0.08066666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.488997395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2227,
+      "loss/crossentropy": 1.8982819005846978,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18387150373309852,
+      "step": 9680
+    },
+    {
+      "epoch": 0.08075,
+      "grad_norm": 30.5,
+      "grad_norm_var": 23.84765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2229,
+      "loss/crossentropy": 2.042688661813736,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19536025915294886,
+      "step": 9690
+    },
+    {
+      "epoch": 0.08083333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 7.214322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1841,
+      "loss/crossentropy": 1.8631189227104188,
+      "loss/hidden": 3.24921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16748407240957022,
+      "step": 9700
+    },
+    {
+      "epoch": 0.08091666666666666,
+      "grad_norm": 38.5,
+      "grad_norm_var": 8.705989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0953,
+      "loss/crossentropy": 2.096615809202194,
+      "loss/hidden": 3.247265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19336612224578859,
+      "step": 9710
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 31.625,
+      "grad_norm_var": 8.949739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0645,
+      "loss/crossentropy": 1.7769379287958145,
+      "loss/hidden": 3.308203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17712265541777014,
+      "step": 9720
+    },
+    {
+      "epoch": 0.08108333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 18.835872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2842,
+      "loss/crossentropy": 2.0219400838017463,
+      "loss/hidden": 3.3078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19118713159114123,
+      "step": 9730
+    },
+    {
+      "epoch": 0.08116666666666666,
+      "grad_norm": 41.5,
+      "grad_norm_var": 22.01015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2142,
+      "loss/crossentropy": 2.1162992775440217,
+      "loss/hidden": 3.45859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19361746832728385,
+      "step": 9740
+    },
+    {
+      "epoch": 0.08125,
+      "grad_norm": 29.375,
+      "grad_norm_var": 10.814583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2149,
+      "loss/crossentropy": 1.854822640120983,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17972449399530888,
+      "step": 9750
+    },
+    {
+      "epoch": 0.08133333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.9228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1515,
+      "loss/crossentropy": 2.042907669395208,
+      "loss/hidden": 3.523046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20937258377671242,
+      "step": 9760
+    },
+    {
+      "epoch": 0.08141666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.533333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1791,
+      "loss/crossentropy": 2.0195691674947738,
+      "loss/hidden": 3.46796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1953078692778945,
+      "step": 9770
+    },
+    {
+      "epoch": 0.0815,
+      "grad_norm": 38.0,
+      "grad_norm_var": 13.008333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1171,
+      "loss/crossentropy": 1.8863514378666877,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18283969424664975,
+      "step": 9780
+    },
+    {
+      "epoch": 0.08158333333333333,
+      "grad_norm": 40.25,
+      "grad_norm_var": 1428.9330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.4353,
+      "loss/crossentropy": 1.9719516918063165,
+      "loss/hidden": 3.469140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2090709051117301,
+      "step": 9790
+    },
+    {
+      "epoch": 0.08166666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.641080729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.181,
+      "loss/crossentropy": 2.0369949251413346,
+      "loss/hidden": 3.294921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18184055741876365,
+      "step": 9800
+    },
+    {
+      "epoch": 0.08175,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.873893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0857,
+      "loss/crossentropy": 2.026154139637947,
+      "loss/hidden": 3.314453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18145214160904288,
+      "step": 9810
+    },
+    {
+      "epoch": 0.08183333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.8608723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2217,
+      "loss/crossentropy": 1.984486585855484,
+      "loss/hidden": 3.393359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2005698978435248,
+      "step": 9820
+    },
+    {
+      "epoch": 0.08191666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 26.6291015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1779,
+      "loss/crossentropy": 2.0041553929448126,
+      "loss/hidden": 3.257421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17574419379234313,
+      "step": 9830
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.23125,
+      "learning_rate": 0.0001,
+      "loss": 7.1577,
+      "loss/crossentropy": 1.8619619227945805,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19841333236545325,
+      "step": 9840
+    },
+    {
+      "epoch": 0.08208333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 15.088541666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2013,
+      "loss/crossentropy": 1.898135930299759,
+      "loss/hidden": 3.440625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854542102664709,
+      "step": 9850
+    },
+    {
+      "epoch": 0.08216666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 12.376822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1536,
+      "loss/crossentropy": 1.9185454472899437,
+      "loss/hidden": 3.417578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20166656840592623,
+      "step": 9860
+    },
+    {
+      "epoch": 0.08225,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.834375,
+      "learning_rate": 0.0001,
+      "loss": 7.2058,
+      "loss/crossentropy": 1.919453853368759,
+      "loss/hidden": 3.411328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1870553806424141,
+      "step": 9870
+    },
+    {
+      "epoch": 0.08233333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 7.107418945754236e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1974,
+      "loss/crossentropy": 1.889936888217926,
+      "loss/hidden": 3.57265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18729659765958787,
+      "step": 9880
+    },
+    {
+      "epoch": 0.08241666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.107418942803543e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2175,
+      "loss/crossentropy": 2.0169277042150497,
+      "loss/hidden": 3.330859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20186357703059912,
+      "step": 9890
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 34.0,
+      "grad_norm_var": 8.82705511779559e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3193,
+      "loss/crossentropy": 2.0031974256038665,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18207758255302905,
+      "step": 9900
+    },
+    {
+      "epoch": 0.08258333333333333,
+      "grad_norm": 43.5,
+      "grad_norm_var": 8.827055118187059e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1468,
+      "loss/crossentropy": 1.8897509559988976,
+      "loss/hidden": 3.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18627819493412973,
+      "step": 9910
+    },
+    {
+      "epoch": 0.08266666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 13.143489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1643,
+      "loss/crossentropy": 1.96635133177042,
+      "loss/hidden": 3.22109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17648211270570754,
+      "step": 9920
+    },
+    {
+      "epoch": 0.08275,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.2447526979635249e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1337,
+      "loss/crossentropy": 1.7700391985476016,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953332625329493,
+      "step": 9930
+    },
+    {
+      "epoch": 0.08283333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 10.516666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 1.9698360793292522,
+      "loss/hidden": 3.2671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18065936379134656,
+      "step": 9940
+    },
+    {
+      "epoch": 0.08291666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.0458333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1922,
+      "loss/crossentropy": 1.9713639967143535,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909080557525158,
+      "step": 9950
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.8082682291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2286,
+      "loss/crossentropy": 1.9554906249046327,
+      "loss/hidden": 3.34296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17579805050045252,
+      "step": 9960
+    },
+    {
+      "epoch": 0.08308333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.6155598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.0742,
+      "loss/crossentropy": 1.9254092440009116,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17711936933919786,
+      "step": 9970
+    },
+    {
+      "epoch": 0.08316666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.0809895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2025,
+      "loss/crossentropy": 2.0318453282117845,
+      "loss/hidden": 3.2765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854577951133251,
+      "step": 9980
+    },
+    {
+      "epoch": 0.08325,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.5134765625,
+      "learning_rate": 0.0001,
+      "loss": 7.117,
+      "loss/crossentropy": 1.913966417312622,
+      "loss/hidden": 3.28984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17688458357006312,
+      "step": 9990
+    },
+    {
+      "epoch": 0.08333333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2705729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1223,
+      "loss/crossentropy": 1.8660702288150788,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17869504541158676,
+      "step": 10000
+    },
+    {
+      "epoch": 0.08341666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 19.8556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1629,
+      "loss/crossentropy": 1.9022649198770523,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17940771766006947,
+      "step": 10010
+    },
+    {
+      "epoch": 0.0835,
+      "grad_norm": 40.75,
+      "grad_norm_var": 7.496875,
+      "learning_rate": 0.0001,
+      "loss": 7.1594,
+      "loss/crossentropy": 1.9832664839923382,
+      "loss/hidden": 3.314453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774651188403368,
+      "step": 10020
+    },
+    {
+      "epoch": 0.08358333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.530143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2348,
+      "loss/crossentropy": 2.045203870534897,
+      "loss/hidden": 3.23125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16557962372899054,
+      "step": 10030
+    },
+    {
+      "epoch": 0.08366666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.680989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3016,
+      "loss/crossentropy": 1.8034087099134921,
+      "loss/hidden": 3.433984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18954076152294874,
+      "step": 10040
+    },
+    {
+      "epoch": 0.08375,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.0646087135446172e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3546,
+      "loss/crossentropy": 2.141270640492439,
+      "loss/hidden": 3.234765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18412422593683003,
+      "step": 10050
+    },
+    {
+      "epoch": 0.08383333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 29.053059895833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1943,
+      "loss/crossentropy": 1.9576030984520911,
+      "loss/hidden": 3.34609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1939836498349905,
+      "step": 10060
+    },
+    {
+      "epoch": 0.08391666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 31.228125,
+      "learning_rate": 0.0001,
+      "loss": 7.1314,
+      "loss/crossentropy": 1.9822376906871795,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18735816478729247,
+      "step": 10070
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 29.875,
+      "grad_norm_var": 7.443489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1807,
+      "loss/crossentropy": 2.019585521519184,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21315673794597387,
+      "step": 10080
+    },
+    {
+      "epoch": 0.08408333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.045572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0985,
+      "loss/crossentropy": 1.9764751970767975,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1869516059756279,
+      "step": 10090
+    },
+    {
+      "epoch": 0.08416666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.175,
+      "learning_rate": 0.0001,
+      "loss": 7.2217,
+      "loss/crossentropy": 1.948393326997757,
+      "loss/hidden": 3.440625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825612099841237,
+      "step": 10100
+    },
+    {
+      "epoch": 0.08425,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.9809895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.263,
+      "loss/crossentropy": 1.8718901932239533,
+      "loss/hidden": 3.432421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1928726764395833,
+      "step": 10110
+    },
+    {
+      "epoch": 0.08433333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 13.518684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2244,
+      "loss/crossentropy": 1.9680954083800315,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19962124936282635,
+      "step": 10120
+    },
+    {
+      "epoch": 0.08441666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.6056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.3134,
+      "loss/crossentropy": 1.9413045309484005,
+      "loss/hidden": 3.4984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19008888639509677,
+      "step": 10130
+    },
+    {
+      "epoch": 0.0845,
+      "grad_norm": 31.25,
+      "grad_norm_var": 10.490625,
+      "learning_rate": 0.0001,
+      "loss": 7.0934,
+      "loss/crossentropy": 1.9629930078983306,
+      "loss/hidden": 3.33046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18266241662204266,
+      "step": 10140
+    },
+    {
+      "epoch": 0.08458333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 19.312434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2018,
+      "loss/crossentropy": 1.9162113338708877,
+      "loss/hidden": 3.483203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2075673297047615,
+      "step": 10150
+    },
+    {
+      "epoch": 0.08466666666666667,
+      "grad_norm": 39.0,
+      "grad_norm_var": 13.939518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1795,
+      "loss/crossentropy": 1.861514998972416,
+      "loss/hidden": 3.321875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18033697009086608,
+      "step": 10160
+    },
+    {
+      "epoch": 0.08475,
+      "grad_norm": 36.25,
+      "grad_norm_var": 11.56875,
+      "learning_rate": 0.0001,
+      "loss": 7.1884,
+      "loss/crossentropy": 1.9989107728004456,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18097838908433914,
+      "step": 10170
+    },
+    {
+      "epoch": 0.08483333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 10.505208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1849,
+      "loss/crossentropy": 1.9304386712610722,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17962286612018943,
+      "step": 10180
+    },
+    {
+      "epoch": 0.08491666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.6291015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2364,
+      "loss/crossentropy": 1.8655005872249604,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17910738606005908,
+      "step": 10190
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 29.0,
+      "grad_norm_var": 8.738541666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0075,
+      "loss/crossentropy": 1.8369006037712097,
+      "loss/hidden": 3.35859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17068145107477903,
+      "step": 10200
+    },
+    {
+      "epoch": 0.08508333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.42890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1341,
+      "loss/crossentropy": 2.033593140542507,
+      "loss/hidden": 3.3546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18672468289732932,
+      "step": 10210
+    },
+    {
+      "epoch": 0.08516666666666667,
+      "grad_norm": 36.5,
+      "grad_norm_var": 8.985384775159471e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3214,
+      "loss/crossentropy": 1.9985199861228466,
+      "loss/hidden": 3.3953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19838361945003272,
+      "step": 10220
+    },
+    {
+      "epoch": 0.08525,
+      "grad_norm": 33.75,
+      "grad_norm_var": 8.98538477310566e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.9306078165769578,
+      "loss/hidden": 3.306640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1709518525749445,
+      "step": 10230
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 29.743489583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2143,
+      "loss/crossentropy": 1.9644243538379669,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20114268753677605,
+      "step": 10240
+    },
+    {
+      "epoch": 0.08541666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.350779033464778e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2055,
+      "loss/crossentropy": 1.9427935935556888,
+      "loss/hidden": 3.247265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036175568588078,
+      "step": 10250
+    },
+    {
+      "epoch": 0.0855,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.108268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0379,
+      "loss/crossentropy": 2.061500659584999,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19168602395802736,
+      "step": 10260
+    },
+    {
+      "epoch": 0.08558333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.306266257317343e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1855,
+      "loss/crossentropy": 1.8393201567232609,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18665622984990476,
+      "step": 10270
+    },
+    {
+      "epoch": 0.08566666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.718489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2163,
+      "loss/crossentropy": 1.9231659054756165,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19700709749013184,
+      "step": 10280
+    },
+    {
+      "epoch": 0.08575,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.31875,
+      "learning_rate": 0.0001,
+      "loss": 7.2146,
+      "loss/crossentropy": 1.9567264869809151,
+      "loss/hidden": 3.45625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20131768714636564,
+      "step": 10290
+    },
+    {
+      "epoch": 0.08583333333333333,
+      "grad_norm": 38.75,
+      "grad_norm_var": 14.565625,
+      "learning_rate": 0.0001,
+      "loss": 7.1047,
+      "loss/crossentropy": 2.0508678406476974,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843091931194067,
+      "step": 10300
+    },
+    {
+      "epoch": 0.08591666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 9.573893229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2799,
+      "loss/crossentropy": 1.948277612030506,
+      "loss/hidden": 3.48203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2040827188640833,
+      "step": 10310
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 35.5,
+      "grad_norm_var": 6.020572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1226,
+      "loss/crossentropy": 1.9011885672807693,
+      "loss/hidden": 3.299609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18988663759082555,
+      "step": 10320
+    },
+    {
+      "epoch": 0.08608333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 13.781705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1359,
+      "loss/crossentropy": 2.008572620898485,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18977851662784814,
+      "step": 10330
+    },
+    {
+      "epoch": 0.08616666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 11.628125,
+      "learning_rate": 0.0001,
+      "loss": 7.1844,
+      "loss/crossentropy": 1.7907821334898473,
+      "loss/hidden": 3.330859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1695950568653643,
+      "step": 10340
+    },
+    {
+      "epoch": 0.08625,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.092643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1184,
+      "loss/crossentropy": 1.9050454296171666,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18444497175514699,
+      "step": 10350
+    },
+    {
+      "epoch": 0.08633333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.190625,
+      "learning_rate": 0.0001,
+      "loss": 7.2375,
+      "loss/crossentropy": 2.076017972826958,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19985819570720195,
+      "step": 10360
+    },
+    {
+      "epoch": 0.08641666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 10.315559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2657,
+      "loss/crossentropy": 1.9178860872983932,
+      "loss/hidden": 3.516796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21050459016114473,
+      "step": 10370
+    },
+    {
+      "epoch": 0.0865,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.971875,
+      "learning_rate": 0.0001,
+      "loss": 7.1385,
+      "loss/crossentropy": 1.7682075582444667,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16693565947934985,
+      "step": 10380
+    },
+    {
+      "epoch": 0.08658333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.417805256710511e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2885,
+      "loss/crossentropy": 1.987515565752983,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1908052433282137,
+      "step": 10390
+    },
+    {
+      "epoch": 0.08666666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 9.037239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1114,
+      "loss/crossentropy": 1.9178199276328087,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1676847252994776,
+      "step": 10400
+    },
+    {
+      "epoch": 0.08675,
+      "grad_norm": 29.625,
+      "grad_norm_var": 6.906705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1645,
+      "loss/crossentropy": 1.9134997978806496,
+      "loss/hidden": 3.31171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1701915455982089,
+      "step": 10410
+    },
+    {
+      "epoch": 0.08683333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.696875,
+      "learning_rate": 0.0001,
+      "loss": 7.2155,
+      "loss/crossentropy": 2.0359968066215517,
+      "loss/hidden": 3.374609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18990014400333166,
+      "step": 10420
+    },
+    {
+      "epoch": 0.08691666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.411393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.132,
+      "loss/crossentropy": 1.8710991702973843,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17285765218548477,
+      "step": 10430
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.795572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1247,
+      "loss/crossentropy": 1.9286628663539886,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18416991382837294,
+      "step": 10440
+    },
+    {
+      "epoch": 0.08708333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.223372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1433,
+      "loss/crossentropy": 2.069103667140007,
+      "loss/hidden": 3.2671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18272748328745364,
+      "step": 10450
+    },
+    {
+      "epoch": 0.08716666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.280208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1994,
+      "loss/crossentropy": 1.9527845323085784,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18390033449977636,
+      "step": 10460
+    },
+    {
+      "epoch": 0.08725,
+      "grad_norm": 31.375,
+      "grad_norm_var": 8.099739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1659,
+      "loss/crossentropy": 1.9093656569719315,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18116378430277108,
+      "step": 10470
+    },
+    {
+      "epoch": 0.08733333333333333,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.997330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1718,
+      "loss/crossentropy": 1.9801187694072724,
+      "loss/hidden": 3.25,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18494955655187367,
+      "step": 10480
+    },
+    {
+      "epoch": 0.08741666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.1259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0675,
+      "loss/crossentropy": 1.9519363179802895,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17371598407626151,
+      "step": 10490
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 28.875,
+      "grad_norm_var": 5.485872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1353,
+      "loss/crossentropy": 1.8069478645920753,
+      "loss/hidden": 3.2921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1630322251468897,
+      "step": 10500
+    },
+    {
+      "epoch": 0.08758333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.436458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1655,
+      "loss/crossentropy": 1.8974637396633625,
+      "loss/hidden": 3.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1867376634851098,
+      "step": 10510
+    },
+    {
+      "epoch": 0.08766666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.425455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1548,
+      "loss/crossentropy": 1.88455178886652,
+      "loss/hidden": 3.362890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16794521678239108,
+      "step": 10520
+    },
+    {
+      "epoch": 0.08775,
+      "grad_norm": 35.75,
+      "grad_norm_var": 6.442708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1584,
+      "loss/crossentropy": 2.0113652512431144,
+      "loss/hidden": 3.270703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18112584240734578,
+      "step": 10530
+    },
+    {
+      "epoch": 0.08783333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.073372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1687,
+      "loss/crossentropy": 1.8693530946969985,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19360871855169534,
+      "step": 10540
+    },
+    {
+      "epoch": 0.08791666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.458072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1976,
+      "loss/crossentropy": 1.9990513175725937,
+      "loss/hidden": 3.45546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21094244308769702,
+      "step": 10550
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 37.75,
+      "grad_norm_var": 16.712955729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0825,
+      "loss/crossentropy": 1.7959557376801967,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17526750043034553,
+      "step": 10560
+    },
+    {
+      "epoch": 0.08808333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 11.593489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1247,
+      "loss/crossentropy": 1.9136253535747527,
+      "loss/hidden": 3.27890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16128470674157142,
+      "step": 10570
+    },
+    {
+      "epoch": 0.08816666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 6.94375,
+      "learning_rate": 0.0001,
+      "loss": 7.1836,
+      "loss/crossentropy": 2.0141227670013904,
+      "loss/hidden": 3.279296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17919235862791538,
+      "step": 10580
+    },
+    {
+      "epoch": 0.08825,
+      "grad_norm": 30.0,
+      "grad_norm_var": 45.81223958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1608,
+      "loss/crossentropy": 1.9507560439407825,
+      "loss/hidden": 3.404296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19365838579833508,
+      "step": 10590
+    },
+    {
+      "epoch": 0.08833333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 45.628580729166664,
+      "learning_rate": 0.0001,
+      "loss": 7.1172,
+      "loss/crossentropy": 2.0076355174183846,
+      "loss/hidden": 3.319921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18096669893711806,
+      "step": 10600
+    },
+    {
+      "epoch": 0.08841666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.121809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1205,
+      "loss/crossentropy": 1.8196946240961551,
+      "loss/hidden": 3.277734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16406874284148215,
+      "step": 10610
+    },
+    {
+      "epoch": 0.0885,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.5353515625,
+      "learning_rate": 0.0001,
+      "loss": 7.149,
+      "loss/crossentropy": 1.9936213076114655,
+      "loss/hidden": 3.2859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1952991671860218,
+      "step": 10620
+    },
+    {
+      "epoch": 0.08858333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 22.788541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.197,
+      "loss/crossentropy": 1.862833446264267,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18173334691673518,
+      "step": 10630
+    },
+    {
+      "epoch": 0.08866666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.67890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1265,
+      "loss/crossentropy": 1.9695732221007347,
+      "loss/hidden": 3.430078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19707035087049007,
+      "step": 10640
+    },
+    {
+      "epoch": 0.08875,
+      "grad_norm": 31.125,
+      "grad_norm_var": 13.308072916666667,
+      "learning_rate": 0.0001,
+      "loss": 6.9949,
+      "loss/crossentropy": 1.9376090109348296,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19217673875391483,
+      "step": 10650
+    },
+    {
+      "epoch": 0.08883333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.0087890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1151,
+      "loss/crossentropy": 1.9215148121118546,
+      "loss/hidden": 3.292578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18200548309832812,
+      "step": 10660
+    },
+    {
+      "epoch": 0.08891666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.190625,
+      "learning_rate": 0.0001,
+      "loss": 7.0804,
+      "loss/crossentropy": 2.0141350597143175,
+      "loss/hidden": 3.2984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18078853078186513,
+      "step": 10670
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 37.75,
+      "grad_norm_var": 18.985872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2503,
+      "loss/crossentropy": 1.8657403737306595,
+      "loss/hidden": 3.476171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18957516588270665,
+      "step": 10680
+    },
+    {
+      "epoch": 0.08908333333333333,
+      "grad_norm": 4060086272.0,
+      "grad_norm_var": 1.0302687665035675e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2715,
+      "loss/crossentropy": 1.919578692317009,
+      "loss/hidden": 3.68515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18535034377127885,
+      "step": 10690
+    },
+    {
+      "epoch": 0.08916666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.0302687671591022e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1533,
+      "loss/crossentropy": 1.9495833858847618,
+      "loss/hidden": 3.2703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19119064919650555,
+      "step": 10700
+    },
+    {
+      "epoch": 0.08925,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.4947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1031,
+      "loss/crossentropy": 1.9701412051916123,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18934717047959565,
+      "step": 10710
+    },
+    {
+      "epoch": 0.08933333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.2113932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1913,
+      "loss/crossentropy": 1.8715698659420013,
+      "loss/hidden": 3.321875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17601410672068596,
+      "step": 10720
+    },
+    {
+      "epoch": 0.08941666666666667,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.802018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1962,
+      "loss/crossentropy": 2.0264753639698028,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19696479327976704,
+      "step": 10730
+    },
+    {
+      "epoch": 0.0895,
+      "grad_norm": 29.0,
+      "grad_norm_var": 7.732291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1175,
+      "loss/crossentropy": 1.8837075635790825,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17116072084754705,
+      "step": 10740
+    },
+    {
+      "epoch": 0.08958333333333333,
+      "grad_norm": 36.5,
+      "grad_norm_var": 4.662434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1497,
+      "loss/crossentropy": 1.9943507939577103,
+      "loss/hidden": 3.333203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18434021826833485,
+      "step": 10750
+    },
+    {
+      "epoch": 0.08966666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.4744140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1849,
+      "loss/crossentropy": 1.9133451774716377,
+      "loss/hidden": 3.297265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18499344643205404,
+      "step": 10760
+    },
+    {
+      "epoch": 0.08975,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.8749348958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1684,
+      "loss/crossentropy": 1.86435616761446,
+      "loss/hidden": 3.325,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18305802326649429,
+      "step": 10770
+    },
+    {
+      "epoch": 0.08983333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 107.4322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2144,
+      "loss/crossentropy": 1.8334193125367164,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18382519632577896,
+      "step": 10780
+    },
+    {
+      "epoch": 0.08991666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 104.815625,
+      "learning_rate": 0.0001,
+      "loss": 7.1307,
+      "loss/crossentropy": 1.8171788975596428,
+      "loss/hidden": 3.255078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17925475109368563,
+      "step": 10790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.71015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2726,
+      "loss/crossentropy": 1.913816411793232,
+      "loss/hidden": 3.346875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17627263125032186,
+      "step": 10800
+    },
+    {
+      "epoch": 0.09008333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.2806640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1423,
+      "loss/crossentropy": 1.9569356814026833,
+      "loss/hidden": 3.430078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20463840384036303,
+      "step": 10810
+    },
+    {
+      "epoch": 0.09016666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.866666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1729,
+      "loss/crossentropy": 2.013381141424179,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19274718966335058,
+      "step": 10820
+    },
+    {
+      "epoch": 0.09025,
+      "grad_norm": 34.0,
+      "grad_norm_var": 9.798143737119905e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.253,
+      "loss/crossentropy": 2.048290731012821,
+      "loss/hidden": 3.301171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188222586363554,
+      "step": 10830
+    },
+    {
+      "epoch": 0.09033333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 9.79814377609547e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2405,
+      "loss/crossentropy": 1.9618924796581267,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816230181604624,
+      "step": 10840
+    },
+    {
+      "epoch": 0.09041666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 220.94212239583334,
+      "learning_rate": 0.0001,
+      "loss": 7.1372,
+      "loss/crossentropy": 1.9004179686307907,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17816591057926418,
+      "step": 10850
+    },
+    {
+      "epoch": 0.0905,
+      "grad_norm": 30.875,
+      "grad_norm_var": 215.15520833333332,
+      "learning_rate": 0.0001,
+      "loss": 7.1717,
+      "loss/crossentropy": 1.996871554851532,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20265654604882002,
+      "step": 10860
+    },
+    {
+      "epoch": 0.09058333333333334,
+      "grad_norm": 37.0,
+      "grad_norm_var": 19.86875,
+      "learning_rate": 0.0001,
+      "loss": 7.1966,
+      "loss/crossentropy": 1.914494514465332,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1783069744706154,
+      "step": 10870
+    },
+    {
+      "epoch": 0.09066666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 38.11145833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2457,
+      "loss/crossentropy": 1.982391807436943,
+      "loss/hidden": 3.253125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19039569403976203,
+      "step": 10880
+    },
+    {
+      "epoch": 0.09075,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.87265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1233,
+      "loss/crossentropy": 1.8585443809628486,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18120319172739982,
+      "step": 10890
+    },
+    {
+      "epoch": 0.09083333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.946875,
+      "learning_rate": 0.0001,
+      "loss": 7.1387,
+      "loss/crossentropy": 1.7904712766408921,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.198022804223001,
+      "step": 10900
+    },
+    {
+      "epoch": 0.09091666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.398893229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1571,
+      "loss/crossentropy": 1.8014109954237938,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1762148378416896,
+      "step": 10910
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 33.0,
+      "grad_norm_var": 20.616080729166665,
+      "learning_rate": 0.0001,
+      "loss": 7.2656,
+      "loss/crossentropy": 1.8642778381705285,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960908764973283,
+      "step": 10920
+    },
+    {
+      "epoch": 0.09108333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 10.373893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1459,
+      "loss/crossentropy": 1.7690608263015748,
+      "loss/hidden": 3.342578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18550572786480188,
+      "step": 10930
+    },
+    {
+      "epoch": 0.09116666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 17.536458333333332,
+      "learning_rate": 0.0001,
+      "loss": 7.1696,
+      "loss/crossentropy": 1.9643425852060319,
+      "loss/hidden": 3.559375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2071039265021682,
+      "step": 10940
+    },
+    {
+      "epoch": 0.09125,
+      "grad_norm": 33.0,
+      "grad_norm_var": 22.759309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1998,
+      "loss/crossentropy": 1.9260810315608978,
+      "loss/hidden": 3.476171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18847699128091336,
+      "step": 10950
+    },
+    {
+      "epoch": 0.09133333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 12.107291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1959,
+      "loss/crossentropy": 1.9213246397674084,
+      "loss/hidden": 3.2453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16759488992393018,
+      "step": 10960
+    },
+    {
+      "epoch": 0.09141666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.0348307291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1403,
+      "loss/crossentropy": 2.061849758028984,
+      "loss/hidden": 3.272265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18976418543606996,
+      "step": 10970
+    },
+    {
+      "epoch": 0.0915,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.4900390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2057,
+      "loss/crossentropy": 1.8701441615819931,
+      "loss/hidden": 3.346875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17094586398452521,
+      "step": 10980
+    },
+    {
+      "epoch": 0.09158333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.49375,
+      "learning_rate": 0.0001,
+      "loss": 7.2646,
+      "loss/crossentropy": 2.016074526309967,
+      "loss/hidden": 3.478515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21079414822161197,
+      "step": 10990
+    },
+    {
+      "epoch": 0.09166666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.7625,
+      "learning_rate": 0.0001,
+      "loss": 7.1929,
+      "loss/crossentropy": 1.9487488180398942,
+      "loss/hidden": 3.300390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17733574695885182,
+      "step": 11000
+    },
+    {
+      "epoch": 0.09175,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.3046223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2034,
+      "loss/crossentropy": 2.0100974045693873,
+      "loss/hidden": 3.246484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17906391359865664,
+      "step": 11010
+    },
+    {
+      "epoch": 0.09183333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.4702473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0553,
+      "loss/crossentropy": 1.8588308840990067,
+      "loss/hidden": 3.27421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17334383996203542,
+      "step": 11020
+    },
+    {
+      "epoch": 0.09191666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.618489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2876,
+      "loss/crossentropy": 1.9602320730686187,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959123121574521,
+      "step": 11030
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.8811848958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2807,
+      "loss/crossentropy": 1.832445289194584,
+      "loss/hidden": 3.33046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16387658324092627,
+      "step": 11040
+    },
+    {
+      "epoch": 0.09208333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.755143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1354,
+      "loss/crossentropy": 1.9954833656549453,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18346874881535769,
+      "step": 11050
+    },
+    {
+      "epoch": 0.09216666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.265559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2109,
+      "loss/crossentropy": 1.9801868200302124,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915831219404936,
+      "step": 11060
+    },
+    {
+      "epoch": 0.09225,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.7041666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2754,
+      "loss/crossentropy": 1.8625975862145423,
+      "loss/hidden": 3.437109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20278072878718376,
+      "step": 11070
+    },
+    {
+      "epoch": 0.09233333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.43515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2916,
+      "loss/crossentropy": 1.9130890280008317,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749176114797592,
+      "step": 11080
+    },
+    {
+      "epoch": 0.09241666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.5957682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3289,
+      "loss/crossentropy": 1.9334603801369667,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18508129734545947,
+      "step": 11090
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.09375,
+      "learning_rate": 0.0001,
+      "loss": 7.2589,
+      "loss/crossentropy": 1.9720475479960442,
+      "loss/hidden": 3.284765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17871356531977653,
+      "step": 11100
+    },
+    {
+      "epoch": 0.09258333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 12.9384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1781,
+      "loss/crossentropy": 2.0212643474340437,
+      "loss/hidden": 3.251953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18279597517102958,
+      "step": 11110
+    },
+    {
+      "epoch": 0.09266666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 17.95625,
+      "learning_rate": 0.0001,
+      "loss": 7.2713,
+      "loss/crossentropy": 2.0117853201925753,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18089143969118596,
+      "step": 11120
+    },
+    {
+      "epoch": 0.09275,
+      "grad_norm": 30.0,
+      "grad_norm_var": 96.61432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1381,
+      "loss/crossentropy": 1.854602186381817,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813962809741497,
+      "step": 11130
+    },
+    {
+      "epoch": 0.09283333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 10.380989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1759,
+      "loss/crossentropy": 1.961889337003231,
+      "loss/hidden": 3.309765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17766086738556625,
+      "step": 11140
+    },
+    {
+      "epoch": 0.09291666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 9.297330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1439,
+      "loss/crossentropy": 1.9535532534122466,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785486549139023,
+      "step": 11150
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 29.5,
+      "grad_norm_var": 8.742122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2287,
+      "loss/crossentropy": 2.0362745851278303,
+      "loss/hidden": 3.298828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879869369789958,
+      "step": 11160
+    },
+    {
+      "epoch": 0.09308333333333334,
+      "grad_norm": 45.0,
+      "grad_norm_var": 14.179622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2009,
+      "loss/crossentropy": 2.0105560675263403,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17597202267497777,
+      "step": 11170
+    },
+    {
+      "epoch": 0.09316666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 597.18125,
+      "learning_rate": 0.0001,
+      "loss": 7.2097,
+      "loss/crossentropy": 1.8762803822755814,
+      "loss/hidden": 3.43203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1950616255402565,
+      "step": 11180
+    },
+    {
+      "epoch": 0.09325,
+      "grad_norm": 32.75,
+      "grad_norm_var": 596.3080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0767,
+      "loss/crossentropy": 1.8652254909276962,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17202208880335093,
+      "step": 11190
+    },
+    {
+      "epoch": 0.09333333333333334,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.546809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1778,
+      "loss/crossentropy": 1.8042904496192933,
+      "loss/hidden": 3.472265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18975660093128682,
+      "step": 11200
+    },
+    {
+      "epoch": 0.09341666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 7.1103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1002,
+      "loss/crossentropy": 1.9169615030288696,
+      "loss/hidden": 3.4875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19444217309355735,
+      "step": 11210
+    },
+    {
+      "epoch": 0.0935,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.169791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.197,
+      "loss/crossentropy": 1.8842434898018836,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19005265682935715,
+      "step": 11220
+    },
+    {
+      "epoch": 0.09358333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.8494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2485,
+      "loss/crossentropy": 1.9873083665966988,
+      "loss/hidden": 3.39375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19263495169579983,
+      "step": 11230
+    },
+    {
+      "epoch": 0.09366666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.23125,
+      "learning_rate": 0.0001,
+      "loss": 7.2567,
+      "loss/crossentropy": 1.9251227468252181,
+      "loss/hidden": 3.35078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1743849113583565,
+      "step": 11240
+    },
+    {
+      "epoch": 0.09375,
+      "grad_norm": 31.375,
+      "grad_norm_var": 7.249563805703734e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2576,
+      "loss/crossentropy": 1.9011923953890801,
+      "loss/hidden": 3.462890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18671644926071168,
+      "step": 11250
+    },
+    {
+      "epoch": 0.09383333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 11.0681640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1955,
+      "loss/crossentropy": 2.006237690150738,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20202471744269132,
+      "step": 11260
+    },
+    {
+      "epoch": 0.09391666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 138.05045572916666,
+      "learning_rate": 0.0001,
+      "loss": 7.2362,
+      "loss/crossentropy": 1.9293814316391944,
+      "loss/hidden": 3.310546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17077183034271,
+      "step": 11270
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.9869140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1492,
+      "loss/crossentropy": 1.9578928805887699,
+      "loss/hidden": 3.320703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841365193016827,
+      "step": 11280
+    },
+    {
+      "epoch": 0.09408333333333334,
+      "grad_norm": 37.0,
+      "grad_norm_var": 26.484830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.125,
+      "loss/crossentropy": 1.9779614746570586,
+      "loss/hidden": 3.2859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809269543737173,
+      "step": 11290
+    },
+    {
+      "epoch": 0.09416666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 8.886393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1136,
+      "loss/crossentropy": 1.8797535948455333,
+      "loss/hidden": 3.2609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1679527286440134,
+      "step": 11300
+    },
+    {
+      "epoch": 0.09425,
+      "grad_norm": 29.0,
+      "grad_norm_var": 60.19479166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2188,
+      "loss/crossentropy": 1.9251924127340316,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857452355325222,
+      "step": 11310
+    },
+    {
+      "epoch": 0.09433333333333334,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.2812342728403206e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.24,
+      "loss/crossentropy": 1.9788130685687064,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18850935772061347,
+      "step": 11320
+    },
+    {
+      "epoch": 0.09441666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.281234162534518e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2208,
+      "loss/crossentropy": 2.0085683524608613,
+      "loss/hidden": 3.351171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17746647857129574,
+      "step": 11330
+    },
+    {
+      "epoch": 0.0945,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.5625,
+      "learning_rate": 0.0001,
+      "loss": 7.1407,
+      "loss/crossentropy": 1.95258212313056,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924839597195387,
+      "step": 11340
+    },
+    {
+      "epoch": 0.09458333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.208268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1655,
+      "loss/crossentropy": 1.9691084608435632,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20145466476678847,
+      "step": 11350
+    },
+    {
+      "epoch": 0.09466666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.5927083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0081,
+      "loss/crossentropy": 2.0321754992008207,
+      "loss/hidden": 3.175390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16929113008081914,
+      "step": 11360
+    },
+    {
+      "epoch": 0.09475,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.6083333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1561,
+      "loss/crossentropy": 1.8121349863708018,
+      "loss/hidden": 3.470703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18886148929595947,
+      "step": 11370
+    },
+    {
+      "epoch": 0.09483333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.6343098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1232,
+      "loss/crossentropy": 1.8912312626838683,
+      "loss/hidden": 3.41875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19765781927853823,
+      "step": 11380
+    },
+    {
+      "epoch": 0.09491666666666666,
+      "grad_norm": 40.25,
+      "grad_norm_var": 7.611083230815104e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1876,
+      "loss/crossentropy": 1.8526273503899575,
+      "loss/hidden": 3.380859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1915534194558859,
+      "step": 11390
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 31.75,
+      "grad_norm_var": 7.611083224671848e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3306,
+      "loss/crossentropy": 2.053269697725773,
+      "loss/hidden": 3.523828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2169021500274539,
+      "step": 11400
+    },
+    {
+      "epoch": 0.09508333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.24765625,
+      "learning_rate": 0.0001,
+      "loss": 7.197,
+      "loss/crossentropy": 1.8971049457788467,
+      "loss/hidden": 3.344140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899229221045971,
+      "step": 11410
+    },
+    {
+      "epoch": 0.09516666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.545833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2505,
+      "loss/crossentropy": 2.0065065011382104,
+      "loss/hidden": 3.257421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18574204593896865,
+      "step": 11420
+    },
+    {
+      "epoch": 0.09525,
+      "grad_norm": 35.75,
+      "grad_norm_var": 4.10625,
+      "learning_rate": 0.0001,
+      "loss": 7.2836,
+      "loss/crossentropy": 2.024059595167637,
+      "loss/hidden": 3.461328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20745400954037904,
+      "step": 11430
+    },
+    {
+      "epoch": 0.09533333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.423958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.138,
+      "loss/crossentropy": 1.8951629042625426,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17089098505675793,
+      "step": 11440
+    },
+    {
+      "epoch": 0.09541666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.730989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.23,
+      "loss/crossentropy": 2.002152493596077,
+      "loss/hidden": 3.3421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19319903161376714,
+      "step": 11450
+    },
+    {
+      "epoch": 0.0955,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.1997395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1999,
+      "loss/crossentropy": 1.9776404619216919,
+      "loss/hidden": 3.406640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815666791051626,
+      "step": 11460
+    },
+    {
+      "epoch": 0.09558333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.759375,
+      "learning_rate": 0.0001,
+      "loss": 7.2827,
+      "loss/crossentropy": 2.030766838788986,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19363442957401275,
+      "step": 11470
+    },
+    {
+      "epoch": 0.09566666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.2181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0091,
+      "loss/crossentropy": 1.6673102363944055,
+      "loss/hidden": 3.241015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16620758287608622,
+      "step": 11480
+    },
+    {
+      "epoch": 0.09575,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.2978515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2373,
+      "loss/crossentropy": 1.9102192282676698,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18641335964202882,
+      "step": 11490
+    },
+    {
+      "epoch": 0.09583333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.9113932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2123,
+      "loss/crossentropy": 1.988862706720829,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1913462642580271,
+      "step": 11500
+    },
+    {
+      "epoch": 0.09591666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.8051432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1625,
+      "loss/crossentropy": 1.9352146543562412,
+      "loss/hidden": 3.305859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18198430854827166,
+      "step": 11510
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.8556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.144,
+      "loss/crossentropy": 2.1403306275606155,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18342356365174056,
+      "step": 11520
+    },
+    {
+      "epoch": 0.09608333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.2358723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0883,
+      "loss/crossentropy": 1.905369622260332,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17095069997012616,
+      "step": 11530
+    },
+    {
+      "epoch": 0.09616666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.1802083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2336,
+      "loss/crossentropy": 1.9543854504823686,
+      "loss/hidden": 3.4578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19613500405102968,
+      "step": 11540
+    },
+    {
+      "epoch": 0.09625,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.1705729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1571,
+      "loss/crossentropy": 1.9872371323406697,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19156625671312213,
+      "step": 11550
+    },
+    {
+      "epoch": 0.09633333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.9067057291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2433,
+      "loss/crossentropy": 1.9041082590818406,
+      "loss/hidden": 3.5109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802822045981884,
+      "step": 11560
+    },
+    {
+      "epoch": 0.09641666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.378059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1836,
+      "loss/crossentropy": 1.8514071375131607,
+      "loss/hidden": 3.428125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843704305589199,
+      "step": 11570
+    },
+    {
+      "epoch": 0.0965,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.169205729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2139,
+      "loss/crossentropy": 1.8199843727052212,
+      "loss/hidden": 3.25078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16416108421981335,
+      "step": 11580
+    },
+    {
+      "epoch": 0.09658333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.3875,
+      "learning_rate": 0.0001,
+      "loss": 7.2476,
+      "loss/crossentropy": 1.8533273190259933,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18512159064412118,
+      "step": 11590
+    },
+    {
+      "epoch": 0.09666666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.273372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2533,
+      "loss/crossentropy": 2.0454796701669693,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19563965629786254,
+      "step": 11600
+    },
+    {
+      "epoch": 0.09675,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.1801432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1836,
+      "loss/crossentropy": 1.9645930647850036,
+      "loss/hidden": 3.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19418679140508174,
+      "step": 11610
+    },
+    {
+      "epoch": 0.09683333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.19765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1132,
+      "loss/crossentropy": 1.8267022013664245,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1694145280867815,
+      "step": 11620
+    },
+    {
+      "epoch": 0.09691666666666666,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.294791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1224,
+      "loss/crossentropy": 1.9234918773174285,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16964951027184724,
+      "step": 11630
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.5166666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1406,
+      "loss/crossentropy": 2.0221316441893578,
+      "loss/hidden": 3.266796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19291031137108802,
+      "step": 11640
+    },
+    {
+      "epoch": 0.09708333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.47890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2739,
+      "loss/crossentropy": 1.893275460600853,
+      "loss/hidden": 3.46875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18795952778309583,
+      "step": 11650
+    },
+    {
+      "epoch": 0.09716666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.626822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1946,
+      "loss/crossentropy": 1.9983355477452278,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20295516289770604,
+      "step": 11660
+    },
+    {
+      "epoch": 0.09725,
+      "grad_norm": 29.875,
+      "grad_norm_var": 25.284309895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.2673,
+      "loss/crossentropy": 1.929758571088314,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18477606382220985,
+      "step": 11670
+    },
+    {
+      "epoch": 0.09733333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 26.224739583333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2434,
+      "loss/crossentropy": 1.981206202507019,
+      "loss/hidden": 3.44375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17995677031576635,
+      "step": 11680
+    },
+    {
+      "epoch": 0.09741666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.1205729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1195,
+      "loss/crossentropy": 1.9690660208463668,
+      "loss/hidden": 3.440234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18974249269813298,
+      "step": 11690
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.8900390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0954,
+      "loss/crossentropy": 1.8680916860699655,
+      "loss/hidden": 3.297265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17862058337777853,
+      "step": 11700
+    },
+    {
+      "epoch": 0.09758333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.328125,
+      "learning_rate": 0.0001,
+      "loss": 7.2492,
+      "loss/crossentropy": 1.9948291823267936,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18580178990960122,
+      "step": 11710
+    },
+    {
+      "epoch": 0.09766666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 21.081184895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.1597,
+      "loss/crossentropy": 1.930022406578064,
+      "loss/hidden": 3.231640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18058926239609718,
+      "step": 11720
+    },
+    {
+      "epoch": 0.09775,
+      "grad_norm": 30.125,
+      "grad_norm_var": 31.927018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0945,
+      "loss/crossentropy": 1.833520622551441,
+      "loss/hidden": 3.534765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18105747178196907,
+      "step": 11730
+    },
+    {
+      "epoch": 0.09783333333333333,
+      "grad_norm": 36.75,
+      "grad_norm_var": 35.34557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2264,
+      "loss/crossentropy": 1.920599752664566,
+      "loss/hidden": 3.5875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20554377418011427,
+      "step": 11740
+    },
+    {
+      "epoch": 0.09791666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 12.3759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.203,
+      "loss/crossentropy": 1.8264538645744324,
+      "loss/hidden": 3.478125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17480854503810406,
+      "step": 11750
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.405989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0892,
+      "loss/crossentropy": 1.849447526782751,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17364850863814354,
+      "step": 11760
+    },
+    {
+      "epoch": 0.09808333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 14.484375,
+      "learning_rate": 0.0001,
+      "loss": 7.2483,
+      "loss/crossentropy": 1.991353189945221,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18836470153182744,
+      "step": 11770
+    },
+    {
+      "epoch": 0.09816666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 10.457291666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2123,
+      "loss/crossentropy": 1.9754149086773396,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1845713852904737,
+      "step": 11780
+    },
+    {
+      "epoch": 0.09825,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.493684895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0911,
+      "loss/crossentropy": 1.8103220909833908,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1820437341928482,
+      "step": 11790
+    },
+    {
+      "epoch": 0.09833333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.655989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.157,
+      "loss/crossentropy": 2.013156922161579,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18750574309378862,
+      "step": 11800
+    },
+    {
+      "epoch": 0.09841666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.821809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1571,
+      "loss/crossentropy": 1.893529909849167,
+      "loss/hidden": 3.4421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17757591046392918,
+      "step": 11810
+    },
+    {
+      "epoch": 0.0985,
+      "grad_norm": 28.75,
+      "grad_norm_var": 10.958072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2017,
+      "loss/crossentropy": 2.1257301539182665,
+      "loss/hidden": 3.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17153541482985019,
+      "step": 11820
+    },
+    {
+      "epoch": 0.09858333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 8.939322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2972,
+      "loss/crossentropy": 2.0279329270124435,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19466775842010975,
+      "step": 11830
+    },
+    {
+      "epoch": 0.09866666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 32.6369140625,
+      "learning_rate": 0.0001,
+      "loss": 7.3001,
+      "loss/crossentropy": 1.9315491870045662,
+      "loss/hidden": 3.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17389588709920645,
+      "step": 11840
+    },
+    {
+      "epoch": 0.09875,
+      "grad_norm": 29.375,
+      "grad_norm_var": 38.05416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1412,
+      "loss/crossentropy": 1.9916979007422924,
+      "loss/hidden": 3.308203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1808990327641368,
+      "step": 11850
+    },
+    {
+      "epoch": 0.09883333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.410872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2075,
+      "loss/crossentropy": 2.017706024646759,
+      "loss/hidden": 3.4109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19204182233661413,
+      "step": 11860
+    },
+    {
+      "epoch": 0.09891666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.340102344174469e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1396,
+      "loss/crossentropy": 1.9543035812675953,
+      "loss/hidden": 3.283203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1944132639095187,
+      "step": 11870
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 31.5,
+      "grad_norm_var": 14.304166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1053,
+      "loss/crossentropy": 1.9666469424962998,
+      "loss/hidden": 3.473046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20341123901307584,
+      "step": 11880
+    },
+    {
+      "epoch": 0.09908333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.380208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1059,
+      "loss/crossentropy": 1.9001951694488526,
+      "loss/hidden": 3.301953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19399545583873987,
+      "step": 11890
+    },
+    {
+      "epoch": 0.09916666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.60625,
+      "learning_rate": 0.0001,
+      "loss": 7.1082,
+      "loss/crossentropy": 1.8756570994853974,
+      "loss/hidden": 3.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1765897661447525,
+      "step": 11900
+    },
+    {
+      "epoch": 0.09925,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.5603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0572,
+      "loss/crossentropy": 1.8405318051576613,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.172229509241879,
+      "step": 11910
+    },
+    {
+      "epoch": 0.09933333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 523.1309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1834,
+      "loss/crossentropy": 1.8818136483430863,
+      "loss/hidden": 3.278515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18015837762504816,
+      "step": 11920
+    },
+    {
+      "epoch": 0.09941666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.847702671604319e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1697,
+      "loss/crossentropy": 1.9347866751253604,
+      "loss/hidden": 3.41640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18266602866351606,
+      "step": 11930
+    },
+    {
+      "epoch": 0.0995,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.8477026714882765e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2432,
+      "loss/crossentropy": 1.9067823722958566,
+      "loss/hidden": 3.466015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19923510327935218,
+      "step": 11940
+    },
+    {
+      "epoch": 0.09958333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 37.52682291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1902,
+      "loss/crossentropy": 2.1187405943870545,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17510780673474075,
+      "step": 11950
+    },
+    {
+      "epoch": 0.09966666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 35.13645833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2112,
+      "loss/crossentropy": 2.0079662546515467,
+      "loss/hidden": 3.35390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1920156165957451,
+      "step": 11960
+    },
+    {
+      "epoch": 0.09975,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.400455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2292,
+      "loss/crossentropy": 2.0223580092191695,
+      "loss/hidden": 3.501171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19307674951851367,
+      "step": 11970
+    },
+    {
+      "epoch": 0.09983333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.802018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1028,
+      "loss/crossentropy": 2.0315533101558687,
+      "loss/hidden": 3.33046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2036166999489069,
+      "step": 11980
+    },
+    {
+      "epoch": 0.09991666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.1455729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1072,
+      "loss/crossentropy": 1.9079547315835952,
+      "loss/hidden": 3.348828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17668173555284739,
+      "step": 11990
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 30.75,
+      "grad_norm_var": 143.31712239583334,
+      "learning_rate": 0.0001,
+      "loss": 7.1285,
+      "loss/crossentropy": 1.9362183973193168,
+      "loss/hidden": 3.38984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19406130611896516,
+      "step": 12000
+    },
+    {
+      "epoch": 0.10008333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 139.22649739583332,
+      "learning_rate": 0.0001,
+      "loss": 7.2505,
+      "loss/crossentropy": 1.92686520665884,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18346634581685067,
+      "step": 12010
+    },
+    {
+      "epoch": 0.10016666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.153580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2547,
+      "loss/crossentropy": 2.0029779627919195,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18490857090801,
+      "step": 12020
+    },
+    {
+      "epoch": 0.10025,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1508,
+      "loss/crossentropy": 1.900698359310627,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18282779939472676,
+      "step": 12030
+    },
+    {
+      "epoch": 0.10033333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.4083702445597435e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1056,
+      "loss/crossentropy": 1.9221736580133437,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18743555713444948,
+      "step": 12040
+    },
+    {
+      "epoch": 0.10041666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.6635416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1285,
+      "loss/crossentropy": 1.9425101041793824,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18018915858119727,
+      "step": 12050
+    },
+    {
+      "epoch": 0.1005,
+      "grad_norm": 31.5,
+      "grad_norm_var": 9.189322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2847,
+      "loss/crossentropy": 1.923607735335827,
+      "loss/hidden": 3.470703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147286396473646,
+      "step": 12060
+    },
+    {
+      "epoch": 0.10058333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 8.940559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2109,
+      "loss/crossentropy": 1.879532267153263,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18306696321815252,
+      "step": 12070
+    },
+    {
+      "epoch": 0.10066666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.014322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2161,
+      "loss/crossentropy": 2.024162586033344,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17965770401060582,
+      "step": 12080
+    },
+    {
+      "epoch": 0.10075,
+      "grad_norm": 34.0,
+      "grad_norm_var": 8.949934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.228,
+      "loss/crossentropy": 1.937704546749592,
+      "loss/hidden": 3.403515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18645930513739586,
+      "step": 12090
+    },
+    {
+      "epoch": 0.10083333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 8.51875,
+      "learning_rate": 0.0001,
+      "loss": 7.1664,
+      "loss/crossentropy": 1.9228780284523963,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19229220058768987,
+      "step": 12100
+    },
+    {
+      "epoch": 0.10091666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.036874281451521e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3095,
+      "loss/crossentropy": 1.918504586070776,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18654032479971647,
+      "step": 12110
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.472330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.183,
+      "loss/crossentropy": 1.9967175737023353,
+      "loss/hidden": 3.31484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994472701102495,
+      "step": 12120
+    },
+    {
+      "epoch": 0.10108333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.7593098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1808,
+      "loss/crossentropy": 2.088398113846779,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18488444089889527,
+      "step": 12130
+    },
+    {
+      "epoch": 0.10116666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 23.390625,
+      "learning_rate": 0.0001,
+      "loss": 7.262,
+      "loss/crossentropy": 1.9120210327208043,
+      "loss/hidden": 3.31953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16619170643389225,
+      "step": 12140
+    },
+    {
+      "epoch": 0.10125,
+      "grad_norm": 28.125,
+      "grad_norm_var": 2.76015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0382,
+      "loss/crossentropy": 1.781062449514866,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1894195105880499,
+      "step": 12150
+    },
+    {
+      "epoch": 0.10133333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.354622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1127,
+      "loss/crossentropy": 1.760746394097805,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18243793863803148,
+      "step": 12160
+    },
+    {
+      "epoch": 0.10141666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.5895833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.8511279866099357,
+      "loss/hidden": 3.4078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18570996094495057,
+      "step": 12170
+    },
+    {
+      "epoch": 0.1015,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.5072916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2137,
+      "loss/crossentropy": 1.956222912669182,
+      "loss/hidden": 3.390234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18348711188882588,
+      "step": 12180
+    },
+    {
+      "epoch": 0.10158333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.6708333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2334,
+      "loss/crossentropy": 1.8933769166469574,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833400011062622,
+      "step": 12190
+    },
+    {
+      "epoch": 0.10166666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.59765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2167,
+      "loss/crossentropy": 1.9778078615665435,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19039046382531524,
+      "step": 12200
+    },
+    {
+      "epoch": 0.10175,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.2197916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.9912208177149295,
+      "loss/hidden": 3.332421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1730046510696411,
+      "step": 12210
+    },
+    {
+      "epoch": 0.10183333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.0697916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1379,
+      "loss/crossentropy": 1.9885770983994007,
+      "loss/hidden": 3.3875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1814948114566505,
+      "step": 12220
+    },
+    {
+      "epoch": 0.10191666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.5020833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.119,
+      "loss/crossentropy": 1.8945732980966568,
+      "loss/hidden": 3.26171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17931772507727145,
+      "step": 12230
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.414518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2614,
+      "loss/crossentropy": 2.017138597369194,
+      "loss/hidden": 3.2484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1819797668606043,
+      "step": 12240
+    },
+    {
+      "epoch": 0.10208333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.1416015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2403,
+      "loss/crossentropy": 1.983961570262909,
+      "loss/hidden": 3.440234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19026287347078324,
+      "step": 12250
+    },
+    {
+      "epoch": 0.10216666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.020247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2413,
+      "loss/crossentropy": 2.027003343403339,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1870082238689065,
+      "step": 12260
+    },
+    {
+      "epoch": 0.10225,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.2635411521926246e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3057,
+      "loss/crossentropy": 1.8065140090882779,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1696597938425839,
+      "step": 12270
+    },
+    {
+      "epoch": 0.10233333333333333,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.6416015625,
+      "learning_rate": 0.0001,
+      "loss": 7.252,
+      "loss/crossentropy": 1.9603703975677491,
+      "loss/hidden": 3.43828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1895959597080946,
+      "step": 12280
+    },
+    {
+      "epoch": 0.10241666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.275,
+      "learning_rate": 0.0001,
+      "loss": 7.1794,
+      "loss/crossentropy": 1.9987883269786835,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18563129249960184,
+      "step": 12290
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.1947916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1426,
+      "loss/crossentropy": 2.0014384269714354,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844819789752364,
+      "step": 12300
+    },
+    {
+      "epoch": 0.10258333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.8509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2895,
+      "loss/crossentropy": 1.8975801825523377,
+      "loss/hidden": 3.331640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19599497634917498,
+      "step": 12310
+    },
+    {
+      "epoch": 0.10266666666666667,
+      "grad_norm": 43.0,
+      "grad_norm_var": 11.52265625,
+      "learning_rate": 0.0001,
+      "loss": 7.219,
+      "loss/crossentropy": 1.9807260885834694,
+      "loss/hidden": 3.291796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19955590814352037,
+      "step": 12320
+    },
+    {
+      "epoch": 0.10275,
+      "grad_norm": 31.625,
+      "grad_norm_var": 13.76015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1334,
+      "loss/crossentropy": 1.9443981781601907,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18899141624569893,
+      "step": 12330
+    },
+    {
+      "epoch": 0.10283333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.0926432291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3424,
+      "loss/crossentropy": 1.9711116895079612,
+      "loss/hidden": 3.5859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1896815186366439,
+      "step": 12340
+    },
+    {
+      "epoch": 0.10291666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.956184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.16,
+      "loss/crossentropy": 1.6733422517776488,
+      "loss/hidden": 3.437890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17213238067924977,
+      "step": 12350
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.196809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1909,
+      "loss/crossentropy": 2.028626634180546,
+      "loss/hidden": 3.316015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17515519298613072,
+      "step": 12360
+    },
+    {
+      "epoch": 0.10308333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.872916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2359,
+      "loss/crossentropy": 1.9783917248249054,
+      "loss/hidden": 3.403515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18638654407113792,
+      "step": 12370
+    },
+    {
+      "epoch": 0.10316666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.9160807291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1366,
+      "loss/crossentropy": 1.867820891737938,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18024916313588618,
+      "step": 12380
+    },
+    {
+      "epoch": 0.10325,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.887239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2216,
+      "loss/crossentropy": 1.9069418162107468,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17871958892792464,
+      "step": 12390
+    },
+    {
+      "epoch": 0.10333333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 10.955143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1852,
+      "loss/crossentropy": 1.8888402715325356,
+      "loss/hidden": 3.5171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1944135267287493,
+      "step": 12400
+    },
+    {
+      "epoch": 0.10341666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.622330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.13,
+      "loss/crossentropy": 1.8761833429336547,
+      "loss/hidden": 3.44296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19608120806515217,
+      "step": 12410
+    },
+    {
+      "epoch": 0.1035,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0345,
+      "loss/crossentropy": 1.8996757566928864,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810669604688883,
+      "step": 12420
+    },
+    {
+      "epoch": 0.10358333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 0.9363932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0908,
+      "loss/crossentropy": 1.951111237704754,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18944073971360922,
+      "step": 12430
+    },
+    {
+      "epoch": 0.10366666666666667,
+      "grad_norm": 37.5,
+      "grad_norm_var": 7.67265625,
+      "learning_rate": 0.0001,
+      "loss": 7.3028,
+      "loss/crossentropy": 2.06214716732502,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1852529702708125,
+      "step": 12440
+    },
+    {
+      "epoch": 0.10375,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.862239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0807,
+      "loss/crossentropy": 2.0049709871411325,
+      "loss/hidden": 3.297265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18642192110419273,
+      "step": 12450
+    },
+    {
+      "epoch": 0.10383333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.1587890625,
+      "learning_rate": 0.0001,
+      "loss": 7.158,
+      "loss/crossentropy": 1.9456622168421744,
+      "loss/hidden": 3.477734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20739529095590115,
+      "step": 12460
+    },
+    {
+      "epoch": 0.10391666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 0.5374348958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0698,
+      "loss/crossentropy": 1.9484572559595108,
+      "loss/hidden": 3.333984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1869681142270565,
+      "step": 12470
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.128580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1697,
+      "loss/crossentropy": 1.9990088611841201,
+      "loss/hidden": 3.35703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18577630929648875,
+      "step": 12480
+    },
+    {
+      "epoch": 0.10408333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.630989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0908,
+      "loss/crossentropy": 1.912271463871002,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17868589647114277,
+      "step": 12490
+    },
+    {
+      "epoch": 0.10416666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.6998681514292026e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.9653410285711288,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19682321604341269,
+      "step": 12500
+    },
+    {
+      "epoch": 0.10425,
+      "grad_norm": 30.125,
+      "grad_norm_var": 9.568684895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1248,
+      "loss/crossentropy": 1.884364977478981,
+      "loss/hidden": 3.519921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1960652332752943,
+      "step": 12510
+    },
+    {
+      "epoch": 0.10433333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.830989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2535,
+      "loss/crossentropy": 2.024576540291309,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20311113484203816,
+      "step": 12520
+    },
+    {
+      "epoch": 0.10441666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.470247395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0843,
+      "loss/crossentropy": 2.014328496158123,
+      "loss/hidden": 3.275390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2006282364949584,
+      "step": 12530
+    },
+    {
+      "epoch": 0.1045,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.540559895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2808,
+      "loss/crossentropy": 1.9064289540052415,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18439610339701176,
+      "step": 12540
+    },
+    {
+      "epoch": 0.10458333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 7.023893229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2478,
+      "loss/crossentropy": 2.0493562422692775,
+      "loss/hidden": 3.327734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19878917997702955,
+      "step": 12550
+    },
+    {
+      "epoch": 0.10466666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 11.77890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2023,
+      "loss/crossentropy": 1.9017269283533096,
+      "loss/hidden": 3.291796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18583347275853157,
+      "step": 12560
+    },
+    {
+      "epoch": 0.10475,
+      "grad_norm": 32.75,
+      "grad_norm_var": 12.576822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2167,
+      "loss/crossentropy": 1.9804055333137511,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20087866000831128,
+      "step": 12570
+    },
+    {
+      "epoch": 0.10483333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 8.23125,
+      "learning_rate": 0.0001,
+      "loss": 7.1885,
+      "loss/crossentropy": 1.947347742319107,
+      "loss/hidden": 3.24453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16737934947013855,
+      "step": 12580
+    },
+    {
+      "epoch": 0.10491666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.5197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2053,
+      "loss/crossentropy": 1.978326364606619,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18662697710096837,
+      "step": 12590
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.31875,
+      "learning_rate": 0.0001,
+      "loss": 7.1406,
+      "loss/crossentropy": 2.0392678931355475,
+      "loss/hidden": 3.528125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19746034014970065,
+      "step": 12600
+    },
+    {
+      "epoch": 0.10508333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.6900390625,
+      "learning_rate": 0.0001,
+      "loss": 7.15,
+      "loss/crossentropy": 1.933197446167469,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17729342449456453,
+      "step": 12610
+    },
+    {
+      "epoch": 0.10516666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 6.513541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1468,
+      "loss/crossentropy": 1.9596221596002579,
+      "loss/hidden": 3.266015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18164707385003567,
+      "step": 12620
+    },
+    {
+      "epoch": 0.10525,
+      "grad_norm": 30.25,
+      "grad_norm_var": 16.212239583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.1643,
+      "loss/crossentropy": 1.8631553776562213,
+      "loss/hidden": 3.509375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20503986105322838,
+      "step": 12630
+    },
+    {
+      "epoch": 0.10533333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 20.940625,
+      "learning_rate": 0.0001,
+      "loss": 7.2662,
+      "loss/crossentropy": 2.0203105390071867,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20416631661355494,
+      "step": 12640
+    },
+    {
+      "epoch": 0.10541666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 10.104622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1907,
+      "loss/crossentropy": 2.040492755174637,
+      "loss/hidden": 3.319140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16889610178768635,
+      "step": 12650
+    },
+    {
+      "epoch": 0.1055,
+      "grad_norm": 32.0,
+      "grad_norm_var": 508.06764322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.1739,
+      "loss/crossentropy": 2.0491758584976196,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18211234211921692,
+      "step": 12660
+    },
+    {
+      "epoch": 0.10558333333333333,
+      "grad_norm": 41.0,
+      "grad_norm_var": 506.99108072916664,
+      "learning_rate": 0.0001,
+      "loss": 7.2787,
+      "loss/crossentropy": 1.9086828008294106,
+      "loss/hidden": 3.4828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19612369406968355,
+      "step": 12670
+    },
+    {
+      "epoch": 0.10566666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 38.83515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2184,
+      "loss/crossentropy": 1.8647835373878479,
+      "loss/hidden": 3.39765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19826739598065615,
+      "step": 12680
+    },
+    {
+      "epoch": 0.10575,
+      "grad_norm": 35.75,
+      "grad_norm_var": 36.9947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1581,
+      "loss/crossentropy": 1.7978095635771751,
+      "loss/hidden": 3.53671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17947316039353609,
+      "step": 12690
+    },
+    {
+      "epoch": 0.10583333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.6885416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.242,
+      "loss/crossentropy": 2.009678316116333,
+      "loss/hidden": 3.370703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1803743414580822,
+      "step": 12700
+    },
+    {
+      "epoch": 0.10591666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.7035807291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1796,
+      "loss/crossentropy": 1.833764837682247,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1640061866492033,
+      "step": 12710
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 30.5,
+      "grad_norm_var": 33.670833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1342,
+      "loss/crossentropy": 2.0209478750824927,
+      "loss/hidden": 3.3109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18588003953918814,
+      "step": 12720
+    },
+    {
+      "epoch": 0.10608333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 33.94140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2657,
+      "loss/crossentropy": 2.1548540562391283,
+      "loss/hidden": 3.478125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19924551136791707,
+      "step": 12730
+    },
+    {
+      "epoch": 0.10616666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.957291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1361,
+      "loss/crossentropy": 2.0598884031176565,
+      "loss/hidden": 3.374609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1739529339596629,
+      "step": 12740
+    },
+    {
+      "epoch": 0.10625,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.7509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1158,
+      "loss/crossentropy": 2.1019793540239333,
+      "loss/hidden": 3.292578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1942106056958437,
+      "step": 12750
+    },
+    {
+      "epoch": 0.10633333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.787955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1677,
+      "loss/crossentropy": 2.0136178761720656,
+      "loss/hidden": 3.27734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18769948203116654,
+      "step": 12760
+    },
+    {
+      "epoch": 0.10641666666666667,
+      "grad_norm": 35.75,
+      "grad_norm_var": 7.758153900988827e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2253,
+      "loss/crossentropy": 1.9753437519073487,
+      "loss/hidden": 3.296484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1741545509546995,
+      "step": 12770
+    },
+    {
+      "epoch": 0.1065,
+      "grad_norm": 32.25,
+      "grad_norm_var": 16.728059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1686,
+      "loss/crossentropy": 1.8049860179424286,
+      "loss/hidden": 3.5328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18819726444780827,
+      "step": 12780
+    },
+    {
+      "epoch": 0.10658333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.0580729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1303,
+      "loss/crossentropy": 1.9733722150325774,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18745031766593456,
+      "step": 12790
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.0518229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2166,
+      "loss/crossentropy": 1.9571923613548279,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886135835200548,
+      "step": 12800
+    },
+    {
+      "epoch": 0.10675,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.2587890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0812,
+      "loss/crossentropy": 2.009890964627266,
+      "loss/hidden": 3.324609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18211095612496137,
+      "step": 12810
+    },
+    {
+      "epoch": 0.10683333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.9332682291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2605,
+      "loss/crossentropy": 2.034913158416748,
+      "loss/hidden": 3.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1808745814487338,
+      "step": 12820
+    },
+    {
+      "epoch": 0.10691666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.6155598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.9772749245166779,
+      "loss/hidden": 3.334765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17896076738834382,
+      "step": 12830
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.3806640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2159,
+      "loss/crossentropy": 1.960922408103943,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17023195046931505,
+      "step": 12840
+    },
+    {
+      "epoch": 0.10708333333333334,
+      "grad_norm": 27.25,
+      "grad_norm_var": 3.863541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0719,
+      "loss/crossentropy": 2.0399751991033552,
+      "loss/hidden": 3.289453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18855413533747195,
+      "step": 12850
+    },
+    {
+      "epoch": 0.10716666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.862434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0622,
+      "loss/crossentropy": 1.9447935424745082,
+      "loss/hidden": 3.22109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16721157040446996,
+      "step": 12860
+    },
+    {
+      "epoch": 0.10725,
+      "grad_norm": 29.25,
+      "grad_norm_var": 25.332747395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.106,
+      "loss/crossentropy": 1.7755009964108468,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17843105010688304,
+      "step": 12870
+    },
+    {
+      "epoch": 0.10733333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 31.406705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1961,
+      "loss/crossentropy": 1.9738472610712052,
+      "loss/hidden": 3.35,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18686946518719197,
+      "step": 12880
+    },
+    {
+      "epoch": 0.10741666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 10.451041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3298,
+      "loss/crossentropy": 1.9548229411244393,
+      "loss/hidden": 3.576171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1992871094495058,
+      "step": 12890
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 33.5,
+      "grad_norm_var": 24.6853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.255,
+      "loss/crossentropy": 1.988719192147255,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18740662932395935,
+      "step": 12900
+    },
+    {
+      "epoch": 0.10758333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 26.306705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 1.999283502995968,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1943864120170474,
+      "step": 12910
+    },
+    {
+      "epoch": 0.10766666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8405598958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.3037,
+      "loss/crossentropy": 1.8890342444181443,
+      "loss/hidden": 3.365234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21169042512774466,
+      "step": 12920
+    },
+    {
+      "epoch": 0.10775,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.9385416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2528,
+      "loss/crossentropy": 1.9625538483262062,
+      "loss/hidden": 3.320703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1784692605957389,
+      "step": 12930
+    },
+    {
+      "epoch": 0.10783333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.4551432291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0633,
+      "loss/crossentropy": 1.9682739168405532,
+      "loss/hidden": 3.37265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937200540676713,
+      "step": 12940
+    },
+    {
+      "epoch": 0.10791666666666666,
+      "grad_norm": 55.75,
+      "grad_norm_var": 446.13645833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1596,
+      "loss/crossentropy": 1.8137162983417512,
+      "loss/hidden": 3.35078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16852411348372698,
+      "step": 12950
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 34.75,
+      "grad_norm_var": 448.1275390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1881,
+      "loss/crossentropy": 1.885158045589924,
+      "loss/hidden": 3.371875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18899058364331722,
+      "step": 12960
+    },
+    {
+      "epoch": 0.10808333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.46881805335309e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2007,
+      "loss/crossentropy": 1.965352798253298,
+      "loss/hidden": 3.403515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19080529361963272,
+      "step": 12970
+    },
+    {
+      "epoch": 0.10816666666666666,
+      "grad_norm": 28.375,
+      "grad_norm_var": 11.783268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1274,
+      "loss/crossentropy": 1.9587459579110145,
+      "loss/hidden": 3.430859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18071820214390755,
+      "step": 12980
+    },
+    {
+      "epoch": 0.10825,
+      "grad_norm": 30.5,
+      "grad_norm_var": 12.4375,
+      "learning_rate": 0.0001,
+      "loss": 7.2373,
+      "loss/crossentropy": 2.0089691922068598,
+      "loss/hidden": 3.2796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.177796680200845,
+      "step": 12990
+    },
+    {
+      "epoch": 0.10833333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.018489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1284,
+      "loss/crossentropy": 1.9447548791766167,
+      "loss/hidden": 3.25078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17130674179643393,
+      "step": 13000
+    },
+    {
+      "epoch": 0.10841666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.478059895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1158,
+      "loss/crossentropy": 1.9497190311551094,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1769522450864315,
+      "step": 13010
+    },
+    {
+      "epoch": 0.1085,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.1830729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0497,
+      "loss/crossentropy": 2.0208690792322157,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18711450472474098,
+      "step": 13020
+    },
+    {
+      "epoch": 0.10858333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 12.246875,
+      "learning_rate": 0.0001,
+      "loss": 7.2283,
+      "loss/crossentropy": 1.8341820068657397,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821217535994947,
+      "step": 13030
+    },
+    {
+      "epoch": 0.10866666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 36.614322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1576,
+      "loss/crossentropy": 1.7387959524989127,
+      "loss/hidden": 3.507421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774640517309308,
+      "step": 13040
+    },
+    {
+      "epoch": 0.10875,
+      "grad_norm": 30.25,
+      "grad_norm_var": 21.225455729166665,
+      "learning_rate": 0.0001,
+      "loss": 7.0853,
+      "loss/crossentropy": 1.9904129430651665,
+      "loss/hidden": 3.28671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18333331272006034,
+      "step": 13050
+    },
+    {
+      "epoch": 0.10883333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 42.6181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0897,
+      "loss/crossentropy": 1.9294225499033928,
+      "loss/hidden": 3.249609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16889635231345893,
+      "step": 13060
+    },
+    {
+      "epoch": 0.10891666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 10.7728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0831,
+      "loss/crossentropy": 1.9553338959813118,
+      "loss/hidden": 3.380859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1850135087966919,
+      "step": 13070
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 35.5,
+      "grad_norm_var": 30.639322916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2265,
+      "loss/crossentropy": 1.848113013803959,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19494788870215415,
+      "step": 13080
+    },
+    {
+      "epoch": 0.10908333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 22.126497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1766,
+      "loss/crossentropy": 1.802114023268223,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17259575948119163,
+      "step": 13090
+    },
+    {
+      "epoch": 0.10916666666666666,
+      "grad_norm": 48.75,
+      "grad_norm_var": 52.8337890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1689,
+      "loss/crossentropy": 1.9827998235821724,
+      "loss/hidden": 3.5109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21250546909868717,
+      "step": 13100
+    },
+    {
+      "epoch": 0.10925,
+      "grad_norm": 30.125,
+      "grad_norm_var": 27.980989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1465,
+      "loss/crossentropy": 1.9074020877480506,
+      "loss/hidden": 3.441015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19721721541136503,
+      "step": 13110
+    },
+    {
+      "epoch": 0.10933333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 6.558333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0929,
+      "loss/crossentropy": 1.9373712703585624,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18476964067667723,
+      "step": 13120
+    },
+    {
+      "epoch": 0.10941666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.1009765625,
+      "learning_rate": 0.0001,
+      "loss": 7.098,
+      "loss/crossentropy": 1.9109055921435356,
+      "loss/hidden": 3.3859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18421365208923818,
+      "step": 13130
+    },
+    {
+      "epoch": 0.1095,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.193489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1484,
+      "loss/crossentropy": 1.9655496269464492,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19097178876399995,
+      "step": 13140
+    },
+    {
+      "epoch": 0.10958333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 13.835872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.124,
+      "loss/crossentropy": 2.060957744717598,
+      "loss/hidden": 3.306640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19035868030041456,
+      "step": 13150
+    },
+    {
+      "epoch": 0.10966666666666666,
+      "grad_norm": 39.25,
+      "grad_norm_var": 15.633333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1147,
+      "loss/crossentropy": 2.010559469461441,
+      "loss/hidden": 3.44296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2029681695625186,
+      "step": 13160
+    },
+    {
+      "epoch": 0.10975,
+      "grad_norm": 33.25,
+      "grad_norm_var": 12.177083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1987,
+      "loss/crossentropy": 1.8469918087124824,
+      "loss/hidden": 3.332421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17849041391164064,
+      "step": 13170
+    },
+    {
+      "epoch": 0.10983333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.736393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1348,
+      "loss/crossentropy": 1.9129036650061608,
+      "loss/hidden": 3.460546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1853424146771431,
+      "step": 13180
+    },
+    {
+      "epoch": 0.10991666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 8.360510341526499e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1106,
+      "loss/crossentropy": 1.7563239932060242,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19243694096803665,
+      "step": 13190
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 28.875,
+      "grad_norm_var": 8.360510341602696e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1245,
+      "loss/crossentropy": 1.8102601557970046,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17791619021445512,
+      "step": 13200
+    },
+    {
+      "epoch": 0.11008333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 7.927018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1177,
+      "loss/crossentropy": 1.953826653957367,
+      "loss/hidden": 3.285546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18068501241505147,
+      "step": 13210
+    },
+    {
+      "epoch": 0.11016666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.443684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1553,
+      "loss/crossentropy": 1.846192567795515,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17584749069064856,
+      "step": 13220
+    },
+    {
+      "epoch": 0.11025,
+      "grad_norm": 28.375,
+      "grad_norm_var": 5.149934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0971,
+      "loss/crossentropy": 1.9693508207798005,
+      "loss/hidden": 3.2765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1745366407558322,
+      "step": 13230
+    },
+    {
+      "epoch": 0.11033333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.89765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2049,
+      "loss/crossentropy": 2.001208884268999,
+      "loss/hidden": 3.4171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20233699418604373,
+      "step": 13240
+    },
+    {
+      "epoch": 0.11041666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.5504557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1502,
+      "loss/crossentropy": 1.9744655668735505,
+      "loss/hidden": 3.2984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17376624681055547,
+      "step": 13250
+    },
+    {
+      "epoch": 0.1105,
+      "grad_norm": 35.5,
+      "grad_norm_var": 679.0707682291667,
+      "learning_rate": 0.0001,
+      "loss": 7.1727,
+      "loss/crossentropy": 1.900152012705803,
+      "loss/hidden": 3.365234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1979702627286315,
+      "step": 13260
+    },
+    {
+      "epoch": 0.11058333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 6.862239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2654,
+      "loss/crossentropy": 2.0497558236122133,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859972968697548,
+      "step": 13270
+    },
+    {
+      "epoch": 0.11066666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.949739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2032,
+      "loss/crossentropy": 1.9490113705396652,
+      "loss/hidden": 3.332421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18092472292482853,
+      "step": 13280
+    },
+    {
+      "epoch": 0.11075,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.69765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0927,
+      "loss/crossentropy": 1.9998493492603302,
+      "loss/hidden": 3.275,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17558108344674112,
+      "step": 13290
+    },
+    {
+      "epoch": 0.11083333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.1108723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2388,
+      "loss/crossentropy": 1.8353417202830316,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18846051972359418,
+      "step": 13300
+    },
+    {
+      "epoch": 0.11091666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.7541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2428,
+      "loss/crossentropy": 2.0540729075670243,
+      "loss/hidden": 3.332421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18476204108446836,
+      "step": 13310
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.1809895833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2189,
+      "loss/crossentropy": 2.0570236518979073,
+      "loss/hidden": 3.4109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20177374817430974,
+      "step": 13320
+    },
+    {
+      "epoch": 0.11108333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.78515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1534,
+      "loss/crossentropy": 1.9106008380651474,
+      "loss/hidden": 3.331640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18688999786972998,
+      "step": 13330
+    },
+    {
+      "epoch": 0.11116666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.4603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1206,
+      "loss/crossentropy": 1.960108731687069,
+      "loss/hidden": 3.4703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18869107980281113,
+      "step": 13340
+    },
+    {
+      "epoch": 0.11125,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.5395182291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.195,
+      "loss/crossentropy": 1.9822386115789414,
+      "loss/hidden": 3.3421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17934756074100733,
+      "step": 13350
+    },
+    {
+      "epoch": 0.11133333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.755989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.154,
+      "loss/crossentropy": 1.8304448828101159,
+      "loss/hidden": 3.469140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1796064030379057,
+      "step": 13360
+    },
+    {
+      "epoch": 0.11141666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 11.447330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1574,
+      "loss/crossentropy": 1.8059681802988052,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19601923916488886,
+      "step": 13370
+    },
+    {
+      "epoch": 0.1115,
+      "grad_norm": 31.0,
+      "grad_norm_var": 11.966080729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2444,
+      "loss/crossentropy": 2.0005480498075485,
+      "loss/hidden": 3.43984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19403798878192902,
+      "step": 13380
+    },
+    {
+      "epoch": 0.11158333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.9973307291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1511,
+      "loss/crossentropy": 2.0060314163565636,
+      "loss/hidden": 3.325390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19443422537297012,
+      "step": 13390
+    },
+    {
+      "epoch": 0.11166666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.4025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.3083,
+      "loss/crossentropy": 1.8162367567420006,
+      "loss/hidden": 3.394140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780845783650875,
+      "step": 13400
+    },
+    {
+      "epoch": 0.11175,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.626041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1307,
+      "loss/crossentropy": 1.9779260143637658,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18656796868890524,
+      "step": 13410
+    },
+    {
+      "epoch": 0.11183333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.9572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2205,
+      "loss/crossentropy": 1.9409028336405754,
+      "loss/hidden": 3.354296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20121930167078972,
+      "step": 13420
+    },
+    {
+      "epoch": 0.11191666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5248.964518229167,
+      "learning_rate": 0.0001,
+      "loss": 7.4033,
+      "loss/crossentropy": 2.0979151248931887,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19585807099938393,
+      "step": 13430
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 28.125,
+      "grad_norm_var": 13.918489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1963,
+      "loss/crossentropy": 1.9601235270500184,
+      "loss/hidden": 3.47734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21391492672264575,
+      "step": 13440
+    },
+    {
+      "epoch": 0.11208333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.388541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1352,
+      "loss/crossentropy": 1.8527611121535301,
+      "loss/hidden": 3.3625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.183625746704638,
+      "step": 13450
+    },
+    {
+      "epoch": 0.11216666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.0634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1637,
+      "loss/crossentropy": 1.8165352158248425,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1754130182787776,
+      "step": 13460
+    },
+    {
+      "epoch": 0.11225,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.7375,
+      "learning_rate": 0.0001,
+      "loss": 7.0687,
+      "loss/crossentropy": 1.7781095772981643,
+      "loss/hidden": 3.262890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17313408087939025,
+      "step": 13470
+    },
+    {
+      "epoch": 0.11233333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.2625,
+      "learning_rate": 0.0001,
+      "loss": 7.2185,
+      "loss/crossentropy": 1.8874119453132152,
+      "loss/hidden": 3.423828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2206179355736822,
+      "step": 13480
+    },
+    {
+      "epoch": 0.11241666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.6525390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1585,
+      "loss/crossentropy": 1.995115227997303,
+      "loss/hidden": 3.4546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20874616373330354,
+      "step": 13490
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 30.0,
+      "grad_norm_var": 22.112955729166668,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.8940367341041564,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19161785752512514,
+      "step": 13500
+    },
+    {
+      "epoch": 0.11258333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 25.092122395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1085,
+      "loss/crossentropy": 1.9849766418337822,
+      "loss/hidden": 3.2921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1772268157452345,
+      "step": 13510
+    },
+    {
+      "epoch": 0.11266666666666666,
+      "grad_norm": 45.5,
+      "grad_norm_var": 33.243489583333336,
+      "learning_rate": 0.0001,
+      "loss": 7.2129,
+      "loss/crossentropy": 1.8361537724733352,
+      "loss/hidden": 3.24296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.170785696990788,
+      "step": 13520
+    },
+    {
+      "epoch": 0.11275,
+      "grad_norm": 41.25,
+      "grad_norm_var": 23.16015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1165,
+      "loss/crossentropy": 1.9720616310834884,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19569863062351942,
+      "step": 13530
+    },
+    {
+      "epoch": 0.11283333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 15.653125,
+      "learning_rate": 0.0001,
+      "loss": 7.2567,
+      "loss/crossentropy": 1.9601970702409743,
+      "loss/hidden": 3.278515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735742162913084,
+      "step": 13540
+    },
+    {
+      "epoch": 0.11291666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 14.1791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.146,
+      "loss/crossentropy": 1.9234497502446175,
+      "loss/hidden": 3.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1829282358288765,
+      "step": 13550
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 36.5,
+      "grad_norm_var": 10.0806640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1246,
+      "loss/crossentropy": 1.8681599967181683,
+      "loss/hidden": 3.325,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16795233618468047,
+      "step": 13560
+    },
+    {
+      "epoch": 0.11308333333333333,
+      "grad_norm": 34.0,
+      "grad_norm_var": 43.34973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1351,
+      "loss/crossentropy": 1.9687900215387344,
+      "loss/hidden": 3.29921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18534957449883221,
+      "step": 13570
+    },
+    {
+      "epoch": 0.11316666666666667,
+      "grad_norm": 27.875,
+      "grad_norm_var": 50.096875,
+      "learning_rate": 0.0001,
+      "loss": 7.0499,
+      "loss/crossentropy": 1.9385055661201478,
+      "loss/hidden": 3.3015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18169882129877807,
+      "step": 13580
+    },
+    {
+      "epoch": 0.11325,
+      "grad_norm": 29.75,
+      "grad_norm_var": 12.1494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2855,
+      "loss/crossentropy": 2.0746473222970963,
+      "loss/hidden": 3.321484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1902322567999363,
+      "step": 13590
+    },
+    {
+      "epoch": 0.11333333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.689322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0863,
+      "loss/crossentropy": 1.990707655251026,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17817136123776436,
+      "step": 13600
+    },
+    {
+      "epoch": 0.11341666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 12.71640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1487,
+      "loss/crossentropy": 2.038090743124485,
+      "loss/hidden": 3.4390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18759762030094862,
+      "step": 13610
+    },
+    {
+      "epoch": 0.1135,
+      "grad_norm": 32.25,
+      "grad_norm_var": 16.467122395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1756,
+      "loss/crossentropy": 1.960167148709297,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17449798323214055,
+      "step": 13620
+    },
+    {
+      "epoch": 0.11358333333333333,
+      "grad_norm": 37.25,
+      "grad_norm_var": 5.09140625,
+      "learning_rate": 0.0001,
+      "loss": 7.135,
+      "loss/crossentropy": 1.8514888763427735,
+      "loss/hidden": 3.314453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17315484527498484,
+      "step": 13630
+    },
+    {
+      "epoch": 0.11366666666666667,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.4541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1022,
+      "loss/crossentropy": 1.953445229679346,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18329770136624574,
+      "step": 13640
+    },
+    {
+      "epoch": 0.11375,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.797330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1751,
+      "loss/crossentropy": 1.9880620568990708,
+      "loss/hidden": 3.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18864249717444181,
+      "step": 13650
+    },
+    {
+      "epoch": 0.11383333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.729622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2608,
+      "loss/crossentropy": 1.9410797566175462,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19353037010878324,
+      "step": 13660
+    },
+    {
+      "epoch": 0.11391666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.629622395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2006,
+      "loss/crossentropy": 2.029595251381397,
+      "loss/hidden": 3.47890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21051803044974804,
+      "step": 13670
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 56.5,
+      "grad_norm_var": 112.04765625,
+      "learning_rate": 0.0001,
+      "loss": 7.3549,
+      "loss/crossentropy": 1.955371654033661,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19862194415181875,
+      "step": 13680
+    },
+    {
+      "epoch": 0.11408333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 110.078125,
+      "learning_rate": 0.0001,
+      "loss": 7.1131,
+      "loss/crossentropy": 1.9256178140640259,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19249056726694108,
+      "step": 13690
+    },
+    {
+      "epoch": 0.11416666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.981184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2024,
+      "loss/crossentropy": 2.0161723107099534,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18494158796966076,
+      "step": 13700
+    },
+    {
+      "epoch": 0.11425,
+      "grad_norm": 35.0,
+      "grad_norm_var": 2.863997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2131,
+      "loss/crossentropy": 1.915719175338745,
+      "loss/hidden": 3.490234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1831233872100711,
+      "step": 13710
+    },
+    {
+      "epoch": 0.11433333333333333,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.46640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2349,
+      "loss/crossentropy": 1.9530277743935585,
+      "loss/hidden": 3.43828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20726575665175914,
+      "step": 13720
+    },
+    {
+      "epoch": 0.11441666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 6.246875,
+      "learning_rate": 0.0001,
+      "loss": 7.242,
+      "loss/crossentropy": 1.9605878829956054,
+      "loss/hidden": 3.385546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1861640816554427,
+      "step": 13730
+    },
+    {
+      "epoch": 0.1145,
+      "grad_norm": 29.625,
+      "grad_norm_var": 26.615625,
+      "learning_rate": 0.0001,
+      "loss": 7.0771,
+      "loss/crossentropy": 1.798363845795393,
+      "loss/hidden": 3.433984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1684430256485939,
+      "step": 13740
+    },
+    {
+      "epoch": 0.11458333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.6472619097323996e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.233,
+      "loss/crossentropy": 1.9076340973377228,
+      "loss/hidden": 3.376953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18004499338567256,
+      "step": 13750
+    },
+    {
+      "epoch": 0.11466666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.6472619110425953e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1815,
+      "loss/crossentropy": 1.9968195796012878,
+      "loss/hidden": 3.33125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20676399134099482,
+      "step": 13760
+    },
+    {
+      "epoch": 0.11475,
+      "grad_norm": 29.75,
+      "grad_norm_var": 8.458072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1808,
+      "loss/crossentropy": 1.830625294148922,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17546319533139468,
+      "step": 13770
+    },
+    {
+      "epoch": 0.11483333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 12.753580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1893,
+      "loss/crossentropy": 1.8140571914613246,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810930219478905,
+      "step": 13780
+    },
+    {
+      "epoch": 0.11491666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 18.028059895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1317,
+      "loss/crossentropy": 2.0302459567785265,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1862118661403656,
+      "step": 13790
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 38.25,
+      "grad_norm_var": 18.2306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1714,
+      "loss/crossentropy": 1.9490053534507752,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20752408523112537,
+      "step": 13800
+    },
+    {
+      "epoch": 0.11508333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.723372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2793,
+      "loss/crossentropy": 2.050610472261906,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22376690562814475,
+      "step": 13810
+    },
+    {
+      "epoch": 0.11516666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.655989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2798,
+      "loss/crossentropy": 2.1291163764894008,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19109589802101254,
+      "step": 13820
+    },
+    {
+      "epoch": 0.11525,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.119791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1679,
+      "loss/crossentropy": 1.965355785191059,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.187576493807137,
+      "step": 13830
+    },
+    {
+      "epoch": 0.11533333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 29.424934895833335,
+      "learning_rate": 0.0001,
+      "loss": 7.1923,
+      "loss/crossentropy": 1.9938945531845094,
+      "loss/hidden": 3.371875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18410432226955892,
+      "step": 13840
+    },
+    {
+      "epoch": 0.11541666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 26.720833333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2132,
+      "loss/crossentropy": 1.8234495922923089,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17247499637305735,
+      "step": 13850
+    },
+    {
+      "epoch": 0.1155,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.87890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1551,
+      "loss/crossentropy": 2.048468878865242,
+      "loss/hidden": 3.308984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749510670080781,
+      "step": 13860
+    },
+    {
+      "epoch": 0.11558333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.2916666666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1284,
+      "loss/crossentropy": 1.8810873091220857,
+      "loss/hidden": 3.259765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16340148337185384,
+      "step": 13870
+    },
+    {
+      "epoch": 0.11566666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.036458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2566,
+      "loss/crossentropy": 1.9686560451984405,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18240359034389259,
+      "step": 13880
+    },
+    {
+      "epoch": 0.11575,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.280143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1305,
+      "loss/crossentropy": 1.786266028881073,
+      "loss/hidden": 3.3828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16755877770483493,
+      "step": 13890
+    },
+    {
+      "epoch": 0.11583333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.338997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2248,
+      "loss/crossentropy": 1.910762904584408,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18704288508743047,
+      "step": 13900
+    },
+    {
+      "epoch": 0.11591666666666667,
+      "grad_norm": 37.5,
+      "grad_norm_var": 13.518684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3601,
+      "loss/crossentropy": 2.003139579296112,
+      "loss/hidden": 3.439453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20703245475888252,
+      "step": 13910
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 29.625,
+      "grad_norm_var": 15.218489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1382,
+      "loss/crossentropy": 1.9374216735363006,
+      "loss/hidden": 3.309765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17754750661551952,
+      "step": 13920
+    },
+    {
+      "epoch": 0.11608333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 12.782291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1944,
+      "loss/crossentropy": 1.8373799830675126,
+      "loss/hidden": 3.415625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18869947120547295,
+      "step": 13930
+    },
+    {
+      "epoch": 0.11616666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.4556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2898,
+      "loss/crossentropy": 1.9491986751556396,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1921183492988348,
+      "step": 13940
+    },
+    {
+      "epoch": 0.11625,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.2075980010763758e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.231,
+      "loss/crossentropy": 1.9398064777255057,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19295795522630216,
+      "step": 13950
+    },
+    {
+      "epoch": 0.11633333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.4147503324281e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3516,
+      "loss/crossentropy": 1.9348388269543648,
+      "loss/hidden": 3.4390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20045128148049116,
+      "step": 13960
+    },
+    {
+      "epoch": 0.11641666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 9.674739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1047,
+      "loss/crossentropy": 1.9415176302194594,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18492716439068319,
+      "step": 13970
+    },
+    {
+      "epoch": 0.1165,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.5872395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1614,
+      "loss/crossentropy": 1.9615104436874389,
+      "loss/hidden": 3.39375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18381981793791055,
+      "step": 13980
+    },
+    {
+      "epoch": 0.11658333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.36640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1376,
+      "loss/crossentropy": 1.8144659370183944,
+      "loss/hidden": 3.3484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17246513133868574,
+      "step": 13990
+    },
+    {
+      "epoch": 0.11666666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.8,
+      "learning_rate": 0.0001,
+      "loss": 7.2223,
+      "loss/crossentropy": 2.0782426699995993,
+      "loss/hidden": 3.210546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17946213204413652,
+      "step": 14000
+    },
+    {
+      "epoch": 0.11675,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.833072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1614,
+      "loss/crossentropy": 1.8217924684286118,
+      "loss/hidden": 3.505078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18446492031216621,
+      "step": 14010
+    },
+    {
+      "epoch": 0.11683333333333333,
+      "grad_norm": 28.375,
+      "grad_norm_var": 31.263997395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1911,
+      "loss/crossentropy": 1.9678502976894379,
+      "loss/hidden": 3.45,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20646918267011644,
+      "step": 14020
+    },
+    {
+      "epoch": 0.11691666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 39.42604166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2663,
+      "loss/crossentropy": 2.0206977039575578,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19073572158813476,
+      "step": 14030
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 30.125,
+      "grad_norm_var": 25.6322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2166,
+      "loss/crossentropy": 1.9004216700792314,
+      "loss/hidden": 3.518359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1931025877594948,
+      "step": 14040
+    },
+    {
+      "epoch": 0.11708333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 19.357291666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1769,
+      "loss/crossentropy": 1.9893005669116974,
+      "loss/hidden": 3.354296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910881519317627,
+      "step": 14050
+    },
+    {
+      "epoch": 0.11716666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 12.623958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1432,
+      "loss/crossentropy": 1.994839932024479,
+      "loss/hidden": 3.21171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17068272717297078,
+      "step": 14060
+    },
+    {
+      "epoch": 0.11725,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2564.4455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.171,
+      "loss/crossentropy": 1.9999317556619645,
+      "loss/hidden": 3.406640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19567822776734828,
+      "step": 14070
+    },
+    {
+      "epoch": 0.11733333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 10.252018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0678,
+      "loss/crossentropy": 1.8994514107704163,
+      "loss/hidden": 3.421484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1832486478611827,
+      "step": 14080
+    },
+    {
+      "epoch": 0.11741666666666667,
+      "grad_norm": 36.0,
+      "grad_norm_var": 10.696875,
+      "learning_rate": 0.0001,
+      "loss": 7.1504,
+      "loss/crossentropy": 1.8676493011415005,
+      "loss/hidden": 3.41171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19122676020488144,
+      "step": 14090
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.0744140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1667,
+      "loss/crossentropy": 2.027650533616543,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18178627286106347,
+      "step": 14100
+    },
+    {
+      "epoch": 0.11758333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.648958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1602,
+      "loss/crossentropy": 1.8069852709770202,
+      "loss/hidden": 3.44453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17312425281852484,
+      "step": 14110
+    },
+    {
+      "epoch": 0.11766666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 6.005143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0619,
+      "loss/crossentropy": 1.7716371312737464,
+      "loss/hidden": 3.4703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182572347112,
+      "step": 14120
+    },
+    {
+      "epoch": 0.11775,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.891666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0916,
+      "loss/crossentropy": 1.9237942054867745,
+      "loss/hidden": 3.4078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18569694012403487,
+      "step": 14130
+    },
+    {
+      "epoch": 0.11783333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.086458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2044,
+      "loss/crossentropy": 1.8212244987487793,
+      "loss/hidden": 3.277734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.169494036026299,
+      "step": 14140
+    },
+    {
+      "epoch": 0.11791666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 6.92265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2012,
+      "loss/crossentropy": 1.9114794224500655,
+      "loss/hidden": 3.4171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19415497798472642,
+      "step": 14150
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.670768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0859,
+      "loss/crossentropy": 1.721632531285286,
+      "loss/hidden": 3.184375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15097201969474555,
+      "step": 14160
+    },
+    {
+      "epoch": 0.11808333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.313541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1225,
+      "loss/crossentropy": 2.088696774840355,
+      "loss/hidden": 3.346484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18932171799242498,
+      "step": 14170
+    },
+    {
+      "epoch": 0.11816666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 16.41015625,
+      "learning_rate": 0.0001,
+      "loss": 7.3067,
+      "loss/crossentropy": 1.9494910374283791,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17766583263874053,
+      "step": 14180
+    },
+    {
+      "epoch": 0.11825,
+      "grad_norm": 30.375,
+      "grad_norm_var": 29.732747395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.2236,
+      "loss/crossentropy": 1.932012678682804,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20607319716364145,
+      "step": 14190
+    },
+    {
+      "epoch": 0.11833333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 28.7462890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1495,
+      "loss/crossentropy": 1.9365191757678986,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19420478381216527,
+      "step": 14200
+    },
+    {
+      "epoch": 0.11841666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.79140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0935,
+      "loss/crossentropy": 2.0120673567056655,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18497201893478632,
+      "step": 14210
+    },
+    {
+      "epoch": 0.1185,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.098893229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1354,
+      "loss/crossentropy": 1.8606435388326645,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1858756672590971,
+      "step": 14220
+    },
+    {
+      "epoch": 0.11858333333333333,
+      "grad_norm": 39.75,
+      "grad_norm_var": 9.964917699411707e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1705,
+      "loss/crossentropy": 1.8670482061803342,
+      "loss/hidden": 3.390234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18284698380157352,
+      "step": 14230
+    },
+    {
+      "epoch": 0.11866666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 9.964917691176191e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1642,
+      "loss/crossentropy": 1.9060709655284882,
+      "loss/hidden": 3.2828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17148990742862225,
+      "step": 14240
+    },
+    {
+      "epoch": 0.11875,
+      "grad_norm": 30.0,
+      "grad_norm_var": 35.580143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1117,
+      "loss/crossentropy": 1.9479292884469033,
+      "loss/hidden": 3.31484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1758895706385374,
+      "step": 14250
+    },
+    {
+      "epoch": 0.11883333333333333,
+      "grad_norm": 41.0,
+      "grad_norm_var": 143.628125,
+      "learning_rate": 0.0001,
+      "loss": 7.3299,
+      "loss/crossentropy": 1.8063128411769866,
+      "loss/hidden": 3.41875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2178534297272563,
+      "step": 14260
+    },
+    {
+      "epoch": 0.11891666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 141.3103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2203,
+      "loss/crossentropy": 1.8228762581944467,
+      "loss/hidden": 3.284765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17942063976079226,
+      "step": 14270
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 41.0,
+      "grad_norm_var": 3.4975024080371066e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.245,
+      "loss/crossentropy": 1.9525840029120445,
+      "loss/hidden": 3.28828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17769890762865542,
+      "step": 14280
+    },
+    {
+      "epoch": 0.11908333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.4975024110680154e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0898,
+      "loss/crossentropy": 1.9266117215156555,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17632283382117747,
+      "step": 14290
+    },
+    {
+      "epoch": 0.11916666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.110416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1917,
+      "loss/crossentropy": 1.906543004512787,
+      "loss/hidden": 3.33828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1790146768093109,
+      "step": 14300
+    },
+    {
+      "epoch": 0.11925,
+      "grad_norm": 34.0,
+      "grad_norm_var": 64.86432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2705,
+      "loss/crossentropy": 1.8918564431369305,
+      "loss/hidden": 3.405859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20345929320901632,
+      "step": 14310
+    },
+    {
+      "epoch": 0.11933333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 67.3259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1777,
+      "loss/crossentropy": 2.017512695491314,
+      "loss/hidden": 3.283203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18107608687132598,
+      "step": 14320
+    },
+    {
+      "epoch": 0.11941666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.0184895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1802,
+      "loss/crossentropy": 2.0612320780754088,
+      "loss/hidden": 3.191796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17330344580113888,
+      "step": 14330
+    },
+    {
+      "epoch": 0.1195,
+      "grad_norm": 31.75,
+      "grad_norm_var": 23.143489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0868,
+      "loss/crossentropy": 1.8566990703344346,
+      "loss/hidden": 3.409375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17643862068653107,
+      "step": 14340
+    },
+    {
+      "epoch": 0.11958333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 22.558072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1485,
+      "loss/crossentropy": 2.041011855006218,
+      "loss/hidden": 3.309375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.188278042152524,
+      "step": 14350
+    },
+    {
+      "epoch": 0.11966666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 21.495572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2321,
+      "loss/crossentropy": 1.9916374891996385,
+      "loss/hidden": 3.4703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18982053957879544,
+      "step": 14360
+    },
+    {
+      "epoch": 0.11975,
+      "grad_norm": 30.25,
+      "grad_norm_var": 20.471875,
+      "learning_rate": 0.0001,
+      "loss": 7.1462,
+      "loss/crossentropy": 1.9886096283793449,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17898899391293527,
+      "step": 14370
+    },
+    {
+      "epoch": 0.11983333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.528580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1523,
+      "loss/crossentropy": 1.944193209707737,
+      "loss/hidden": 3.46953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1991956029087305,
+      "step": 14380
+    },
+    {
+      "epoch": 0.11991666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.4322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2451,
+      "loss/crossentropy": 1.9098138615489006,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19298403412103654,
+      "step": 14390
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.250455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2018,
+      "loss/crossentropy": 1.9501084595918656,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19287684448063375,
+      "step": 14400
+    },
+    {
+      "epoch": 0.12008333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 9.88135481613956e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3321,
+      "loss/crossentropy": 1.8319449417293072,
+      "loss/hidden": 3.39765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899934284389019,
+      "step": 14410
+    },
+    {
+      "epoch": 0.12016666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.881354814648485e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1123,
+      "loss/crossentropy": 1.9026117384433747,
+      "loss/hidden": 3.468359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18749726992100477,
+      "step": 14420
+    },
+    {
+      "epoch": 0.12025,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.230143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0995,
+      "loss/crossentropy": 1.9893973782658576,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2014074932783842,
+      "step": 14430
+    },
+    {
+      "epoch": 0.12033333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.41640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1294,
+      "loss/crossentropy": 1.9313647992908956,
+      "loss/hidden": 3.312890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17697440776973963,
+      "step": 14440
+    },
+    {
+      "epoch": 0.12041666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.11875,
+      "learning_rate": 0.0001,
+      "loss": 7.1813,
+      "loss/crossentropy": 1.8936800889670848,
+      "loss/hidden": 3.349609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18409691452980043,
+      "step": 14450
+    },
+    {
+      "epoch": 0.1205,
+      "grad_norm": 31.5,
+      "grad_norm_var": 9.175,
+      "learning_rate": 0.0001,
+      "loss": 7.1984,
+      "loss/crossentropy": 2.0225543528795242,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18387900274246932,
+      "step": 14460
+    },
+    {
+      "epoch": 0.12058333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.69140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1377,
+      "loss/crossentropy": 1.9767469987273216,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17938012722879648,
+      "step": 14470
+    },
+    {
+      "epoch": 0.12066666666666667,
+      "grad_norm": 47.5,
+      "grad_norm_var": 4.3917132191176685e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1756,
+      "loss/crossentropy": 1.816517861187458,
+      "loss/hidden": 3.365625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18571343477815389,
+      "step": 14480
+    },
+    {
+      "epoch": 0.12075,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.391713209094331e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1932,
+      "loss/crossentropy": 1.9697727650403976,
+      "loss/hidden": 3.3375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18909659199416637,
+      "step": 14490
+    },
+    {
+      "epoch": 0.12083333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.729166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2369,
+      "loss/crossentropy": 1.866569310426712,
+      "loss/hidden": 3.3046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1849963691085577,
+      "step": 14500
+    },
+    {
+      "epoch": 0.12091666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.814518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.274,
+      "loss/crossentropy": 1.969509779661894,
+      "loss/hidden": 3.23828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17163542015478014,
+      "step": 14510
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 31.75,
+      "grad_norm_var": 12.976041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0409,
+      "loss/crossentropy": 1.7333900466561318,
+      "loss/hidden": 3.42265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1648394949734211,
+      "step": 14520
+    },
+    {
+      "epoch": 0.12108333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.207747395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1404,
+      "loss/crossentropy": 1.996240859106183,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18585094418376685,
+      "step": 14530
+    },
+    {
+      "epoch": 0.12116666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.90625,
+      "learning_rate": 0.0001,
+      "loss": 7.2275,
+      "loss/crossentropy": 1.8614683374762535,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18676308058202268,
+      "step": 14540
+    },
+    {
+      "epoch": 0.12125,
+      "grad_norm": 36.0,
+      "grad_norm_var": 2.5364583333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.0675,
+      "loss/crossentropy": 1.8792560204863549,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1713730487972498,
+      "step": 14550
+    },
+    {
+      "epoch": 0.12133333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.553125,
+      "learning_rate": 0.0001,
+      "loss": 7.1637,
+      "loss/crossentropy": 1.8288133054971696,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1782605094835162,
+      "step": 14560
+    },
+    {
+      "epoch": 0.12141666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.1249348958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2267,
+      "loss/crossentropy": 1.897878061234951,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17876417245715856,
+      "step": 14570
+    },
+    {
+      "epoch": 0.1215,
+      "grad_norm": 30.25,
+      "grad_norm_var": 20.378125,
+      "learning_rate": 0.0001,
+      "loss": 7.193,
+      "loss/crossentropy": 1.952752670645714,
+      "loss/hidden": 3.26875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18517019506543875,
+      "step": 14580
+    },
+    {
+      "epoch": 0.12158333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 20.071875,
+      "learning_rate": 0.0001,
+      "loss": 7.1497,
+      "loss/crossentropy": 1.9442493468523026,
+      "loss/hidden": 3.4890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19208364766091107,
+      "step": 14590
+    },
+    {
+      "epoch": 0.12166666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.985872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2874,
+      "loss/crossentropy": 1.9756167352199554,
+      "loss/hidden": 3.311328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18927722051739693,
+      "step": 14600
+    },
+    {
+      "epoch": 0.12175,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.980143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2523,
+      "loss/crossentropy": 1.9072391271591187,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19173142705112695,
+      "step": 14610
+    },
+    {
+      "epoch": 0.12183333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.966681454147099e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2014,
+      "loss/crossentropy": 1.965491108596325,
+      "loss/hidden": 3.327734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19286179542541504,
+      "step": 14620
+    },
+    {
+      "epoch": 0.12191666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.895247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2268,
+      "loss/crossentropy": 1.9338607162237167,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1848919352516532,
+      "step": 14630
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.933268229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 1.9111004769802094,
+      "loss/hidden": 3.3390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1890100993216038,
+      "step": 14640
+    },
+    {
+      "epoch": 0.12208333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.657747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2738,
+      "loss/crossentropy": 2.1023745089769363,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946237213909626,
+      "step": 14650
+    },
+    {
+      "epoch": 0.12216666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 20.634830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0866,
+      "loss/crossentropy": 1.9427856147289275,
+      "loss/hidden": 3.30625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17716602217406036,
+      "step": 14660
+    },
+    {
+      "epoch": 0.12225,
+      "grad_norm": 29.625,
+      "grad_norm_var": 18.592122395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1371,
+      "loss/crossentropy": 1.9776272669434547,
+      "loss/hidden": 3.44609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20407221037894488,
+      "step": 14670
+    },
+    {
+      "epoch": 0.12233333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.4358723958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1476,
+      "loss/crossentropy": 1.985303920507431,
+      "loss/hidden": 3.35703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19023565910756587,
+      "step": 14680
+    },
+    {
+      "epoch": 0.12241666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 5.272330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1009,
+      "loss/crossentropy": 1.9959974050521851,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1887131780385971,
+      "step": 14690
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 35.5,
+      "grad_norm_var": 30.77890625,
+      "learning_rate": 0.0001,
+      "loss": 7.3223,
+      "loss/crossentropy": 2.046718167513609,
+      "loss/hidden": 3.394140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1997451340779662,
+      "step": 14700
+    },
+    {
+      "epoch": 0.12258333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.470572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1242,
+      "loss/crossentropy": 1.9147102862596512,
+      "loss/hidden": 3.5296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18637195322662592,
+      "step": 14710
+    },
+    {
+      "epoch": 0.12266666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.830208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2496,
+      "loss/crossentropy": 1.8110238060355186,
+      "loss/hidden": 3.4078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19058231487870217,
+      "step": 14720
+    },
+    {
+      "epoch": 0.12275,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.8822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2131,
+      "loss/crossentropy": 1.9766608387231828,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801937036216259,
+      "step": 14730
+    },
+    {
+      "epoch": 0.12283333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.878059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2105,
+      "loss/crossentropy": 1.985130463540554,
+      "loss/hidden": 3.35,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18413505535572766,
+      "step": 14740
+    },
+    {
+      "epoch": 0.12291666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 185.87395833333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2064,
+      "loss/crossentropy": 1.8777363255620003,
+      "loss/hidden": 3.36875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795031763613224,
+      "step": 14750
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 31.375,
+      "grad_norm_var": 32.70520833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1825,
+      "loss/crossentropy": 1.9649272993206979,
+      "loss/hidden": 3.493359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19079206679016353,
+      "step": 14760
+    },
+    {
+      "epoch": 0.12308333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 13.602083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1624,
+      "loss/crossentropy": 1.9600737869739533,
+      "loss/hidden": 3.227734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17734416034072636,
+      "step": 14770
+    },
+    {
+      "epoch": 0.12316666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.6639973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1586,
+      "loss/crossentropy": 1.947436621785164,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19724200554192067,
+      "step": 14780
+    },
+    {
+      "epoch": 0.12325,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.8059895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1369,
+      "loss/crossentropy": 2.0453428626060486,
+      "loss/hidden": 3.26484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18231261000037194,
+      "step": 14790
+    },
+    {
+      "epoch": 0.12333333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 7.842643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1427,
+      "loss/crossentropy": 1.9762895375490188,
+      "loss/hidden": 3.336328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17903975918889045,
+      "step": 14800
+    },
+    {
+      "epoch": 0.12341666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.314322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1992,
+      "loss/crossentropy": 2.0383625090122224,
+      "loss/hidden": 3.283203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843028925359249,
+      "step": 14810
+    },
+    {
+      "epoch": 0.1235,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.4541666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.312,
+      "loss/crossentropy": 1.9732710406184197,
+      "loss/hidden": 3.465234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18139485251158477,
+      "step": 14820
+    },
+    {
+      "epoch": 0.12358333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 28.378580729166668,
+      "learning_rate": 0.0001,
+      "loss": 7.1507,
+      "loss/crossentropy": 1.9309901729226113,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17788593601435423,
+      "step": 14830
+    },
+    {
+      "epoch": 0.12366666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 29.370572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2475,
+      "loss/crossentropy": 1.9890019282698632,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18680631071329118,
+      "step": 14840
+    },
+    {
+      "epoch": 0.12375,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.253850637962227e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1857,
+      "loss/crossentropy": 2.015805447101593,
+      "loss/hidden": 3.269140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17531456090509892,
+      "step": 14850
+    },
+    {
+      "epoch": 0.12383333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.870247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2117,
+      "loss/crossentropy": 1.9835448637604713,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18218748420476913,
+      "step": 14860
+    },
+    {
+      "epoch": 0.12391666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.6541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.225,
+      "loss/crossentropy": 2.1301302671432496,
+      "loss/hidden": 3.333203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19686594158411025,
+      "step": 14870
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.537955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2111,
+      "loss/crossentropy": 1.9723198503255843,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19141894802451134,
+      "step": 14880
+    },
+    {
+      "epoch": 0.12408333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.896875,
+      "learning_rate": 0.0001,
+      "loss": 7.2045,
+      "loss/crossentropy": 1.9322915732860566,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1867530271410942,
+      "step": 14890
+    },
+    {
+      "epoch": 0.12416666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.15,
+      "learning_rate": 0.0001,
+      "loss": 7.2908,
+      "loss/crossentropy": 1.9152386605739593,
+      "loss/hidden": 3.458203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18486501164734365,
+      "step": 14900
+    },
+    {
+      "epoch": 0.12425,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.801495314067672e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2906,
+      "loss/crossentropy": 1.9992727652192115,
+      "loss/hidden": 3.396875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22329286392778158,
+      "step": 14910
+    },
+    {
+      "epoch": 0.12433333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 38.412434895833336,
+      "learning_rate": 0.0001,
+      "loss": 7.123,
+      "loss/crossentropy": 2.0338529318571092,
+      "loss/hidden": 3.305859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19171512462198734,
+      "step": 14920
+    },
+    {
+      "epoch": 0.12441666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.92265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1557,
+      "loss/crossentropy": 1.8733994647860528,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18240153342485427,
+      "step": 14930
+    },
+    {
+      "epoch": 0.1245,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.637434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2487,
+      "loss/crossentropy": 1.9415182277560235,
+      "loss/hidden": 3.44375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18578971400856972,
+      "step": 14940
+    },
+    {
+      "epoch": 0.12458333333333334,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.4802083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2336,
+      "loss/crossentropy": 1.9551210507750512,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19206518828868865,
+      "step": 14950
+    },
+    {
+      "epoch": 0.12466666666666666,
+      "grad_norm": 62.25,
+      "grad_norm_var": 59.60983072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2245,
+      "loss/crossentropy": 1.937887005507946,
+      "loss/hidden": 3.33515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19873388316482304,
+      "step": 14960
+    },
+    {
+      "epoch": 0.12475,
+      "grad_norm": 32.0,
+      "grad_norm_var": 61.50416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2057,
+      "loss/crossentropy": 2.0197895735502245,
+      "loss/hidden": 3.28828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18216200023889542,
+      "step": 14970
+    },
+    {
+      "epoch": 0.12483333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.6212890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1761,
+      "loss/crossentropy": 1.994225400686264,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810664651915431,
+      "step": 14980
+    },
+    {
+      "epoch": 0.12491666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 12.065625,
+      "learning_rate": 0.0001,
+      "loss": 7.0706,
+      "loss/crossentropy": 2.0393955439329146,
+      "loss/hidden": 3.276171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17954570725560187,
+      "step": 14990
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 32.0,
+      "grad_norm_var": 15.471809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2026,
+      "loss/crossentropy": 1.930550955235958,
+      "loss/hidden": 3.466015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18915690928697587,
+      "step": 15000
+    },
+    {
+      "epoch": 0.12508333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.3760416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1162,
+      "loss/crossentropy": 1.9567875742912293,
+      "loss/hidden": 3.440625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.175424394197762,
+      "step": 15010
+    },
+    {
+      "epoch": 0.12516666666666668,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.698757016452684e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.4099,
+      "loss/crossentropy": 1.9233490712940693,
+      "loss/hidden": 3.6171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20609719324856995,
+      "step": 15020
+    },
+    {
+      "epoch": 0.12525,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.798958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1441,
+      "loss/crossentropy": 1.8938697695732116,
+      "loss/hidden": 3.472265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19456320125609636,
+      "step": 15030
+    },
+    {
+      "epoch": 0.12533333333333332,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.9643229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1739,
+      "loss/crossentropy": 2.014716199040413,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1990583285689354,
+      "step": 15040
+    },
+    {
+      "epoch": 0.12541666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.37265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2679,
+      "loss/crossentropy": 1.9679681181907653,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18550205416977406,
+      "step": 15050
+    },
+    {
+      "epoch": 0.1255,
+      "grad_norm": 29.625,
+      "grad_norm_var": 11951.704622395833,
+      "learning_rate": 0.0001,
+      "loss": 7.1669,
+      "loss/crossentropy": 1.8434071607887745,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17588050682097672,
+      "step": 15060
+    },
+    {
+      "epoch": 0.12558333333333332,
+      "grad_norm": 30.75,
+      "grad_norm_var": 44.6666015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0803,
+      "loss/crossentropy": 1.9289218172430993,
+      "loss/hidden": 3.396875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17938923817127944,
+      "step": 15070
+    },
+    {
+      "epoch": 0.12566666666666668,
+      "grad_norm": 29.75,
+      "grad_norm_var": 11.401497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1891,
+      "loss/crossentropy": 1.9962383687496186,
+      "loss/hidden": 3.376953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17921186946332454,
+      "step": 15080
+    },
+    {
+      "epoch": 0.12575,
+      "grad_norm": 30.0,
+      "grad_norm_var": 10.1697265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1641,
+      "loss/crossentropy": 2.0570418909192085,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2009740699082613,
+      "step": 15090
+    },
+    {
+      "epoch": 0.12583333333333332,
+      "grad_norm": 34.75,
+      "grad_norm_var": 6.41780526291913e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2767,
+      "loss/crossentropy": 1.98907128572464,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1829331198707223,
+      "step": 15100
+    },
+    {
+      "epoch": 0.12591666666666668,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.41780525988158e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.067,
+      "loss/crossentropy": 1.8924177050590516,
+      "loss/hidden": 3.38203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17696069460362196,
+      "step": 15110
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 32.75,
+      "grad_norm_var": 10.72265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2268,
+      "loss/crossentropy": 1.8721692830324173,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17582578407600521,
+      "step": 15120
+    },
+    {
+      "epoch": 0.12608333333333333,
+      "grad_norm": 57.75,
+      "grad_norm_var": 194.5119140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1612,
+      "loss/crossentropy": 1.9719461858272553,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17765378914773464,
+      "step": 15130
+    },
+    {
+      "epoch": 0.12616666666666668,
+      "grad_norm": 38.75,
+      "grad_norm_var": 51.21145833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2458,
+      "loss/crossentropy": 1.8766948290169239,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18227943358942866,
+      "step": 15140
+    },
+    {
+      "epoch": 0.12625,
+      "grad_norm": 30.125,
+      "grad_norm_var": 27.754166666666666,
+      "learning_rate": 0.0001,
+      "loss": 6.9738,
+      "loss/crossentropy": 1.8035222582519055,
+      "loss/hidden": 3.216796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16646662391722203,
+      "step": 15150
+    },
+    {
+      "epoch": 0.12633333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 31.848372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0791,
+      "loss/crossentropy": 1.749423536658287,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16647220477461816,
+      "step": 15160
+    },
+    {
+      "epoch": 0.12641666666666668,
+      "grad_norm": 28.875,
+      "grad_norm_var": 56.77493489583333,
+      "learning_rate": 0.0001,
+      "loss": 7.1518,
+      "loss/crossentropy": 1.940764057636261,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19319361187517642,
+      "step": 15170
+    },
+    {
+      "epoch": 0.1265,
+      "grad_norm": 31.625,
+      "grad_norm_var": 33.430989583333336,
+      "learning_rate": 0.0001,
+      "loss": 7.133,
+      "loss/crossentropy": 1.9622076898813248,
+      "loss/hidden": 3.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.172550798766315,
+      "step": 15180
+    },
+    {
+      "epoch": 0.12658333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 32.53515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1377,
+      "loss/crossentropy": 1.9135102063417435,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18686997145414352,
+      "step": 15190
+    },
+    {
+      "epoch": 0.12666666666666668,
+      "grad_norm": 34.0,
+      "grad_norm_var": 37.08125,
+      "learning_rate": 0.0001,
+      "loss": 7.1447,
+      "loss/crossentropy": 1.9013337537646293,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19119456205517055,
+      "step": 15200
+    },
+    {
+      "epoch": 0.12675,
+      "grad_norm": 34.75,
+      "grad_norm_var": 1.3595944868521882e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2682,
+      "loss/crossentropy": 2.009699907898903,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1828319203108549,
+      "step": 15210
+    },
+    {
+      "epoch": 0.12683333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.0818451890705882e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2749,
+      "loss/crossentropy": 1.889081773161888,
+      "loss/hidden": 3.81171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22749623227864504,
+      "step": 15220
+    },
+    {
+      "epoch": 0.12691666666666668,
+      "grad_norm": 28.5,
+      "grad_norm_var": 9.47890625,
+      "learning_rate": 0.0001,
+      "loss": 7.149,
+      "loss/crossentropy": 2.046559324860573,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19004294071346522,
+      "step": 15230
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.628059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1536,
+      "loss/crossentropy": 1.9196593284606933,
+      "loss/hidden": 3.441796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19016770981252193,
+      "step": 15240
+    },
+    {
+      "epoch": 0.12708333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 4.834375,
+      "learning_rate": 0.0001,
+      "loss": 7.1153,
+      "loss/crossentropy": 1.9865416973829269,
+      "loss/hidden": 3.429296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18435781188309192,
+      "step": 15250
+    },
+    {
+      "epoch": 0.12716666666666668,
+      "grad_norm": 36.5,
+      "grad_norm_var": 2.6399273310722464e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2194,
+      "loss/crossentropy": 2.040367802977562,
+      "loss/hidden": 3.394921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18219520393759012,
+      "step": 15260
+    },
+    {
+      "epoch": 0.12725,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.826497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1216,
+      "loss/crossentropy": 1.9420078113675117,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17638408057391644,
+      "step": 15270
+    },
+    {
+      "epoch": 0.12733333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 6.365625,
+      "learning_rate": 0.0001,
+      "loss": 7.1515,
+      "loss/crossentropy": 2.0508300185203554,
+      "loss/hidden": 3.2765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18873732509091495,
+      "step": 15280
+    },
+    {
+      "epoch": 0.12741666666666668,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.594205729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1325,
+      "loss/crossentropy": 1.8841113954782487,
+      "loss/hidden": 3.466796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20680903624743224,
+      "step": 15290
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.5625,
+      "learning_rate": 0.0001,
+      "loss": 7.0077,
+      "loss/crossentropy": 1.9704409554600715,
+      "loss/hidden": 3.259375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18083923123776913,
+      "step": 15300
+    },
+    {
+      "epoch": 0.12758333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.6025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0905,
+      "loss/crossentropy": 1.9114600375294686,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17417905516922474,
+      "step": 15310
+    },
+    {
+      "epoch": 0.12766666666666668,
+      "grad_norm": 30.875,
+      "grad_norm_var": 55.520572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0758,
+      "loss/crossentropy": 1.911510793864727,
+      "loss/hidden": 3.2515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17712097559124232,
+      "step": 15320
+    },
+    {
+      "epoch": 0.12775,
+      "grad_norm": 28.0,
+      "grad_norm_var": 5.686393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0649,
+      "loss/crossentropy": 1.867690885066986,
+      "loss/hidden": 3.266015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18837859556078912,
+      "step": 15330
+    },
+    {
+      "epoch": 0.12783333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.5747395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1822,
+      "loss/crossentropy": 1.8442138597369193,
+      "loss/hidden": 3.3859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17477424200624228,
+      "step": 15340
+    },
+    {
+      "epoch": 0.12791666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.1083333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1386,
+      "loss/crossentropy": 1.899175336956978,
+      "loss/hidden": 3.309765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16746481824666262,
+      "step": 15350
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 29.25,
+      "grad_norm_var": 5.400455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1742,
+      "loss/crossentropy": 1.917744728922844,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18900063615292312,
+      "step": 15360
+    },
+    {
+      "epoch": 0.12808333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.995768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0668,
+      "loss/crossentropy": 2.036524084210396,
+      "loss/hidden": 3.255078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18470738660544156,
+      "step": 15370
+    },
+    {
+      "epoch": 0.12816666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.7962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1984,
+      "loss/crossentropy": 1.9222897566854953,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17857631081715225,
+      "step": 15380
+    },
+    {
+      "epoch": 0.12825,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.356184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0456,
+      "loss/crossentropy": 1.9531288266181945,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17672130912542344,
+      "step": 15390
+    },
+    {
+      "epoch": 0.12833333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.80390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1455,
+      "loss/crossentropy": 1.9161741420626641,
+      "loss/hidden": 3.43203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1759538996964693,
+      "step": 15400
+    },
+    {
+      "epoch": 0.12841666666666668,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.869205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0978,
+      "loss/crossentropy": 1.9341224372386931,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18735274989157916,
+      "step": 15410
+    },
+    {
+      "epoch": 0.1285,
+      "grad_norm": 33.75,
+      "grad_norm_var": 8.350455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1612,
+      "loss/crossentropy": 2.0203406751155852,
+      "loss/hidden": 3.336328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18966775350272655,
+      "step": 15420
+    },
+    {
+      "epoch": 0.12858333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 16.042708333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.173,
+      "loss/crossentropy": 2.000143977999687,
+      "loss/hidden": 3.264453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17995088547468185,
+      "step": 15430
+    },
+    {
+      "epoch": 0.12866666666666668,
+      "grad_norm": 35.75,
+      "grad_norm_var": 5.626822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1843,
+      "loss/crossentropy": 1.9581128850579261,
+      "loss/hidden": 3.38203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1853957787156105,
+      "step": 15440
+    },
+    {
+      "epoch": 0.12875,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.47265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2325,
+      "loss/crossentropy": 1.8682523369789124,
+      "loss/hidden": 3.43984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20011475328356026,
+      "step": 15450
+    },
+    {
+      "epoch": 0.12883333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.126822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0661,
+      "loss/crossentropy": 1.9018582656979561,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17348549887537956,
+      "step": 15460
+    },
+    {
+      "epoch": 0.12891666666666668,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.7212890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2532,
+      "loss/crossentropy": 2.0075298234820367,
+      "loss/hidden": 3.437109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19473833702504634,
+      "step": 15470
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.572330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1726,
+      "loss/crossentropy": 1.9899009227752686,
+      "loss/hidden": 3.430078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20826521869748832,
+      "step": 15480
+    },
+    {
+      "epoch": 0.12908333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.590625,
+      "learning_rate": 0.0001,
+      "loss": 7.3254,
+      "loss/crossentropy": 1.9966183453798294,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20746800526976586,
+      "step": 15490
+    },
+    {
+      "epoch": 0.12916666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.45390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1495,
+      "loss/crossentropy": 1.9456835612654686,
+      "loss/hidden": 3.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1936283141374588,
+      "step": 15500
+    },
+    {
+      "epoch": 0.12925,
+      "grad_norm": 28.75,
+      "grad_norm_var": 45.63645833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2269,
+      "loss/crossentropy": 1.9356549561023713,
+      "loss/hidden": 3.472265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1964884826913476,
+      "step": 15510
+    },
+    {
+      "epoch": 0.12933333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 17.330208333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1962,
+      "loss/crossentropy": 2.0024677157402038,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897840365767479,
+      "step": 15520
+    },
+    {
+      "epoch": 0.12941666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.674934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.046,
+      "loss/crossentropy": 1.9206486582756042,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18070366214960815,
+      "step": 15530
+    },
+    {
+      "epoch": 0.1295,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.604166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2061,
+      "loss/crossentropy": 1.990031922608614,
+      "loss/hidden": 3.3078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1833953645080328,
+      "step": 15540
+    },
+    {
+      "epoch": 0.12958333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.7556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1962,
+      "loss/crossentropy": 1.9080590605735779,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18792618289589882,
+      "step": 15550
+    },
+    {
+      "epoch": 0.12966666666666668,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.15390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2062,
+      "loss/crossentropy": 2.01756657063961,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19796357899904252,
+      "step": 15560
+    },
+    {
+      "epoch": 0.12975,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.372916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0979,
+      "loss/crossentropy": 1.944714032113552,
+      "loss/hidden": 3.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17425163146108388,
+      "step": 15570
+    },
+    {
+      "epoch": 0.12983333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.1892317582454513e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3316,
+      "loss/crossentropy": 2.009487435221672,
+      "loss/hidden": 3.540625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19467739202082157,
+      "step": 15580
+    },
+    {
+      "epoch": 0.12991666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 11.234830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1979,
+      "loss/crossentropy": 1.984761607646942,
+      "loss/hidden": 3.284375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17868782561272384,
+      "step": 15590
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.7535807291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1806,
+      "loss/crossentropy": 1.954510636627674,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859197087585926,
+      "step": 15600
+    },
+    {
+      "epoch": 0.13008333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.5259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1472,
+      "loss/crossentropy": 2.0174022622406484,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749066274613142,
+      "step": 15610
+    },
+    {
+      "epoch": 0.13016666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.512434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1799,
+      "loss/crossentropy": 2.027938242256641,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19273827634751797,
+      "step": 15620
+    },
+    {
+      "epoch": 0.13025,
+      "grad_norm": 30.375,
+      "grad_norm_var": 0.8603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2191,
+      "loss/crossentropy": 1.922875054180622,
+      "loss/hidden": 3.31796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18529862463474273,
+      "step": 15630
+    },
+    {
+      "epoch": 0.13033333333333333,
+      "grad_norm": 39.0,
+      "grad_norm_var": 6.31875,
+      "learning_rate": 0.0001,
+      "loss": 7.2483,
+      "loss/crossentropy": 2.0646585285663606,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19696942400187253,
+      "step": 15640
+    },
+    {
+      "epoch": 0.13041666666666665,
+      "grad_norm": 31.75,
+      "grad_norm_var": 10.042708333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1909,
+      "loss/crossentropy": 1.97297203540802,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1981798242777586,
+      "step": 15650
+    },
+    {
+      "epoch": 0.1305,
+      "grad_norm": 31.375,
+      "grad_norm_var": 8.3509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0978,
+      "loss/crossentropy": 1.9443943321704864,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927189402282238,
+      "step": 15660
+    },
+    {
+      "epoch": 0.13058333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.998372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1333,
+      "loss/crossentropy": 1.9420040637254714,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19153628256171942,
+      "step": 15670
+    },
+    {
+      "epoch": 0.13066666666666665,
+      "grad_norm": 35.75,
+      "grad_norm_var": 7.718684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2449,
+      "loss/crossentropy": 2.033763536810875,
+      "loss/hidden": 3.308984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19441983066499233,
+      "step": 15680
+    },
+    {
+      "epoch": 0.13075,
+      "grad_norm": 36.75,
+      "grad_norm_var": 1.0388009762571683e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.3474,
+      "loss/crossentropy": 1.947944739460945,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1805997768417001,
+      "step": 15690
+    },
+    {
+      "epoch": 0.13083333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.0388009773995393e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2303,
+      "loss/crossentropy": 1.96535335034132,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18672094820067286,
+      "step": 15700
+    },
+    {
+      "epoch": 0.13091666666666665,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.0927083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1291,
+      "loss/crossentropy": 1.8772934928536416,
+      "loss/hidden": 3.3109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17811705311760306,
+      "step": 15710
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 32.25,
+      "grad_norm_var": 19.464322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2074,
+      "loss/crossentropy": 1.9448927730321883,
+      "loss/hidden": 3.2828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18854384291917087,
+      "step": 15720
+    },
+    {
+      "epoch": 0.13108333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 30.85390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1755,
+      "loss/crossentropy": 1.8886026624590158,
+      "loss/hidden": 3.45390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18559149792417884,
+      "step": 15730
+    },
+    {
+      "epoch": 0.13116666666666665,
+      "grad_norm": 31.75,
+      "grad_norm_var": 30.999739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1293,
+      "loss/crossentropy": 1.9122322157025338,
+      "loss/hidden": 3.53203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20887614991515874,
+      "step": 15740
+    },
+    {
+      "epoch": 0.13125,
+      "grad_norm": 46.5,
+      "grad_norm_var": 15.768489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2532,
+      "loss/crossentropy": 1.889380306005478,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18477531764656305,
+      "step": 15750
+    },
+    {
+      "epoch": 0.13133333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 17.316666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1737,
+      "loss/crossentropy": 2.007970739901066,
+      "loss/hidden": 3.361328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17606964632868766,
+      "step": 15760
+    },
+    {
+      "epoch": 0.13141666666666665,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.4462890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1454,
+      "loss/crossentropy": 1.9743964850902558,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17718939799815417,
+      "step": 15770
+    },
+    {
+      "epoch": 0.1315,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.8202473958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1243,
+      "loss/crossentropy": 1.9149464294314384,
+      "loss/hidden": 3.34609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18673150297254323,
+      "step": 15780
+    },
+    {
+      "epoch": 0.13158333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.0572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1071,
+      "loss/crossentropy": 1.8895102962851524,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19650962334126235,
+      "step": 15790
+    },
+    {
+      "epoch": 0.13166666666666665,
+      "grad_norm": 29.875,
+      "grad_norm_var": 19.887434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2393,
+      "loss/crossentropy": 1.934299600124359,
+      "loss/hidden": 3.449609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18527342919260265,
+      "step": 15800
+    },
+    {
+      "epoch": 0.13175,
+      "grad_norm": 30.5,
+      "grad_norm_var": 20.559830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2594,
+      "loss/crossentropy": 2.0676255375146866,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1885557908564806,
+      "step": 15810
+    },
+    {
+      "epoch": 0.13183333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 16.451822916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.0991,
+      "loss/crossentropy": 1.9720468699932099,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1806204194203019,
+      "step": 15820
+    },
+    {
+      "epoch": 0.13191666666666665,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.4233723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2149,
+      "loss/crossentropy": 1.937279088050127,
+      "loss/hidden": 3.3859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1986948948353529,
+      "step": 15830
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.216080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1224,
+      "loss/crossentropy": 1.9335426777601241,
+      "loss/hidden": 3.250390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18058245107531548,
+      "step": 15840
+    },
+    {
+      "epoch": 0.13208333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 45.91868489583333,
+      "learning_rate": 0.0001,
+      "loss": 7.1376,
+      "loss/crossentropy": 1.9211553812026978,
+      "loss/hidden": 3.280859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1789987886324525,
+      "step": 15850
+    },
+    {
+      "epoch": 0.13216666666666665,
+      "grad_norm": 39.0,
+      "grad_norm_var": 38.236393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1365,
+      "loss/crossentropy": 1.9883318156003953,
+      "loss/hidden": 3.471484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1840870851650834,
+      "step": 15860
+    },
+    {
+      "epoch": 0.13225,
+      "grad_norm": 41.75,
+      "grad_norm_var": 1.6044777139071877e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2617,
+      "loss/crossentropy": 2.0764810025691984,
+      "loss/hidden": 3.24453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1871108701452613,
+      "step": 15870
+    },
+    {
+      "epoch": 0.13233333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.6044777144455268e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1396,
+      "loss/crossentropy": 2.001447968930006,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18228637967258693,
+      "step": 15880
+    },
+    {
+      "epoch": 0.13241666666666665,
+      "grad_norm": 28.0,
+      "grad_norm_var": 20.817122395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0422,
+      "loss/crossentropy": 2.065645283460617,
+      "loss/hidden": 3.221875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17658126037567853,
+      "step": 15890
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.318684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0876,
+      "loss/crossentropy": 2.0048751294612885,
+      "loss/hidden": 3.190625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16993176201358437,
+      "step": 15900
+    },
+    {
+      "epoch": 0.13258333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.062239583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1936,
+      "loss/crossentropy": 1.992582231760025,
+      "loss/hidden": 3.288671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18070014007389545,
+      "step": 15910
+    },
+    {
+      "epoch": 0.13266666666666665,
+      "grad_norm": 29.0,
+      "grad_norm_var": 6.130989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0372,
+      "loss/crossentropy": 1.9080621749162674,
+      "loss/hidden": 3.430859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18981511779129506,
+      "step": 15920
+    },
+    {
+      "epoch": 0.13275,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.7666015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1492,
+      "loss/crossentropy": 1.9066756546497345,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19044621121138333,
+      "step": 15930
+    },
+    {
+      "epoch": 0.13283333333333333,
+      "grad_norm": 35.5,
+      "grad_norm_var": 7.058072916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1766,
+      "loss/crossentropy": 2.022434741258621,
+      "loss/hidden": 3.3375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846444124355912,
+      "step": 15940
+    },
+    {
+      "epoch": 0.13291666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.379622395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0968,
+      "loss/crossentropy": 1.9119803443551064,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18046996165066959,
+      "step": 15950
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.94140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1163,
+      "loss/crossentropy": 2.130883114039898,
+      "loss/hidden": 3.281640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20072123007848858,
+      "step": 15960
+    },
+    {
+      "epoch": 0.13308333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.257747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1689,
+      "loss/crossentropy": 1.9111644223332405,
+      "loss/hidden": 3.465234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18044994678348303,
+      "step": 15970
+    },
+    {
+      "epoch": 0.13316666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 11.488541666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1912,
+      "loss/crossentropy": 1.8651348814368247,
+      "loss/hidden": 3.2296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1784767468459904,
+      "step": 15980
+    },
+    {
+      "epoch": 0.13325,
+      "grad_norm": 31.25,
+      "grad_norm_var": 19.318489583333335,
+      "learning_rate": 0.0001,
+      "loss": 7.182,
+      "loss/crossentropy": 2.019369959831238,
+      "loss/hidden": 3.448828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1955235343426466,
+      "step": 15990
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 18.403059895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1219,
+      "loss/crossentropy": 1.951167756319046,
+      "loss/hidden": 3.415234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18693782836198808,
+      "step": 16000
+    },
+    {
+      "epoch": 0.13341666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.444791666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1177,
+      "loss/crossentropy": 1.9451298780739308,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18884576605632902,
+      "step": 16010
+    },
+    {
+      "epoch": 0.1335,
+      "grad_norm": 31.625,
+      "grad_norm_var": 5.2509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0904,
+      "loss/crossentropy": 1.9699655503034592,
+      "loss/hidden": 3.294921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18711993899196386,
+      "step": 16020
+    },
+    {
+      "epoch": 0.13358333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 7.306184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1864,
+      "loss/crossentropy": 2.0398706987500193,
+      "loss/hidden": 3.255078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17892837729305028,
+      "step": 16030
+    },
+    {
+      "epoch": 0.13366666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2186848958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1511,
+      "loss/crossentropy": 2.0942850202322005,
+      "loss/hidden": 3.25625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18929117489606143,
+      "step": 16040
+    },
+    {
+      "epoch": 0.13375,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.0104166666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1738,
+      "loss/crossentropy": 1.8911314755678177,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18707896452397108,
+      "step": 16050
+    },
+    {
+      "epoch": 0.13383333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.287955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2119,
+      "loss/crossentropy": 1.9806726440787314,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18338480899110438,
+      "step": 16060
+    },
+    {
+      "epoch": 0.13391666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.6916015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1687,
+      "loss/crossentropy": 1.9684767544269561,
+      "loss/hidden": 3.283203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18100943136960268,
+      "step": 16070
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.442122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1606,
+      "loss/crossentropy": 1.8862227275967598,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18479395639151336,
+      "step": 16080
+    },
+    {
+      "epoch": 0.13408333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.35625,
+      "learning_rate": 0.0001,
+      "loss": 7.1319,
+      "loss/crossentropy": 2.037122744321823,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19270640648901463,
+      "step": 16090
+    },
+    {
+      "epoch": 0.13416666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.3393229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1462,
+      "loss/crossentropy": 1.9014740332961082,
+      "loss/hidden": 3.408203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20305475890636443,
+      "step": 16100
+    },
+    {
+      "epoch": 0.13425,
+      "grad_norm": 29.0,
+      "grad_norm_var": 2.8270833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2143,
+      "loss/crossentropy": 1.9933300539851189,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20600338969379664,
+      "step": 16110
+    },
+    {
+      "epoch": 0.13433333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 9.6416015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2357,
+      "loss/crossentropy": 1.9651847302913665,
+      "loss/hidden": 3.319140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18921629693359138,
+      "step": 16120
+    },
+    {
+      "epoch": 0.13441666666666666,
+      "grad_norm": 35.5,
+      "grad_norm_var": 2.6426432291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.207,
+      "loss/crossentropy": 1.9286102376878262,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1868777284398675,
+      "step": 16130
+    },
+    {
+      "epoch": 0.1345,
+      "grad_norm": 37.25,
+      "grad_norm_var": 29.420572916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2056,
+      "loss/crossentropy": 1.9314211502671241,
+      "loss/hidden": 3.483984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20501322001218797,
+      "step": 16140
+    },
+    {
+      "epoch": 0.13458333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 537.5603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1686,
+      "loss/crossentropy": 2.025644926726818,
+      "loss/hidden": 3.348046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.193122448772192,
+      "step": 16150
+    },
+    {
+      "epoch": 0.13466666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.946809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1749,
+      "loss/crossentropy": 1.8911065384745598,
+      "loss/hidden": 3.521875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18477482963353395,
+      "step": 16160
+    },
+    {
+      "epoch": 0.13475,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.7733723958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2075,
+      "loss/crossentropy": 2.027583548426628,
+      "loss/hidden": 3.24140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18013087175786496,
+      "step": 16170
+    },
+    {
+      "epoch": 0.13483333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.6184208882531485e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2689,
+      "loss/crossentropy": 1.9891380831599235,
+      "loss/hidden": 3.31328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18181416615843773,
+      "step": 16180
+    },
+    {
+      "epoch": 0.13491666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.619205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.146,
+      "loss/crossentropy": 1.9221839740872384,
+      "loss/hidden": 3.38046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18138978648930787,
+      "step": 16190
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 43.0,
+      "grad_norm_var": 13.951822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1264,
+      "loss/crossentropy": 2.033442145586014,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1908789176493883,
+      "step": 16200
+    },
+    {
+      "epoch": 0.13508333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.408268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3406,
+      "loss/crossentropy": 1.9518346801400184,
+      "loss/hidden": 3.4265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20574640445411205,
+      "step": 16210
+    },
+    {
+      "epoch": 0.13516666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.698372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0818,
+      "loss/crossentropy": 1.8279100894927978,
+      "loss/hidden": 3.374609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16689587775617837,
+      "step": 16220
+    },
+    {
+      "epoch": 0.13525,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.53807565806266e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2268,
+      "loss/crossentropy": 2.125316482782364,
+      "loss/hidden": 3.30390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20031116604804994,
+      "step": 16230
+    },
+    {
+      "epoch": 0.13533333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 0.9604166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1722,
+      "loss/crossentropy": 1.9833017945289613,
+      "loss/hidden": 3.385546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19948794972151518,
+      "step": 16240
+    },
+    {
+      "epoch": 0.13541666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.184375,
+      "learning_rate": 0.0001,
+      "loss": 7.1387,
+      "loss/crossentropy": 1.9802332893013954,
+      "loss/hidden": 3.308984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1790276074782014,
+      "step": 16250
+    },
+    {
+      "epoch": 0.1355,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.8869140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0948,
+      "loss/crossentropy": 1.9125315710902213,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17763164471834897,
+      "step": 16260
+    },
+    {
+      "epoch": 0.13558333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 56.84765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2524,
+      "loss/crossentropy": 1.899502444267273,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19081068132072687,
+      "step": 16270
+    },
+    {
+      "epoch": 0.13566666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.7739583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.22,
+      "loss/crossentropy": 2.019946187734604,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17925603445619345,
+      "step": 16280
+    },
+    {
+      "epoch": 0.13575,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.7363932291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2329,
+      "loss/crossentropy": 2.1178683906793596,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19205426890403032,
+      "step": 16290
+    },
+    {
+      "epoch": 0.13583333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.3129557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1618,
+      "loss/crossentropy": 1.8536071583628655,
+      "loss/hidden": 3.236328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.199834194034338,
+      "step": 16300
+    },
+    {
+      "epoch": 0.13591666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.3707682291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0803,
+      "loss/crossentropy": 1.9322735711932182,
+      "loss/hidden": 3.405078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857559671625495,
+      "step": 16310
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.1041666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1092,
+      "loss/crossentropy": 1.875018187612295,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17738636545836925,
+      "step": 16320
+    },
+    {
+      "epoch": 0.13608333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.340625,
+      "learning_rate": 0.0001,
+      "loss": 7.1927,
+      "loss/crossentropy": 1.8174295820295812,
+      "loss/hidden": 3.530859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1959748630411923,
+      "step": 16330
+    },
+    {
+      "epoch": 0.13616666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.83125,
+      "learning_rate": 0.0001,
+      "loss": 7.2945,
+      "loss/crossentropy": 1.8757753998041153,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18712248653173447,
+      "step": 16340
+    },
+    {
+      "epoch": 0.13625,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.7462890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1948,
+      "loss/crossentropy": 1.9482501089572906,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17816255632787942,
+      "step": 16350
+    },
+    {
+      "epoch": 0.13633333333333333,
+      "grad_norm": 28.625,
+      "grad_norm_var": 1.8952473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2756,
+      "loss/crossentropy": 1.9245134707540275,
+      "loss/hidden": 3.233203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16692193718627096,
+      "step": 16360
+    },
+    {
+      "epoch": 0.13641666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.713997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2218,
+      "loss/crossentropy": 1.9683432579040527,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18291530329734088,
+      "step": 16370
+    },
+    {
+      "epoch": 0.1365,
+      "grad_norm": 34.25,
+      "grad_norm_var": 4.533072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2604,
+      "loss/crossentropy": 1.8656769856810569,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18199607012793423,
+      "step": 16380
+    },
+    {
+      "epoch": 0.13658333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.274934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.143,
+      "loss/crossentropy": 2.0562920920550822,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18838359927758574,
+      "step": 16390
+    },
+    {
+      "epoch": 0.13666666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.082291666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.4106,
+      "loss/crossentropy": 1.880037860572338,
+      "loss/hidden": 3.45390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22908992245793341,
+      "step": 16400
+    },
+    {
+      "epoch": 0.13675,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.725,
+      "learning_rate": 0.0001,
+      "loss": 7.2986,
+      "loss/crossentropy": 1.8582791283726692,
+      "loss/hidden": 3.466796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19821546822786332,
+      "step": 16410
+    },
+    {
+      "epoch": 0.13683333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.106705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2276,
+      "loss/crossentropy": 1.965068084001541,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18812269438058138,
+      "step": 16420
+    },
+    {
+      "epoch": 0.13691666666666666,
+      "grad_norm": 36.0,
+      "grad_norm_var": 2.4833333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.321,
+      "loss/crossentropy": 2.0821048587560655,
+      "loss/hidden": 3.32578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19172437936067582,
+      "step": 16430
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.912239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1378,
+      "loss/crossentropy": 2.0249056473374365,
+      "loss/hidden": 3.49453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1955235879868269,
+      "step": 16440
+    },
+    {
+      "epoch": 0.13708333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.97265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1802,
+      "loss/crossentropy": 1.9519851967692374,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18446833454072475,
+      "step": 16450
+    },
+    {
+      "epoch": 0.13716666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.9205729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2567,
+      "loss/crossentropy": 2.0515071928501127,
+      "loss/hidden": 3.221484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18094625659286975,
+      "step": 16460
+    },
+    {
+      "epoch": 0.13725,
+      "grad_norm": 37.75,
+      "grad_norm_var": 21.0447265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2549,
+      "loss/crossentropy": 1.8245956741273404,
+      "loss/hidden": 3.459375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19014889169484378,
+      "step": 16470
+    },
+    {
+      "epoch": 0.13733333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 26.6134765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2259,
+      "loss/crossentropy": 2.1028824627399443,
+      "loss/hidden": 3.35859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18351248502731324,
+      "step": 16480
+    },
+    {
+      "epoch": 0.13741666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 10.264322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1711,
+      "loss/crossentropy": 1.9383347466588021,
+      "loss/hidden": 3.225,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17141741700470448,
+      "step": 16490
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 31.5,
+      "grad_norm_var": 58.34583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.165,
+      "loss/crossentropy": 1.766675853729248,
+      "loss/hidden": 3.516796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1954731835052371,
+      "step": 16500
+    },
+    {
+      "epoch": 0.13758333333333334,
+      "grad_norm": 43.25,
+      "grad_norm_var": 10.204166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1487,
+      "loss/crossentropy": 2.0143339335918427,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953774724155663,
+      "step": 16510
+    },
+    {
+      "epoch": 0.13766666666666666,
+      "grad_norm": 39.75,
+      "grad_norm_var": 1.2447526976241708e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1666,
+      "loss/crossentropy": 2.1544763147830963,
+      "loss/hidden": 3.44453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21087858639657497,
+      "step": 16520
+    },
+    {
+      "epoch": 0.13775,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.2447526976706578e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1727,
+      "loss/crossentropy": 2.115397298336029,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19533645510673522,
+      "step": 16530
+    },
+    {
+      "epoch": 0.13783333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 10.255208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.161,
+      "loss/crossentropy": 1.7955206647515296,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16244610734283924,
+      "step": 16540
+    },
+    {
+      "epoch": 0.13791666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.6372395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1505,
+      "loss/crossentropy": 1.9106352940201758,
+      "loss/hidden": 3.32109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17099260557442902,
+      "step": 16550
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 32.0,
+      "grad_norm_var": 10.684375,
+      "learning_rate": 0.0001,
+      "loss": 7.2617,
+      "loss/crossentropy": 1.9729534566402436,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1935939616523683,
+      "step": 16560
+    },
+    {
+      "epoch": 0.13808333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 0.8103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1139,
+      "loss/crossentropy": 1.9311561360955238,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17554287016391754,
+      "step": 16570
+    },
+    {
+      "epoch": 0.13816666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.2020833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1122,
+      "loss/crossentropy": 1.8604721650481224,
+      "loss/hidden": 3.287109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17922435896471142,
+      "step": 16580
+    },
+    {
+      "epoch": 0.13825,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.3893229166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1541,
+      "loss/crossentropy": 1.8901896685361863,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17984781051054596,
+      "step": 16590
+    },
+    {
+      "epoch": 0.13833333333333334,
+      "grad_norm": 36.0,
+      "grad_norm_var": 8.6306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2119,
+      "loss/crossentropy": 1.9950422018766403,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17552002966403962,
+      "step": 16600
+    },
+    {
+      "epoch": 0.13841666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.494205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1255,
+      "loss/crossentropy": 2.0939732491970062,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20039153583347796,
+      "step": 16610
+    },
+    {
+      "epoch": 0.1385,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.6478515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1838,
+      "loss/crossentropy": 1.959303469210863,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18559585232287645,
+      "step": 16620
+    },
+    {
+      "epoch": 0.13858333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.59140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2226,
+      "loss/crossentropy": 1.9500691398978234,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.171992280613631,
+      "step": 16630
+    },
+    {
+      "epoch": 0.13866666666666666,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.5014973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1067,
+      "loss/crossentropy": 1.9014824159443378,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17935178810730576,
+      "step": 16640
+    },
+    {
+      "epoch": 0.13875,
+      "grad_norm": 30.875,
+      "grad_norm_var": 13.858268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1515,
+      "loss/crossentropy": 1.9320432528853417,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.184787742421031,
+      "step": 16650
+    },
+    {
+      "epoch": 0.13883333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 13.001041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1034,
+      "loss/crossentropy": 1.9003541678190232,
+      "loss/hidden": 3.47734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19456628821790217,
+      "step": 16660
+    },
+    {
+      "epoch": 0.13891666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.5686848958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1777,
+      "loss/crossentropy": 2.040318152308464,
+      "loss/hidden": 3.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19157934691756964,
+      "step": 16670
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.255989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1691,
+      "loss/crossentropy": 1.8969489634037018,
+      "loss/hidden": 3.260546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16939087118953466,
+      "step": 16680
+    },
+    {
+      "epoch": 0.13908333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.186393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2077,
+      "loss/crossentropy": 1.8697502434253692,
+      "loss/hidden": 3.44921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18470772933214902,
+      "step": 16690
+    },
+    {
+      "epoch": 0.13916666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.923958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2238,
+      "loss/crossentropy": 1.9544553384184837,
+      "loss/hidden": 3.351171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19111399538815022,
+      "step": 16700
+    },
+    {
+      "epoch": 0.13925,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.709830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1427,
+      "loss/crossentropy": 1.9969765067100524,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18862808551639318,
+      "step": 16710
+    },
+    {
+      "epoch": 0.13933333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.6122395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1844,
+      "loss/crossentropy": 2.017768216133118,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19979458674788475,
+      "step": 16720
+    },
+    {
+      "epoch": 0.13941666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.53807565701356e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3168,
+      "loss/crossentropy": 2.089987801015377,
+      "loss/hidden": 3.36640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1985271617770195,
+      "step": 16730
+    },
+    {
+      "epoch": 0.1395,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.1775390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0846,
+      "loss/crossentropy": 1.8769831657409668,
+      "loss/hidden": 3.358984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1834286769852042,
+      "step": 16740
+    },
+    {
+      "epoch": 0.13958333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.5764973958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2574,
+      "loss/crossentropy": 2.0436582267284393,
+      "loss/hidden": 3.458984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18217622712254525,
+      "step": 16750
+    },
+    {
+      "epoch": 0.13966666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 8.341666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1923,
+      "loss/crossentropy": 2.016822022199631,
+      "loss/hidden": 3.300390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17654962390661239,
+      "step": 16760
+    },
+    {
+      "epoch": 0.13975,
+      "grad_norm": 31.25,
+      "grad_norm_var": 5.278125,
+      "learning_rate": 0.0001,
+      "loss": 7.1375,
+      "loss/crossentropy": 2.0330835223197936,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18024676851928234,
+      "step": 16770
+    },
+    {
+      "epoch": 0.13983333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.449739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1755,
+      "loss/crossentropy": 1.9075776189565659,
+      "loss/hidden": 3.3796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180111500993371,
+      "step": 16780
+    },
+    {
+      "epoch": 0.13991666666666666,
+      "grad_norm": 36.75,
+      "grad_norm_var": 6.958333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0967,
+      "loss/crossentropy": 1.8716136440634727,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19005781207233668,
+      "step": 16790
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 34.0,
+      "grad_norm_var": 7.789518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1581,
+      "loss/crossentropy": 1.902388795465231,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16655305847525598,
+      "step": 16800
+    },
+    {
+      "epoch": 0.14008333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.976822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2142,
+      "loss/crossentropy": 1.979095134139061,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19661365021020175,
+      "step": 16810
+    },
+    {
+      "epoch": 0.14016666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.1561848958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1482,
+      "loss/crossentropy": 1.8501420453190804,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20047238562256098,
+      "step": 16820
+    },
+    {
+      "epoch": 0.14025,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.081184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2969,
+      "loss/crossentropy": 2.0721643194556236,
+      "loss/hidden": 3.268359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18362869489938022,
+      "step": 16830
+    },
+    {
+      "epoch": 0.14033333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 3.114518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1583,
+      "loss/crossentropy": 2.022780400514603,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18333950098603963,
+      "step": 16840
+    },
+    {
+      "epoch": 0.14041666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.677018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1397,
+      "loss/crossentropy": 2.0585033267736437,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18751913998275996,
+      "step": 16850
+    },
+    {
+      "epoch": 0.1405,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.827055114272375e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1745,
+      "loss/crossentropy": 2.0010389149188996,
+      "loss/hidden": 3.3078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18487434647977352,
+      "step": 16860
+    },
+    {
+      "epoch": 0.14058333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.617708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1279,
+      "loss/crossentropy": 1.9822161227464676,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21053801085799934,
+      "step": 16870
+    },
+    {
+      "epoch": 0.14066666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 91.3125,
+      "learning_rate": 0.0001,
+      "loss": 7.2491,
+      "loss/crossentropy": 1.929964379966259,
+      "loss/hidden": 3.408984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18784339427948,
+      "step": 16880
+    },
+    {
+      "epoch": 0.14075,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.7270833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1116,
+      "loss/crossentropy": 1.97335384786129,
+      "loss/hidden": 3.279296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18938699476420878,
+      "step": 16890
+    },
+    {
+      "epoch": 0.14083333333333334,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.313541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1442,
+      "loss/crossentropy": 2.059920623898506,
+      "loss/hidden": 3.321875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18405720070004464,
+      "step": 16900
+    },
+    {
+      "epoch": 0.14091666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.6322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1629,
+      "loss/crossentropy": 1.8624357432127,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18418188840150834,
+      "step": 16910
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.003059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1561,
+      "loss/crossentropy": 1.9761703863739968,
+      "loss/hidden": 3.43125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19616197869181634,
+      "step": 16920
+    },
+    {
+      "epoch": 0.14108333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 0.9624348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1872,
+      "loss/crossentropy": 1.9086257696151734,
+      "loss/hidden": 3.31484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17154292277991773,
+      "step": 16930
+    },
+    {
+      "epoch": 0.14116666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.6072916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1423,
+      "loss/crossentropy": 1.8942563936114312,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899500945582986,
+      "step": 16940
+    },
+    {
+      "epoch": 0.14125,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.8608723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1097,
+      "loss/crossentropy": 1.886419241130352,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17658573389053345,
+      "step": 16950
+    },
+    {
+      "epoch": 0.14133333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 97.02682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0219,
+      "loss/crossentropy": 1.9107058852910996,
+      "loss/hidden": 3.395703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18192462101578713,
+      "step": 16960
+    },
+    {
+      "epoch": 0.14141666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 115.42076822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2228,
+      "loss/crossentropy": 1.9597920164465905,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18831583354622125,
+      "step": 16970
+    },
+    {
+      "epoch": 0.1415,
+      "grad_norm": 34.25,
+      "grad_norm_var": 24.053125,
+      "learning_rate": 0.0001,
+      "loss": 7.0959,
+      "loss/crossentropy": 1.8494370967149734,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1837020918726921,
+      "step": 16980
+    },
+    {
+      "epoch": 0.14158333333333334,
+      "grad_norm": 42.5,
+      "grad_norm_var": 49.18932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0798,
+      "loss/crossentropy": 1.89827319085598,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19195458963513373,
+      "step": 16990
+    },
+    {
+      "epoch": 0.14166666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 47.3556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1699,
+      "loss/crossentropy": 1.9777143388986587,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18385167326778173,
+      "step": 17000
+    },
+    {
+      "epoch": 0.14175,
+      "grad_norm": 31.0,
+      "grad_norm_var": 14.075,
+      "learning_rate": 0.0001,
+      "loss": 7.1896,
+      "loss/crossentropy": 1.9535424515604973,
+      "loss/hidden": 3.27109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1725752430036664,
+      "step": 17010
+    },
+    {
+      "epoch": 0.14183333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 15.742708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2216,
+      "loss/crossentropy": 1.8578058376908302,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.201280995644629,
+      "step": 17020
+    },
+    {
+      "epoch": 0.14191666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 14.935416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1872,
+      "loss/crossentropy": 1.8223856076598168,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19270891025662423,
+      "step": 17030
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 31.375,
+      "grad_norm_var": 15.419205729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2758,
+      "loss/crossentropy": 1.9710450127720833,
+      "loss/hidden": 3.446875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18656891975551843,
+      "step": 17040
+    },
+    {
+      "epoch": 0.14208333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 11.42265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1801,
+      "loss/crossentropy": 2.0855958491563795,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19650328084826468,
+      "step": 17050
+    },
+    {
+      "epoch": 0.14216666666666666,
+      "grad_norm": 37.0,
+      "grad_norm_var": 13.508268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1392,
+      "loss/crossentropy": 1.8866979256272316,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19344396144151688,
+      "step": 17060
+    },
+    {
+      "epoch": 0.14225,
+      "grad_norm": 31.375,
+      "grad_norm_var": 487.46868489583335,
+      "learning_rate": 0.0001,
+      "loss": 7.2703,
+      "loss/crossentropy": 1.9285213589668273,
+      "loss/hidden": 3.426953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1891774756833911,
+      "step": 17070
+    },
+    {
+      "epoch": 0.14233333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7566853029885734e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.096,
+      "loss/crossentropy": 1.7998098939657212,
+      "loss/hidden": 3.44765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16677596103399991,
+      "step": 17080
+    },
+    {
+      "epoch": 0.14241666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.4139973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.057,
+      "loss/crossentropy": 1.9627402544021606,
+      "loss/hidden": 3.271484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18540838565677403,
+      "step": 17090
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.487434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1591,
+      "loss/crossentropy": 1.9734266802668572,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19736174698919057,
+      "step": 17100
+    },
+    {
+      "epoch": 0.14258333333333334,
+      "grad_norm": 37.0,
+      "grad_norm_var": 9.829622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.154,
+      "loss/crossentropy": 1.9343258753418922,
+      "loss/hidden": 3.36328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18086709678173066,
+      "step": 17110
+    },
+    {
+      "epoch": 0.14266666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 16.243489583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.0218,
+      "loss/crossentropy": 2.0700346119701862,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18307855501770973,
+      "step": 17120
+    },
+    {
+      "epoch": 0.14275,
+      "grad_norm": 38.0,
+      "grad_norm_var": 29.5759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0509,
+      "loss/crossentropy": 1.9485379591584207,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16534922439605,
+      "step": 17130
+    },
+    {
+      "epoch": 0.14283333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 14.601822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.165,
+      "loss/crossentropy": 1.9386911273002625,
+      "loss/hidden": 3.43203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19081108085811138,
+      "step": 17140
+    },
+    {
+      "epoch": 0.14291666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.61640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1197,
+      "loss/crossentropy": 1.9144646424800158,
+      "loss/hidden": 3.1703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1634495006175712,
+      "step": 17150
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 31.25,
+      "grad_norm_var": 9.670572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1047,
+      "loss/crossentropy": 2.0530686244368552,
+      "loss/hidden": 3.276953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18403382133692503,
+      "step": 17160
+    },
+    {
+      "epoch": 0.14308333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 6.658072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2083,
+      "loss/crossentropy": 2.024194931983948,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18329261541366576,
+      "step": 17170
+    },
+    {
+      "epoch": 0.14316666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.056517365020842e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1973,
+      "loss/crossentropy": 2.0247035413980483,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897724865935743,
+      "step": 17180
+    },
+    {
+      "epoch": 0.14325,
+      "grad_norm": 32.5,
+      "grad_norm_var": 21.5619140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1413,
+      "loss/crossentropy": 1.796980158984661,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16852799169719218,
+      "step": 17190
+    },
+    {
+      "epoch": 0.14333333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.705143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1374,
+      "loss/crossentropy": 1.9237392783164977,
+      "loss/hidden": 3.274609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1800888005644083,
+      "step": 17200
+    },
+    {
+      "epoch": 0.14341666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 17.9259765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0886,
+      "loss/crossentropy": 1.9101226203143598,
+      "loss/hidden": 3.1921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18074699472635986,
+      "step": 17210
+    },
+    {
+      "epoch": 0.1435,
+      "grad_norm": 33.0,
+      "grad_norm_var": 18.924739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1298,
+      "loss/crossentropy": 1.98937831223011,
+      "loss/hidden": 3.277734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18572146277874707,
+      "step": 17220
+    },
+    {
+      "epoch": 0.14358333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 14.577083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0321,
+      "loss/crossentropy": 1.825450524687767,
+      "loss/hidden": 3.259765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17630416359752416,
+      "step": 17230
+    },
+    {
+      "epoch": 0.14366666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.491080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2045,
+      "loss/crossentropy": 1.879987331479788,
+      "loss/hidden": 3.423828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024867407977581,
+      "step": 17240
+    },
+    {
+      "epoch": 0.14375,
+      "grad_norm": 35.75,
+      "grad_norm_var": 5.249739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.054,
+      "loss/crossentropy": 2.0020218193531036,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17012976426631213,
+      "step": 17250
+    },
+    {
+      "epoch": 0.14383333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 6.011458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1973,
+      "loss/crossentropy": 1.9204036131501199,
+      "loss/hidden": 3.40546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909493926912546,
+      "step": 17260
+    },
+    {
+      "epoch": 0.14391666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.3384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1264,
+      "loss/crossentropy": 1.865542894601822,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17411553859710693,
+      "step": 17270
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.6145833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1145,
+      "loss/crossentropy": 1.9514264270663262,
+      "loss/hidden": 3.433984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1903881285339594,
+      "step": 17280
+    },
+    {
+      "epoch": 0.14408333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.161393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.218,
+      "loss/crossentropy": 2.0100998997688295,
+      "loss/hidden": 3.278125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1798895327374339,
+      "step": 17290
+    },
+    {
+      "epoch": 0.14416666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 13151.9978515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2477,
+      "loss/crossentropy": 1.8427374526858329,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953713685274125,
+      "step": 17300
+    },
+    {
+      "epoch": 0.14425,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.9145833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1257,
+      "loss/crossentropy": 1.9385129079222678,
+      "loss/hidden": 3.310546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17996535655111073,
+      "step": 17310
+    },
+    {
+      "epoch": 0.14433333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.1372395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1302,
+      "loss/crossentropy": 1.828189279139042,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1745473574846983,
+      "step": 17320
+    },
+    {
+      "epoch": 0.14441666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.0729166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1226,
+      "loss/crossentropy": 1.9782569497823714,
+      "loss/hidden": 3.41171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19845657665282487,
+      "step": 17330
+    },
+    {
+      "epoch": 0.1445,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.7837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0521,
+      "loss/crossentropy": 1.8818863868713378,
+      "loss/hidden": 3.358984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18264735229313372,
+      "step": 17340
+    },
+    {
+      "epoch": 0.14458333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 10.716666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1386,
+      "loss/crossentropy": 1.9800301864743233,
+      "loss/hidden": 3.30546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17297747079283,
+      "step": 17350
+    },
+    {
+      "epoch": 0.14466666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 10.617708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2054,
+      "loss/crossentropy": 1.9759259164333343,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18966067116707563,
+      "step": 17360
+    },
+    {
+      "epoch": 0.14475,
+      "grad_norm": 31.125,
+      "grad_norm_var": 9.964917696541754e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2349,
+      "loss/crossentropy": 1.787013690173626,
+      "loss/hidden": 3.38203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18244071789085864,
+      "step": 17370
+    },
+    {
+      "epoch": 0.14483333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 9.964917703654245e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1623,
+      "loss/crossentropy": 2.0726762622594834,
+      "loss/hidden": 3.407421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18806409742683172,
+      "step": 17380
+    },
+    {
+      "epoch": 0.14491666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.8934895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1651,
+      "loss/crossentropy": 1.8631496280431747,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17424067426472903,
+      "step": 17390
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 30.75,
+      "grad_norm_var": 1.6478515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1501,
+      "loss/crossentropy": 1.966734354197979,
+      "loss/hidden": 3.34921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20803162902593614,
+      "step": 17400
+    },
+    {
+      "epoch": 0.14508333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.3968098958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1774,
+      "loss/crossentropy": 1.9725135400891305,
+      "loss/hidden": 3.477734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18822983466088772,
+      "step": 17410
+    },
+    {
+      "epoch": 0.14516666666666667,
+      "grad_norm": 1845493760.0,
+      "grad_norm_var": 2.1286544379006166e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2558,
+      "loss/crossentropy": 2.095953916758299,
+      "loss/hidden": 3.591796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19755497556179763,
+      "step": 17420
+    },
+    {
+      "epoch": 0.14525,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.1286544105642525e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2601,
+      "loss/crossentropy": 1.991393756866455,
+      "loss/hidden": 3.35234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20299724359065294,
+      "step": 17430
+    },
+    {
+      "epoch": 0.14533333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 25.880989583333335,
+      "learning_rate": 0.0001,
+      "loss": 7.0827,
+      "loss/crossentropy": 1.8140402466058732,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18777146693319083,
+      "step": 17440
+    },
+    {
+      "epoch": 0.14541666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 7.7150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1715,
+      "loss/crossentropy": 1.8402201429009437,
+      "loss/hidden": 3.5078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1878326991572976,
+      "step": 17450
+    },
+    {
+      "epoch": 0.1455,
+      "grad_norm": 29.625,
+      "grad_norm_var": 7.626041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0576,
+      "loss/crossentropy": 1.7638488478958607,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15995034966617822,
+      "step": 17460
+    },
+    {
+      "epoch": 0.14558333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.8872395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1051,
+      "loss/crossentropy": 1.9325864657759666,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19336935002356767,
+      "step": 17470
+    },
+    {
+      "epoch": 0.14566666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3.4650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0814,
+      "loss/crossentropy": 2.065740743279457,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18579264339059592,
+      "step": 17480
+    },
+    {
+      "epoch": 0.14575,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.8447916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2145,
+      "loss/crossentropy": 1.8700942993164062,
+      "loss/hidden": 3.476953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20072167459875345,
+      "step": 17490
+    },
+    {
+      "epoch": 0.14583333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 22.821875,
+      "learning_rate": 0.0001,
+      "loss": 7.2092,
+      "loss/crossentropy": 1.9041163980960847,
+      "loss/hidden": 3.44765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19332583406940101,
+      "step": 17500
+    },
+    {
+      "epoch": 0.14591666666666667,
+      "grad_norm": 45.75,
+      "grad_norm_var": 37.670247395833336,
+      "learning_rate": 0.0001,
+      "loss": 7.0822,
+      "loss/crossentropy": 1.9003212735056878,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18156563993543387,
+      "step": 17510
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 31.5,
+      "grad_norm_var": 18.989518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1233,
+      "loss/crossentropy": 1.9977738589048386,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18585671707987786,
+      "step": 17520
+    },
+    {
+      "epoch": 0.14608333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 27.2306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.256,
+      "loss/crossentropy": 2.00472939312458,
+      "loss/hidden": 3.4859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20386974159628152,
+      "step": 17530
+    },
+    {
+      "epoch": 0.14616666666666667,
+      "grad_norm": 36.25,
+      "grad_norm_var": 35.21848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2672,
+      "loss/crossentropy": 2.014525604248047,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855944911018014,
+      "step": 17540
+    },
+    {
+      "epoch": 0.14625,
+      "grad_norm": 31.5,
+      "grad_norm_var": 17.86015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1351,
+      "loss/crossentropy": 2.005857099592686,
+      "loss/hidden": 3.2546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17380460985004903,
+      "step": 17550
+    },
+    {
+      "epoch": 0.14633333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 15.252018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2025,
+      "loss/crossentropy": 1.9624576315283775,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17943943087011577,
+      "step": 17560
+    },
+    {
+      "epoch": 0.14641666666666667,
+      "grad_norm": 40.0,
+      "grad_norm_var": 14.9572265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2244,
+      "loss/crossentropy": 1.9506747543811798,
+      "loss/hidden": 3.3234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802437637001276,
+      "step": 17570
+    },
+    {
+      "epoch": 0.1465,
+      "grad_norm": 29.625,
+      "grad_norm_var": 239.52395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1606,
+      "loss/crossentropy": 1.9469669878482818,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19329970814287661,
+      "step": 17580
+    },
+    {
+      "epoch": 0.14658333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 78.3775390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1506,
+      "loss/crossentropy": 1.8958590522408485,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21433875244110823,
+      "step": 17590
+    },
+    {
+      "epoch": 0.14666666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.085416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1012,
+      "loss/crossentropy": 1.9255237102508544,
+      "loss/hidden": 3.408984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1944601196795702,
+      "step": 17600
+    },
+    {
+      "epoch": 0.14675,
+      "grad_norm": 30.375,
+      "grad_norm_var": 17.66015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1638,
+      "loss/crossentropy": 1.9022264942526816,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19645339213311672,
+      "step": 17610
+    },
+    {
+      "epoch": 0.14683333333333334,
+      "grad_norm": 42.25,
+      "grad_norm_var": 26.916666666666668,
+      "learning_rate": 0.0001,
+      "loss": 7.1554,
+      "loss/crossentropy": 1.914298328012228,
+      "loss/hidden": 3.556640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20535127203911543,
+      "step": 17620
+    },
+    {
+      "epoch": 0.14691666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 19.845572916666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1462,
+      "loss/crossentropy": 2.0465259671211244,
+      "loss/hidden": 3.415625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1973290266469121,
+      "step": 17630
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 33.5,
+      "grad_norm_var": 10.6837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1758,
+      "loss/crossentropy": 2.008231112360954,
+      "loss/hidden": 3.333984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18332547824829817,
+      "step": 17640
+    },
+    {
+      "epoch": 0.14708333333333334,
+      "grad_norm": 36.75,
+      "grad_norm_var": 5.793489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1968,
+      "loss/crossentropy": 1.9246063113212586,
+      "loss/hidden": 3.46328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22379896901547908,
+      "step": 17650
+    },
+    {
+      "epoch": 0.14716666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 12.828580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1939,
+      "loss/crossentropy": 1.9574020236730576,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18317737132310868,
+      "step": 17660
+    },
+    {
+      "epoch": 0.14725,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.798893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1082,
+      "loss/crossentropy": 1.9085025578737258,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1798366179689765,
+      "step": 17670
+    },
+    {
+      "epoch": 0.14733333333333334,
+      "grad_norm": 34.25,
+      "grad_norm_var": 21.945833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2413,
+      "loss/crossentropy": 1.9746336936950684,
+      "loss/hidden": 3.255078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17803266569972037,
+      "step": 17680
+    },
+    {
+      "epoch": 0.14741666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 22.093489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0406,
+      "loss/crossentropy": 1.8079740449786186,
+      "loss/hidden": 3.34296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16634609289467334,
+      "step": 17690
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 29.75,
+      "grad_norm_var": 61.49733072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.1336,
+      "loss/crossentropy": 1.906479087471962,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865730356425047,
+      "step": 17700
+    },
+    {
+      "epoch": 0.14758333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 7.256184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.101,
+      "loss/crossentropy": 1.9061211571097374,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17805896177887917,
+      "step": 17710
+    },
+    {
+      "epoch": 0.14766666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.011194234879258e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2017,
+      "loss/crossentropy": 2.0191638290882112,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18407542910426855,
+      "step": 17720
+    },
+    {
+      "epoch": 0.14775,
+      "grad_norm": 35.0,
+      "grad_norm_var": 6.867643229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1837,
+      "loss/crossentropy": 1.8662656158208848,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18083018343895674,
+      "step": 17730
+    },
+    {
+      "epoch": 0.14783333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.495833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.216,
+      "loss/crossentropy": 1.8949942022562027,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17319657057523727,
+      "step": 17740
+    },
+    {
+      "epoch": 0.14791666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.7291015625,
+      "learning_rate": 0.0001,
+      "loss": 7.154,
+      "loss/crossentropy": 1.9429595813155174,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18627694062888622,
+      "step": 17750
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.8125,
+      "learning_rate": 0.0001,
+      "loss": 7.1615,
+      "loss/crossentropy": 1.8885102063417434,
+      "loss/hidden": 3.373828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1861933596432209,
+      "step": 17760
+    },
+    {
+      "epoch": 0.14808333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.402083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1382,
+      "loss/crossentropy": 1.902276399731636,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18343093506991864,
+      "step": 17770
+    },
+    {
+      "epoch": 0.14816666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.8268229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1709,
+      "loss/crossentropy": 2.073600899428129,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18779241442680358,
+      "step": 17780
+    },
+    {
+      "epoch": 0.14825,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.6343098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1528,
+      "loss/crossentropy": 1.9148548379540444,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1788425948470831,
+      "step": 17790
+    },
+    {
+      "epoch": 0.14833333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.43515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3325,
+      "loss/crossentropy": 1.9912944473326206,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1773607784882188,
+      "step": 17800
+    },
+    {
+      "epoch": 0.14841666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 4.814322916666667,
+      "learning_rate": 0.0001,
+      "loss": 6.9437,
+      "loss/crossentropy": 1.8436437964439392,
+      "loss/hidden": 3.301953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17089982144534588,
+      "step": 17810
+    },
+    {
+      "epoch": 0.1485,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.025,
+      "learning_rate": 0.0001,
+      "loss": 7.1493,
+      "loss/crossentropy": 1.9381831735372543,
+      "loss/hidden": 3.290234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18582655768841505,
+      "step": 17820
+    },
+    {
+      "epoch": 0.14858333333333335,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.8739583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2667,
+      "loss/crossentropy": 1.9228506922721862,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18991653826087712,
+      "step": 17830
+    },
+    {
+      "epoch": 0.14866666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.7462890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2378,
+      "loss/crossentropy": 2.031201511621475,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20182929262518884,
+      "step": 17840
+    },
+    {
+      "epoch": 0.14875,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.7427083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.223,
+      "loss/crossentropy": 1.9969941288232804,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18042377084493638,
+      "step": 17850
+    },
+    {
+      "epoch": 0.14883333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.024739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1729,
+      "loss/crossentropy": 1.7874360233545303,
+      "loss/hidden": 3.4359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18858081717044115,
+      "step": 17860
+    },
+    {
+      "epoch": 0.14891666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2135416666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2091,
+      "loss/crossentropy": 1.9868085607886314,
+      "loss/hidden": 3.244921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1737012831494212,
+      "step": 17870
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.47265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0867,
+      "loss/crossentropy": 1.8363221064209938,
+      "loss/hidden": 3.28984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886484021320939,
+      "step": 17880
+    },
+    {
+      "epoch": 0.14908333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.915559895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1471,
+      "loss/crossentropy": 1.94259662181139,
+      "loss/hidden": 3.42890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18277982641011475,
+      "step": 17890
+    },
+    {
+      "epoch": 0.14916666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.874739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0762,
+      "loss/crossentropy": 1.7676879599690438,
+      "loss/hidden": 3.407421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1791672594845295,
+      "step": 17900
+    },
+    {
+      "epoch": 0.14925,
+      "grad_norm": 31.625,
+      "grad_norm_var": 8.827055113802614e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.307,
+      "loss/crossentropy": 2.068084768950939,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18616199484094978,
+      "step": 17910
+    },
+    {
+      "epoch": 0.14933333333333335,
+      "grad_norm": 31.0,
+      "grad_norm_var": 17.724934895833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1507,
+      "loss/crossentropy": 1.9794188559055328,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19609470050781966,
+      "step": 17920
+    },
+    {
+      "epoch": 0.14941666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.0330729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1232,
+      "loss/crossentropy": 1.9650203216820956,
+      "loss/hidden": 3.291015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1673107784241438,
+      "step": 17930
+    },
+    {
+      "epoch": 0.1495,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.758214059255224e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3226,
+      "loss/crossentropy": 1.9549292869865895,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1828561224974692,
+      "step": 17940
+    },
+    {
+      "epoch": 0.14958333333333335,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.758214058570154e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1397,
+      "loss/crossentropy": 1.9911674067378045,
+      "loss/hidden": 3.36875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18351277131587268,
+      "step": 17950
+    },
+    {
+      "epoch": 0.14966666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.5389973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1527,
+      "loss/crossentropy": 1.9845456242561341,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17963197641074657,
+      "step": 17960
+    },
+    {
+      "epoch": 0.14975,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.35390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1345,
+      "loss/crossentropy": 2.0877785712480543,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1777985269203782,
+      "step": 17970
+    },
+    {
+      "epoch": 0.14983333333333335,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.81640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0952,
+      "loss/crossentropy": 1.843693608045578,
+      "loss/hidden": 3.315625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17312157526612282,
+      "step": 17980
+    },
+    {
+      "epoch": 0.14991666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5697265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1143,
+      "loss/crossentropy": 1.961574801802635,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18399861454963684,
+      "step": 17990
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.767708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2438,
+      "loss/crossentropy": 2.0191131353378298,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19609930831938982,
+      "step": 18000
+    },
+    {
+      "epoch": 0.15008333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.99375,
+      "learning_rate": 0.0001,
+      "loss": 7.1917,
+      "loss/crossentropy": 1.9451420933008194,
+      "loss/hidden": 3.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18419922776520253,
+      "step": 18010
+    },
+    {
+      "epoch": 0.15016666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.305989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2295,
+      "loss/crossentropy": 2.0177721232175827,
+      "loss/hidden": 3.272265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17394812162965537,
+      "step": 18020
+    },
+    {
+      "epoch": 0.15025,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.678580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2477,
+      "loss/crossentropy": 1.8986812591552735,
+      "loss/hidden": 3.501953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18476867936551572,
+      "step": 18030
+    },
+    {
+      "epoch": 0.15033333333333335,
+      "grad_norm": 29.125,
+      "grad_norm_var": 17.316080729166668,
+      "learning_rate": 0.0001,
+      "loss": 7.1183,
+      "loss/crossentropy": 1.8986513808369636,
+      "loss/hidden": 3.31171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17607410848140717,
+      "step": 18040
+    },
+    {
+      "epoch": 0.15041666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.488997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2586,
+      "loss/crossentropy": 2.0480462282896044,
+      "loss/hidden": 3.45,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1936732977628708,
+      "step": 18050
+    },
+    {
+      "epoch": 0.1505,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1877,
+      "loss/crossentropy": 2.00994084328413,
+      "loss/hidden": 3.358984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19193823914974928,
+      "step": 18060
+    },
+    {
+      "epoch": 0.15058333333333335,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.4275390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1406,
+      "loss/crossentropy": 2.051914116740227,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18952101226896048,
+      "step": 18070
+    },
+    {
+      "epoch": 0.15066666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.3427083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1944,
+      "loss/crossentropy": 1.9460471779108048,
+      "loss/hidden": 3.441015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19146472681313753,
+      "step": 18080
+    },
+    {
+      "epoch": 0.15075,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.22890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1124,
+      "loss/crossentropy": 1.93141558021307,
+      "loss/hidden": 3.351171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17704984191805123,
+      "step": 18090
+    },
+    {
+      "epoch": 0.15083333333333335,
+      "grad_norm": 39.5,
+      "grad_norm_var": 6.823958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1793,
+      "loss/crossentropy": 1.8866180911660195,
+      "loss/hidden": 3.412109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1794743249192834,
+      "step": 18100
+    },
+    {
+      "epoch": 0.15091666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.896840472489101e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2938,
+      "loss/crossentropy": 2.0666448324918747,
+      "loss/hidden": 3.516015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1951539421454072,
+      "step": 18110
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 31.875,
+      "grad_norm_var": 43.15182291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3008,
+      "loss/crossentropy": 1.9401115417480468,
+      "loss/hidden": 3.292578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1967674797400832,
+      "step": 18120
+    },
+    {
+      "epoch": 0.15108333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.4580729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1148,
+      "loss/crossentropy": 1.9141343221068383,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1805819789879024,
+      "step": 18130
+    },
+    {
+      "epoch": 0.15116666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.8416666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1355,
+      "loss/crossentropy": 1.9958683744072914,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1887912791222334,
+      "step": 18140
+    },
+    {
+      "epoch": 0.15125,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.7330729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1369,
+      "loss/crossentropy": 1.8992652669548988,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18042057007551193,
+      "step": 18150
+    },
+    {
+      "epoch": 0.15133333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.2379557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1323,
+      "loss/crossentropy": 2.0010840713977815,
+      "loss/hidden": 3.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19548068027943372,
+      "step": 18160
+    },
+    {
+      "epoch": 0.15141666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.2978515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0299,
+      "loss/crossentropy": 1.9053284496068954,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19124095179140568,
+      "step": 18170
+    },
+    {
+      "epoch": 0.1515,
+      "grad_norm": 30.125,
+      "grad_norm_var": 10.84140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1916,
+      "loss/crossentropy": 2.0646612286567687,
+      "loss/hidden": 3.349609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18376285918056964,
+      "step": 18180
+    },
+    {
+      "epoch": 0.15158333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 9.955208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2234,
+      "loss/crossentropy": 1.7416272193193436,
+      "loss/hidden": 3.521484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19293987099081278,
+      "step": 18190
+    },
+    {
+      "epoch": 0.15166666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.207747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1507,
+      "loss/crossentropy": 1.9817813321948052,
+      "loss/hidden": 3.371484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18243551589548587,
+      "step": 18200
+    },
+    {
+      "epoch": 0.15175,
+      "grad_norm": 36.5,
+      "grad_norm_var": 5.3244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1802,
+      "loss/crossentropy": 1.8642112880945205,
+      "loss/hidden": 3.436328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844145068898797,
+      "step": 18210
+    },
+    {
+      "epoch": 0.15183333333333332,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.6416015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1226,
+      "loss/crossentropy": 1.8745998591184616,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19363704677671195,
+      "step": 18220
+    },
+    {
+      "epoch": 0.15191666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.2997395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1281,
+      "loss/crossentropy": 1.8652560412883759,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18754334505647421,
+      "step": 18230
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.728580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0264,
+      "loss/crossentropy": 1.8875459350645543,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17601182907819748,
+      "step": 18240
+    },
+    {
+      "epoch": 0.15208333333333332,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.8291015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1833,
+      "loss/crossentropy": 1.8843158677220344,
+      "loss/hidden": 3.425390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18848291654139757,
+      "step": 18250
+    },
+    {
+      "epoch": 0.15216666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 5.529622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.234,
+      "loss/crossentropy": 1.9466109693050384,
+      "loss/hidden": 3.4421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18400811534374953,
+      "step": 18260
+    },
+    {
+      "epoch": 0.15225,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.528059895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.207,
+      "loss/crossentropy": 1.9581806391477585,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19348377268761396,
+      "step": 18270
+    },
+    {
+      "epoch": 0.15233333333333332,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.707291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.104,
+      "loss/crossentropy": 1.9371780931949616,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918162316083908,
+      "step": 18280
+    },
+    {
+      "epoch": 0.15241666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 24.672916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1675,
+      "loss/crossentropy": 1.9870081901550294,
+      "loss/hidden": 3.425,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18982220739126204,
+      "step": 18290
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.3041666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1735,
+      "loss/crossentropy": 2.003659498691559,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18024815227836372,
+      "step": 18300
+    },
+    {
+      "epoch": 0.15258333333333332,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.9124348958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1513,
+      "loss/crossentropy": 1.7380959689617157,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18031938448548318,
+      "step": 18310
+    },
+    {
+      "epoch": 0.15266666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.2718098958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2646,
+      "loss/crossentropy": 1.9350053653120995,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.175525313988328,
+      "step": 18320
+    },
+    {
+      "epoch": 0.15275,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.2931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1911,
+      "loss/crossentropy": 1.9946703404188155,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1877204915508628,
+      "step": 18330
+    },
+    {
+      "epoch": 0.15283333333333332,
+      "grad_norm": 36.75,
+      "grad_norm_var": 4.20390625,
+      "learning_rate": 0.0001,
+      "loss": 7.157,
+      "loss/crossentropy": 1.8865586653351785,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18652425203472375,
+      "step": 18340
+    },
+    {
+      "epoch": 0.15291666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.653580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 2.0678406953811646,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19146652407944204,
+      "step": 18350
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.4999801311166902e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2295,
+      "loss/crossentropy": 1.8809891670942307,
+      "loss/hidden": 3.32578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18959175925701857,
+      "step": 18360
+    },
+    {
+      "epoch": 0.15308333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.4999801314432865e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1526,
+      "loss/crossentropy": 2.0591250747442245,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18697906974703074,
+      "step": 18370
+    },
+    {
+      "epoch": 0.15316666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.019205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.129,
+      "loss/crossentropy": 1.8456390529870987,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780494086444378,
+      "step": 18380
+    },
+    {
+      "epoch": 0.15325,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.0819897936463965e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2584,
+      "loss/crossentropy": 1.9993614748120307,
+      "loss/hidden": 3.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1919097865000367,
+      "step": 18390
+    },
+    {
+      "epoch": 0.15333333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.147916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1617,
+      "loss/crossentropy": 1.9496661603450776,
+      "loss/hidden": 3.433203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1817046355456114,
+      "step": 18400
+    },
+    {
+      "epoch": 0.15341666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.731184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1401,
+      "loss/crossentropy": 1.9402483448386192,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18009484149515628,
+      "step": 18410
+    },
+    {
+      "epoch": 0.1535,
+      "grad_norm": 32.0,
+      "grad_norm_var": 37.67076822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2456,
+      "loss/crossentropy": 1.9319218501448632,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17397410683333875,
+      "step": 18420
+    },
+    {
+      "epoch": 0.15358333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.918684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2419,
+      "loss/crossentropy": 1.9003607004880905,
+      "loss/hidden": 3.47578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19959020651876927,
+      "step": 18430
+    },
+    {
+      "epoch": 0.15366666666666667,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.362239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.165,
+      "loss/crossentropy": 1.9196560755372047,
+      "loss/hidden": 3.576953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2123214827850461,
+      "step": 18440
+    },
+    {
+      "epoch": 0.15375,
+      "grad_norm": 29.125,
+      "grad_norm_var": 13.280989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2098,
+      "loss/crossentropy": 2.019237519800663,
+      "loss/hidden": 3.323828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18751068767160178,
+      "step": 18450
+    },
+    {
+      "epoch": 0.15383333333333332,
+      "grad_norm": 38.5,
+      "grad_norm_var": 109.59576822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.0833,
+      "loss/crossentropy": 1.8818476274609566,
+      "loss/hidden": 3.338671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17979788575321437,
+      "step": 18460
+    },
+    {
+      "epoch": 0.15391666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 110.88932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1183,
+      "loss/crossentropy": 1.955721442401409,
+      "loss/hidden": 3.41640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.205044118873775,
+      "step": 18470
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.471809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1223,
+      "loss/crossentropy": 2.0897390320897102,
+      "loss/hidden": 3.423046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1786660948768258,
+      "step": 18480
+    },
+    {
+      "epoch": 0.15408333333333332,
+      "grad_norm": 30.375,
+      "grad_norm_var": 7.439322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1016,
+      "loss/crossentropy": 1.9049207262694836,
+      "loss/hidden": 3.273046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.178234576433897,
+      "step": 18490
+    },
+    {
+      "epoch": 0.15416666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.9056411222964096e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1234,
+      "loss/crossentropy": 1.926728616654873,
+      "loss/hidden": 3.370703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19228889402002097,
+      "step": 18500
+    },
+    {
+      "epoch": 0.15425,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.5728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1798,
+      "loss/crossentropy": 1.919435414671898,
+      "loss/hidden": 3.54609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2025587311014533,
+      "step": 18510
+    },
+    {
+      "epoch": 0.15433333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.5749348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1353,
+      "loss/crossentropy": 2.029505896568298,
+      "loss/hidden": 3.2453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18726584110409022,
+      "step": 18520
+    },
+    {
+      "epoch": 0.15441666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.2739583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.088,
+      "loss/crossentropy": 1.9236000820994377,
+      "loss/hidden": 3.3015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17061380632221698,
+      "step": 18530
+    },
+    {
+      "epoch": 0.1545,
+      "grad_norm": 29.375,
+      "grad_norm_var": 28.944791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0934,
+      "loss/crossentropy": 1.9393044471740724,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17598522771149874,
+      "step": 18540
+    },
+    {
+      "epoch": 0.15458333333333332,
+      "grad_norm": 30.875,
+      "grad_norm_var": 30.981184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.13,
+      "loss/crossentropy": 1.9169656589627266,
+      "loss/hidden": 3.40078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1925279337912798,
+      "step": 18550
+    },
+    {
+      "epoch": 0.15466666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 44.264322916666664,
+      "learning_rate": 0.0001,
+      "loss": 7.3271,
+      "loss/crossentropy": 2.0036231935024262,
+      "loss/hidden": 3.4765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2271136650815606,
+      "step": 18560
+    },
+    {
+      "epoch": 0.15475,
+      "grad_norm": 31.625,
+      "grad_norm_var": 9.6025390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0954,
+      "loss/crossentropy": 1.8454778000712395,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17759868446737528,
+      "step": 18570
+    },
+    {
+      "epoch": 0.15483333333333332,
+      "grad_norm": 32.75,
+      "grad_norm_var": 4.664583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1809,
+      "loss/crossentropy": 1.9348541289567946,
+      "loss/hidden": 3.442578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18641308657824993,
+      "step": 18580
+    },
+    {
+      "epoch": 0.15491666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.56875,
+      "learning_rate": 0.0001,
+      "loss": 7.1607,
+      "loss/crossentropy": 1.7871997088193894,
+      "loss/hidden": 3.397265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18637465024366975,
+      "step": 18590
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.346875,
+      "learning_rate": 0.0001,
+      "loss": 7.2942,
+      "loss/crossentropy": 1.861001893877983,
+      "loss/hidden": 3.44453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21215323433279992,
+      "step": 18600
+    },
+    {
+      "epoch": 0.15508333333333332,
+      "grad_norm": 30.875,
+      "grad_norm_var": 9.088997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0886,
+      "loss/crossentropy": 1.957412102073431,
+      "loss/hidden": 3.236328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16389200901612638,
+      "step": 18610
+    },
+    {
+      "epoch": 0.15516666666666667,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.0143229166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1151,
+      "loss/crossentropy": 2.0088422290980814,
+      "loss/hidden": 3.265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17890582252293824,
+      "step": 18620
+    },
+    {
+      "epoch": 0.15525,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.590625,
+      "learning_rate": 0.0001,
+      "loss": 7.0701,
+      "loss/crossentropy": 2.0099326848983763,
+      "loss/hidden": 3.319921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17842565421015025,
+      "step": 18630
+    },
+    {
+      "epoch": 0.15533333333333332,
+      "grad_norm": 35.0,
+      "grad_norm_var": 28.328125,
+      "learning_rate": 0.0001,
+      "loss": 7.124,
+      "loss/crossentropy": 1.7827160835266114,
+      "loss/hidden": 3.393359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18283120319247245,
+      "step": 18640
+    },
+    {
+      "epoch": 0.15541666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.6520182291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0117,
+      "loss/crossentropy": 2.010917329788208,
+      "loss/hidden": 3.327734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17667978219687938,
+      "step": 18650
+    },
+    {
+      "epoch": 0.1555,
+      "grad_norm": 38.75,
+      "grad_norm_var": 8517.058333333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2355,
+      "loss/crossentropy": 2.025261604785919,
+      "loss/hidden": 3.315234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18550921231508255,
+      "step": 18660
+    },
+    {
+      "epoch": 0.15558333333333332,
+      "grad_norm": 30.75,
+      "grad_norm_var": 8463.995572916667,
+      "learning_rate": 0.0001,
+      "loss": 7.2151,
+      "loss/crossentropy": 2.06703442633152,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19284330494701862,
+      "step": 18670
+    },
+    {
+      "epoch": 0.15566666666666668,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.0759765625,
+      "learning_rate": 0.0001,
+      "loss": 6.9788,
+      "loss/crossentropy": 1.807583911716938,
+      "loss/hidden": 3.33125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17016871124505997,
+      "step": 18680
+    },
+    {
+      "epoch": 0.15575,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.38125,
+      "learning_rate": 0.0001,
+      "loss": 7.087,
+      "loss/crossentropy": 1.9083680778741836,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1921663861721754,
+      "step": 18690
+    },
+    {
+      "epoch": 0.15583333333333332,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.4759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1001,
+      "loss/crossentropy": 1.911944080889225,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17480471022427083,
+      "step": 18700
+    },
+    {
+      "epoch": 0.15591666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.2635411526750395e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1885,
+      "loss/crossentropy": 1.9704201996326447,
+      "loss/hidden": 3.4328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19554522894322873,
+      "step": 18710
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.2635411521598392e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1698,
+      "loss/crossentropy": 1.9283909045159817,
+      "loss/hidden": 3.371875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1946629300713539,
+      "step": 18720
+    },
+    {
+      "epoch": 0.15608333333333332,
+      "grad_norm": 2181038080.0,
+      "grad_norm_var": 2.97307935669573e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2524,
+      "loss/crossentropy": 1.897528025507927,
+      "loss/hidden": 3.387109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1786580251529813,
+      "step": 18730
+    },
+    {
+      "epoch": 0.15616666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.973079352151901e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2131,
+      "loss/crossentropy": 1.8891230538487433,
+      "loss/hidden": 3.42109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1921351669356227,
+      "step": 18740
+    },
+    {
+      "epoch": 0.15625,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.58125,
+      "learning_rate": 0.0001,
+      "loss": 7.1114,
+      "loss/crossentropy": 2.071422126889229,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17793976347893475,
+      "step": 18750
+    },
+    {
+      "epoch": 0.15633333333333332,
+      "grad_norm": 54.25,
+      "grad_norm_var": 37.56432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2142,
+      "loss/crossentropy": 2.0677654266357424,
+      "loss/hidden": 3.41328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2160305231809616,
+      "step": 18760
+    },
+    {
+      "epoch": 0.15641666666666668,
+      "grad_norm": 33.75,
+      "grad_norm_var": 53.588541666666664,
+      "learning_rate": 0.0001,
+      "loss": 7.2731,
+      "loss/crossentropy": 1.9512261405587197,
+      "loss/hidden": 3.51171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20266925655305384,
+      "step": 18770
+    },
+    {
+      "epoch": 0.1565,
+      "grad_norm": 37.25,
+      "grad_norm_var": 44.020572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1831,
+      "loss/crossentropy": 2.075438083708286,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17665863800793885,
+      "step": 18780
+    },
+    {
+      "epoch": 0.15658333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 25.195572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1971,
+      "loss/crossentropy": 1.9097235754132271,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18935867827385663,
+      "step": 18790
+    },
+    {
+      "epoch": 0.15666666666666668,
+      "grad_norm": 201.0,
+      "grad_norm_var": 1782.6150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.225,
+      "loss/crossentropy": 1.9649134397506713,
+      "loss/hidden": 3.426953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19300914984196424,
+      "step": 18800
+    },
+    {
+      "epoch": 0.15675,
+      "grad_norm": 38.25,
+      "grad_norm_var": 1751.4410807291667,
+      "learning_rate": 0.0001,
+      "loss": 7.2445,
+      "loss/crossentropy": 1.9446261554956437,
+      "loss/hidden": 3.497265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2123917168006301,
+      "step": 18810
+    },
+    {
+      "epoch": 0.15683333333333332,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.1439318797079021e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.0584,
+      "loss/crossentropy": 1.9147826530039311,
+      "loss/hidden": 3.482421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.199047539383173,
+      "step": 18820
+    },
+    {
+      "epoch": 0.15691666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.1439318796633376e+18,
+      "learning_rate": 0.0001,
+      "loss": 6.9888,
+      "loss/crossentropy": 1.9809871926903724,
+      "loss/hidden": 3.291015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1731796983629465,
+      "step": 18830
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 34.75,
+      "grad_norm_var": 12.208072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1636,
+      "loss/crossentropy": 2.0335353165864944,
+      "loss/hidden": 3.283203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1914576642215252,
+      "step": 18840
+    },
+    {
+      "epoch": 0.15708333333333332,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.5970024021887555e+17,
+      "learning_rate": 0.0001,
+      "loss": 6.9943,
+      "loss/crossentropy": 1.9717642769217492,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1991662908345461,
+      "step": 18850
+    },
+    {
+      "epoch": 0.15716666666666668,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.5970024012544746e+17,
+      "learning_rate": 0.0001,
+      "loss": 6.9865,
+      "loss/crossentropy": 1.9509471356868744,
+      "loss/hidden": 3.251953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1842729203402996,
+      "step": 18860
+    },
+    {
+      "epoch": 0.15725,
+      "grad_norm": 31.375,
+      "grad_norm_var": 9.492122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0098,
+      "loss/crossentropy": 1.9393299847841263,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1816339949145913,
+      "step": 18870
+    },
+    {
+      "epoch": 0.15733333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 8.276822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0566,
+      "loss/crossentropy": 1.9912663742899894,
+      "loss/hidden": 3.279296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18120460938662292,
+      "step": 18880
+    },
+    {
+      "epoch": 0.15741666666666668,
+      "grad_norm": 37.0,
+      "grad_norm_var": 5.810872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1196,
+      "loss/crossentropy": 1.8049977391958236,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1642103621736169,
+      "step": 18890
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 31.625,
+      "grad_norm_var": 7.890559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0605,
+      "loss/crossentropy": 1.935225610435009,
+      "loss/hidden": 3.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18383098654448987,
+      "step": 18900
+    },
+    {
+      "epoch": 0.15758333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.726832732101936e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1589,
+      "loss/crossentropy": 1.9154966719448567,
+      "loss/hidden": 3.303515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18004091046750545,
+      "step": 18910
+    },
+    {
+      "epoch": 0.15766666666666668,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.72683273120986e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0353,
+      "loss/crossentropy": 1.8733932368457318,
+      "loss/hidden": 3.221875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1733988095074892,
+      "step": 18920
+    },
+    {
+      "epoch": 0.15775,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.941080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0849,
+      "loss/crossentropy": 2.009824226796627,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19008228778839112,
+      "step": 18930
+    },
+    {
+      "epoch": 0.15783333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 8.831705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1377,
+      "loss/crossentropy": 1.989619529247284,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19383876267820596,
+      "step": 18940
+    },
+    {
+      "epoch": 0.15791666666666668,
+      "grad_norm": 31.5,
+      "grad_norm_var": 19.837434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0665,
+      "loss/crossentropy": 1.9502414911985397,
+      "loss/hidden": 3.361328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18037104681134225,
+      "step": 18950
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 36.0,
+      "grad_norm_var": 19.893489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0697,
+      "loss/crossentropy": 2.1159609287977217,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17629108000546695,
+      "step": 18960
+    },
+    {
+      "epoch": 0.15808333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 103.2212890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1771,
+      "loss/crossentropy": 2.060988609492779,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20584311038255693,
+      "step": 18970
+    },
+    {
+      "epoch": 0.15816666666666668,
+      "grad_norm": 28.125,
+      "grad_norm_var": 99.20807291666667,
+      "learning_rate": 0.0001,
+      "loss": 6.9853,
+      "loss/crossentropy": 1.9348157882690429,
+      "loss/hidden": 3.264453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1714519090950489,
+      "step": 18980
+    },
+    {
+      "epoch": 0.15825,
+      "grad_norm": 28.375,
+      "grad_norm_var": 5.849934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0564,
+      "loss/crossentropy": 2.082990960776806,
+      "loss/hidden": 3.2703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17705970890820028,
+      "step": 18990
+    },
+    {
+      "epoch": 0.15833333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.298372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.197,
+      "loss/crossentropy": 1.8553043477237225,
+      "loss/hidden": 3.441015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18200365789234638,
+      "step": 19000
+    },
+    {
+      "epoch": 0.15841666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.729622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0939,
+      "loss/crossentropy": 1.9618657492101192,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18808676647022365,
+      "step": 19010
+    },
+    {
+      "epoch": 0.1585,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.292122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.163,
+      "loss/crossentropy": 1.9934092611074448,
+      "loss/hidden": 3.40859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1916485806927085,
+      "step": 19020
+    },
+    {
+      "epoch": 0.15858333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.093489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1244,
+      "loss/crossentropy": 1.9670961856842042,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937435757368803,
+      "step": 19030
+    },
+    {
+      "epoch": 0.15866666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 5.455208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1548,
+      "loss/crossentropy": 1.9508709952235221,
+      "loss/hidden": 3.276171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17919052559882404,
+      "step": 19040
+    },
+    {
+      "epoch": 0.15875,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.095833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2455,
+      "loss/crossentropy": 1.9306133687496185,
+      "loss/hidden": 3.296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19505799002945423,
+      "step": 19050
+    },
+    {
+      "epoch": 0.15883333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.0559709499153014e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1775,
+      "loss/crossentropy": 1.9510480538010597,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18996794521808624,
+      "step": 19060
+    },
+    {
+      "epoch": 0.15891666666666668,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.306266257719297e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1621,
+      "loss/crossentropy": 1.8554041370749474,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17033051270991564,
+      "step": 19070
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 44.75,
+      "grad_norm_var": 9.306266258724182e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0453,
+      "loss/crossentropy": 1.9016903087496757,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1839282332919538,
+      "step": 19080
+    },
+    {
+      "epoch": 0.15908333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 28.802083333333332,
+      "learning_rate": 0.0001,
+      "loss": 7.1295,
+      "loss/crossentropy": 2.0031721755862235,
+      "loss/hidden": 3.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18534120228141546,
+      "step": 19090
+    },
+    {
+      "epoch": 0.15916666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 11.561393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1754,
+      "loss/crossentropy": 1.8480873089283705,
+      "loss/hidden": 3.40390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18209592271596192,
+      "step": 19100
+    },
+    {
+      "epoch": 0.15925,
+      "grad_norm": 38.0,
+      "grad_norm_var": 12.589322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1554,
+      "loss/crossentropy": 1.907559299468994,
+      "loss/hidden": 3.27265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1801645690575242,
+      "step": 19110
+    },
+    {
+      "epoch": 0.15933333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.322330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.137,
+      "loss/crossentropy": 1.9169094741344452,
+      "loss/hidden": 3.498046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18878476433455943,
+      "step": 19120
+    },
+    {
+      "epoch": 0.15941666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.72890625,
+      "learning_rate": 0.0001,
+      "loss": 7.153,
+      "loss/crossentropy": 1.922089895606041,
+      "loss/hidden": 3.284375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17243139669299126,
+      "step": 19130
+    },
+    {
+      "epoch": 0.1595,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.9457682291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1331,
+      "loss/crossentropy": 1.8787727117538453,
+      "loss/hidden": 3.35390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18439795188605784,
+      "step": 19140
+    },
+    {
+      "epoch": 0.15958333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.346875,
+      "learning_rate": 0.0001,
+      "loss": 7.0518,
+      "loss/crossentropy": 1.9286392159759997,
+      "loss/hidden": 3.261328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17249171026051044,
+      "step": 19150
+    },
+    {
+      "epoch": 0.15966666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.483333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1433,
+      "loss/crossentropy": 1.9940631791949273,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1849735563620925,
+      "step": 19160
+    },
+    {
+      "epoch": 0.15975,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.730208333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1279,
+      "loss/crossentropy": 1.9090391889214515,
+      "loss/hidden": 3.32109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1874424744397402,
+      "step": 19170
+    },
+    {
+      "epoch": 0.15983333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.5518229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0568,
+      "loss/crossentropy": 1.876129651069641,
+      "loss/hidden": 3.184765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16148128025233746,
+      "step": 19180
+    },
+    {
+      "epoch": 0.15991666666666668,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.8379557291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1634,
+      "loss/crossentropy": 1.8751564130187035,
+      "loss/hidden": 3.262109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16824828516691923,
+      "step": 19190
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 38.25,
+      "grad_norm_var": 6.12265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1151,
+      "loss/crossentropy": 2.001160889863968,
+      "loss/hidden": 3.409375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18779610935598612,
+      "step": 19200
+    },
+    {
+      "epoch": 0.16008333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 6.72890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1415,
+      "loss/crossentropy": 1.9250038638710976,
+      "loss/hidden": 3.485546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19871903136372565,
+      "step": 19210
+    },
+    {
+      "epoch": 0.16016666666666668,
+      "grad_norm": 36.25,
+      "grad_norm_var": 5.299739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1255,
+      "loss/crossentropy": 1.8138721562922,
+      "loss/hidden": 3.2765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1769710690714419,
+      "step": 19220
+    },
+    {
+      "epoch": 0.16025,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.2705729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0938,
+      "loss/crossentropy": 1.961549961566925,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19108945187181234,
+      "step": 19230
+    },
+    {
+      "epoch": 0.16033333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.466080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2194,
+      "loss/crossentropy": 1.9689617395401,
+      "loss/hidden": 3.511328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21945166327059268,
+      "step": 19240
+    },
+    {
+      "epoch": 0.16041666666666668,
+      "grad_norm": 38.25,
+      "grad_norm_var": 40.0869140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2296,
+      "loss/crossentropy": 2.0724002063274383,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2176670690998435,
+      "step": 19250
+    },
+    {
+      "epoch": 0.1605,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.2635411499913142e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1193,
+      "loss/crossentropy": 1.8169685006141663,
+      "loss/hidden": 3.338671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17037609480321408,
+      "step": 19260
+    },
+    {
+      "epoch": 0.16058333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 14.370247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1751,
+      "loss/crossentropy": 1.9744362197816372,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19431510157883167,
+      "step": 19270
+    },
+    {
+      "epoch": 0.16066666666666668,
+      "grad_norm": 31.875,
+      "grad_norm_var": 542.7309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1446,
+      "loss/crossentropy": 1.9297940462827683,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2029770040884614,
+      "step": 19280
+    },
+    {
+      "epoch": 0.16075,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.109830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1553,
+      "loss/crossentropy": 1.9910588189959526,
+      "loss/hidden": 3.414453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19653251189738513,
+      "step": 19290
+    },
+    {
+      "epoch": 0.16083333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.0619140625,
+      "learning_rate": 0.0001,
+      "loss": 7.134,
+      "loss/crossentropy": 1.8709509029984475,
+      "loss/hidden": 3.269921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1742756930179894,
+      "step": 19300
+    },
+    {
+      "epoch": 0.16091666666666668,
+      "grad_norm": 27.875,
+      "grad_norm_var": 9.157291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0646,
+      "loss/crossentropy": 1.874299732595682,
+      "loss/hidden": 3.2515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17179571595042945,
+      "step": 19310
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 36.5,
+      "grad_norm_var": 12.7166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.092,
+      "loss/crossentropy": 1.9718119740486144,
+      "loss/hidden": 3.415625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1848680291324854,
+      "step": 19320
+    },
+    {
+      "epoch": 0.16108333333333333,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.727018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1349,
+      "loss/crossentropy": 1.7799494415521622,
+      "loss/hidden": 3.298828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1672179948538542,
+      "step": 19330
+    },
+    {
+      "epoch": 0.16116666666666668,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.6176432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2313,
+      "loss/crossentropy": 1.978345349431038,
+      "loss/hidden": 3.38046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17690238170325756,
+      "step": 19340
+    },
+    {
+      "epoch": 0.16125,
+      "grad_norm": 34.25,
+      "grad_norm_var": 4.27890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1434,
+      "loss/crossentropy": 1.9939823731780053,
+      "loss/hidden": 3.44296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19046493843197823,
+      "step": 19350
+    },
+    {
+      "epoch": 0.16133333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.212239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1586,
+      "loss/crossentropy": 1.9346661597490311,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18260504975914954,
+      "step": 19360
+    },
+    {
+      "epoch": 0.16141666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.6478515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0929,
+      "loss/crossentropy": 1.999790969491005,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1820996723137796,
+      "step": 19370
+    },
+    {
+      "epoch": 0.1615,
+      "grad_norm": 34.5,
+      "grad_norm_var": 19.816080729166668,
+      "learning_rate": 0.0001,
+      "loss": 7.194,
+      "loss/crossentropy": 1.7856683239340783,
+      "loss/hidden": 3.33515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17992732618004084,
+      "step": 19380
+    },
+    {
+      "epoch": 0.16158333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.111881692959737e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2683,
+      "loss/crossentropy": 1.8499102681875228,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17964148223400117,
+      "step": 19390
+    },
+    {
+      "epoch": 0.16166666666666665,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.1118816957489485e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0911,
+      "loss/crossentropy": 1.9728141754865647,
+      "loss/hidden": 3.485546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19145925920456647,
+      "step": 19400
+    },
+    {
+      "epoch": 0.16175,
+      "grad_norm": 28.375,
+      "grad_norm_var": 19.3369140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1253,
+      "loss/crossentropy": 1.8109243243932724,
+      "loss/hidden": 3.34296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18406782802194357,
+      "step": 19410
+    },
+    {
+      "epoch": 0.16183333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 9.468818065597661e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1747,
+      "loss/crossentropy": 1.920046366751194,
+      "loss/hidden": 3.39765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846001248806715,
+      "step": 19420
+    },
+    {
+      "epoch": 0.16191666666666665,
+      "grad_norm": 31.25,
+      "grad_norm_var": 9.468818059232106e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2025,
+      "loss/crossentropy": 1.8375889554619789,
+      "loss/hidden": 3.365234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18694970067590475,
+      "step": 19430
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 29.75,
+      "grad_norm_var": 14.615559895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2081,
+      "loss/crossentropy": 1.93913331925869,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18600110933184624,
+      "step": 19440
+    },
+    {
+      "epoch": 0.16208333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 7.43125,
+      "learning_rate": 0.0001,
+      "loss": 7.1934,
+      "loss/crossentropy": 1.9657758876681328,
+      "loss/hidden": 3.432421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2021039716899395,
+      "step": 19450
+    },
+    {
+      "epoch": 0.16216666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.8497395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1895,
+      "loss/crossentropy": 1.9433988243341447,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2006751226261258,
+      "step": 19460
+    },
+    {
+      "epoch": 0.16225,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.4268229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0751,
+      "loss/crossentropy": 2.0031242236495017,
+      "loss/hidden": 3.28046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18491288553923368,
+      "step": 19470
+    },
+    {
+      "epoch": 0.16233333333333333,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.72265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1263,
+      "loss/crossentropy": 2.046249917149544,
+      "loss/hidden": 3.404296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19146426543593406,
+      "step": 19480
+    },
+    {
+      "epoch": 0.16241666666666665,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.201041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1503,
+      "loss/crossentropy": 2.02630864828825,
+      "loss/hidden": 3.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18987087812274694,
+      "step": 19490
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 31.375,
+      "grad_norm_var": 4.7525390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2131,
+      "loss/crossentropy": 2.0087099388241767,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19217790365219117,
+      "step": 19500
+    },
+    {
+      "epoch": 0.16258333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.217122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1103,
+      "loss/crossentropy": 1.8748227655887604,
+      "loss/hidden": 3.432421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19003468416631222,
+      "step": 19510
+    },
+    {
+      "epoch": 0.16266666666666665,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.8823036756625114e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3116,
+      "loss/crossentropy": 2.0296377152204514,
+      "loss/hidden": 3.305078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18557718340307475,
+      "step": 19520
+    },
+    {
+      "epoch": 0.16275,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.45,
+      "learning_rate": 0.0001,
+      "loss": 7.1314,
+      "loss/crossentropy": 1.9304698213934899,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17999850492924452,
+      "step": 19530
+    },
+    {
+      "epoch": 0.16283333333333333,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.1676432291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.3362,
+      "loss/crossentropy": 2.0195963978767395,
+      "loss/hidden": 3.3625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1860950912348926,
+      "step": 19540
+    },
+    {
+      "epoch": 0.16291666666666665,
+      "grad_norm": 36.5,
+      "grad_norm_var": 10.092708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1396,
+      "loss/crossentropy": 1.9251476585865022,
+      "loss/hidden": 3.483203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21049285661429168,
+      "step": 19550
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.985384779978026e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1538,
+      "loss/crossentropy": 1.9182105913758278,
+      "loss/hidden": 3.254296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16572684291750192,
+      "step": 19560
+    },
+    {
+      "epoch": 0.16308333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.985384789694131e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1545,
+      "loss/crossentropy": 1.8906398624181748,
+      "loss/hidden": 3.450390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18660023249685764,
+      "step": 19570
+    },
+    {
+      "epoch": 0.16316666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 26.758072916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1804,
+      "loss/crossentropy": 1.920269176363945,
+      "loss/hidden": 3.348046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18976410180330278,
+      "step": 19580
+    },
+    {
+      "epoch": 0.16325,
+      "grad_norm": 28.75,
+      "grad_norm_var": 27.570572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0518,
+      "loss/crossentropy": 2.0110680162906647,
+      "loss/hidden": 3.33828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927089313045144,
+      "step": 19590
+    },
+    {
+      "epoch": 0.16333333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.6489583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0142,
+      "loss/crossentropy": 1.9435459971427917,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735059792175889,
+      "step": 19600
+    },
+    {
+      "epoch": 0.16341666666666665,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.4681640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2289,
+      "loss/crossentropy": 1.908098728954792,
+      "loss/hidden": 3.4015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19317488931119442,
+      "step": 19610
+    },
+    {
+      "epoch": 0.1635,
+      "grad_norm": 28.625,
+      "grad_norm_var": 2.79140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2269,
+      "loss/crossentropy": 1.948152408003807,
+      "loss/hidden": 3.381640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18927929010242223,
+      "step": 19620
+    },
+    {
+      "epoch": 0.16358333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.8330729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1724,
+      "loss/crossentropy": 2.040789821743965,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822492267936468,
+      "step": 19630
+    },
+    {
+      "epoch": 0.16366666666666665,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.372330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1894,
+      "loss/crossentropy": 1.8017787858843803,
+      "loss/hidden": 3.45859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17243175562471152,
+      "step": 19640
+    },
+    {
+      "epoch": 0.16375,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.3634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1692,
+      "loss/crossentropy": 1.9264248684048653,
+      "loss/hidden": 3.3484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17540650218725204,
+      "step": 19650
+    },
+    {
+      "epoch": 0.16383333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.7478515625,
+      "learning_rate": 0.0001,
+      "loss": 7.3067,
+      "loss/crossentropy": 1.9077580362558364,
+      "loss/hidden": 3.476953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19736877121031285,
+      "step": 19660
+    },
+    {
+      "epoch": 0.16391666666666665,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.020833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1173,
+      "loss/crossentropy": 2.0139551982283592,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19549070075154304,
+      "step": 19670
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 35.25,
+      "grad_norm_var": 7.895572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1562,
+      "loss/crossentropy": 2.0356906935572625,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1790664467960596,
+      "step": 19680
+    },
+    {
+      "epoch": 0.16408333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 13.9400390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1223,
+      "loss/crossentropy": 1.945811128616333,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1818154737353325,
+      "step": 19690
+    },
+    {
+      "epoch": 0.16416666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.365559895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2644,
+      "loss/crossentropy": 1.9962551683187484,
+      "loss/hidden": 3.331640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18522231467068195,
+      "step": 19700
+    },
+    {
+      "epoch": 0.16425,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.6447916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2116,
+      "loss/crossentropy": 1.9646634325385093,
+      "loss/hidden": 3.553515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20782224107533692,
+      "step": 19710
+    },
+    {
+      "epoch": 0.16433333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.251822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3307,
+      "loss/crossentropy": 2.0151273593306542,
+      "loss/hidden": 3.44375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20999886188656092,
+      "step": 19720
+    },
+    {
+      "epoch": 0.16441666666666666,
+      "grad_norm": 36.75,
+      "grad_norm_var": 1.3015402721170163e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2361,
+      "loss/crossentropy": 2.0667057156562807,
+      "loss/hidden": 3.27109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18990101981908084,
+      "step": 19730
+    },
+    {
+      "epoch": 0.1645,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.3015402724164897e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1281,
+      "loss/crossentropy": 1.8923170864582062,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18683328703045846,
+      "step": 19740
+    },
+    {
+      "epoch": 0.16458333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.8893229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0114,
+      "loss/crossentropy": 1.9253984332084655,
+      "loss/hidden": 3.219140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16697643585503102,
+      "step": 19750
+    },
+    {
+      "epoch": 0.16466666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.881705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2907,
+      "loss/crossentropy": 1.9410972431302072,
+      "loss/hidden": 3.491796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20904589667916298,
+      "step": 19760
+    },
+    {
+      "epoch": 0.16475,
+      "grad_norm": 31.75,
+      "grad_norm_var": 15.8728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2186,
+      "loss/crossentropy": 1.9214755564928054,
+      "loss/hidden": 3.435546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1759912172332406,
+      "step": 19770
+    },
+    {
+      "epoch": 0.16483333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.61875,
+      "learning_rate": 0.0001,
+      "loss": 7.0637,
+      "loss/crossentropy": 1.9233395352959632,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17473995946347715,
+      "step": 19780
+    },
+    {
+      "epoch": 0.16491666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.9916666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1867,
+      "loss/crossentropy": 1.9047813713550568,
+      "loss/hidden": 3.305859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15949498675763607,
+      "step": 19790
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.3760416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0988,
+      "loss/crossentropy": 1.744388183951378,
+      "loss/hidden": 3.4359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18206265531480312,
+      "step": 19800
+    },
+    {
+      "epoch": 0.16508333333333333,
+      "grad_norm": 44.5,
+      "grad_norm_var": 13.893489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2228,
+      "loss/crossentropy": 2.050296738743782,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20759957507252694,
+      "step": 19810
+    },
+    {
+      "epoch": 0.16516666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 14.695247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1679,
+      "loss/crossentropy": 1.9715024009346962,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835065135732293,
+      "step": 19820
+    },
+    {
+      "epoch": 0.16525,
+      "grad_norm": 60.5,
+      "grad_norm_var": 2620.2853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1783,
+      "loss/crossentropy": 1.9325301676988602,
+      "loss/hidden": 3.405078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2006636902689934,
+      "step": 19830
+    },
+    {
+      "epoch": 0.16533333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2581.5372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1592,
+      "loss/crossentropy": 1.9409628361463547,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17196089420467614,
+      "step": 19840
+    },
+    {
+      "epoch": 0.16541666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 4.483072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1818,
+      "loss/crossentropy": 1.8903601258993148,
+      "loss/hidden": 3.305859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17661178326234223,
+      "step": 19850
+    },
+    {
+      "epoch": 0.1655,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.1718098958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0268,
+      "loss/crossentropy": 1.8702323392033577,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17986167576164008,
+      "step": 19860
+    },
+    {
+      "epoch": 0.16558333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.4830729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1093,
+      "loss/crossentropy": 1.968776412308216,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1839535530656576,
+      "step": 19870
+    },
+    {
+      "epoch": 0.16566666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.232291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1128,
+      "loss/crossentropy": 2.02837230861187,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18482233323156833,
+      "step": 19880
+    },
+    {
+      "epoch": 0.16575,
+      "grad_norm": 29.375,
+      "grad_norm_var": 12.9822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1985,
+      "loss/crossentropy": 1.937032762169838,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886212131008506,
+      "step": 19890
+    },
+    {
+      "epoch": 0.16583333333333333,
+      "grad_norm": 28.5,
+      "grad_norm_var": 13.646809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1062,
+      "loss/crossentropy": 1.7979896873235703,
+      "loss/hidden": 3.416796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17600254379212857,
+      "step": 19900
+    },
+    {
+      "epoch": 0.16591666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5559895833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1592,
+      "loss/crossentropy": 2.0396345034241676,
+      "loss/hidden": 3.25703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18164590019732713,
+      "step": 19910
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.474934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2535,
+      "loss/crossentropy": 1.8868967339396476,
+      "loss/hidden": 3.36953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17275471705943346,
+      "step": 19920
+    },
+    {
+      "epoch": 0.16608333333333333,
+      "grad_norm": 35.75,
+      "grad_norm_var": 10.042122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2469,
+      "loss/crossentropy": 1.9820918217301369,
+      "loss/hidden": 3.44140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978895004838705,
+      "step": 19930
+    },
+    {
+      "epoch": 0.16616666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.80625,
+      "learning_rate": 0.0001,
+      "loss": 7.214,
+      "loss/crossentropy": 1.9374962270259857,
+      "loss/hidden": 3.48046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19119593799114226,
+      "step": 19940
+    },
+    {
+      "epoch": 0.16625,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.9468098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1681,
+      "loss/crossentropy": 2.0098665446043014,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19282399471849204,
+      "step": 19950
+    },
+    {
+      "epoch": 0.16633333333333333,
+      "grad_norm": 27.375,
+      "grad_norm_var": 1.2809895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1022,
+      "loss/crossentropy": 1.9588414117693902,
+      "loss/hidden": 3.349609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17622464820742606,
+      "step": 19960
+    },
+    {
+      "epoch": 0.16641666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.8264973958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2131,
+      "loss/crossentropy": 2.025524842739105,
+      "loss/hidden": 3.346484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19479350578039883,
+      "step": 19970
+    },
+    {
+      "epoch": 0.1665,
+      "grad_norm": 29.25,
+      "grad_norm_var": 2.124739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2225,
+      "loss/crossentropy": 1.9848186463117599,
+      "loss/hidden": 3.37265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18212208710610867,
+      "step": 19980
+    },
+    {
+      "epoch": 0.16658333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.873893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1403,
+      "loss/crossentropy": 1.9571938082575797,
+      "loss/hidden": 3.4078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18815288674086333,
+      "step": 19990
+    },
+    {
+      "epoch": 0.16666666666666666,
+      "grad_norm": 34.5,
+      "grad_norm_var": 21.687434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.3035,
+      "loss/crossentropy": 1.974594208598137,
+      "loss/hidden": 3.439453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18790525905787944,
+      "step": 20000
+    },
+    {
+      "epoch": 0.16675,
+      "grad_norm": 30.125,
+      "grad_norm_var": 41.26145833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1667,
+      "loss/crossentropy": 1.9736678391695022,
+      "loss/hidden": 3.3578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20029252283275129,
+      "step": 20010
+    },
+    {
+      "epoch": 0.16683333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 30.714518229166668,
+      "learning_rate": 0.0001,
+      "loss": 7.2184,
+      "loss/crossentropy": 1.967193180322647,
+      "loss/hidden": 3.39140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17278012707829477,
+      "step": 20020
+    },
+    {
+      "epoch": 0.16691666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 7.071875,
+      "learning_rate": 0.0001,
+      "loss": 7.0742,
+      "loss/crossentropy": 1.944534134864807,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17957909852266313,
+      "step": 20030
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 29.125,
+      "grad_norm_var": 6.534375,
+      "learning_rate": 0.0001,
+      "loss": 7.09,
+      "loss/crossentropy": 1.9741789191961288,
+      "loss/hidden": 3.2390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16954710464924574,
+      "step": 20040
+    },
+    {
+      "epoch": 0.16708333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.44140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1204,
+      "loss/crossentropy": 1.9629234582185746,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17836790028959512,
+      "step": 20050
+    },
+    {
+      "epoch": 0.16716666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.839518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1787,
+      "loss/crossentropy": 1.9100020214915276,
+      "loss/hidden": 3.2796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17498023826628922,
+      "step": 20060
+    },
+    {
+      "epoch": 0.16725,
+      "grad_norm": 30.375,
+      "grad_norm_var": 22.652083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2583,
+      "loss/crossentropy": 1.7778278335928916,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17571195494383574,
+      "step": 20070
+    },
+    {
+      "epoch": 0.16733333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.143931880233763e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2601,
+      "loss/crossentropy": 2.07454999089241,
+      "loss/hidden": 3.483203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1844800652936101,
+      "step": 20080
+    },
+    {
+      "epoch": 0.16741666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.1439318803808257e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1344,
+      "loss/crossentropy": 1.9219613991677762,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17868009638041257,
+      "step": 20090
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.343489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0743,
+      "loss/crossentropy": 1.8762156173586846,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17499885708093643,
+      "step": 20100
+    },
+    {
+      "epoch": 0.16758333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 12.0603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0511,
+      "loss/crossentropy": 2.0174882501363753,
+      "loss/hidden": 3.43671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18392845503985883,
+      "step": 20110
+    },
+    {
+      "epoch": 0.16766666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 11.819205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2065,
+      "loss/crossentropy": 2.1494825392961503,
+      "loss/hidden": 3.2671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18850625175982713,
+      "step": 20120
+    },
+    {
+      "epoch": 0.16775,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.034375,
+      "learning_rate": 0.0001,
+      "loss": 7.1086,
+      "loss/crossentropy": 1.9701856628060341,
+      "loss/hidden": 3.269140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16942516155540943,
+      "step": 20130
+    },
+    {
+      "epoch": 0.16783333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.695572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0342,
+      "loss/crossentropy": 1.9697773188352585,
+      "loss/hidden": 3.309375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1962472852319479,
+      "step": 20140
+    },
+    {
+      "epoch": 0.16791666666666666,
+      "grad_norm": 35.25,
+      "grad_norm_var": 54.09557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3786,
+      "loss/crossentropy": 1.9167721211910247,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18927976656705142,
+      "step": 20150
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 32.0,
+      "grad_norm_var": 56.07076822916667,
+      "learning_rate": 0.0001,
+      "loss": 7.1066,
+      "loss/crossentropy": 1.7909273773431778,
+      "loss/hidden": 3.38984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813417287543416,
+      "step": 20160
+    },
+    {
+      "epoch": 0.16808333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.4458333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0968,
+      "loss/crossentropy": 1.7690905332565308,
+      "loss/hidden": 3.4453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1775512784719467,
+      "step": 20170
+    },
+    {
+      "epoch": 0.16816666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.189231759076972e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2083,
+      "loss/crossentropy": 1.9566049858927728,
+      "loss/hidden": 3.35859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17438329700380564,
+      "step": 20180
+    },
+    {
+      "epoch": 0.16825,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.426822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1032,
+      "loss/crossentropy": 1.967501977086067,
+      "loss/hidden": 3.45390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19263167642056941,
+      "step": 20190
+    },
+    {
+      "epoch": 0.16833333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.69375,
+      "learning_rate": 0.0001,
+      "loss": 7.0984,
+      "loss/crossentropy": 2.0709970980882644,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19457380268722774,
+      "step": 20200
+    },
+    {
+      "epoch": 0.16841666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.050455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1901,
+      "loss/crossentropy": 1.8898002624511718,
+      "loss/hidden": 3.538671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22220217995345592,
+      "step": 20210
+    },
+    {
+      "epoch": 0.1685,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.474739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1791,
+      "loss/crossentropy": 1.9248455747961999,
+      "loss/hidden": 3.33828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1824678654782474,
+      "step": 20220
+    },
+    {
+      "epoch": 0.16858333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.1802083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0963,
+      "loss/crossentropy": 1.7882749281823636,
+      "loss/hidden": 3.336328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1796683069318533,
+      "step": 20230
+    },
+    {
+      "epoch": 0.16866666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.2405598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1632,
+      "loss/crossentropy": 1.9421565659344195,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18475861977785826,
+      "step": 20240
+    },
+    {
+      "epoch": 0.16875,
+      "grad_norm": 30.75,
+      "grad_norm_var": 0.9936848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2453,
+      "loss/crossentropy": 2.0783223837614058,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21532182805240155,
+      "step": 20250
+    },
+    {
+      "epoch": 0.16883333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.5197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1993,
+      "loss/crossentropy": 1.937929594516754,
+      "loss/hidden": 3.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17956498302519322,
+      "step": 20260
+    },
+    {
+      "epoch": 0.16891666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 23.054622395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1453,
+      "loss/crossentropy": 1.9024322986602784,
+      "loss/hidden": 3.273046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180820725671947,
+      "step": 20270
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 31.5,
+      "grad_norm_var": 25.4181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2571,
+      "loss/crossentropy": 1.9528051786124707,
+      "loss/hidden": 3.293359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19371688459068537,
+      "step": 20280
+    },
+    {
+      "epoch": 0.16908333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 342.08899739583336,
+      "learning_rate": 0.0001,
+      "loss": 7.2479,
+      "loss/crossentropy": 1.9640430465340615,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17461762819439172,
+      "step": 20290
+    },
+    {
+      "epoch": 0.16916666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3035.5494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2843,
+      "loss/crossentropy": 1.9877676889300346,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19341197293251752,
+      "step": 20300
+    },
+    {
+      "epoch": 0.16925,
+      "grad_norm": 3472883712.0,
+      "grad_norm_var": 7.538075596093421e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.3376,
+      "loss/crossentropy": 2.1059855103492735,
+      "loss/hidden": 3.572265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21047109067440034,
+      "step": 20310
+    },
+    {
+      "epoch": 0.16933333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 7.538075652165993e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.138,
+      "loss/crossentropy": 1.9671943917870522,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17307756394147872,
+      "step": 20320
+    },
+    {
+      "epoch": 0.16941666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.8270833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1866,
+      "loss/crossentropy": 1.929891800135374,
+      "loss/hidden": 3.479296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20060783661901951,
+      "step": 20330
+    },
+    {
+      "epoch": 0.1695,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.364322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1916,
+      "loss/crossentropy": 1.9168181657791137,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918028002604842,
+      "step": 20340
+    },
+    {
+      "epoch": 0.16958333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 1.0217717414514307e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2237,
+      "loss/crossentropy": 1.9711541652679443,
+      "loss/hidden": 3.451171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19848111569881438,
+      "step": 20350
+    },
+    {
+      "epoch": 0.16966666666666666,
+      "grad_norm": 34.0,
+      "grad_norm_var": 13.803580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2322,
+      "loss/crossentropy": 1.9038699388504028,
+      "loss/hidden": 3.373828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1952126882970333,
+      "step": 20360
+    },
+    {
+      "epoch": 0.16975,
+      "grad_norm": 39.5,
+      "grad_norm_var": 19.155989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2434,
+      "loss/crossentropy": 2.0539934992790223,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17916535902768374,
+      "step": 20370
+    },
+    {
+      "epoch": 0.16983333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 7.022330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2082,
+      "loss/crossentropy": 1.984449103474617,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18695555161684752,
+      "step": 20380
+    },
+    {
+      "epoch": 0.16991666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.2,
+      "learning_rate": 0.0001,
+      "loss": 7.1182,
+      "loss/crossentropy": 2.041994962096214,
+      "loss/hidden": 3.303515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18608700148761273,
+      "step": 20390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.499739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.171,
+      "loss/crossentropy": 1.9582897216081618,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17891748137772084,
+      "step": 20400
+    },
+    {
+      "epoch": 0.17008333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 4.721875,
+      "learning_rate": 0.0001,
+      "loss": 7.0476,
+      "loss/crossentropy": 2.0181909441947936,
+      "loss/hidden": 3.35234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1739883167669177,
+      "step": 20410
+    },
+    {
+      "epoch": 0.17016666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 7.967122395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2752,
+      "loss/crossentropy": 1.8659672304987907,
+      "loss/hidden": 3.456640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22026129364967345,
+      "step": 20420
+    },
+    {
+      "epoch": 0.17025,
+      "grad_norm": 39.0,
+      "grad_norm_var": 11.0556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1633,
+      "loss/crossentropy": 1.9782256945967673,
+      "loss/hidden": 3.2625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19208774864673614,
+      "step": 20430
+    },
+    {
+      "epoch": 0.17033333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 7.506184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2281,
+      "loss/crossentropy": 1.8934879742562771,
+      "loss/hidden": 3.471875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1947283485904336,
+      "step": 20440
+    },
+    {
+      "epoch": 0.17041666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.29765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2757,
+      "loss/crossentropy": 2.0173911824822426,
+      "loss/hidden": 3.555078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19392079655081035,
+      "step": 20450
+    },
+    {
+      "epoch": 0.1705,
+      "grad_norm": 30.125,
+      "grad_norm_var": 7.293489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1701,
+      "loss/crossentropy": 1.9113088577985764,
+      "loss/hidden": 3.37890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1782403675839305,
+      "step": 20460
+    },
+    {
+      "epoch": 0.17058333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.89765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1926,
+      "loss/crossentropy": 1.919259138405323,
+      "loss/hidden": 3.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20750626511871814,
+      "step": 20470
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 37.14375,
+      "learning_rate": 0.0001,
+      "loss": 7.2705,
+      "loss/crossentropy": 1.7986478582024574,
+      "loss/hidden": 3.53828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1986080376431346,
+      "step": 20480
+    },
+    {
+      "epoch": 0.17075,
+      "grad_norm": 39.75,
+      "grad_norm_var": 10.286393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0464,
+      "loss/crossentropy": 2.0147682279348373,
+      "loss/hidden": 3.28671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17417476493865253,
+      "step": 20490
+    },
+    {
+      "epoch": 0.17083333333333334,
+      "grad_norm": 28.5,
+      "grad_norm_var": 24.27890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1267,
+      "loss/crossentropy": 1.9885205298662185,
+      "loss/hidden": 3.2625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17579306047409773,
+      "step": 20500
+    },
+    {
+      "epoch": 0.17091666666666666,
+      "grad_norm": 38.5,
+      "grad_norm_var": 43.1228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2361,
+      "loss/crossentropy": 1.9530100539326667,
+      "loss/hidden": 3.42265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.212185837328434,
+      "step": 20510
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 34.75,
+      "grad_norm_var": 27.574739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2593,
+      "loss/crossentropy": 1.9855598241090775,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20208176802843808,
+      "step": 20520
+    },
+    {
+      "epoch": 0.17108333333333334,
+      "grad_norm": 28.25,
+      "grad_norm_var": 22.65625,
+      "learning_rate": 0.0001,
+      "loss": 7.1722,
+      "loss/crossentropy": 1.9474289268255234,
+      "loss/hidden": 3.371875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18569063413888215,
+      "step": 20530
+    },
+    {
+      "epoch": 0.17116666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.452018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2125,
+      "loss/crossentropy": 1.8864032834768296,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841401696205139,
+      "step": 20540
+    },
+    {
+      "epoch": 0.17125,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.8358723958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0693,
+      "loss/crossentropy": 1.9060160622000695,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18349910359829663,
+      "step": 20550
+    },
+    {
+      "epoch": 0.17133333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 30.594791666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0694,
+      "loss/crossentropy": 1.8780332520604133,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18763979207724332,
+      "step": 20560
+    },
+    {
+      "epoch": 0.17141666666666666,
+      "grad_norm": 39.0,
+      "grad_norm_var": 7.61108322249081e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2846,
+      "loss/crossentropy": 1.9709115386009217,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893330669030547,
+      "step": 20570
+    },
+    {
+      "epoch": 0.1715,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.611083228888521e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0982,
+      "loss/crossentropy": 2.0085377737879755,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18956856336444616,
+      "step": 20580
+    },
+    {
+      "epoch": 0.17158333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.298958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1639,
+      "loss/crossentropy": 2.0797108948230743,
+      "loss/hidden": 3.403125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19589026495814324,
+      "step": 20590
+    },
+    {
+      "epoch": 0.17166666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.215625,
+      "learning_rate": 0.0001,
+      "loss": 7.2052,
+      "loss/crossentropy": 1.8209454715251923,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17919619288295507,
+      "step": 20600
+    },
+    {
+      "epoch": 0.17175,
+      "grad_norm": 29.25,
+      "grad_norm_var": 0.91875,
+      "learning_rate": 0.0001,
+      "loss": 7.1336,
+      "loss/crossentropy": 1.9698986455798149,
+      "loss/hidden": 3.491796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18794093392789363,
+      "step": 20610
+    },
+    {
+      "epoch": 0.17183333333333334,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.7385416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1662,
+      "loss/crossentropy": 1.9401928976178169,
+      "loss/hidden": 3.287890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17406183741986753,
+      "step": 20620
+    },
+    {
+      "epoch": 0.17191666666666666,
+      "grad_norm": 28.25,
+      "grad_norm_var": 2.0869140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0602,
+      "loss/crossentropy": 1.9672116324305535,
+      "loss/hidden": 3.290234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19270303882658482,
+      "step": 20630
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.205143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2236,
+      "loss/crossentropy": 1.9349315941333771,
+      "loss/hidden": 3.4390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18944280669093133,
+      "step": 20640
+    },
+    {
+      "epoch": 0.17208333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.0561848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1739,
+      "loss/crossentropy": 1.8996097341179847,
+      "loss/hidden": 3.29921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16837234497070314,
+      "step": 20650
+    },
+    {
+      "epoch": 0.17216666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.4056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 1.9734445691108704,
+      "loss/hidden": 3.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18427527118474246,
+      "step": 20660
+    },
+    {
+      "epoch": 0.17225,
+      "grad_norm": 35.75,
+      "grad_norm_var": 3.64140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1723,
+      "loss/crossentropy": 1.868157309293747,
+      "loss/hidden": 3.41015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17660889271646737,
+      "step": 20670
+    },
+    {
+      "epoch": 0.17233333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.82890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1102,
+      "loss/crossentropy": 1.9386316768825054,
+      "loss/hidden": 3.268359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17882515136152505,
+      "step": 20680
+    },
+    {
+      "epoch": 0.17241666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.29765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1921,
+      "loss/crossentropy": 1.957555428147316,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19079423155635594,
+      "step": 20690
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.099739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2904,
+      "loss/crossentropy": 1.908998817205429,
+      "loss/hidden": 3.48515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19699111133813857,
+      "step": 20700
+    },
+    {
+      "epoch": 0.17258333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 4.080989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2089,
+      "loss/crossentropy": 2.011483873426914,
+      "loss/hidden": 3.34140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18217164035886527,
+      "step": 20710
+    },
+    {
+      "epoch": 0.17266666666666666,
+      "grad_norm": 49.0,
+      "grad_norm_var": 22.1119140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1806,
+      "loss/crossentropy": 1.969153216481209,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17806398253887892,
+      "step": 20720
+    },
+    {
+      "epoch": 0.17275,
+      "grad_norm": 34.75,
+      "grad_norm_var": 29.743489583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2103,
+      "loss/crossentropy": 1.907319189608097,
+      "loss/hidden": 3.229296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749850135296583,
+      "step": 20730
+    },
+    {
+      "epoch": 0.17283333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.069205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.167,
+      "loss/crossentropy": 2.0715465426445006,
+      "loss/hidden": 3.406640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20546231959015132,
+      "step": 20740
+    },
+    {
+      "epoch": 0.17291666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 8.131987855085405e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1696,
+      "loss/crossentropy": 1.9828820422291755,
+      "loss/hidden": 3.61171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20777845261618494,
+      "step": 20750
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.13198785286854e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0541,
+      "loss/crossentropy": 2.006457194685936,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18682518638670445,
+      "step": 20760
+    },
+    {
+      "epoch": 0.17308333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.7291015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1458,
+      "loss/crossentropy": 1.905017113685608,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17664384599775076,
+      "step": 20770
+    },
+    {
+      "epoch": 0.17316666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.9921223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1497,
+      "loss/crossentropy": 1.9181560665369033,
+      "loss/hidden": 3.42109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18414331339299678,
+      "step": 20780
+    },
+    {
+      "epoch": 0.17325,
+      "grad_norm": 33.25,
+      "grad_norm_var": 15.006184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.164,
+      "loss/crossentropy": 2.0797928765416147,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18293892052024602,
+      "step": 20790
+    },
+    {
+      "epoch": 0.17333333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 15.470572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2319,
+      "loss/crossentropy": 1.8644145905971528,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17734147123992444,
+      "step": 20800
+    },
+    {
+      "epoch": 0.17341666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 5.196875,
+      "learning_rate": 0.0001,
+      "loss": 7.1502,
+      "loss/crossentropy": 2.0136256992816923,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18733342494815589,
+      "step": 20810
+    },
+    {
+      "epoch": 0.1735,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.206184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2597,
+      "loss/crossentropy": 1.926319620013237,
+      "loss/hidden": 3.452734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19622033033519984,
+      "step": 20820
+    },
+    {
+      "epoch": 0.17358333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.6143229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.237,
+      "loss/crossentropy": 2.0856179744005203,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19681357759982349,
+      "step": 20830
+    },
+    {
+      "epoch": 0.17366666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.5999348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2027,
+      "loss/crossentropy": 1.854654923081398,
+      "loss/hidden": 3.41640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18734797462821007,
+      "step": 20840
+    },
+    {
+      "epoch": 0.17375,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.0014973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1428,
+      "loss/crossentropy": 1.8657638885080814,
+      "loss/hidden": 3.284765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17107422454282642,
+      "step": 20850
+    },
+    {
+      "epoch": 0.17383333333333334,
+      "grad_norm": 37.0,
+      "grad_norm_var": 20.97265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2543,
+      "loss/crossentropy": 1.9382992327213286,
+      "loss/hidden": 3.44453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19008651226758957,
+      "step": 20860
+    },
+    {
+      "epoch": 0.17391666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 22.148372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1382,
+      "loss/crossentropy": 1.8269093707203865,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18238500589504839,
+      "step": 20870
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.575455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1999,
+      "loss/crossentropy": 1.8812201246619225,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922521373257041,
+      "step": 20880
+    },
+    {
+      "epoch": 0.17408333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 12.518489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2116,
+      "loss/crossentropy": 2.0255912609398363,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19205207144841552,
+      "step": 20890
+    },
+    {
+      "epoch": 0.17416666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.7504557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0997,
+      "loss/crossentropy": 1.9907532975077629,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19411338102072478,
+      "step": 20900
+    },
+    {
+      "epoch": 0.17425,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.286458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2029,
+      "loss/crossentropy": 1.9424551278352737,
+      "loss/hidden": 3.298046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18332043793052435,
+      "step": 20910
+    },
+    {
+      "epoch": 0.17433333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.42890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0455,
+      "loss/crossentropy": 1.9261737681925297,
+      "loss/hidden": 3.36640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17205108515918255,
+      "step": 20920
+    },
+    {
+      "epoch": 0.17441666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.2177083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1948,
+      "loss/crossentropy": 1.9312485739588738,
+      "loss/hidden": 3.332421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18051387835294008,
+      "step": 20930
+    },
+    {
+      "epoch": 0.1745,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.626822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1633,
+      "loss/crossentropy": 1.997141894698143,
+      "loss/hidden": 3.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18722971267998217,
+      "step": 20940
+    },
+    {
+      "epoch": 0.17458333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.4270833333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1534,
+      "loss/crossentropy": 1.9865899249911307,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18692340292036533,
+      "step": 20950
+    },
+    {
+      "epoch": 0.17466666666666666,
+      "grad_norm": 29.5,
+      "grad_norm_var": 5.658268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0797,
+      "loss/crossentropy": 1.9483724921941756,
+      "loss/hidden": 3.17890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1703987043350935,
+      "step": 20960
+    },
+    {
+      "epoch": 0.17475,
+      "grad_norm": 30.5,
+      "grad_norm_var": 38.084309895833336,
+      "learning_rate": 0.0001,
+      "loss": 7.2583,
+      "loss/crossentropy": 1.8192249581217765,
+      "loss/hidden": 3.455859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19130104836076497,
+      "step": 20970
+    },
+    {
+      "epoch": 0.17483333333333334,
+      "grad_norm": 39.75,
+      "grad_norm_var": 63.06920572916667,
+      "learning_rate": 0.0001,
+      "loss": 7.0819,
+      "loss/crossentropy": 1.8219539567828178,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18972553666681052,
+      "step": 20980
+    },
+    {
+      "epoch": 0.17491666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 30.492643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0089,
+      "loss/crossentropy": 1.8958367705345154,
+      "loss/hidden": 3.375390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16689416002482177,
+      "step": 20990
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.2895833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.196,
+      "loss/crossentropy": 2.0764365196228027,
+      "loss/hidden": 3.32109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21450832188129426,
+      "step": 21000
+    },
+    {
+      "epoch": 0.17508333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.6885416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.3037,
+      "loss/crossentropy": 1.9507210105657578,
+      "loss/hidden": 3.24921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17072254456579686,
+      "step": 21010
+    },
+    {
+      "epoch": 0.17516666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.6791666666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0873,
+      "loss/crossentropy": 1.9367124151438475,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17976205619052052,
+      "step": 21020
+    },
+    {
+      "epoch": 0.17525,
+      "grad_norm": 39.5,
+      "grad_norm_var": 25.1375,
+      "learning_rate": 0.0001,
+      "loss": 7.2196,
+      "loss/crossentropy": 1.9442919731140136,
+      "loss/hidden": 3.447265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.197759541682899,
+      "step": 21030
+    },
+    {
+      "epoch": 0.17533333333333334,
+      "grad_norm": 40.75,
+      "grad_norm_var": 27.83125,
+      "learning_rate": 0.0001,
+      "loss": 7.0902,
+      "loss/crossentropy": 2.0014768242836,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18321804963052274,
+      "step": 21040
+    },
+    {
+      "epoch": 0.17541666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 17.644791666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0555,
+      "loss/crossentropy": 1.9349362432956696,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17762522976845502,
+      "step": 21050
+    },
+    {
+      "epoch": 0.1755,
+      "grad_norm": 37.0,
+      "grad_norm_var": 153.59765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2561,
+      "loss/crossentropy": 1.8269620671868325,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18688229732215406,
+      "step": 21060
+    },
+    {
+      "epoch": 0.17558333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 153.53639322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.0717,
+      "loss/crossentropy": 1.9161958619952202,
+      "loss/hidden": 3.2078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18175008203834295,
+      "step": 21070
+    },
+    {
+      "epoch": 0.17566666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 9.633072916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0166,
+      "loss/crossentropy": 1.944762909412384,
+      "loss/hidden": 3.3109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18148341085761785,
+      "step": 21080
+    },
+    {
+      "epoch": 0.17575,
+      "grad_norm": 39.25,
+      "grad_norm_var": 11.1603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1211,
+      "loss/crossentropy": 2.0051519334316255,
+      "loss/hidden": 3.260546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19057092368602752,
+      "step": 21090
+    },
+    {
+      "epoch": 0.17583333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.741080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0708,
+      "loss/crossentropy": 1.9906994640827178,
+      "loss/hidden": 3.253515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1924733016639948,
+      "step": 21100
+    },
+    {
+      "epoch": 0.17591666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.559830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0727,
+      "loss/crossentropy": 1.8093304127454757,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16857612747699022,
+      "step": 21110
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.362239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0325,
+      "loss/crossentropy": 1.9151247292757034,
+      "loss/hidden": 3.33515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17153692077845334,
+      "step": 21120
+    },
+    {
+      "epoch": 0.17608333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 7.142643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0412,
+      "loss/crossentropy": 1.915633451938629,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17531707808375357,
+      "step": 21130
+    },
+    {
+      "epoch": 0.17616666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.745833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0504,
+      "loss/crossentropy": 1.8117642849683762,
+      "loss/hidden": 3.325390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1695346875116229,
+      "step": 21140
+    },
+    {
+      "epoch": 0.17625,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.55,
+      "learning_rate": 0.0001,
+      "loss": 7.0864,
+      "loss/crossentropy": 2.054150219261646,
+      "loss/hidden": 3.284375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17537228874862193,
+      "step": 21150
+    },
+    {
+      "epoch": 0.17633333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 16.0625,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.9306156545877458,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17940887417644263,
+      "step": 21160
+    },
+    {
+      "epoch": 0.17641666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 18.293489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0259,
+      "loss/crossentropy": 1.9809558287262916,
+      "loss/hidden": 3.311328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17992461659014225,
+      "step": 21170
+    },
+    {
+      "epoch": 0.1765,
+      "grad_norm": 30.0,
+      "grad_norm_var": 15.62265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1482,
+      "loss/crossentropy": 1.9801484018564224,
+      "loss/hidden": 3.467578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21556537672877313,
+      "step": 21180
+    },
+    {
+      "epoch": 0.17658333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.769791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1123,
+      "loss/crossentropy": 1.9046063914895057,
+      "loss/hidden": 3.290234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18891190122812987,
+      "step": 21190
+    },
+    {
+      "epoch": 0.17666666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.4192057291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1404,
+      "loss/crossentropy": 1.9384473890066147,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893476253375411,
+      "step": 21200
+    },
+    {
+      "epoch": 0.17675,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.034309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0929,
+      "loss/crossentropy": 1.8499881118535995,
+      "loss/hidden": 3.28359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1762183213606477,
+      "step": 21210
+    },
+    {
+      "epoch": 0.17683333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 20.12265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1438,
+      "loss/crossentropy": 1.7578793555498122,
+      "loss/hidden": 3.410546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17032650909386576,
+      "step": 21220
+    },
+    {
+      "epoch": 0.17691666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 16.9134765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1549,
+      "loss/crossentropy": 1.8758591383695602,
+      "loss/hidden": 3.247265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18351999539881944,
+      "step": 21230
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.0014973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0902,
+      "loss/crossentropy": 1.8791117697954178,
+      "loss/hidden": 3.306640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17864123359322548,
+      "step": 21240
+    },
+    {
+      "epoch": 0.17708333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 9.9197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1855,
+      "loss/crossentropy": 1.9066469356417657,
+      "loss/hidden": 3.266015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.185793387144804,
+      "step": 21250
+    },
+    {
+      "epoch": 0.17716666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 9.613541666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2545,
+      "loss/crossentropy": 2.0106510326266287,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.200165849737823,
+      "step": 21260
+    },
+    {
+      "epoch": 0.17725,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.333072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0743,
+      "loss/crossentropy": 2.0421891048550607,
+      "loss/hidden": 3.312890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17994126975536345,
+      "step": 21270
+    },
+    {
+      "epoch": 0.17733333333333334,
+      "grad_norm": 28.625,
+      "grad_norm_var": 8.732291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1346,
+      "loss/crossentropy": 2.0621480025351047,
+      "loss/hidden": 3.261328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17075770646333693,
+      "step": 21280
+    },
+    {
+      "epoch": 0.17741666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 5.127083333333333,
+      "learning_rate": 0.0001,
+      "loss": 6.9973,
+      "loss/crossentropy": 1.9399575263261795,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19377149622887374,
+      "step": 21290
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.237239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1829,
+      "loss/crossentropy": 1.866205206513405,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18856921121478082,
+      "step": 21300
+    },
+    {
+      "epoch": 0.17758333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.244205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1293,
+      "loss/crossentropy": 2.0109529733657836,
+      "loss/hidden": 3.449609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19027695059776306,
+      "step": 21310
+    },
+    {
+      "epoch": 0.17766666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.00390625,
+      "learning_rate": 0.0001,
+      "loss": 7.112,
+      "loss/crossentropy": 1.872106634825468,
+      "loss/hidden": 3.17265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1676292611286044,
+      "step": 21320
+    },
+    {
+      "epoch": 0.17775,
+      "grad_norm": 34.75,
+      "grad_norm_var": 3.3837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0871,
+      "loss/crossentropy": 1.920562918484211,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18120457511395216,
+      "step": 21330
+    },
+    {
+      "epoch": 0.17783333333333334,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.119791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1034,
+      "loss/crossentropy": 1.860015056282282,
+      "loss/hidden": 3.408984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18130568116903306,
+      "step": 21340
+    },
+    {
+      "epoch": 0.17791666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.5660807291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0764,
+      "loss/crossentropy": 1.9515722021460533,
+      "loss/hidden": 3.283984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20127151608467103,
+      "step": 21350
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 31.75,
+      "grad_norm_var": 14.317643229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1788,
+      "loss/crossentropy": 1.840681430697441,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20178202018141747,
+      "step": 21360
+    },
+    {
+      "epoch": 0.17808333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 15.29765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2874,
+      "loss/crossentropy": 1.9817566379904747,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20724791213870047,
+      "step": 21370
+    },
+    {
+      "epoch": 0.17816666666666667,
+      "grad_norm": 27.375,
+      "grad_norm_var": 6.7087890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0899,
+      "loss/crossentropy": 1.9329698652029037,
+      "loss/hidden": 3.4796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19680206775665282,
+      "step": 21380
+    },
+    {
+      "epoch": 0.17825,
+      "grad_norm": 32.5,
+      "grad_norm_var": 25.580208333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2182,
+      "loss/crossentropy": 1.7622111141681671,
+      "loss/hidden": 3.269140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16894294111989439,
+      "step": 21390
+    },
+    {
+      "epoch": 0.17833333333333334,
+      "grad_norm": 41.5,
+      "grad_norm_var": 25.19375,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 1.9464396089315414,
+      "loss/hidden": 3.31171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2048334326595068,
+      "step": 21400
+    },
+    {
+      "epoch": 0.17841666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.649934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1754,
+      "loss/crossentropy": 1.9862362071871758,
+      "loss/hidden": 3.30625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.186565706320107,
+      "step": 21410
+    },
+    {
+      "epoch": 0.1785,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.847330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2133,
+      "loss/crossentropy": 1.837697020173073,
+      "loss/hidden": 3.2875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18464917819947005,
+      "step": 21420
+    },
+    {
+      "epoch": 0.17858333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.787434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1901,
+      "loss/crossentropy": 1.9632452219724654,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18861688170582055,
+      "step": 21430
+    },
+    {
+      "epoch": 0.17866666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.65390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0743,
+      "loss/crossentropy": 1.9409735351800919,
+      "loss/hidden": 3.284375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16855674423277378,
+      "step": 21440
+    },
+    {
+      "epoch": 0.17875,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.6494140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2735,
+      "loss/crossentropy": 1.880461908876896,
+      "loss/hidden": 3.509375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17760127065703274,
+      "step": 21450
+    },
+    {
+      "epoch": 0.17883333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 11.239518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2565,
+      "loss/crossentropy": 1.8518391624093056,
+      "loss/hidden": 3.4703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20302680414170027,
+      "step": 21460
+    },
+    {
+      "epoch": 0.17891666666666667,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.6864583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1103,
+      "loss/crossentropy": 1.8603545751422643,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17260765563696623,
+      "step": 21470
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.11640625,
+      "learning_rate": 0.0001,
+      "loss": 7.181,
+      "loss/crossentropy": 1.935846994072199,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18108982145786284,
+      "step": 21480
+    },
+    {
+      "epoch": 0.17908333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.349739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0254,
+      "loss/crossentropy": 1.847939160466194,
+      "loss/hidden": 3.411328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18489929512143136,
+      "step": 21490
+    },
+    {
+      "epoch": 0.17916666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.283072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2079,
+      "loss/crossentropy": 1.9450252279639244,
+      "loss/hidden": 3.28203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18005480505526067,
+      "step": 21500
+    },
+    {
+      "epoch": 0.17925,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.559375,
+      "learning_rate": 0.0001,
+      "loss": 7.1295,
+      "loss/crossentropy": 1.8692103013396264,
+      "loss/hidden": 3.403515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1927676422521472,
+      "step": 21510
+    },
+    {
+      "epoch": 0.17933333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.107418946983692e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1829,
+      "loss/crossentropy": 1.8750286191701888,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18964456245303155,
+      "step": 21520
+    },
+    {
+      "epoch": 0.17941666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 73.08639322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.1951,
+      "loss/crossentropy": 1.9951496042311192,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17876034323126078,
+      "step": 21530
+    },
+    {
+      "epoch": 0.1795,
+      "grad_norm": 31.25,
+      "grad_norm_var": 74.24348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2168,
+      "loss/crossentropy": 2.008412830531597,
+      "loss/hidden": 3.375390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18957524821162225,
+      "step": 21540
+    },
+    {
+      "epoch": 0.17958333333333334,
+      "grad_norm": 29.75,
+      "grad_norm_var": 14.365559895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1493,
+      "loss/crossentropy": 1.8941684760153294,
+      "loss/hidden": 3.281640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17081814613193275,
+      "step": 21550
+    },
+    {
+      "epoch": 0.17966666666666667,
+      "grad_norm": 28.625,
+      "grad_norm_var": 13.04140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1053,
+      "loss/crossentropy": 1.9805061787366867,
+      "loss/hidden": 3.372265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17744793705642223,
+      "step": 21560
+    },
+    {
+      "epoch": 0.17975,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.7197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1105,
+      "loss/crossentropy": 1.8534754283726216,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17784328125417231,
+      "step": 21570
+    },
+    {
+      "epoch": 0.17983333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.207598001557148e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1082,
+      "loss/crossentropy": 1.8273938789963722,
+      "loss/hidden": 3.253125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1753040511161089,
+      "step": 21580
+    },
+    {
+      "epoch": 0.17991666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.248958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2893,
+      "loss/crossentropy": 1.8355638459324837,
+      "loss/hidden": 3.441796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18569518942385912,
+      "step": 21590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 37.5,
+      "grad_norm_var": 3.858268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1866,
+      "loss/crossentropy": 2.0116271018981933,
+      "loss/hidden": 3.278125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18747425377368926,
+      "step": 21600
+    },
+    {
+      "epoch": 0.18008333333333335,
+      "grad_norm": 27.25,
+      "grad_norm_var": 5.5041015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1206,
+      "loss/crossentropy": 1.876588474214077,
+      "loss/hidden": 3.27421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1746151703875512,
+      "step": 21610
+    },
+    {
+      "epoch": 0.18016666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.8525390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1844,
+      "loss/crossentropy": 1.9754797160625457,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18899267297238112,
+      "step": 21620
+    },
+    {
+      "epoch": 0.18025,
+      "grad_norm": 29.75,
+      "grad_norm_var": 35.775455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2164,
+      "loss/crossentropy": 1.9548555478453635,
+      "loss/hidden": 3.437109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1899010267108679,
+      "step": 21630
+    },
+    {
+      "epoch": 0.18033333333333335,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.1864583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2192,
+      "loss/crossentropy": 1.9254567980766297,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18021824043244122,
+      "step": 21640
+    },
+    {
+      "epoch": 0.18041666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.1968098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1924,
+      "loss/crossentropy": 2.030823102593422,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.189572992362082,
+      "step": 21650
+    },
+    {
+      "epoch": 0.1805,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.6197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2448,
+      "loss/crossentropy": 1.8883960962295532,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1849999286234379,
+      "step": 21660
+    },
+    {
+      "epoch": 0.18058333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 7.2462890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1476,
+      "loss/crossentropy": 2.062503894418478,
+      "loss/hidden": 3.28984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18000608943402768,
+      "step": 21670
+    },
+    {
+      "epoch": 0.18066666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 7.6587890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1875,
+      "loss/crossentropy": 1.9085218161344528,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1774459894746542,
+      "step": 21680
+    },
+    {
+      "epoch": 0.18075,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.165625,
+      "learning_rate": 0.0001,
+      "loss": 7.1292,
+      "loss/crossentropy": 2.033360409736633,
+      "loss/hidden": 3.2703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19001046791672707,
+      "step": 21690
+    },
+    {
+      "epoch": 0.18083333333333335,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.3520182291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.3108,
+      "loss/crossentropy": 1.9485228419303895,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1983707632869482,
+      "step": 21700
+    },
+    {
+      "epoch": 0.18091666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.9434895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1976,
+      "loss/crossentropy": 1.9755011923611163,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2008528908714652,
+      "step": 21710
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.995768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1452,
+      "loss/crossentropy": 2.0429214388132095,
+      "loss/hidden": 3.396875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19865372963249683,
+      "step": 21720
+    },
+    {
+      "epoch": 0.18108333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.030143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1246,
+      "loss/crossentropy": 2.007198192179203,
+      "loss/hidden": 3.390234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19710920918732883,
+      "step": 21730
+    },
+    {
+      "epoch": 0.18116666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.04375,
+      "learning_rate": 0.0001,
+      "loss": 7.242,
+      "loss/crossentropy": 1.9074750356376171,
+      "loss/hidden": 3.28203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16688416637480258,
+      "step": 21740
+    },
+    {
+      "epoch": 0.18125,
+      "grad_norm": 28.5,
+      "grad_norm_var": 2.5228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.2612,
+      "loss/crossentropy": 2.0462217807769774,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1881243022158742,
+      "step": 21750
+    },
+    {
+      "epoch": 0.18133333333333335,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.5580729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2617,
+      "loss/crossentropy": 1.8171587988734246,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17960255667567254,
+      "step": 21760
+    },
+    {
+      "epoch": 0.18141666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.1848307291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1653,
+      "loss/crossentropy": 1.776734608411789,
+      "loss/hidden": 3.48203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18971325419843196,
+      "step": 21770
+    },
+    {
+      "epoch": 0.1815,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.66875,
+      "learning_rate": 0.0001,
+      "loss": 7.1115,
+      "loss/crossentropy": 1.9086107723414898,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18301735278218983,
+      "step": 21780
+    },
+    {
+      "epoch": 0.18158333333333335,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.819205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.3371,
+      "loss/crossentropy": 1.908405600488186,
+      "loss/hidden": 3.411328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19641327355057,
+      "step": 21790
+    },
+    {
+      "epoch": 0.18166666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.97890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0643,
+      "loss/crossentropy": 1.9705938309431077,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1787406163290143,
+      "step": 21800
+    },
+    {
+      "epoch": 0.18175,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.05390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1203,
+      "loss/crossentropy": 2.159313699603081,
+      "loss/hidden": 3.28828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19056724552065135,
+      "step": 21810
+    },
+    {
+      "epoch": 0.18183333333333335,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.255989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0628,
+      "loss/crossentropy": 2.032162272930145,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19196782782673835,
+      "step": 21820
+    },
+    {
+      "epoch": 0.18191666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.522330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.22,
+      "loss/crossentropy": 1.8344575047492981,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18553915787488223,
+      "step": 21830
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.39765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1151,
+      "loss/crossentropy": 1.9267772898077964,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1810599728487432,
+      "step": 21840
+    },
+    {
+      "epoch": 0.18208333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.071875,
+      "learning_rate": 0.0001,
+      "loss": 7.2166,
+      "loss/crossentropy": 1.981689989566803,
+      "loss/hidden": 3.282421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1776347430422902,
+      "step": 21850
+    },
+    {
+      "epoch": 0.18216666666666667,
+      "grad_norm": 65.0,
+      "grad_norm_var": 3946.815625,
+      "learning_rate": 0.0001,
+      "loss": 7.2695,
+      "loss/crossentropy": 1.9034842535853387,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1790511528030038,
+      "step": 21860
+    },
+    {
+      "epoch": 0.18225,
+      "grad_norm": 34.0,
+      "grad_norm_var": 3946.3113932291667,
+      "learning_rate": 0.0001,
+      "loss": 7.1144,
+      "loss/crossentropy": 1.8658113181591034,
+      "loss/hidden": 3.4390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17436937838792801,
+      "step": 21870
+    },
+    {
+      "epoch": 0.18233333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.0791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2798,
+      "loss/crossentropy": 1.9000372186303138,
+      "loss/hidden": 3.43984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18413659315556288,
+      "step": 21880
+    },
+    {
+      "epoch": 0.18241666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.231184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1192,
+      "loss/crossentropy": 1.9039353042840959,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17424365859478713,
+      "step": 21890
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 29.375,
+      "grad_norm_var": 103.3447265625,
+      "learning_rate": 0.0001,
+      "loss": 7.164,
+      "loss/crossentropy": 1.9990982174873353,
+      "loss/hidden": 3.478515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21232158020138742,
+      "step": 21900
+    },
+    {
+      "epoch": 0.18258333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 144.79680989583332,
+      "learning_rate": 0.0001,
+      "loss": 7.1039,
+      "loss/crossentropy": 1.991016660630703,
+      "loss/hidden": 3.34140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18620871379971504,
+      "step": 21910
+    },
+    {
+      "epoch": 0.18266666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 45.228125,
+      "learning_rate": 0.0001,
+      "loss": 7.1249,
+      "loss/crossentropy": 1.9497494012117387,
+      "loss/hidden": 3.365625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19447645582258702,
+      "step": 21920
+    },
+    {
+      "epoch": 0.18275,
+      "grad_norm": 29.375,
+      "grad_norm_var": 37.72265625,
+      "learning_rate": 0.0001,
+      "loss": 7.159,
+      "loss/crossentropy": 1.9128453850746154,
+      "loss/hidden": 3.350390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18499513305723667,
+      "step": 21930
+    },
+    {
+      "epoch": 0.18283333333333332,
+      "grad_norm": 34.75,
+      "grad_norm_var": 23.9056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1145,
+      "loss/crossentropy": 1.893994764983654,
+      "loss/hidden": 3.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16739220172166824,
+      "step": 21940
+    },
+    {
+      "epoch": 0.18291666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 23.7791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.231,
+      "loss/crossentropy": 1.9981656968593597,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19310638960450888,
+      "step": 21950
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.124739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1937,
+      "loss/crossentropy": 1.9169929325580597,
+      "loss/hidden": 3.279296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17555649243295193,
+      "step": 21960
+    },
+    {
+      "epoch": 0.18308333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.1660807291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1859,
+      "loss/crossentropy": 1.9669486790895463,
+      "loss/hidden": 3.4546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1969633696600795,
+      "step": 21970
+    },
+    {
+      "epoch": 0.18316666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.8684895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0411,
+      "loss/crossentropy": 1.779083925485611,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17282047271728515,
+      "step": 21980
+    },
+    {
+      "epoch": 0.18325,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.7791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0728,
+      "loss/crossentropy": 1.8291324481368065,
+      "loss/hidden": 3.536328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18751801066100598,
+      "step": 21990
+    },
+    {
+      "epoch": 0.18333333333333332,
+      "grad_norm": 31.375,
+      "grad_norm_var": 0.6124348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1559,
+      "loss/crossentropy": 2.0308694422245024,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17829547077417374,
+      "step": 22000
+    },
+    {
+      "epoch": 0.18341666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.6740678037445024e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2092,
+      "loss/crossentropy": 1.8981263637542725,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18251936007291078,
+      "step": 22010
+    },
+    {
+      "epoch": 0.1835,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.674067801636515e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.191,
+      "loss/crossentropy": 1.9962030783295632,
+      "loss/hidden": 3.40078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18759495932608844,
+      "step": 22020
+    },
+    {
+      "epoch": 0.18358333333333332,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.6208333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2085,
+      "loss/crossentropy": 2.1387590736150743,
+      "loss/hidden": 3.235546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813628438860178,
+      "step": 22030
+    },
+    {
+      "epoch": 0.18366666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.478125,
+      "learning_rate": 0.0001,
+      "loss": 7.2195,
+      "loss/crossentropy": 1.9104967445135117,
+      "loss/hidden": 3.458984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1847192894667387,
+      "step": 22040
+    },
+    {
+      "epoch": 0.18375,
+      "grad_norm": 37.0,
+      "grad_norm_var": 298.31451822916665,
+      "learning_rate": 0.0001,
+      "loss": 7.4271,
+      "loss/crossentropy": 1.997025790810585,
+      "loss/hidden": 3.484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21325294878333806,
+      "step": 22050
+    },
+    {
+      "epoch": 0.18383333333333332,
+      "grad_norm": 31.25,
+      "grad_norm_var": 307.12265625,
+      "learning_rate": 0.0001,
+      "loss": 7.192,
+      "loss/crossentropy": 1.9597471974790097,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20086586745455862,
+      "step": 22060
+    },
+    {
+      "epoch": 0.18391666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 483.44765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2893,
+      "loss/crossentropy": 1.9892470613121986,
+      "loss/hidden": 3.2390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16612087301909922,
+      "step": 22070
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.693489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2789,
+      "loss/crossentropy": 2.033068688213825,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1896548074670136,
+      "step": 22080
+    },
+    {
+      "epoch": 0.18408333333333332,
+      "grad_norm": 28.5,
+      "grad_norm_var": 3.3509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1815,
+      "loss/crossentropy": 1.9157329767942428,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19151314422488214,
+      "step": 22090
+    },
+    {
+      "epoch": 0.18416666666666667,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.8372395833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1136,
+      "loss/crossentropy": 1.7613531425595284,
+      "loss/hidden": 3.359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17461738027632237,
+      "step": 22100
+    },
+    {
+      "epoch": 0.18425,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.421809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.104,
+      "loss/crossentropy": 1.8239581808447838,
+      "loss/hidden": 3.22109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17822336852550508,
+      "step": 22110
+    },
+    {
+      "epoch": 0.18433333333333332,
+      "grad_norm": 4999610368.0,
+      "grad_norm_var": 1.8396535701223857e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2562,
+      "loss/crossentropy": 2.032672482728958,
+      "loss/hidden": 3.380859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18458173088729382,
+      "step": 22120
+    },
+    {
+      "epoch": 0.18441666666666667,
+      "grad_norm": 28.75,
+      "grad_norm_var": 1.839653570192378e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1793,
+      "loss/crossentropy": 1.9058007255196572,
+      "loss/hidden": 3.255859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846909010782838,
+      "step": 22130
+    },
+    {
+      "epoch": 0.1845,
+      "grad_norm": 30.875,
+      "grad_norm_var": 20.204622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3172,
+      "loss/crossentropy": 1.7383491396903992,
+      "loss/hidden": 3.4734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18420727290213107,
+      "step": 22140
+    },
+    {
+      "epoch": 0.18458333333333332,
+      "grad_norm": 30.75,
+      "grad_norm_var": 6.981184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1186,
+      "loss/crossentropy": 1.8524989560246468,
+      "loss/hidden": 3.353515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841329261660576,
+      "step": 22150
+    },
+    {
+      "epoch": 0.18466666666666667,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.792643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1784,
+      "loss/crossentropy": 2.0768550515174864,
+      "loss/hidden": 3.3015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18903156593441964,
+      "step": 22160
+    },
+    {
+      "epoch": 0.18475,
+      "grad_norm": 30.625,
+      "grad_norm_var": 7.145572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1584,
+      "loss/crossentropy": 1.994654531776905,
+      "loss/hidden": 3.41796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17800347432494162,
+      "step": 22170
+    },
+    {
+      "epoch": 0.18483333333333332,
+      "grad_norm": 31.25,
+      "grad_norm_var": 2.676497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2383,
+      "loss/crossentropy": 1.9594170451164246,
+      "loss/hidden": 3.394921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18395585883408785,
+      "step": 22180
+    },
+    {
+      "epoch": 0.18491666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.350779033796827e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2087,
+      "loss/crossentropy": 1.8478225871920586,
+      "loss/hidden": 3.360546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17307716123759748,
+      "step": 22190
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 35.75,
+      "grad_norm_var": 8.704166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2124,
+      "loss/crossentropy": 1.9063466116786003,
+      "loss/hidden": 3.51640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19347276352345943,
+      "step": 22200
+    },
+    {
+      "epoch": 0.18508333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 6.151811412705062e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2525,
+      "loss/crossentropy": 1.9729831591248512,
+      "loss/hidden": 3.5765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19688243959099055,
+      "step": 22210
+    },
+    {
+      "epoch": 0.18516666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 6.15181141205145e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1636,
+      "loss/crossentropy": 1.9547577396035194,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961274055764079,
+      "step": 22220
+    },
+    {
+      "epoch": 0.18525,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.315625,
+      "learning_rate": 0.0001,
+      "loss": 7.0614,
+      "loss/crossentropy": 1.8509338155388833,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19308034833520651,
+      "step": 22230
+    },
+    {
+      "epoch": 0.18533333333333332,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.2452473958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1437,
+      "loss/crossentropy": 2.0442804262042045,
+      "loss/hidden": 3.4171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17812191285192966,
+      "step": 22240
+    },
+    {
+      "epoch": 0.18541666666666667,
+      "grad_norm": 35.25,
+      "grad_norm_var": 7.934830729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2369,
+      "loss/crossentropy": 1.830637451261282,
+      "loss/hidden": 3.534765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20697210887446998,
+      "step": 22250
+    },
+    {
+      "epoch": 0.1855,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.44140625,
+      "learning_rate": 0.0001,
+      "loss": 7.187,
+      "loss/crossentropy": 1.8408591210842133,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18112487625330687,
+      "step": 22260
+    },
+    {
+      "epoch": 0.18558333333333332,
+      "grad_norm": 32.75,
+      "grad_norm_var": 17.448958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0347,
+      "loss/crossentropy": 1.7636138372123242,
+      "loss/hidden": 3.333984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16783975344151258,
+      "step": 22270
+    },
+    {
+      "epoch": 0.18566666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 18.480143229166668,
+      "learning_rate": 0.0001,
+      "loss": 7.0986,
+      "loss/crossentropy": 1.970015862584114,
+      "loss/hidden": 3.395703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1851288428530097,
+      "step": 22280
+    },
+    {
+      "epoch": 0.18575,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.3705729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1961,
+      "loss/crossentropy": 1.9914399236440659,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1902041558176279,
+      "step": 22290
+    },
+    {
+      "epoch": 0.18583333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 10.545572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.147,
+      "loss/crossentropy": 2.02308244779706,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19159749764949083,
+      "step": 22300
+    },
+    {
+      "epoch": 0.18591666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.279166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1949,
+      "loss/crossentropy": 2.025154173374176,
+      "loss/hidden": 3.24765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16830889787524939,
+      "step": 22310
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.1395833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2059,
+      "loss/crossentropy": 2.0100185997784137,
+      "loss/hidden": 3.370703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18132574241608382,
+      "step": 22320
+    },
+    {
+      "epoch": 0.18608333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.238541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1658,
+      "loss/crossentropy": 2.0642576068639755,
+      "loss/hidden": 3.419921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19487057253718376,
+      "step": 22330
+    },
+    {
+      "epoch": 0.18616666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.9254557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1191,
+      "loss/crossentropy": 1.9411341682076455,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19180229846388103,
+      "step": 22340
+    },
+    {
+      "epoch": 0.18625,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.3051432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0705,
+      "loss/crossentropy": 2.052951917052269,
+      "loss/hidden": 3.306640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17878592237830163,
+      "step": 22350
+    },
+    {
+      "epoch": 0.18633333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.22265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0938,
+      "loss/crossentropy": 2.028418146073818,
+      "loss/hidden": 3.269140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18585060331970454,
+      "step": 22360
+    },
+    {
+      "epoch": 0.18641666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.906631956050915e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2286,
+      "loss/crossentropy": 1.922825987637043,
+      "loss/hidden": 3.367578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19583759624511005,
+      "step": 22370
+    },
+    {
+      "epoch": 0.1865,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.906631949419021e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1389,
+      "loss/crossentropy": 1.8866703882813454,
+      "loss/hidden": 3.4140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19007880557328463,
+      "step": 22380
+    },
+    {
+      "epoch": 0.18658333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.9830729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1525,
+      "loss/crossentropy": 2.032797488570213,
+      "loss/hidden": 3.330859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1895093174651265,
+      "step": 22390
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 30.375,
+      "grad_norm_var": 1.8770182291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1148,
+      "loss/crossentropy": 1.9339124277234077,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18026548773050308,
+      "step": 22400
+    },
+    {
+      "epoch": 0.18675,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5893229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0513,
+      "loss/crossentropy": 1.8591840133070945,
+      "loss/hidden": 3.417578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17713533174246548,
+      "step": 22410
+    },
+    {
+      "epoch": 0.18683333333333332,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.16015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2072,
+      "loss/crossentropy": 2.0397066488862037,
+      "loss/hidden": 3.407421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1848060205578804,
+      "step": 22420
+    },
+    {
+      "epoch": 0.18691666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 1.3747395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1741,
+      "loss/crossentropy": 1.9008927434682845,
+      "loss/hidden": 3.282421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16786355432122946,
+      "step": 22430
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.908333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.155,
+      "loss/crossentropy": 1.9573460966348648,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18546217102557422,
+      "step": 22440
+    },
+    {
+      "epoch": 0.18708333333333332,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.30390625,
+      "learning_rate": 0.0001,
+      "loss": 7.204,
+      "loss/crossentropy": 1.8952807098627091,
+      "loss/hidden": 3.471484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968030147254467,
+      "step": 22450
+    },
+    {
+      "epoch": 0.18716666666666668,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.8811848958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1798,
+      "loss/crossentropy": 1.965817154198885,
+      "loss/hidden": 3.333984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18468138631433248,
+      "step": 22460
+    },
+    {
+      "epoch": 0.18725,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.51640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1595,
+      "loss/crossentropy": 1.9224350303411484,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18262785244733096,
+      "step": 22470
+    },
+    {
+      "epoch": 0.18733333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.3426432291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0754,
+      "loss/crossentropy": 1.9634606420993805,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18272318951785566,
+      "step": 22480
+    },
+    {
+      "epoch": 0.18741666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.394791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1032,
+      "loss/crossentropy": 1.8750592201948166,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2044448995962739,
+      "step": 22490
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.847916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1546,
+      "loss/crossentropy": 1.9921681955456734,
+      "loss/hidden": 3.33828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18482560459524394,
+      "step": 22500
+    },
+    {
+      "epoch": 0.18758333333333332,
+      "grad_norm": 29.625,
+      "grad_norm_var": 7.07265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1545,
+      "loss/crossentropy": 1.81395875364542,
+      "loss/hidden": 3.2640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17643945179879666,
+      "step": 22510
+    },
+    {
+      "epoch": 0.18766666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 10.3728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1316,
+      "loss/crossentropy": 1.9426890328526496,
+      "loss/hidden": 3.353125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1746394332498312,
+      "step": 22520
+    },
+    {
+      "epoch": 0.18775,
+      "grad_norm": 28.875,
+      "grad_norm_var": 7.320572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.179,
+      "loss/crossentropy": 1.9067205019295215,
+      "loss/hidden": 3.57890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20180891193449496,
+      "step": 22530
+    },
+    {
+      "epoch": 0.18783333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.792708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2004,
+      "loss/crossentropy": 1.8880975291132926,
+      "loss/hidden": 3.26015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1830406405031681,
+      "step": 22540
+    },
+    {
+      "epoch": 0.18791666666666668,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.6702473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1685,
+      "loss/crossentropy": 1.9224606469273566,
+      "loss/hidden": 3.305859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18565046451985837,
+      "step": 22550
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.419791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1482,
+      "loss/crossentropy": 1.8401526108384132,
+      "loss/hidden": 3.44375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1876707063987851,
+      "step": 22560
+    },
+    {
+      "epoch": 0.18808333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 10.328580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1859,
+      "loss/crossentropy": 2.0541805654764174,
+      "loss/hidden": 3.316015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18220127169042827,
+      "step": 22570
+    },
+    {
+      "epoch": 0.18816666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.4830729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1605,
+      "loss/crossentropy": 1.9627857252955436,
+      "loss/hidden": 3.333984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19063445292413234,
+      "step": 22580
+    },
+    {
+      "epoch": 0.18825,
+      "grad_norm": 29.75,
+      "grad_norm_var": 2.195572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1384,
+      "loss/crossentropy": 1.8928602121770381,
+      "loss/hidden": 3.2609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17759551513008773,
+      "step": 22590
+    },
+    {
+      "epoch": 0.18833333333333332,
+      "grad_norm": 37.0,
+      "grad_norm_var": 7.53807566008851e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1494,
+      "loss/crossentropy": 1.9004074782133102,
+      "loss/hidden": 3.280078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19086783099919558,
+      "step": 22600
+    },
+    {
+      "epoch": 0.18841666666666668,
+      "grad_norm": 30.875,
+      "grad_norm_var": 7.538075650284849e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1784,
+      "loss/crossentropy": 1.9848673433065414,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19275220930576326,
+      "step": 22610
+    },
+    {
+      "epoch": 0.1885,
+      "grad_norm": 39.75,
+      "grad_norm_var": 41.72265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2255,
+      "loss/crossentropy": 1.8599149212241173,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18047851473093032,
+      "step": 22620
+    },
+    {
+      "epoch": 0.18858333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 38.51223958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1558,
+      "loss/crossentropy": 1.8665556862950325,
+      "loss/hidden": 3.421484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17745948843657972,
+      "step": 22630
+    },
+    {
+      "epoch": 0.18866666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 24.801041666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.186,
+      "loss/crossentropy": 2.030363583564758,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18138584941625596,
+      "step": 22640
+    },
+    {
+      "epoch": 0.18875,
+      "grad_norm": 30.125,
+      "grad_norm_var": 12.4134765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0666,
+      "loss/crossentropy": 1.9326512590050697,
+      "loss/hidden": 3.516796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19353458732366563,
+      "step": 22650
+    },
+    {
+      "epoch": 0.18883333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 26.887239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1368,
+      "loss/crossentropy": 1.9909636095166205,
+      "loss/hidden": 3.4421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19027703274041413,
+      "step": 22660
+    },
+    {
+      "epoch": 0.18891666666666668,
+      "grad_norm": 28.625,
+      "grad_norm_var": 11.929166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0782,
+      "loss/crossentropy": 1.7553676299750804,
+      "loss/hidden": 3.369921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18036161381751298,
+      "step": 22670
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 29.375,
+      "grad_norm_var": 22.187434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0577,
+      "loss/crossentropy": 1.8762573644518852,
+      "loss/hidden": 3.35625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19104025475680828,
+      "step": 22680
+    },
+    {
+      "epoch": 0.18908333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 19.361458333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1287,
+      "loss/crossentropy": 1.9260812915861607,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1766460970044136,
+      "step": 22690
+    },
+    {
+      "epoch": 0.18916666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 15.76015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0496,
+      "loss/crossentropy": 1.9115044504404068,
+      "loss/hidden": 3.30546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17360460506752134,
+      "step": 22700
+    },
+    {
+      "epoch": 0.18925,
+      "grad_norm": 37.25,
+      "grad_norm_var": 20.909830729166668,
+      "learning_rate": 0.0001,
+      "loss": 7.2371,
+      "loss/crossentropy": 2.0428875207901003,
+      "loss/hidden": 3.25546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17103400435298682,
+      "step": 22710
+    },
+    {
+      "epoch": 0.18933333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 10.848958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2291,
+      "loss/crossentropy": 2.08243068754673,
+      "loss/hidden": 3.323828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17972611039876937,
+      "step": 22720
+    },
+    {
+      "epoch": 0.18941666666666668,
+      "grad_norm": 31.375,
+      "grad_norm_var": 9.322916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.131,
+      "loss/crossentropy": 1.9026483923196793,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18596989195793867,
+      "step": 22730
+    },
+    {
+      "epoch": 0.1895,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.2676432291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0143,
+      "loss/crossentropy": 1.866141827404499,
+      "loss/hidden": 3.26640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19081275463104247,
+      "step": 22740
+    },
+    {
+      "epoch": 0.18958333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.4775390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1368,
+      "loss/crossentropy": 1.9652056694030762,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20104524549096822,
+      "step": 22750
+    },
+    {
+      "epoch": 0.18966666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.361458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0189,
+      "loss/crossentropy": 1.9203747354447842,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1821503307670355,
+      "step": 22760
+    },
+    {
+      "epoch": 0.18975,
+      "grad_norm": 30.375,
+      "grad_norm_var": 14.617708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0716,
+      "loss/crossentropy": 1.9551649019122124,
+      "loss/hidden": 3.3484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846253264695406,
+      "step": 22770
+    },
+    {
+      "epoch": 0.18983333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 2.436393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2189,
+      "loss/crossentropy": 1.970412702858448,
+      "loss/hidden": 3.246875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18902099095284938,
+      "step": 22780
+    },
+    {
+      "epoch": 0.18991666666666668,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.508333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1886,
+      "loss/crossentropy": 1.9552744179964066,
+      "loss/hidden": 3.336328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18718259762972594,
+      "step": 22790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 27.5,
+      "grad_norm_var": 8.207810176720307e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2032,
+      "loss/crossentropy": 1.9800579234957696,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18155794367194175,
+      "step": 22800
+    },
+    {
+      "epoch": 0.19008333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.139322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0441,
+      "loss/crossentropy": 1.9766518473625183,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18110963702201843,
+      "step": 22810
+    },
+    {
+      "epoch": 0.19016666666666668,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.2075980016212508e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1213,
+      "loss/crossentropy": 2.010078339278698,
+      "loss/hidden": 3.18828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1759355364367366,
+      "step": 22820
+    },
+    {
+      "epoch": 0.19025,
+      "grad_norm": 3204448256.0,
+      "grad_norm_var": 1.7319987723268086e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2402,
+      "loss/crossentropy": 1.9352431520819664,
+      "loss/hidden": 3.4015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20097873844206332,
+      "step": 22830
+    },
+    {
+      "epoch": 0.19033333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.417805263286305e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.043,
+      "loss/crossentropy": 1.9038178950548172,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17204765416681767,
+      "step": 22840
+    },
+    {
+      "epoch": 0.19041666666666668,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.533072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1196,
+      "loss/crossentropy": 1.844023048877716,
+      "loss/hidden": 3.45703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19323448538780214,
+      "step": 22850
+    },
+    {
+      "epoch": 0.1905,
+      "grad_norm": 25.5,
+      "grad_norm_var": 29.4384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0827,
+      "loss/crossentropy": 1.9496434807777405,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18853728882968426,
+      "step": 22860
+    },
+    {
+      "epoch": 0.19058333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.912434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2292,
+      "loss/crossentropy": 2.046557179093361,
+      "loss/hidden": 3.250390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18114321380853654,
+      "step": 22870
+    },
+    {
+      "epoch": 0.19066666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 49.46608072916667,
+      "learning_rate": 0.0001,
+      "loss": 7.077,
+      "loss/crossentropy": 2.0590321317315103,
+      "loss/hidden": 3.22890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17168218810111285,
+      "step": 22880
+    },
+    {
+      "epoch": 0.19075,
+      "grad_norm": 31.625,
+      "grad_norm_var": 22.447330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1501,
+      "loss/crossentropy": 1.958970120549202,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1676177540794015,
+      "step": 22890
+    },
+    {
+      "epoch": 0.19083333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 21.99140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0674,
+      "loss/crossentropy": 1.86142338514328,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18389992639422417,
+      "step": 22900
+    },
+    {
+      "epoch": 0.19091666666666668,
+      "grad_norm": 28.75,
+      "grad_norm_var": 17.1759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.117,
+      "loss/crossentropy": 2.1035741597414015,
+      "loss/hidden": 3.334765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18425506744533776,
+      "step": 22910
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 40.25,
+      "grad_norm_var": 8.537239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2247,
+      "loss/crossentropy": 1.9087388664484024,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1998944602906704,
+      "step": 22920
+    },
+    {
+      "epoch": 0.19108333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.975455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0875,
+      "loss/crossentropy": 1.998468678817153,
+      "loss/hidden": 3.3890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19227500013075768,
+      "step": 22930
+    },
+    {
+      "epoch": 0.19116666666666668,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.19765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2364,
+      "loss/crossentropy": 2.0985035330057142,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19702797383069992,
+      "step": 22940
+    },
+    {
+      "epoch": 0.19125,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.4697916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2475,
+      "loss/crossentropy": 1.952008032798767,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17378232274204491,
+      "step": 22950
+    },
+    {
+      "epoch": 0.19133333333333333,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.638997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2261,
+      "loss/crossentropy": 1.9933818072080611,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1859418511390686,
+      "step": 22960
+    },
+    {
+      "epoch": 0.19141666666666668,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.084141647314289e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1035,
+      "loss/crossentropy": 1.8527798980474472,
+      "loss/hidden": 3.260546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17463534083217383,
+      "step": 22970
+    },
+    {
+      "epoch": 0.1915,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.262239583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2276,
+      "loss/crossentropy": 1.9707329601049424,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19758185762912034,
+      "step": 22980
+    },
+    {
+      "epoch": 0.19158333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 5.1150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0783,
+      "loss/crossentropy": 1.795823697000742,
+      "loss/hidden": 3.40859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1741282207891345,
+      "step": 22990
+    },
+    {
+      "epoch": 0.19166666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.7894725423136774e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1981,
+      "loss/crossentropy": 1.9167943388223647,
+      "loss/hidden": 3.426171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18773870598524808,
+      "step": 23000
+    },
+    {
+      "epoch": 0.19175,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.789472540064482e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1615,
+      "loss/crossentropy": 1.967720426991582,
+      "loss/hidden": 3.39140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19772576363757252,
+      "step": 23010
+    },
+    {
+      "epoch": 0.19183333333333333,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.0895833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1917,
+      "loss/crossentropy": 1.9111167326569558,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1820769850164652,
+      "step": 23020
+    },
+    {
+      "epoch": 0.19191666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.87265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1311,
+      "loss/crossentropy": 1.9490511178970338,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19133517425507307,
+      "step": 23030
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 28.625,
+      "grad_norm_var": 4.198958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0937,
+      "loss/crossentropy": 1.9124677032232285,
+      "loss/hidden": 3.500390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2015688754618168,
+      "step": 23040
+    },
+    {
+      "epoch": 0.19208333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.9875,
+      "learning_rate": 0.0001,
+      "loss": 7.0889,
+      "loss/crossentropy": 1.9894707791507245,
+      "loss/hidden": 3.2859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17659089881926776,
+      "step": 23050
+    },
+    {
+      "epoch": 0.19216666666666668,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.578125,
+      "learning_rate": 0.0001,
+      "loss": 7.0669,
+      "loss/crossentropy": 1.986474347114563,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19992576763033867,
+      "step": 23060
+    },
+    {
+      "epoch": 0.19225,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.715559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1445,
+      "loss/crossentropy": 2.1189512491226195,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19541212003678082,
+      "step": 23070
+    },
+    {
+      "epoch": 0.19233333333333333,
+      "grad_norm": 42.75,
+      "grad_norm_var": 35.693684895833336,
+      "learning_rate": 0.0001,
+      "loss": 7.1767,
+      "loss/crossentropy": 2.051038008928299,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1860709184780717,
+      "step": 23080
+    },
+    {
+      "epoch": 0.19241666666666668,
+      "grad_norm": 28.25,
+      "grad_norm_var": 43.83958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0974,
+      "loss/crossentropy": 1.956616322696209,
+      "loss/hidden": 3.248046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17528742086142302,
+      "step": 23090
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.3625,
+      "learning_rate": 0.0001,
+      "loss": 7.1497,
+      "loss/crossentropy": 1.9142398178577422,
+      "loss/hidden": 3.28203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1778088515624404,
+      "step": 23100
+    },
+    {
+      "epoch": 0.19258333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.5181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0064,
+      "loss/crossentropy": 1.7949944779276847,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17725951205939056,
+      "step": 23110
+    },
+    {
+      "epoch": 0.19266666666666668,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.1884765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0897,
+      "loss/crossentropy": 1.8402652472257615,
+      "loss/hidden": 3.40390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20032060872763396,
+      "step": 23120
+    },
+    {
+      "epoch": 0.19275,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.9872395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.141,
+      "loss/crossentropy": 2.013263949751854,
+      "loss/hidden": 3.3875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18622555695474147,
+      "step": 23130
+    },
+    {
+      "epoch": 0.19283333333333333,
+      "grad_norm": 51.0,
+      "grad_norm_var": 1.0646087137853704e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.069,
+      "loss/crossentropy": 1.8066285640001296,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1797430979087949,
+      "step": 23140
+    },
+    {
+      "epoch": 0.19291666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.0646087128051615e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.0799,
+      "loss/crossentropy": 1.8005072593688964,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17465029526501893,
+      "step": 23150
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.2619140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1555,
+      "loss/crossentropy": 1.9099337190389634,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953379787504672,
+      "step": 23160
+    },
+    {
+      "epoch": 0.19308333333333333,
+      "grad_norm": 27.5,
+      "grad_norm_var": 1.1973681081194912e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1271,
+      "loss/crossentropy": 1.9409714683890342,
+      "loss/hidden": 3.363671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17417194042354822,
+      "step": 23170
+    },
+    {
+      "epoch": 0.19316666666666665,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.197368107081401e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1448,
+      "loss/crossentropy": 2.077232673764229,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18673932887613773,
+      "step": 23180
+    },
+    {
+      "epoch": 0.19325,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.393684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1395,
+      "loss/crossentropy": 2.0895217835903166,
+      "loss/hidden": 3.348046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841981064528227,
+      "step": 23190
+    },
+    {
+      "epoch": 0.19333333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.3603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1589,
+      "loss/crossentropy": 2.0004743456840517,
+      "loss/hidden": 3.3890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.200113507732749,
+      "step": 23200
+    },
+    {
+      "epoch": 0.19341666666666665,
+      "grad_norm": 28.25,
+      "grad_norm_var": 4.842643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0975,
+      "loss/crossentropy": 1.903498537838459,
+      "loss/hidden": 3.463671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19584673251956702,
+      "step": 23210
+    },
+    {
+      "epoch": 0.1935,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.7764973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1154,
+      "loss/crossentropy": 1.9488500237464905,
+      "loss/hidden": 3.421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1934487085789442,
+      "step": 23220
+    },
+    {
+      "epoch": 0.19358333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.3285807291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1419,
+      "loss/crossentropy": 1.9744546920061112,
+      "loss/hidden": 3.405859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18422229308634996,
+      "step": 23230
+    },
+    {
+      "epoch": 0.19366666666666665,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.98515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0563,
+      "loss/crossentropy": 1.8374863535165786,
+      "loss/hidden": 3.39375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17604625821113587,
+      "step": 23240
+    },
+    {
+      "epoch": 0.19375,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.9744140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0827,
+      "loss/crossentropy": 1.8100355714559555,
+      "loss/hidden": 3.455859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18963099773973227,
+      "step": 23250
+    },
+    {
+      "epoch": 0.19383333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.6931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0827,
+      "loss/crossentropy": 1.923319613188505,
+      "loss/hidden": 3.411328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18460817066952587,
+      "step": 23260
+    },
+    {
+      "epoch": 0.19391666666666665,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.158072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.183,
+      "loss/crossentropy": 2.001956608146429,
+      "loss/hidden": 3.31875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18520703986287118,
+      "step": 23270
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.7499348958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1266,
+      "loss/crossentropy": 1.9800696864724159,
+      "loss/hidden": 3.226171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1657674415037036,
+      "step": 23280
+    },
+    {
+      "epoch": 0.19408333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.5900390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2029,
+      "loss/crossentropy": 1.8416034236550332,
+      "loss/hidden": 3.3078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17592809088528155,
+      "step": 23290
+    },
+    {
+      "epoch": 0.19416666666666665,
+      "grad_norm": 30.625,
+      "grad_norm_var": 13.195247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1334,
+      "loss/crossentropy": 1.8498968943953513,
+      "loss/hidden": 3.293359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16715408097952605,
+      "step": 23300
+    },
+    {
+      "epoch": 0.19425,
+      "grad_norm": 28.0,
+      "grad_norm_var": 14.753580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0578,
+      "loss/crossentropy": 1.884926898777485,
+      "loss/hidden": 3.292578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16549744512885808,
+      "step": 23310
+    },
+    {
+      "epoch": 0.19433333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.721809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2334,
+      "loss/crossentropy": 2.00539381057024,
+      "loss/hidden": 3.346484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18194886576384306,
+      "step": 23320
+    },
+    {
+      "epoch": 0.19441666666666665,
+      "grad_norm": 28.375,
+      "grad_norm_var": 1.5343098958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0648,
+      "loss/crossentropy": 1.8296189844608306,
+      "loss/hidden": 3.43828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18926206212490798,
+      "step": 23330
+    },
+    {
+      "epoch": 0.1945,
+      "grad_norm": 29.75,
+      "grad_norm_var": 1.3228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1345,
+      "loss/crossentropy": 1.9275019466876984,
+      "loss/hidden": 3.342578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17628537472337485,
+      "step": 23340
+    },
+    {
+      "epoch": 0.19458333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.6747395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1341,
+      "loss/crossentropy": 1.9246577464044095,
+      "loss/hidden": 3.453515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17828861866146326,
+      "step": 23350
+    },
+    {
+      "epoch": 0.19466666666666665,
+      "grad_norm": 29.5,
+      "grad_norm_var": 3.8197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2637,
+      "loss/crossentropy": 1.9230331972241401,
+      "loss/hidden": 3.402734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19247340671718122,
+      "step": 23360
+    },
+    {
+      "epoch": 0.19475,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.0228515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1089,
+      "loss/crossentropy": 1.8846649765968322,
+      "loss/hidden": 3.275390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17611517924815417,
+      "step": 23370
+    },
+    {
+      "epoch": 0.19483333333333333,
+      "grad_norm": 30.375,
+      "grad_norm_var": 35.7072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1075,
+      "loss/crossentropy": 1.8721251353621482,
+      "loss/hidden": 3.440234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.179658723436296,
+      "step": 23380
+    },
+    {
+      "epoch": 0.19491666666666665,
+      "grad_norm": 34.0,
+      "grad_norm_var": 35.4197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1088,
+      "loss/crossentropy": 1.8221307545900345,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18375540971755983,
+      "step": 23390
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.3934895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1561,
+      "loss/crossentropy": 1.9274753212928772,
+      "loss/hidden": 3.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16697789635509253,
+      "step": 23400
+    },
+    {
+      "epoch": 0.19508333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.033333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0739,
+      "loss/crossentropy": 1.8710621312260627,
+      "loss/hidden": 3.404296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.179140686057508,
+      "step": 23410
+    },
+    {
+      "epoch": 0.19516666666666665,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.42265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2232,
+      "loss/crossentropy": 1.7723108381032944,
+      "loss/hidden": 3.373828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17169499825686216,
+      "step": 23420
+    },
+    {
+      "epoch": 0.19525,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.5572916666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.2783,
+      "loss/crossentropy": 1.9991915047168731,
+      "loss/hidden": 3.196875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17540288120508193,
+      "step": 23430
+    },
+    {
+      "epoch": 0.19533333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.5302083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2136,
+      "loss/crossentropy": 1.9357626229524612,
+      "loss/hidden": 3.412890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1945048970170319,
+      "step": 23440
+    },
+    {
+      "epoch": 0.19541666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.8625,
+      "learning_rate": 0.0001,
+      "loss": 7.1444,
+      "loss/crossentropy": 1.90290175229311,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20517104025930166,
+      "step": 23450
+    },
+    {
+      "epoch": 0.1955,
+      "grad_norm": 32.75,
+      "grad_norm_var": 33.24765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2197,
+      "loss/crossentropy": 2.007819926738739,
+      "loss/hidden": 3.338671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19451946951448917,
+      "step": 23460
+    },
+    {
+      "epoch": 0.19558333333333333,
+      "grad_norm": 29.375,
+      "grad_norm_var": 22.71015625,
+      "learning_rate": 0.0001,
+      "loss": 7.097,
+      "loss/crossentropy": 1.8512465931475162,
+      "loss/hidden": 3.36953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1956011440604925,
+      "step": 23470
+    },
+    {
+      "epoch": 0.19566666666666666,
+      "grad_norm": 28.125,
+      "grad_norm_var": 20.540559895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1445,
+      "loss/crossentropy": 1.853680929541588,
+      "loss/hidden": 3.372265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1656801901757717,
+      "step": 23480
+    },
+    {
+      "epoch": 0.19575,
+      "grad_norm": 29.125,
+      "grad_norm_var": 13.266080729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.153,
+      "loss/crossentropy": 1.7230026595294476,
+      "loss/hidden": 3.373828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15728132547810675,
+      "step": 23490
+    },
+    {
+      "epoch": 0.19583333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 12.170572916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0333,
+      "loss/crossentropy": 1.9370465368032455,
+      "loss/hidden": 3.247265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748226959258318,
+      "step": 23500
+    },
+    {
+      "epoch": 0.19591666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 7.194205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0415,
+      "loss/crossentropy": 1.8787044361233711,
+      "loss/hidden": 3.2671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1690024146810174,
+      "step": 23510
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.630208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1086,
+      "loss/crossentropy": 1.8538292430341243,
+      "loss/hidden": 3.376953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1688903359696269,
+      "step": 23520
+    },
+    {
+      "epoch": 0.19608333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 8.469791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0455,
+      "loss/crossentropy": 1.9870653390884399,
+      "loss/hidden": 3.494921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2119064349681139,
+      "step": 23530
+    },
+    {
+      "epoch": 0.19616666666666666,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.806705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1123,
+      "loss/crossentropy": 2.061756807565689,
+      "loss/hidden": 3.243359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17822170779109,
+      "step": 23540
+    },
+    {
+      "epoch": 0.19625,
+      "grad_norm": 28.75,
+      "grad_norm_var": 7.7916015625,
+      "learning_rate": 0.0001,
+      "loss": 7.028,
+      "loss/crossentropy": 1.9402014419436455,
+      "loss/hidden": 3.315234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17425654269754887,
+      "step": 23550
+    },
+    {
+      "epoch": 0.19633333333333333,
+      "grad_norm": 35.0,
+      "grad_norm_var": 10.795833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1189,
+      "loss/crossentropy": 2.0241565346717834,
+      "loss/hidden": 3.257421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17957337964326142,
+      "step": 23560
+    },
+    {
+      "epoch": 0.19641666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 3.095572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.124,
+      "loss/crossentropy": 2.088246059417725,
+      "loss/hidden": 3.298046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17748520839959384,
+      "step": 23570
+    },
+    {
+      "epoch": 0.1965,
+      "grad_norm": 32.25,
+      "grad_norm_var": 2.914322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1499,
+      "loss/crossentropy": 1.8447012618184089,
+      "loss/hidden": 3.348828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18112432733178138,
+      "step": 23580
+    },
+    {
+      "epoch": 0.19658333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.4275390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1401,
+      "loss/crossentropy": 1.8653517186641693,
+      "loss/hidden": 3.3953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19222842212766408,
+      "step": 23590
+    },
+    {
+      "epoch": 0.19666666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.159309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1769,
+      "loss/crossentropy": 2.002972137928009,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18960873428732156,
+      "step": 23600
+    },
+    {
+      "epoch": 0.19675,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.573372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.229,
+      "loss/crossentropy": 2.018272486329079,
+      "loss/hidden": 3.46484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18584080561995506,
+      "step": 23610
+    },
+    {
+      "epoch": 0.19683333333333333,
+      "grad_norm": 28.75,
+      "grad_norm_var": 10.698893229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0502,
+      "loss/crossentropy": 1.95031518638134,
+      "loss/hidden": 3.372265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18469448164105415,
+      "step": 23620
+    },
+    {
+      "epoch": 0.19691666666666666,
+      "grad_norm": 27.375,
+      "grad_norm_var": 33.94557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0532,
+      "loss/crossentropy": 1.8544048748910427,
+      "loss/hidden": 3.3890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17560015842318535,
+      "step": 23630
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 31.5,
+      "grad_norm_var": 20.636393229166668,
+      "learning_rate": 0.0001,
+      "loss": 7.1463,
+      "loss/crossentropy": 1.9214880645275116,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21309775728732347,
+      "step": 23640
+    },
+    {
+      "epoch": 0.19708333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.005989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1996,
+      "loss/crossentropy": 1.9441819377243519,
+      "loss/hidden": 3.419140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1897228709422052,
+      "step": 23650
+    },
+    {
+      "epoch": 0.19716666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.4660807291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2203,
+      "loss/crossentropy": 2.0642124876379966,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886973474174738,
+      "step": 23660
+    },
+    {
+      "epoch": 0.19725,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.7622395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.111,
+      "loss/crossentropy": 2.0513039007782936,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17776992712169887,
+      "step": 23670
+    },
+    {
+      "epoch": 0.19733333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.813997395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1475,
+      "loss/crossentropy": 1.8886918246746063,
+      "loss/hidden": 3.49921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893557518720627,
+      "step": 23680
+    },
+    {
+      "epoch": 0.19741666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 420.31223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1846,
+      "loss/crossentropy": 1.856307938694954,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18057366888970136,
+      "step": 23690
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 32.25,
+      "grad_norm_var": 68.44837239583333,
+      "learning_rate": 0.0001,
+      "loss": 7.2016,
+      "loss/crossentropy": 2.0575849920511247,
+      "loss/hidden": 3.33359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795424113050103,
+      "step": 23700
+    },
+    {
+      "epoch": 0.19758333333333333,
+      "grad_norm": 29.0,
+      "grad_norm_var": 111.88125,
+      "learning_rate": 0.0001,
+      "loss": 6.9827,
+      "loss/crossentropy": 1.951817236840725,
+      "loss/hidden": 3.353515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18553240727633238,
+      "step": 23710
+    },
+    {
+      "epoch": 0.19766666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.9389973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1103,
+      "loss/crossentropy": 1.9200981929898262,
+      "loss/hidden": 3.311328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18008090406656266,
+      "step": 23720
+    },
+    {
+      "epoch": 0.19775,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.5822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.089,
+      "loss/crossentropy": 1.9260015197098255,
+      "loss/hidden": 3.252734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18560872338712214,
+      "step": 23730
+    },
+    {
+      "epoch": 0.19783333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 10.95390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1658,
+      "loss/crossentropy": 1.9970799408853055,
+      "loss/hidden": 3.450390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2176352821290493,
+      "step": 23740
+    },
+    {
+      "epoch": 0.19791666666666666,
+      "grad_norm": 36.25,
+      "grad_norm_var": 18.259309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2144,
+      "loss/crossentropy": 1.8849347412586213,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18868403285741805,
+      "step": 23750
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 34.75,
+      "grad_norm_var": 16.668684895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0928,
+      "loss/crossentropy": 1.9804028034210206,
+      "loss/hidden": 3.3171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1838974365964532,
+      "step": 23760
+    },
+    {
+      "epoch": 0.19808333333333333,
+      "grad_norm": 42.5,
+      "grad_norm_var": 8.906044032299828e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0901,
+      "loss/crossentropy": 1.8409292250871658,
+      "loss/hidden": 3.23984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16110074240714312,
+      "step": 23770
+    },
+    {
+      "epoch": 0.19816666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 8.906044031867291e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0656,
+      "loss/crossentropy": 1.843617521226406,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18220640923827885,
+      "step": 23780
+    },
+    {
+      "epoch": 0.19825,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.9145182291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0182,
+      "loss/crossentropy": 1.967862430214882,
+      "loss/hidden": 3.4046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843209033831954,
+      "step": 23790
+    },
+    {
+      "epoch": 0.19833333333333333,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.3655598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1121,
+      "loss/crossentropy": 1.9053588643670083,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17454120609909296,
+      "step": 23800
+    },
+    {
+      "epoch": 0.19841666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 8.109830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1376,
+      "loss/crossentropy": 1.8701864689588548,
+      "loss/hidden": 3.306640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18011666256934405,
+      "step": 23810
+    },
+    {
+      "epoch": 0.1985,
+      "grad_norm": 29.875,
+      "grad_norm_var": 7.37265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1353,
+      "loss/crossentropy": 1.9539224848151207,
+      "loss/hidden": 3.24609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17535881465300918,
+      "step": 23820
+    },
+    {
+      "epoch": 0.19858333333333333,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.714583333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1497,
+      "loss/crossentropy": 1.9830836653709412,
+      "loss/hidden": 3.426171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19156263265758752,
+      "step": 23830
+    },
+    {
+      "epoch": 0.19866666666666666,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.341080729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2067,
+      "loss/crossentropy": 1.938089296966791,
+      "loss/hidden": 3.35234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19800030943006278,
+      "step": 23840
+    },
+    {
+      "epoch": 0.19875,
+      "grad_norm": 31.75,
+      "grad_norm_var": 24.0181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1203,
+      "loss/crossentropy": 2.0104876697063445,
+      "loss/hidden": 3.4578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18114052731543778,
+      "step": 23850
+    },
+    {
+      "epoch": 0.19883333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 7.542708333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0638,
+      "loss/crossentropy": 1.859134241938591,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17276255525648593,
+      "step": 23860
+    },
+    {
+      "epoch": 0.19891666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 5.383333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0868,
+      "loss/crossentropy": 1.8435207307338715,
+      "loss/hidden": 3.35703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18017004895955324,
+      "step": 23870
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 31.375,
+      "grad_norm_var": 8.4900390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1717,
+      "loss/crossentropy": 2.088299649953842,
+      "loss/hidden": 3.3109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1837046964094043,
+      "step": 23880
+    },
+    {
+      "epoch": 0.19908333333333333,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.7177083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1603,
+      "loss/crossentropy": 2.0537067532539366,
+      "loss/hidden": 3.447265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2020148303359747,
+      "step": 23890
+    },
+    {
+      "epoch": 0.19916666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 3.3707682291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0853,
+      "loss/crossentropy": 1.9481948032975196,
+      "loss/hidden": 3.408203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18937265686690807,
+      "step": 23900
+    },
+    {
+      "epoch": 0.19925,
+      "grad_norm": 29.625,
+      "grad_norm_var": 200.7806640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2249,
+      "loss/crossentropy": 2.009596602618694,
+      "loss/hidden": 3.425390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19282790664583443,
+      "step": 23910
+    },
+    {
+      "epoch": 0.19933333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 301.1302083333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1286,
+      "loss/crossentropy": 1.974546992778778,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18816668055951596,
+      "step": 23920
+    },
+    {
+      "epoch": 0.19941666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.413541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0799,
+      "loss/crossentropy": 1.9964633017778397,
+      "loss/hidden": 3.21875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1824707703664899,
+      "step": 23930
+    },
+    {
+      "epoch": 0.1995,
+      "grad_norm": 33.25,
+      "grad_norm_var": 3.5927083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2296,
+      "loss/crossentropy": 1.8595583960413933,
+      "loss/hidden": 3.38671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18287263475358487,
+      "step": 23940
+    },
+    {
+      "epoch": 0.19958333333333333,
+      "grad_norm": 29.875,
+      "grad_norm_var": 2.8139973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1691,
+      "loss/crossentropy": 1.9636321395635605,
+      "loss/hidden": 3.475,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2030277678743005,
+      "step": 23950
+    },
+    {
+      "epoch": 0.19966666666666666,
+      "grad_norm": 39.75,
+      "grad_norm_var": 115.03587239583334,
+      "learning_rate": 0.0001,
+      "loss": 7.1208,
+      "loss/crossentropy": 1.8875990755856038,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17906233705580235,
+      "step": 23960
+    },
+    {
+      "epoch": 0.19975,
+      "grad_norm": 30.375,
+      "grad_norm_var": 113.53723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1603,
+      "loss/crossentropy": 1.9678095310926438,
+      "loss/hidden": 3.454296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19034356717020273,
+      "step": 23970
+    },
+    {
+      "epoch": 0.19983333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.0733723958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1142,
+      "loss/crossentropy": 2.028428369760513,
+      "loss/hidden": 3.271484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18276078328490258,
+      "step": 23980
+    },
+    {
+      "epoch": 0.19991666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.3455729166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0855,
+      "loss/crossentropy": 1.9124814704060555,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1763345655053854,
+      "step": 23990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.3619140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1424,
+      "loss/crossentropy": 1.8087715968489646,
+      "loss/hidden": 3.415234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18131402209401132,
+      "step": 24000
+    },
+    {
+      "epoch": 0.20008333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.8533923314597894e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1307,
+      "loss/crossentropy": 1.8288925424218179,
+      "loss/hidden": 3.34140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1789520868100226,
+      "step": 24010
+    },
+    {
+      "epoch": 0.20016666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 6.5009765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1366,
+      "loss/crossentropy": 1.974527582526207,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19389855545014142,
+      "step": 24020
+    },
+    {
+      "epoch": 0.20025,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.9166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.2266,
+      "loss/crossentropy": 1.917670765519142,
+      "loss/hidden": 3.27890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17423527911305428,
+      "step": 24030
+    },
+    {
+      "epoch": 0.20033333333333334,
+      "grad_norm": 27.75,
+      "grad_norm_var": 2.2030598958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0352,
+      "loss/crossentropy": 2.0017741471529007,
+      "loss/hidden": 3.253515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1740729859098792,
+      "step": 24040
+    },
+    {
+      "epoch": 0.20041666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.3728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.9003146201372147,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1825672946870327,
+      "step": 24050
+    },
+    {
+      "epoch": 0.2005,
+      "grad_norm": 31.75,
+      "grad_norm_var": 1.6559895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1441,
+      "loss/crossentropy": 1.90264290869236,
+      "loss/hidden": 3.36953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18069742880761624,
+      "step": 24060
+    },
+    {
+      "epoch": 0.20058333333333334,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.4447265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0894,
+      "loss/crossentropy": 1.9436378315091134,
+      "loss/hidden": 3.412890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17049392499029636,
+      "step": 24070
+    },
+    {
+      "epoch": 0.20066666666666666,
+      "grad_norm": 28.75,
+      "grad_norm_var": 2.4759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1527,
+      "loss/crossentropy": 1.9713781118392943,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1764097599312663,
+      "step": 24080
+    },
+    {
+      "epoch": 0.20075,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.6962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1332,
+      "loss/crossentropy": 1.9597785338759421,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18891141805797815,
+      "step": 24090
+    },
+    {
+      "epoch": 0.20083333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.575455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1322,
+      "loss/crossentropy": 2.08375011831522,
+      "loss/hidden": 3.369140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20871647130697965,
+      "step": 24100
+    },
+    {
+      "epoch": 0.20091666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.2768229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1177,
+      "loss/crossentropy": 1.8224686101078986,
+      "loss/hidden": 3.31796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16840583961457015,
+      "step": 24110
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 33.25,
+      "grad_norm_var": 0.9809895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1248,
+      "loss/crossentropy": 2.027084432542324,
+      "loss/hidden": 3.31171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19435680489987134,
+      "step": 24120
+    },
+    {
+      "epoch": 0.20108333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 2.5994140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1273,
+      "loss/crossentropy": 1.8061633482575417,
+      "loss/hidden": 3.3609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17516813725233077,
+      "step": 24130
+    },
+    {
+      "epoch": 0.20116666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.630143229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1682,
+      "loss/crossentropy": 1.9366648003458977,
+      "loss/hidden": 3.32109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16930126007646323,
+      "step": 24140
+    },
+    {
+      "epoch": 0.20125,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.858268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1846,
+      "loss/crossentropy": 2.0362853042781355,
+      "loss/hidden": 3.46640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2465898199006915,
+      "step": 24150
+    },
+    {
+      "epoch": 0.20133333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.44765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1306,
+      "loss/crossentropy": 1.9558606788516044,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18335945438593626,
+      "step": 24160
+    },
+    {
+      "epoch": 0.20141666666666666,
+      "grad_norm": 27.5,
+      "grad_norm_var": 5.218489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0197,
+      "loss/crossentropy": 1.9337234787642956,
+      "loss/hidden": 3.22578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16627709660679102,
+      "step": 24170
+    },
+    {
+      "epoch": 0.2015,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.1989583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1687,
+      "loss/crossentropy": 1.9588148534297942,
+      "loss/hidden": 3.390234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18814067970961332,
+      "step": 24180
+    },
+    {
+      "epoch": 0.20158333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 1.5103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0557,
+      "loss/crossentropy": 1.9633461952209472,
+      "loss/hidden": 3.224609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18487815856933593,
+      "step": 24190
+    },
+    {
+      "epoch": 0.20166666666666666,
+      "grad_norm": 28.5,
+      "grad_norm_var": 7.471809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2808,
+      "loss/crossentropy": 2.0600421875715256,
+      "loss/hidden": 3.40625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19816921409219504,
+      "step": 24200
+    },
+    {
+      "epoch": 0.20175,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.831184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0446,
+      "loss/crossentropy": 1.8207935333251952,
+      "loss/hidden": 3.31328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1720402402803302,
+      "step": 24210
+    },
+    {
+      "epoch": 0.20183333333333334,
+      "grad_norm": 27.375,
+      "grad_norm_var": 5.836393229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1984,
+      "loss/crossentropy": 1.9900528326630593,
+      "loss/hidden": 3.3484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17300470899790527,
+      "step": 24220
+    },
+    {
+      "epoch": 0.20191666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.924934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3198,
+      "loss/crossentropy": 2.0276302605867387,
+      "loss/hidden": 3.416015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19135519210249186,
+      "step": 24230
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.8518229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2097,
+      "loss/crossentropy": 1.9625100061297416,
+      "loss/hidden": 3.35390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20093309227377176,
+      "step": 24240
+    },
+    {
+      "epoch": 0.20208333333333334,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.0150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1708,
+      "loss/crossentropy": 2.0575515031814575,
+      "loss/hidden": 3.302734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17778213545680047,
+      "step": 24250
+    },
+    {
+      "epoch": 0.20216666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 98.7197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1914,
+      "loss/crossentropy": 2.0048379465937614,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19123283866792917,
+      "step": 24260
+    },
+    {
+      "epoch": 0.20225,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.4343098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1246,
+      "loss/crossentropy": 2.0060950994491575,
+      "loss/hidden": 3.3515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19556776378303767,
+      "step": 24270
+    },
+    {
+      "epoch": 0.20233333333333334,
+      "grad_norm": 28.75,
+      "grad_norm_var": 3.414322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1205,
+      "loss/crossentropy": 1.950710302591324,
+      "loss/hidden": 3.4125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18979913424700498,
+      "step": 24280
+    },
+    {
+      "epoch": 0.20241666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 43.37708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0519,
+      "loss/crossentropy": 2.118720631301403,
+      "loss/hidden": 3.240625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18266464760527015,
+      "step": 24290
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.1791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1763,
+      "loss/crossentropy": 1.9891113504767417,
+      "loss/hidden": 3.428125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18398433905094863,
+      "step": 24300
+    },
+    {
+      "epoch": 0.20258333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 6.159375,
+      "learning_rate": 0.0001,
+      "loss": 7.1656,
+      "loss/crossentropy": 1.984175093472004,
+      "loss/hidden": 3.351171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18647950924932957,
+      "step": 24310
+    },
+    {
+      "epoch": 0.20266666666666666,
+      "grad_norm": 30.25,
+      "grad_norm_var": 6.0072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1409,
+      "loss/crossentropy": 1.9776911959052086,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1836971465498209,
+      "step": 24320
+    },
+    {
+      "epoch": 0.20275,
+      "grad_norm": 30.375,
+      "grad_norm_var": 9.474934895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1711,
+      "loss/crossentropy": 1.96462944149971,
+      "loss/hidden": 3.35859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1854661911725998,
+      "step": 24330
+    },
+    {
+      "epoch": 0.20283333333333334,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.0104166666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1952,
+      "loss/crossentropy": 2.0494933426380157,
+      "loss/hidden": 3.32578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19954477138817311,
+      "step": 24340
+    },
+    {
+      "epoch": 0.20291666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.537434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1341,
+      "loss/crossentropy": 2.041038918495178,
+      "loss/hidden": 3.349609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18079897109419107,
+      "step": 24350
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 37.5,
+      "grad_norm_var": 5.006705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.123,
+      "loss/crossentropy": 1.8938631147146225,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17263554744422435,
+      "step": 24360
+    },
+    {
+      "epoch": 0.20308333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.4244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2849,
+      "loss/crossentropy": 1.9228694915771485,
+      "loss/hidden": 3.43125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21245141606777906,
+      "step": 24370
+    },
+    {
+      "epoch": 0.20316666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 3.9400390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1392,
+      "loss/crossentropy": 2.0178455233573915,
+      "loss/hidden": 3.4,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953028045594693,
+      "step": 24380
+    },
+    {
+      "epoch": 0.20325,
+      "grad_norm": 31.125,
+      "grad_norm_var": 3.8046223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2986,
+      "loss/crossentropy": 1.935243132710457,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17596080116927623,
+      "step": 24390
+    },
+    {
+      "epoch": 0.20333333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 13.487434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2366,
+      "loss/crossentropy": 2.03190698325634,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18737903684377671,
+      "step": 24400
+    },
+    {
+      "epoch": 0.20341666666666666,
+      "grad_norm": 31.5,
+      "grad_norm_var": 12.267643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1426,
+      "loss/crossentropy": 1.9489700302481652,
+      "loss/hidden": 3.276171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17259762901812792,
+      "step": 24410
+    },
+    {
+      "epoch": 0.2035,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.7369140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2323,
+      "loss/crossentropy": 1.9775184512138366,
+      "loss/hidden": 3.33203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.187724363617599,
+      "step": 24420
+    },
+    {
+      "epoch": 0.20358333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.2358723958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2819,
+      "loss/crossentropy": 1.8757319539785384,
+      "loss/hidden": 3.40390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19592621624469758,
+      "step": 24430
+    },
+    {
+      "epoch": 0.20366666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.2728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1381,
+      "loss/crossentropy": 1.9956869542598725,
+      "loss/hidden": 3.337109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20254410691559316,
+      "step": 24440
+    },
+    {
+      "epoch": 0.20375,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.367204476918782e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1574,
+      "loss/crossentropy": 2.094840782880783,
+      "loss/hidden": 3.274609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17596059069037437,
+      "step": 24450
+    },
+    {
+      "epoch": 0.20383333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.6934895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1418,
+      "loss/crossentropy": 1.9458136305212974,
+      "loss/hidden": 3.35,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19071349818259478,
+      "step": 24460
+    },
+    {
+      "epoch": 0.20391666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.923958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2266,
+      "loss/crossentropy": 2.005078434944153,
+      "loss/hidden": 3.25703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18951187189668417,
+      "step": 24470
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 28.375,
+      "grad_norm_var": 4.108072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.083,
+      "loss/crossentropy": 1.9002643913030624,
+      "loss/hidden": 3.313671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17683612387627362,
+      "step": 24480
+    },
+    {
+      "epoch": 0.20408333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.5426432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0417,
+      "loss/crossentropy": 2.1399880915880205,
+      "loss/hidden": 3.458984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19764817208051683,
+      "step": 24490
+    },
+    {
+      "epoch": 0.20416666666666666,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.312434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0894,
+      "loss/crossentropy": 1.7228893265128136,
+      "loss/hidden": 3.260546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1478549338877201,
+      "step": 24500
+    },
+    {
+      "epoch": 0.20425,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.7895182291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0956,
+      "loss/crossentropy": 2.0117953419685364,
+      "loss/hidden": 3.362890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19534504376351833,
+      "step": 24510
+    },
+    {
+      "epoch": 0.20433333333333334,
+      "grad_norm": 34.5,
+      "grad_norm_var": 3.65625,
+      "learning_rate": 0.0001,
+      "loss": 7.234,
+      "loss/crossentropy": 1.9248805910348892,
+      "loss/hidden": 3.439453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1795705934986472,
+      "step": 24520
+    },
+    {
+      "epoch": 0.20441666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 8.215625,
+      "learning_rate": 0.0001,
+      "loss": 7.221,
+      "loss/crossentropy": 1.9820847675204276,
+      "loss/hidden": 3.422265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1918133044615388,
+      "step": 24530
+    },
+    {
+      "epoch": 0.2045,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.888997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1258,
+      "loss/crossentropy": 1.7113145127892495,
+      "loss/hidden": 3.457421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16541892644017936,
+      "step": 24540
+    },
+    {
+      "epoch": 0.20458333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.87265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 1.820174442231655,
+      "loss/hidden": 3.424609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18078403491526843,
+      "step": 24550
+    },
+    {
+      "epoch": 0.20466666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 3.95625,
+      "learning_rate": 0.0001,
+      "loss": 7.1113,
+      "loss/crossentropy": 1.8690563261508941,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1913584502413869,
+      "step": 24560
+    },
+    {
+      "epoch": 0.20475,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.9879557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1441,
+      "loss/crossentropy": 2.0876524686813354,
+      "loss/hidden": 3.276953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1778487155213952,
+      "step": 24570
+    },
+    {
+      "epoch": 0.20483333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.4436848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1073,
+      "loss/crossentropy": 1.9638664782047273,
+      "loss/hidden": 3.302734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17659037169069053,
+      "step": 24580
+    },
+    {
+      "epoch": 0.20491666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.763541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1646,
+      "loss/crossentropy": 2.020820362865925,
+      "loss/hidden": 3.332421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19661013316363096,
+      "step": 24590
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 29.0,
+      "grad_norm_var": 1.5447916666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.08,
+      "loss/crossentropy": 1.7996208399534226,
+      "loss/hidden": 3.488671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17888092510402204,
+      "step": 24600
+    },
+    {
+      "epoch": 0.20508333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.1229166666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1831,
+      "loss/crossentropy": 1.8567791432142258,
+      "loss/hidden": 3.3375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18007622212171553,
+      "step": 24610
+    },
+    {
+      "epoch": 0.20516666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.985384794749667e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2379,
+      "loss/crossentropy": 1.8702650167047978,
+      "loss/hidden": 3.319921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1779907383956015,
+      "step": 24620
+    },
+    {
+      "epoch": 0.20525,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.985384782821765e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2801,
+      "loss/crossentropy": 1.9305368185043335,
+      "loss/hidden": 3.480859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033628160133958,
+      "step": 24630
+    },
+    {
+      "epoch": 0.20533333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 15.673893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0719,
+      "loss/crossentropy": 1.8519299775362015,
+      "loss/hidden": 3.2703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18226739028468728,
+      "step": 24640
+    },
+    {
+      "epoch": 0.20541666666666666,
+      "grad_norm": 31.75,
+      "grad_norm_var": 6.764518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1595,
+      "loss/crossentropy": 1.8040445856750011,
+      "loss/hidden": 3.34921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17378551475703716,
+      "step": 24650
+    },
+    {
+      "epoch": 0.2055,
+      "grad_norm": 30.625,
+      "grad_norm_var": 6.4931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0883,
+      "loss/crossentropy": 1.8724176734685898,
+      "loss/hidden": 3.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18044158741831778,
+      "step": 24660
+    },
+    {
+      "epoch": 0.20558333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.424739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1104,
+      "loss/crossentropy": 2.01869605332613,
+      "loss/hidden": 3.310546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19050698727369308,
+      "step": 24670
+    },
+    {
+      "epoch": 0.20566666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 5.374934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 2.0353800505399704,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18586320634931325,
+      "step": 24680
+    },
+    {
+      "epoch": 0.20575,
+      "grad_norm": 30.0,
+      "grad_norm_var": 14.883333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1338,
+      "loss/crossentropy": 1.9757086634635925,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18433211985975503,
+      "step": 24690
+    },
+    {
+      "epoch": 0.20583333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.80390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1675,
+      "loss/crossentropy": 1.8736904069781304,
+      "loss/hidden": 3.364453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18352918177843094,
+      "step": 24700
+    },
+    {
+      "epoch": 0.20591666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 5.710416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1699,
+      "loss/crossentropy": 2.0840922206640244,
+      "loss/hidden": 3.309765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1845611134544015,
+      "step": 24710
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 34.75,
+      "grad_norm_var": 4.999739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1381,
+      "loss/crossentropy": 2.022069463133812,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18830919321626424,
+      "step": 24720
+    },
+    {
+      "epoch": 0.20608333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 8.207810171020248e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1988,
+      "loss/crossentropy": 2.071345128118992,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18941074553877116,
+      "step": 24730
+    },
+    {
+      "epoch": 0.20616666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.20781014497362e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1314,
+      "loss/crossentropy": 1.8681358382105828,
+      "loss/hidden": 3.308203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1860715825110674,
+      "step": 24740
+    },
+    {
+      "epoch": 0.20625,
+      "grad_norm": 31.875,
+      "grad_norm_var": 66.82337239583333,
+      "learning_rate": 0.0001,
+      "loss": 7.2221,
+      "loss/crossentropy": 1.9414967089891433,
+      "loss/hidden": 3.398828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20399159397929906,
+      "step": 24750
+    },
+    {
+      "epoch": 0.20633333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 11.073893229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1983,
+      "loss/crossentropy": 2.012427291274071,
+      "loss/hidden": 3.319921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18468363527208567,
+      "step": 24760
+    },
+    {
+      "epoch": 0.20641666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 16.555208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1121,
+      "loss/crossentropy": 1.9452892273664475,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865605752915144,
+      "step": 24770
+    },
+    {
+      "epoch": 0.2065,
+      "grad_norm": 32.25,
+      "grad_norm_var": 372.00514322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.1066,
+      "loss/crossentropy": 2.010504364967346,
+      "loss/hidden": 3.419140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.204042093642056,
+      "step": 24780
+    },
+    {
+      "epoch": 0.20658333333333334,
+      "grad_norm": 35.0,
+      "grad_norm_var": 373.59055989583334,
+      "learning_rate": 0.0001,
+      "loss": 7.0687,
+      "loss/crossentropy": 1.9241214014589787,
+      "loss/hidden": 3.345703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17118938229978084,
+      "step": 24790
+    },
+    {
+      "epoch": 0.20666666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.832216992890403e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1991,
+      "loss/crossentropy": 2.064511862397194,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18973149117082358,
+      "step": 24800
+    },
+    {
+      "epoch": 0.20675,
+      "grad_norm": 32.75,
+      "grad_norm_var": 7.832216989608534e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1622,
+      "loss/crossentropy": 1.9538428053259849,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1822197688743472,
+      "step": 24810
+    },
+    {
+      "epoch": 0.20683333333333334,
+      "grad_norm": 31.5,
+      "grad_norm_var": 1.3143229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.111,
+      "loss/crossentropy": 2.044131575524807,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18155855555087327,
+      "step": 24820
+    },
+    {
+      "epoch": 0.20691666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 14.174739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1572,
+      "loss/crossentropy": 1.8844437763094901,
+      "loss/hidden": 3.553515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20544027350842953,
+      "step": 24830
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 29.0,
+      "grad_norm_var": 25.5375,
+      "learning_rate": 0.0001,
+      "loss": 7.042,
+      "loss/crossentropy": 1.958268365263939,
+      "loss/hidden": 3.18671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16223876178264618,
+      "step": 24840
+    },
+    {
+      "epoch": 0.20708333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.277083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1529,
+      "loss/crossentropy": 2.0024541571736334,
+      "loss/hidden": 3.26875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18734329380095005,
+      "step": 24850
+    },
+    {
+      "epoch": 0.20716666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.601822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0551,
+      "loss/crossentropy": 1.9296492159366607,
+      "loss/hidden": 3.295703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1752838682383299,
+      "step": 24860
+    },
+    {
+      "epoch": 0.20725,
+      "grad_norm": 32.0,
+      "grad_norm_var": 3.658333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1374,
+      "loss/crossentropy": 1.982964886724949,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19609220344573258,
+      "step": 24870
+    },
+    {
+      "epoch": 0.20733333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 3.73515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0746,
+      "loss/crossentropy": 1.976727682352066,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18361312262713908,
+      "step": 24880
+    },
+    {
+      "epoch": 0.20741666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.2166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0456,
+      "loss/crossentropy": 1.9868737056851387,
+      "loss/hidden": 3.365625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19255165457725526,
+      "step": 24890
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.145247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1099,
+      "loss/crossentropy": 2.035132426023483,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18659703060984612,
+      "step": 24900
+    },
+    {
+      "epoch": 0.20758333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.54765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1294,
+      "loss/crossentropy": 1.8259663090109826,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.181023951433599,
+      "step": 24910
+    },
+    {
+      "epoch": 0.20766666666666667,
+      "grad_norm": 31.125,
+      "grad_norm_var": 5.623372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1055,
+      "loss/crossentropy": 1.9692690193653106,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.194866755977273,
+      "step": 24920
+    },
+    {
+      "epoch": 0.20775,
+      "grad_norm": 3288334336.0,
+      "grad_norm_var": 6.758214062680572e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1842,
+      "loss/crossentropy": 1.9793336600065232,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17870378121733665,
+      "step": 24930
+    },
+    {
+      "epoch": 0.20783333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 6.758214050212306e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0963,
+      "loss/crossentropy": 2.0708570897579195,
+      "loss/hidden": 3.315625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1977537278085947,
+      "step": 24940
+    },
+    {
+      "epoch": 0.20791666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 11.88515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0924,
+      "loss/crossentropy": 1.9046525001525878,
+      "loss/hidden": 3.3203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18452330026775599,
+      "step": 24950
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 28.875,
+      "grad_norm_var": 8.989518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0409,
+      "loss/crossentropy": 1.9976745814085006,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18242889642715454,
+      "step": 24960
+    },
+    {
+      "epoch": 0.20808333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 14.1962890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1514,
+      "loss/crossentropy": 1.9621400028467177,
+      "loss/hidden": 3.2390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17213822351768612,
+      "step": 24970
+    },
+    {
+      "epoch": 0.20816666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 10.27265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1193,
+      "loss/crossentropy": 1.9589134439826013,
+      "loss/hidden": 3.309765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19393112175166607,
+      "step": 24980
+    },
+    {
+      "epoch": 0.20825,
+      "grad_norm": 43.0,
+      "grad_norm_var": 8.375522288491388e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2098,
+      "loss/crossentropy": 1.9795193165540694,
+      "loss/hidden": 3.246875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17827889509499073,
+      "step": 24990
+    },
+    {
+      "epoch": 0.20833333333333334,
+      "grad_norm": 28.0,
+      "grad_norm_var": 8.375522275709247e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2248,
+      "loss/crossentropy": 1.948588028550148,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922807006165385,
+      "step": 25000
+    },
+    {
+      "epoch": 0.20841666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 87.18098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1852,
+      "loss/crossentropy": 1.9399778559803962,
+      "loss/hidden": 3.368359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19269431456923486,
+      "step": 25010
+    },
+    {
+      "epoch": 0.2085,
+      "grad_norm": 29.125,
+      "grad_norm_var": 92.69270833333333,
+      "learning_rate": 0.0001,
+      "loss": 6.9931,
+      "loss/crossentropy": 1.9720507934689522,
+      "loss/hidden": 3.17265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16831290321424602,
+      "step": 25020
+    },
+    {
+      "epoch": 0.20858333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 4.2853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0702,
+      "loss/crossentropy": 1.9350756019353867,
+      "loss/hidden": 3.2953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17938519679009915,
+      "step": 25030
+    },
+    {
+      "epoch": 0.20866666666666667,
+      "grad_norm": 39.0,
+      "grad_norm_var": 7.7431640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 2.0559972673654556,
+      "loss/hidden": 3.428515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20921088419854642,
+      "step": 25040
+    },
+    {
+      "epoch": 0.20875,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.03515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0779,
+      "loss/crossentropy": 1.829003243893385,
+      "loss/hidden": 3.248046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17074189763516187,
+      "step": 25050
+    },
+    {
+      "epoch": 0.20883333333333334,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.408072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1134,
+      "loss/crossentropy": 1.8196294449269772,
+      "loss/hidden": 3.508203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19238717798143626,
+      "step": 25060
+    },
+    {
+      "epoch": 0.20891666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.7072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0227,
+      "loss/crossentropy": 1.9454280734062195,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17801475822925567,
+      "step": 25070
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 32.25,
+      "grad_norm_var": 5.513541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.117,
+      "loss/crossentropy": 1.9178467139601707,
+      "loss/hidden": 3.265234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19138377234339715,
+      "step": 25080
+    },
+    {
+      "epoch": 0.20908333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 31.24140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0378,
+      "loss/crossentropy": 1.9574884802103043,
+      "loss/hidden": 3.513671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19021328445523977,
+      "step": 25090
+    },
+    {
+      "epoch": 0.20916666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.854622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2256,
+      "loss/crossentropy": 1.9145000010728837,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17541245128959418,
+      "step": 25100
+    },
+    {
+      "epoch": 0.20925,
+      "grad_norm": 31.625,
+      "grad_norm_var": 10.954166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1853,
+      "loss/crossentropy": 2.0749082922935487,
+      "loss/hidden": 3.399609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855158470571041,
+      "step": 25110
+    },
+    {
+      "epoch": 0.20933333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 8.583268229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1697,
+      "loss/crossentropy": 1.8130038946866989,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17791857589036225,
+      "step": 25120
+    },
+    {
+      "epoch": 0.20941666666666667,
+      "grad_norm": 30.375,
+      "grad_norm_var": 7.321163998892655e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2162,
+      "loss/crossentropy": 1.9671566575765609,
+      "loss/hidden": 3.368359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20469687711447476,
+      "step": 25130
+    },
+    {
+      "epoch": 0.2095,
+      "grad_norm": 32.5,
+      "grad_norm_var": 7.321164002992588e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1738,
+      "loss/crossentropy": 1.9266522794961929,
+      "loss/hidden": 3.48203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19139146395027637,
+      "step": 25140
+    },
+    {
+      "epoch": 0.20958333333333334,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.9009765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1491,
+      "loss/crossentropy": 1.8566777423024177,
+      "loss/hidden": 3.3484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17281682156026362,
+      "step": 25150
+    },
+    {
+      "epoch": 0.20966666666666667,
+      "grad_norm": 33.75,
+      "grad_norm_var": 4.073372395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0843,
+      "loss/crossentropy": 1.8607637360692024,
+      "loss/hidden": 3.37265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1887682205066085,
+      "step": 25160
+    },
+    {
+      "epoch": 0.20975,
+      "grad_norm": 28.5,
+      "grad_norm_var": 5.263997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0454,
+      "loss/crossentropy": 1.9886658847332002,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19260253924876453,
+      "step": 25170
+    },
+    {
+      "epoch": 0.20983333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 2.8916666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0784,
+      "loss/crossentropy": 1.9509775355458259,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16865229979157448,
+      "step": 25180
+    },
+    {
+      "epoch": 0.20991666666666667,
+      "grad_norm": 31.5,
+      "grad_norm_var": 3.2143229166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0934,
+      "loss/crossentropy": 1.9439710229635239,
+      "loss/hidden": 3.39609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19392619840800762,
+      "step": 25190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 34.25,
+      "grad_norm_var": 74.621875,
+      "learning_rate": 0.0001,
+      "loss": 7.0778,
+      "loss/crossentropy": 1.8896284699440002,
+      "loss/hidden": 3.409765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19547482319176196,
+      "step": 25200
+    },
+    {
+      "epoch": 0.21008333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 75.67265625,
+      "learning_rate": 0.0001,
+      "loss": 7.092,
+      "loss/crossentropy": 1.9811456218361854,
+      "loss/hidden": 3.29609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18618320431560278,
+      "step": 25210
+    },
+    {
+      "epoch": 0.21016666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 40.146875,
+      "learning_rate": 0.0001,
+      "loss": 7.1346,
+      "loss/crossentropy": 2.0292749404907227,
+      "loss/hidden": 3.213671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18399430457502602,
+      "step": 25220
+    },
+    {
+      "epoch": 0.21025,
+      "grad_norm": 30.625,
+      "grad_norm_var": 28.217643229166665,
+      "learning_rate": 0.0001,
+      "loss": 7.1022,
+      "loss/crossentropy": 1.8599492721259594,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17453321143984796,
+      "step": 25230
+    },
+    {
+      "epoch": 0.21033333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 25.283268229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0385,
+      "loss/crossentropy": 1.8925228387117385,
+      "loss/hidden": 3.189453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16890909858047962,
+      "step": 25240
+    },
+    {
+      "epoch": 0.21041666666666667,
+      "grad_norm": 37.25,
+      "grad_norm_var": 47.6759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0845,
+      "loss/crossentropy": 1.966729325056076,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932036241516471,
+      "step": 25250
+    },
+    {
+      "epoch": 0.2105,
+      "grad_norm": 29.625,
+      "grad_norm_var": 46.242122395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.081,
+      "loss/crossentropy": 1.9529634580016135,
+      "loss/hidden": 3.403515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18720184760168196,
+      "step": 25260
+    },
+    {
+      "epoch": 0.21058333333333334,
+      "grad_norm": 36.0,
+      "grad_norm_var": 19.396809895833332,
+      "learning_rate": 0.0001,
+      "loss": 7.0403,
+      "loss/crossentropy": 1.9244548723101615,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17208703532814978,
+      "step": 25270
+    },
+    {
+      "epoch": 0.21066666666666667,
+      "grad_norm": 38.25,
+      "grad_norm_var": 19.56875,
+      "learning_rate": 0.0001,
+      "loss": 7.1547,
+      "loss/crossentropy": 1.9328706055879592,
+      "loss/hidden": 3.3078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18445951752364637,
+      "step": 25280
+    },
+    {
+      "epoch": 0.21075,
+      "grad_norm": 32.75,
+      "grad_norm_var": 20.224739583333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1928,
+      "loss/crossentropy": 1.972021296620369,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17415558286011218,
+      "step": 25290
+    },
+    {
+      "epoch": 0.21083333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 13.026822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0582,
+      "loss/crossentropy": 1.932970654964447,
+      "loss/hidden": 3.39453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17541877925395966,
+      "step": 25300
+    },
+    {
+      "epoch": 0.21091666666666667,
+      "grad_norm": 28.875,
+      "grad_norm_var": 6.620995008298572e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0805,
+      "loss/crossentropy": 1.9781615152955054,
+      "loss/hidden": 3.526171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1858063733205199,
+      "step": 25310
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 37.0,
+      "grad_norm_var": 4.948958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1074,
+      "loss/crossentropy": 1.928247222304344,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18724925485439597,
+      "step": 25320
+    },
+    {
+      "epoch": 0.21108333333333335,
+      "grad_norm": 30.25,
+      "grad_norm_var": 4.5853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.137,
+      "loss/crossentropy": 1.9109850838780402,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.171034119091928,
+      "step": 25330
+    },
+    {
+      "epoch": 0.21116666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.230208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1606,
+      "loss/crossentropy": 1.904369044303894,
+      "loss/hidden": 3.394140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19373000264167786,
+      "step": 25340
+    },
+    {
+      "epoch": 0.21125,
+      "grad_norm": 35.75,
+      "grad_norm_var": 5.365625,
+      "learning_rate": 0.0001,
+      "loss": 7.2361,
+      "loss/crossentropy": 1.888484315574169,
+      "loss/hidden": 3.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17328087855130433,
+      "step": 25350
+    },
+    {
+      "epoch": 0.21133333333333335,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.289322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1383,
+      "loss/crossentropy": 2.0366394743323326,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18470681011676787,
+      "step": 25360
+    },
+    {
+      "epoch": 0.21141666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.072330729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2262,
+      "loss/crossentropy": 2.0070769459009172,
+      "loss/hidden": 3.2109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17460782285779713,
+      "step": 25370
+    },
+    {
+      "epoch": 0.2115,
+      "grad_norm": 29.5,
+      "grad_norm_var": 11.576822916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0762,
+      "loss/crossentropy": 2.045001819729805,
+      "loss/hidden": 3.324609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17671128325164318,
+      "step": 25380
+    },
+    {
+      "epoch": 0.21158333333333335,
+      "grad_norm": 32.0,
+      "grad_norm_var": 248.24765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2381,
+      "loss/crossentropy": 1.898574111610651,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18400687146931888,
+      "step": 25390
+    },
+    {
+      "epoch": 0.21166666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 16.687239583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1892,
+      "loss/crossentropy": 1.9410957738757133,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17161535676568745,
+      "step": 25400
+    },
+    {
+      "epoch": 0.21175,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.1083252957626849e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1997,
+      "loss/crossentropy": 1.9703290075063706,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18135641608387232,
+      "step": 25410
+    },
+    {
+      "epoch": 0.21183333333333335,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.762955729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1225,
+      "loss/crossentropy": 1.98159536793828,
+      "loss/hidden": 3.278125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1778076863847673,
+      "step": 25420
+    },
+    {
+      "epoch": 0.21191666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 11.2681640625,
+      "learning_rate": 0.0001,
+      "loss": 7.0513,
+      "loss/crossentropy": 1.918212030082941,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18460816824808718,
+      "step": 25430
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.12890625,
+      "learning_rate": 0.0001,
+      "loss": 7.204,
+      "loss/crossentropy": 2.1052523128688336,
+      "loss/hidden": 3.392578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18522454258054494,
+      "step": 25440
+    },
+    {
+      "epoch": 0.21208333333333335,
+      "grad_norm": 29.875,
+      "grad_norm_var": 11.436458333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1675,
+      "loss/crossentropy": 1.7871118426322936,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18628909941762686,
+      "step": 25450
+    },
+    {
+      "epoch": 0.21216666666666667,
+      "grad_norm": 28.25,
+      "grad_norm_var": 13.183333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0629,
+      "loss/crossentropy": 1.908453330397606,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18894175458699464,
+      "step": 25460
+    },
+    {
+      "epoch": 0.21225,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.7869140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0976,
+      "loss/crossentropy": 1.8985973209142686,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17754668686538935,
+      "step": 25470
+    },
+    {
+      "epoch": 0.21233333333333335,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.202083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1479,
+      "loss/crossentropy": 1.9730120718479156,
+      "loss/hidden": 3.227734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17484562089666725,
+      "step": 25480
+    },
+    {
+      "epoch": 0.21241666666666667,
+      "grad_norm": 35.5,
+      "grad_norm_var": 19122.812239583334,
+      "learning_rate": 0.0001,
+      "loss": 7.0801,
+      "loss/crossentropy": 1.9286968521773815,
+      "loss/hidden": 3.290625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19581134598702193,
+      "step": 25490
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 30.5,
+      "grad_norm_var": 19088.560872395832,
+      "learning_rate": 0.0001,
+      "loss": 7.0211,
+      "loss/crossentropy": 2.0136352449655535,
+      "loss/hidden": 3.218359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17283423747867346,
+      "step": 25500
+    },
+    {
+      "epoch": 0.21258333333333335,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.345833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0227,
+      "loss/crossentropy": 1.8381054192781447,
+      "loss/hidden": 3.28984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17706162091344596,
+      "step": 25510
+    },
+    {
+      "epoch": 0.21266666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.263997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0427,
+      "loss/crossentropy": 1.9322356529533864,
+      "loss/hidden": 3.39140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17899765176698565,
+      "step": 25520
+    },
+    {
+      "epoch": 0.21275,
+      "grad_norm": 31.5,
+      "grad_norm_var": 4.489518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0227,
+      "loss/crossentropy": 1.8393995121121407,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18383765537291766,
+      "step": 25530
+    },
+    {
+      "epoch": 0.21283333333333335,
+      "grad_norm": 36.75,
+      "grad_norm_var": 8.427083333333334,
+      "learning_rate": 0.0001,
+      "loss": 6.9922,
+      "loss/crossentropy": 2.011235086619854,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18281199112534524,
+      "step": 25540
+    },
+    {
+      "epoch": 0.21291666666666667,
+      "grad_norm": 40.25,
+      "grad_norm_var": 4.56007043697799e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1691,
+      "loss/crossentropy": 1.9823110282421113,
+      "loss/hidden": 3.3625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18033917862921953,
+      "step": 25550
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.560070423162827e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1826,
+      "loss/crossentropy": 2.1367952108383177,
+      "loss/hidden": 3.408984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19303575865924358,
+      "step": 25560
+    },
+    {
+      "epoch": 0.21308333333333335,
+      "grad_norm": 30.875,
+      "grad_norm_var": 38.8181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1053,
+      "loss/crossentropy": 2.124553856253624,
+      "loss/hidden": 3.260546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20232048891484739,
+      "step": 25570
+    },
+    {
+      "epoch": 0.21316666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.4306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.15,
+      "loss/crossentropy": 2.11268679946661,
+      "loss/hidden": 3.4265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19093305207788944,
+      "step": 25580
+    },
+    {
+      "epoch": 0.21325,
+      "grad_norm": 27.5,
+      "grad_norm_var": 4.917643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0787,
+      "loss/crossentropy": 1.9116844087839127,
+      "loss/hidden": 3.44296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879325482994318,
+      "step": 25590
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.172330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0496,
+      "loss/crossentropy": 2.025663011521101,
+      "loss/hidden": 3.258203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1786449492909014,
+      "step": 25600
+    },
+    {
+      "epoch": 0.21341666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 7.178315460298234e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2267,
+      "loss/crossentropy": 1.9342816039919852,
+      "loss/hidden": 3.3578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17539975084364415,
+      "step": 25610
+    },
+    {
+      "epoch": 0.2135,
+      "grad_norm": 31.875,
+      "grad_norm_var": 7.178315451402116e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1626,
+      "loss/crossentropy": 1.9340494081377984,
+      "loss/hidden": 3.3140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17948428280651568,
+      "step": 25620
+    },
+    {
+      "epoch": 0.21358333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 395.50774739583335,
+      "learning_rate": 0.0001,
+      "loss": 7.2443,
+      "loss/crossentropy": 1.9228625565767288,
+      "loss/hidden": 3.283203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18448996152728797,
+      "step": 25630
+    },
+    {
+      "epoch": 0.21366666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 416.14348958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.0459,
+      "loss/crossentropy": 1.92437741458416,
+      "loss/hidden": 3.241796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1690640352666378,
+      "step": 25640
+    },
+    {
+      "epoch": 0.21375,
+      "grad_norm": 34.5,
+      "grad_norm_var": 798.0837890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1346,
+      "loss/crossentropy": 1.9923802688717842,
+      "loss/hidden": 3.290234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17280960585922003,
+      "step": 25650
+    },
+    {
+      "epoch": 0.21383333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 761.17265625,
+      "learning_rate": 0.0001,
+      "loss": 7.2159,
+      "loss/crossentropy": 1.9633371859788895,
+      "loss/hidden": 3.466796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19184397533535957,
+      "step": 25660
+    },
+    {
+      "epoch": 0.21391666666666667,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.426822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1263,
+      "loss/crossentropy": 1.9480347037315369,
+      "loss/hidden": 3.420703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1982917645946145,
+      "step": 25670
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.09765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0907,
+      "loss/crossentropy": 1.9039387851953506,
+      "loss/hidden": 3.326953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16853407472372056,
+      "step": 25680
+    },
+    {
+      "epoch": 0.21408333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 69.26223958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1743,
+      "loss/crossentropy": 2.0335462763905525,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17979698916897177,
+      "step": 25690
+    },
+    {
+      "epoch": 0.21416666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 66.20807291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1652,
+      "loss/crossentropy": 1.946664607524872,
+      "loss/hidden": 3.41875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18638503700494766,
+      "step": 25700
+    },
+    {
+      "epoch": 0.21425,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.3,
+      "learning_rate": 0.0001,
+      "loss": 7.088,
+      "loss/crossentropy": 1.8436803206801415,
+      "loss/hidden": 3.25625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17681604884564878,
+      "step": 25710
+    },
+    {
+      "epoch": 0.21433333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 3.7978515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1686,
+      "loss/crossentropy": 1.9453876182436942,
+      "loss/hidden": 3.351171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19447771292179822,
+      "step": 25720
+    },
+    {
+      "epoch": 0.21441666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.85625,
+      "learning_rate": 0.0001,
+      "loss": 7.1755,
+      "loss/crossentropy": 2.122290536761284,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19864110387861728,
+      "step": 25730
+    },
+    {
+      "epoch": 0.2145,
+      "grad_norm": 31.75,
+      "grad_norm_var": 855.7393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1676,
+      "loss/crossentropy": 1.8324330769479276,
+      "loss/hidden": 3.391796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18320727040991186,
+      "step": 25740
+    },
+    {
+      "epoch": 0.21458333333333332,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.551822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0341,
+      "loss/crossentropy": 1.9923615895211697,
+      "loss/hidden": 3.431640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18184567978605629,
+      "step": 25750
+    },
+    {
+      "epoch": 0.21466666666666667,
+      "grad_norm": 30.125,
+      "grad_norm_var": 44.16848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.088,
+      "loss/crossentropy": 1.850052720308304,
+      "loss/hidden": 3.287890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16342033576220275,
+      "step": 25760
+    },
+    {
+      "epoch": 0.21475,
+      "grad_norm": 28.0,
+      "grad_norm_var": 51.75807291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2021,
+      "loss/crossentropy": 1.9317815765738486,
+      "loss/hidden": 3.535546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2023057235404849,
+      "step": 25770
+    },
+    {
+      "epoch": 0.21483333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 30.89375,
+      "learning_rate": 0.0001,
+      "loss": 7.134,
+      "loss/crossentropy": 1.8493023939430713,
+      "loss/hidden": 3.401171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19848920740187168,
+      "step": 25780
+    },
+    {
+      "epoch": 0.21491666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 19.592643229166665,
+      "learning_rate": 0.0001,
+      "loss": 7.1253,
+      "loss/crossentropy": 2.0072242319583893,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21463057920336723,
+      "step": 25790
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 33.25,
+      "grad_norm_var": 16.598958333333332,
+      "learning_rate": 0.0001,
+      "loss": 7.1017,
+      "loss/crossentropy": 1.9874620258808136,
+      "loss/hidden": 3.521484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20829055029898882,
+      "step": 25800
+    },
+    {
+      "epoch": 0.21508333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 15.401497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1207,
+      "loss/crossentropy": 1.9078074529767037,
+      "loss/hidden": 3.433203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19013442024588584,
+      "step": 25810
+    },
+    {
+      "epoch": 0.21516666666666667,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.3072265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1997,
+      "loss/crossentropy": 2.0049204498529436,
+      "loss/hidden": 3.470703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19886590000241994,
+      "step": 25820
+    },
+    {
+      "epoch": 0.21525,
+      "grad_norm": 33.25,
+      "grad_norm_var": 6.27890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2335,
+      "loss/crossentropy": 1.9146713368594646,
+      "loss/hidden": 3.423828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18319228757172823,
+      "step": 25830
+    },
+    {
+      "epoch": 0.21533333333333332,
+      "grad_norm": 31.25,
+      "grad_norm_var": 8.908333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0223,
+      "loss/crossentropy": 2.0367395117878915,
+      "loss/hidden": 3.2734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1756534393876791,
+      "step": 25840
+    },
+    {
+      "epoch": 0.21541666666666667,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.882747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.164,
+      "loss/crossentropy": 1.8338091112673283,
+      "loss/hidden": 3.421484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18362853918224573,
+      "step": 25850
+    },
+    {
+      "epoch": 0.2155,
+      "grad_norm": 37.0,
+      "grad_norm_var": 13.187955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1333,
+      "loss/crossentropy": 1.9123078510165215,
+      "loss/hidden": 3.45,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19269103035330773,
+      "step": 25860
+    },
+    {
+      "epoch": 0.21558333333333332,
+      "grad_norm": 31.375,
+      "grad_norm_var": 6.103059895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1767,
+      "loss/crossentropy": 2.0201970860362053,
+      "loss/hidden": 3.415234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19183636493980885,
+      "step": 25870
+    },
+    {
+      "epoch": 0.21566666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 13.277018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1609,
+      "loss/crossentropy": 1.8547651540488004,
+      "loss/hidden": 3.404296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1679443832486868,
+      "step": 25880
+    },
+    {
+      "epoch": 0.21575,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.151041666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0926,
+      "loss/crossentropy": 2.0803298011422156,
+      "loss/hidden": 3.247265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1797352595254779,
+      "step": 25890
+    },
+    {
+      "epoch": 0.21583333333333332,
+      "grad_norm": 30.875,
+      "grad_norm_var": 18.735416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1702,
+      "loss/crossentropy": 1.9131331622600556,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18170833624899388,
+      "step": 25900
+    },
+    {
+      "epoch": 0.21591666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 17.452018229166665,
+      "learning_rate": 0.0001,
+      "loss": 7.1517,
+      "loss/crossentropy": 2.0173383384943007,
+      "loss/hidden": 3.333203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18107989095151425,
+      "step": 25910
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 34.5,
+      "grad_norm_var": 1.0995116101642924e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1967,
+      "loss/crossentropy": 1.9246201008558272,
+      "loss/hidden": 3.32734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19072177391499281,
+      "step": 25920
+    },
+    {
+      "epoch": 0.21608333333333332,
+      "grad_norm": 34.75,
+      "grad_norm_var": 10.6056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1391,
+      "loss/crossentropy": 1.8887877836823463,
+      "loss/hidden": 3.289453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17543924301862718,
+      "step": 25930
+    },
+    {
+      "epoch": 0.21616666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 11.1791015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1043,
+      "loss/crossentropy": 2.204209664463997,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2206350963562727,
+      "step": 25940
+    },
+    {
+      "epoch": 0.21625,
+      "grad_norm": 30.875,
+      "grad_norm_var": 8.325,
+      "learning_rate": 0.0001,
+      "loss": 7.0854,
+      "loss/crossentropy": 2.0052800476551056,
+      "loss/hidden": 3.312109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17595654912292957,
+      "step": 25950
+    },
+    {
+      "epoch": 0.21633333333333332,
+      "grad_norm": 33.25,
+      "grad_norm_var": 7.323893229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1875,
+      "loss/crossentropy": 2.016774645447731,
+      "loss/hidden": 3.31875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2024704683572054,
+      "step": 25960
+    },
+    {
+      "epoch": 0.21641666666666667,
+      "grad_norm": 30.25,
+      "grad_norm_var": 148.74375,
+      "learning_rate": 0.0001,
+      "loss": 7.0827,
+      "loss/crossentropy": 2.008812907338142,
+      "loss/hidden": 3.229296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17170650362968445,
+      "step": 25970
+    },
+    {
+      "epoch": 0.2165,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.534375,
+      "learning_rate": 0.0001,
+      "loss": 7.0917,
+      "loss/crossentropy": 1.8728063225746154,
+      "loss/hidden": 3.365234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19558557737618684,
+      "step": 25980
+    },
+    {
+      "epoch": 0.21658333333333332,
+      "grad_norm": 31.625,
+      "grad_norm_var": 16.273958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1836,
+      "loss/crossentropy": 1.8522960014641285,
+      "loss/hidden": 3.32265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16913470225408672,
+      "step": 25990
+    },
+    {
+      "epoch": 0.21666666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 30.343489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0301,
+      "loss/crossentropy": 2.0298531755805014,
+      "loss/hidden": 3.303125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17906388211995364,
+      "step": 26000
+    },
+    {
+      "epoch": 0.21675,
+      "grad_norm": 30.25,
+      "grad_norm_var": 30.810416666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0837,
+      "loss/crossentropy": 1.9665327347815036,
+      "loss/hidden": 3.34375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18679081229493022,
+      "step": 26010
+    },
+    {
+      "epoch": 0.21683333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.2916015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0869,
+      "loss/crossentropy": 1.9430747598409652,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19298500344157218,
+      "step": 26020
+    },
+    {
+      "epoch": 0.21691666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 4.506184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1374,
+      "loss/crossentropy": 1.933246725797653,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1922842614352703,
+      "step": 26030
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.23515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0094,
+      "loss/crossentropy": 1.9978767067193985,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19021354094147683,
+      "step": 26040
+    },
+    {
+      "epoch": 0.21708333333333332,
+      "grad_norm": 35.0,
+      "grad_norm_var": 1.0048832435387018e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1676,
+      "loss/crossentropy": 1.9970777198672294,
+      "loss/hidden": 3.38125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20220550931990147,
+      "step": 26050
+    },
+    {
+      "epoch": 0.21716666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.004883242979007e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.0989,
+      "loss/crossentropy": 2.0598263129591943,
+      "loss/hidden": 3.35234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20695621930062771,
+      "step": 26060
+    },
+    {
+      "epoch": 0.21725,
+      "grad_norm": 39.0,
+      "grad_norm_var": 5.891447062916498e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2289,
+      "loss/crossentropy": 2.019556447863579,
+      "loss/hidden": 3.6109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20935273617506028,
+      "step": 26070
+    },
+    {
+      "epoch": 0.21733333333333332,
+      "grad_norm": 56.75,
+      "grad_norm_var": 1.0527867572541807e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2277,
+      "loss/crossentropy": 2.0993714898824694,
+      "loss/hidden": 3.227734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1966920578852296,
+      "step": 26080
+    },
+    {
+      "epoch": 0.21741666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 5.387606850434937e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0297,
+      "loss/crossentropy": 2.009559527039528,
+      "loss/hidden": 3.237890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1835018103942275,
+      "step": 26090
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 30.875,
+      "grad_norm_var": 12.79375,
+      "learning_rate": 0.0001,
+      "loss": 7.0001,
+      "loss/crossentropy": 1.8580125749111176,
+      "loss/hidden": 3.328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17266698190942406,
+      "step": 26100
+    },
+    {
+      "epoch": 0.21758333333333332,
+      "grad_norm": 29.0,
+      "grad_norm_var": 10.256184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0723,
+      "loss/crossentropy": 2.0379395455121996,
+      "loss/hidden": 3.300390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19339074939489365,
+      "step": 26110
+    },
+    {
+      "epoch": 0.21766666666666667,
+      "grad_norm": 31.75,
+      "grad_norm_var": 6.170247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0905,
+      "loss/crossentropy": 1.841700490564108,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19040193650871515,
+      "step": 26120
+    },
+    {
+      "epoch": 0.21775,
+      "grad_norm": 30.75,
+      "grad_norm_var": 7.7322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0799,
+      "loss/crossentropy": 2.026270292699337,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18778208680450917,
+      "step": 26130
+    },
+    {
+      "epoch": 0.21783333333333332,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.1125,
+      "learning_rate": 0.0001,
+      "loss": 7.1324,
+      "loss/crossentropy": 1.7643813356757163,
+      "loss/hidden": 3.421484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18530586510896682,
+      "step": 26140
+    },
+    {
+      "epoch": 0.21791666666666668,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.5389973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.214,
+      "loss/crossentropy": 1.8893267840147019,
+      "loss/hidden": 3.456640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19192402586340904,
+      "step": 26150
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 30.25,
+      "grad_norm_var": 1.189231758963376e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.9081555992364883,
+      "loss/hidden": 3.386328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17880377853289248,
+      "step": 26160
+    },
+    {
+      "epoch": 0.21808333333333332,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.189231759104235e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.0941,
+      "loss/crossentropy": 1.964975292980671,
+      "loss/hidden": 3.181640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1678079629316926,
+      "step": 26170
+    },
+    {
+      "epoch": 0.21816666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.503599521674036e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1317,
+      "loss/crossentropy": 2.0014721184968947,
+      "loss/hidden": 3.34921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1726706327870488,
+      "step": 26180
+    },
+    {
+      "epoch": 0.21825,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.714583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1675,
+      "loss/crossentropy": 1.9244608849287033,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17802753616124392,
+      "step": 26190
+    },
+    {
+      "epoch": 0.21833333333333332,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.3436848958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1113,
+      "loss/crossentropy": 2.0229026291519405,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1830245525110513,
+      "step": 26200
+    },
+    {
+      "epoch": 0.21841666666666668,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.8988932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1451,
+      "loss/crossentropy": 1.8356290817260743,
+      "loss/hidden": 3.401953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18080337569117547,
+      "step": 26210
+    },
+    {
+      "epoch": 0.2185,
+      "grad_norm": 33.75,
+      "grad_norm_var": 2.77265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0808,
+      "loss/crossentropy": 1.8363368675112723,
+      "loss/hidden": 3.46328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182987941801548,
+      "step": 26220
+    },
+    {
+      "epoch": 0.21858333333333332,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.61640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1074,
+      "loss/crossentropy": 1.8397347405552864,
+      "loss/hidden": 3.416015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18784533441066742,
+      "step": 26230
+    },
+    {
+      "epoch": 0.21866666666666668,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.342643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2053,
+      "loss/crossentropy": 2.015132397413254,
+      "loss/hidden": 3.290625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18501140512526035,
+      "step": 26240
+    },
+    {
+      "epoch": 0.21875,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.687239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1734,
+      "loss/crossentropy": 2.061004877090454,
+      "loss/hidden": 3.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19302913583815098,
+      "step": 26250
+    },
+    {
+      "epoch": 0.21883333333333332,
+      "grad_norm": 30.0,
+      "grad_norm_var": 4.459375,
+      "learning_rate": 0.0001,
+      "loss": 7.0696,
+      "loss/crossentropy": 1.997168380022049,
+      "loss/hidden": 3.259765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17417351640760897,
+      "step": 26260
+    },
+    {
+      "epoch": 0.21891666666666668,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.7108723958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.144,
+      "loss/crossentropy": 2.062076824903488,
+      "loss/hidden": 3.308984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20755025148391723,
+      "step": 26270
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 35.0,
+      "grad_norm_var": 5.011458333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.196,
+      "loss/crossentropy": 2.023951905965805,
+      "loss/hidden": 3.311328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18756244108080863,
+      "step": 26280
+    },
+    {
+      "epoch": 0.21908333333333332,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.932747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1389,
+      "loss/crossentropy": 1.9033431261777878,
+      "loss/hidden": 3.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17559341639280318,
+      "step": 26290
+    },
+    {
+      "epoch": 0.21916666666666668,
+      "grad_norm": 34.0,
+      "grad_norm_var": 2.095572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.2244,
+      "loss/crossentropy": 2.0101781740784643,
+      "loss/hidden": 3.337890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17239926904439926,
+      "step": 26300
+    },
+    {
+      "epoch": 0.21925,
+      "grad_norm": 30.375,
+      "grad_norm_var": 3.3238932291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1053,
+      "loss/crossentropy": 1.8694226652383805,
+      "loss/hidden": 3.28203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18007475342601537,
+      "step": 26310
+    },
+    {
+      "epoch": 0.21933333333333332,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.26015625,
+      "learning_rate": 0.0001,
+      "loss": 7.1395,
+      "loss/crossentropy": 1.8729426562786102,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18007828257977962,
+      "step": 26320
+    },
+    {
+      "epoch": 0.21941666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.5306640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1023,
+      "loss/crossentropy": 2.0202317640185354,
+      "loss/hidden": 3.323828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18215413931757213,
+      "step": 26330
+    },
+    {
+      "epoch": 0.2195,
+      "grad_norm": 29.25,
+      "grad_norm_var": 4.008268229166666,
+      "learning_rate": 0.0001,
+      "loss": 6.9877,
+      "loss/crossentropy": 2.0027072548866274,
+      "loss/hidden": 3.16640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16499932818114757,
+      "step": 26340
+    },
+    {
+      "epoch": 0.21958333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.3244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0226,
+      "loss/crossentropy": 1.8629272803664207,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18851205352693795,
+      "step": 26350
+    },
+    {
+      "epoch": 0.21966666666666668,
+      "grad_norm": 29.625,
+      "grad_norm_var": 1.4559895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1105,
+      "loss/crossentropy": 2.0280668079853057,
+      "loss/hidden": 3.222265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18105703499168158,
+      "step": 26360
+    },
+    {
+      "epoch": 0.21975,
+      "grad_norm": 31.75,
+      "grad_norm_var": 2.792708333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1119,
+      "loss/crossentropy": 1.6670363761484623,
+      "loss/hidden": 3.432421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17590470481663942,
+      "step": 26370
+    },
+    {
+      "epoch": 0.21983333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.8843098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2215,
+      "loss/crossentropy": 1.9094415605068207,
+      "loss/hidden": 3.454296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18737881034612655,
+      "step": 26380
+    },
+    {
+      "epoch": 0.21991666666666668,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.5895182291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0799,
+      "loss/crossentropy": 1.8960991755127907,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865002630278468,
+      "step": 26390
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.3208333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1602,
+      "loss/crossentropy": 1.9633339315652847,
+      "loss/hidden": 3.312890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18124619126319885,
+      "step": 26400
+    },
+    {
+      "epoch": 0.22008333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.2593098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0539,
+      "loss/crossentropy": 2.0058771520853043,
+      "loss/hidden": 3.3328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17581757642328738,
+      "step": 26410
+    },
+    {
+      "epoch": 0.22016666666666668,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.0056640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1481,
+      "loss/crossentropy": 1.9701676651835442,
+      "loss/hidden": 3.233984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17133039943873882,
+      "step": 26420
+    },
+    {
+      "epoch": 0.22025,
+      "grad_norm": 31.75,
+      "grad_norm_var": 5.996875,
+      "learning_rate": 0.0001,
+      "loss": 7.1269,
+      "loss/crossentropy": 1.9138137474656105,
+      "loss/hidden": 3.29375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1669824853539467,
+      "step": 26430
+    },
+    {
+      "epoch": 0.22033333333333333,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.237239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1854,
+      "loss/crossentropy": 1.9141460880637169,
+      "loss/hidden": 3.298046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1793738404288888,
+      "step": 26440
+    },
+    {
+      "epoch": 0.22041666666666668,
+      "grad_norm": 33.0,
+      "grad_norm_var": 7.218684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1202,
+      "loss/crossentropy": 1.8779874980449676,
+      "loss/hidden": 3.346875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18356089480221272,
+      "step": 26450
+    },
+    {
+      "epoch": 0.2205,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.205208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0175,
+      "loss/crossentropy": 1.9437492229044437,
+      "loss/hidden": 3.321484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1828056387603283,
+      "step": 26460
+    },
+    {
+      "epoch": 0.22058333333333333,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.502083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1743,
+      "loss/crossentropy": 1.9961276315152645,
+      "loss/hidden": 3.31484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17449145764112473,
+      "step": 26470
+    },
+    {
+      "epoch": 0.22066666666666668,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.7041666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1055,
+      "loss/crossentropy": 1.8990372709929944,
+      "loss/hidden": 3.31015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16790245426818728,
+      "step": 26480
+    },
+    {
+      "epoch": 0.22075,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.255143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1504,
+      "loss/crossentropy": 1.9570870250463486,
+      "loss/hidden": 3.416796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18750158753246068,
+      "step": 26490
+    },
+    {
+      "epoch": 0.22083333333333333,
+      "grad_norm": 31.625,
+      "grad_norm_var": 0.7655598958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1812,
+      "loss/crossentropy": 1.8565197855234146,
+      "loss/hidden": 3.324609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17969967592507602,
+      "step": 26500
+    },
+    {
+      "epoch": 0.22091666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.12265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1748,
+      "loss/crossentropy": 1.8801551394164562,
+      "loss/hidden": 3.42109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2159306754358113,
+      "step": 26510
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.0434895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1232,
+      "loss/crossentropy": 1.9958307459950446,
+      "loss/hidden": 3.248828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1785376450046897,
+      "step": 26520
+    },
+    {
+      "epoch": 0.22108333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 15.187434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0774,
+      "loss/crossentropy": 1.9507863707840443,
+      "loss/hidden": 3.26328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1734126929193735,
+      "step": 26530
+    },
+    {
+      "epoch": 0.22116666666666668,
+      "grad_norm": 32.25,
+      "grad_norm_var": 14.004622395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1798,
+      "loss/crossentropy": 1.867694914340973,
+      "loss/hidden": 3.469921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19041500594466926,
+      "step": 26540
+    },
+    {
+      "epoch": 0.22125,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.137434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1633,
+      "loss/crossentropy": 1.9365580722689628,
+      "loss/hidden": 3.351953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1893499379977584,
+      "step": 26550
+    },
+    {
+      "epoch": 0.22133333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 13.30625,
+      "learning_rate": 0.0001,
+      "loss": 7.1095,
+      "loss/crossentropy": 1.966548129916191,
+      "loss/hidden": 3.368359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19449863843619825,
+      "step": 26560
+    },
+    {
+      "epoch": 0.22141666666666668,
+      "grad_norm": 35.5,
+      "grad_norm_var": 12.856184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1793,
+      "loss/crossentropy": 1.8481765255331992,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19486072529107332,
+      "step": 26570
+    },
+    {
+      "epoch": 0.2215,
+      "grad_norm": 31.125,
+      "grad_norm_var": 6.389322916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0911,
+      "loss/crossentropy": 2.0334127128124235,
+      "loss/hidden": 3.415234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1994987793266773,
+      "step": 26580
+    },
+    {
+      "epoch": 0.22158333333333333,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.607291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0973,
+      "loss/crossentropy": 2.0624680534005164,
+      "loss/hidden": 3.229296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.169341991096735,
+      "step": 26590
+    },
+    {
+      "epoch": 0.22166666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.3108723958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1,
+      "loss/crossentropy": 1.9950445130467416,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1768806418403983,
+      "step": 26600
+    },
+    {
+      "epoch": 0.22175,
+      "grad_norm": 34.25,
+      "grad_norm_var": 2.9622395833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1013,
+      "loss/crossentropy": 1.9238395437598228,
+      "loss/hidden": 3.359765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1898207848891616,
+      "step": 26610
+    },
+    {
+      "epoch": 0.22183333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 4.3431640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1913,
+      "loss/crossentropy": 1.8916297569870948,
+      "loss/hidden": 3.4671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886750441044569,
+      "step": 26620
+    },
+    {
+      "epoch": 0.22191666666666668,
+      "grad_norm": 29.125,
+      "grad_norm_var": 2.1809895833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1778,
+      "loss/crossentropy": 2.025717315077782,
+      "loss/hidden": 3.4109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1861115347594023,
+      "step": 26630
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.3067057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1642,
+      "loss/crossentropy": 1.9157853037118913,
+      "loss/hidden": 3.37109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18798933206126095,
+      "step": 26640
+    },
+    {
+      "epoch": 0.22208333333333333,
+      "grad_norm": 33.5,
+      "grad_norm_var": 3.19765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1412,
+      "loss/crossentropy": 1.8923842176795005,
+      "loss/hidden": 3.301171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16866083201020957,
+      "step": 26650
+    },
+    {
+      "epoch": 0.22216666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 5.862955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1035,
+      "loss/crossentropy": 1.9416986256837845,
+      "loss/hidden": 3.3640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18185625565238298,
+      "step": 26660
+    },
+    {
+      "epoch": 0.22225,
+      "grad_norm": 35.5,
+      "grad_norm_var": 4.592643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2416,
+      "loss/crossentropy": 1.8928313814103603,
+      "loss/hidden": 3.382421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19859238546341657,
+      "step": 26670
+    },
+    {
+      "epoch": 0.22233333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 8.583333333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0952,
+      "loss/crossentropy": 1.888912557065487,
+      "loss/hidden": 3.275390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17804007064551114,
+      "step": 26680
+    },
+    {
+      "epoch": 0.22241666666666668,
+      "grad_norm": 552.0,
+      "grad_norm_var": 8.207810014325414e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.237,
+      "loss/crossentropy": 1.9477786734700202,
+      "loss/hidden": 3.58046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21397521644830703,
+      "step": 26690
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 35.5,
+      "grad_norm_var": 16572.968489583334,
+      "learning_rate": 0.0001,
+      "loss": 7.363,
+      "loss/crossentropy": 1.9811367318034172,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19772923346608878,
+      "step": 26700
+    },
+    {
+      "epoch": 0.22258333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 229.37395833333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2295,
+      "loss/crossentropy": 1.9286298885941506,
+      "loss/hidden": 3.446875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19154851231724024,
+      "step": 26710
+    },
+    {
+      "epoch": 0.22266666666666668,
+      "grad_norm": 3489660928.0,
+      "grad_norm_var": 7.611083219982615e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1736,
+      "loss/crossentropy": 1.972748449444771,
+      "loss/hidden": 3.430078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18883435167372226,
+      "step": 26720
+    },
+    {
+      "epoch": 0.22275,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.61108322252716e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0354,
+      "loss/crossentropy": 1.9736176609992981,
+      "loss/hidden": 3.299609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1675051363185048,
+      "step": 26730
+    },
+    {
+      "epoch": 0.22283333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.56007043320766e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1593,
+      "loss/crossentropy": 1.9751226410269738,
+      "loss/hidden": 3.3609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18904283102601765,
+      "step": 26740
+    },
+    {
+      "epoch": 0.22291666666666668,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.560070427045704e+17,
+      "learning_rate": 0.0001,
+      "loss": 6.9946,
+      "loss/crossentropy": 1.98078473508358,
+      "loss/hidden": 3.516015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20173397604376078,
+      "step": 26750
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 33.0,
+      "grad_norm_var": 6.894791666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1006,
+      "loss/crossentropy": 2.0676103800535204,
+      "loss/hidden": 3.31875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19137653280049563,
+      "step": 26760
+    },
+    {
+      "epoch": 0.22308333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.483268229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0951,
+      "loss/crossentropy": 1.9552440628409387,
+      "loss/hidden": 3.298046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18239242024719715,
+      "step": 26770
+    },
+    {
+      "epoch": 0.22316666666666668,
+      "grad_norm": 30.5,
+      "grad_norm_var": 3.45390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1308,
+      "loss/crossentropy": 2.0006344258785247,
+      "loss/hidden": 3.459375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19593478310853243,
+      "step": 26780
+    },
+    {
+      "epoch": 0.22325,
+      "grad_norm": 31.5,
+      "grad_norm_var": 11.3384765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0961,
+      "loss/crossentropy": 2.0171128578484057,
+      "loss/hidden": 3.3546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20159837659448385,
+      "step": 26790
+    },
+    {
+      "epoch": 0.22333333333333333,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.99375,
+      "learning_rate": 0.0001,
+      "loss": 7.0688,
+      "loss/crossentropy": 1.9728857278823853,
+      "loss/hidden": 3.26328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17641023490577937,
+      "step": 26800
+    },
+    {
+      "epoch": 0.22341666666666668,
+      "grad_norm": 31.375,
+      "grad_norm_var": 11.620572916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.008,
+      "loss/crossentropy": 1.9222103528678418,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19041021568700672,
+      "step": 26810
+    },
+    {
+      "epoch": 0.2235,
+      "grad_norm": 31.375,
+      "grad_norm_var": 10.164322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2689,
+      "loss/crossentropy": 2.0734514385461806,
+      "loss/hidden": 3.415234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20475540310144424,
+      "step": 26820
+    },
+    {
+      "epoch": 0.22358333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.5622564692555203e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2896,
+      "loss/crossentropy": 1.9720015287399293,
+      "loss/hidden": 3.64296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19303508531302213,
+      "step": 26830
+    },
+    {
+      "epoch": 0.22366666666666668,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.674067798503021e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1668,
+      "loss/crossentropy": 1.9627857737243175,
+      "loss/hidden": 3.217578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17705758824013174,
+      "step": 26840
+    },
+    {
+      "epoch": 0.22375,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.6740677965089786e+17,
+      "learning_rate": 0.0001,
+      "loss": 6.9486,
+      "loss/crossentropy": 1.8676376521587372,
+      "loss/hidden": 3.27265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18021242152899503,
+      "step": 26850
+    },
+    {
+      "epoch": 0.22383333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 20.801497395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0687,
+      "loss/crossentropy": 1.875405278801918,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17537363171577453,
+      "step": 26860
+    },
+    {
+      "epoch": 0.22391666666666668,
+      "grad_norm": 30.75,
+      "grad_norm_var": 10.4197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0327,
+      "loss/crossentropy": 1.867275634407997,
+      "loss/hidden": 3.37421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17917759250849485,
+      "step": 26870
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.7854166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.06,
+      "loss/crossentropy": 1.9366668835282326,
+      "loss/hidden": 3.4375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19871108792722225,
+      "step": 26880
+    },
+    {
+      "epoch": 0.22408333333333333,
+      "grad_norm": 30.625,
+      "grad_norm_var": 2.6184895833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1004,
+      "loss/crossentropy": 1.8226452022790909,
+      "loss/hidden": 3.415234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1930592915043235,
+      "step": 26890
+    },
+    {
+      "epoch": 0.22416666666666665,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.968489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2214,
+      "loss/crossentropy": 1.9352462098002434,
+      "loss/hidden": 3.303125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17417251504957676,
+      "step": 26900
+    },
+    {
+      "epoch": 0.22425,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.1780598958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1831,
+      "loss/crossentropy": 1.940615999698639,
+      "loss/hidden": 3.366796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1868426801636815,
+      "step": 26910
+    },
+    {
+      "epoch": 0.22433333333333333,
+      "grad_norm": 31.875,
+      "grad_norm_var": 2.95625,
+      "learning_rate": 0.0001,
+      "loss": 7.1105,
+      "loss/crossentropy": 2.0782107532024385,
+      "loss/hidden": 3.297265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1860398143529892,
+      "step": 26920
+    },
+    {
+      "epoch": 0.22441666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.37265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0741,
+      "loss/crossentropy": 1.9800411209464073,
+      "loss/hidden": 3.42578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18416937310248613,
+      "step": 26930
+    },
+    {
+      "epoch": 0.2245,
+      "grad_norm": 33.0,
+      "grad_norm_var": 1.76015625,
+      "learning_rate": 0.0001,
+      "loss": 7.134,
+      "loss/crossentropy": 1.898789981007576,
+      "loss/hidden": 3.459375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19875736422836782,
+      "step": 26940
+    },
+    {
+      "epoch": 0.22458333333333333,
+      "grad_norm": 31.5,
+      "grad_norm_var": 8.8322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.9381616115570068,
+      "loss/hidden": 3.3015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17116798926144838,
+      "step": 26950
+    },
+    {
+      "epoch": 0.22466666666666665,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.918684895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1153,
+      "loss/crossentropy": 1.9235254988074302,
+      "loss/hidden": 3.424609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18356654215604068,
+      "step": 26960
+    },
+    {
+      "epoch": 0.22475,
+      "grad_norm": 35.5,
+      "grad_norm_var": 10.7150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.2398,
+      "loss/crossentropy": 1.7901165753602981,
+      "loss/hidden": 3.4125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19185024350881577,
+      "step": 26970
+    },
+    {
+      "epoch": 0.22483333333333333,
+      "grad_norm": 37.75,
+      "grad_norm_var": 10.780208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1673,
+      "loss/crossentropy": 2.0093143239617346,
+      "loss/hidden": 3.31328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18232030887156725,
+      "step": 26980
+    },
+    {
+      "epoch": 0.22491666666666665,
+      "grad_norm": 29.75,
+      "grad_norm_var": 4.804622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0636,
+      "loss/crossentropy": 1.8473149925470351,
+      "loss/hidden": 3.39140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17704327832907438,
+      "step": 26990
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 30.25,
+      "grad_norm_var": 2.582747395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1686,
+      "loss/crossentropy": 1.9777016460895538,
+      "loss/hidden": 3.38359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1996795428916812,
+      "step": 27000
+    },
+    {
+      "epoch": 0.22508333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.096809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0697,
+      "loss/crossentropy": 1.8574656262993812,
+      "loss/hidden": 3.448828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.180445446819067,
+      "step": 27010
+    },
+    {
+      "epoch": 0.22516666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.7301432291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1423,
+      "loss/crossentropy": 2.0723973736166954,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18677765987813472,
+      "step": 27020
+    },
+    {
+      "epoch": 0.22525,
+      "grad_norm": 31.875,
+      "grad_norm_var": 17.837434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2528,
+      "loss/crossentropy": 1.951155810058117,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18211656399071216,
+      "step": 27030
+    },
+    {
+      "epoch": 0.22533333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 18.4603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1759,
+      "loss/crossentropy": 2.0808217883110047,
+      "loss/hidden": 3.256640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1857470976188779,
+      "step": 27040
+    },
+    {
+      "epoch": 0.22541666666666665,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.4893229166666666,
+      "learning_rate": 0.0001,
+      "loss": 7.071,
+      "loss/crossentropy": 1.9631942689418793,
+      "loss/hidden": 3.377734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19388457369059325,
+      "step": 27050
+    },
+    {
+      "epoch": 0.2255,
+      "grad_norm": 46.25,
+      "grad_norm_var": 21.520833333333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2422,
+      "loss/crossentropy": 2.084809072315693,
+      "loss/hidden": 3.447265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2006871158257127,
+      "step": 27060
+    },
+    {
+      "epoch": 0.22558333333333333,
+      "grad_norm": 28.0,
+      "grad_norm_var": 22.7853515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1252,
+      "loss/crossentropy": 1.8997162014245987,
+      "loss/hidden": 3.4109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18645735513418912,
+      "step": 27070
+    },
+    {
+      "epoch": 0.22566666666666665,
+      "grad_norm": 30.75,
+      "grad_norm_var": 7.701822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.053,
+      "loss/crossentropy": 1.9379786774516106,
+      "loss/hidden": 3.30234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1755607735365629,
+      "step": 27080
+    },
+    {
+      "epoch": 0.22575,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.848372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2039,
+      "loss/crossentropy": 1.8419730752706527,
+      "loss/hidden": 3.45234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18232451155781745,
+      "step": 27090
+    },
+    {
+      "epoch": 0.22583333333333333,
+      "grad_norm": 31.25,
+      "grad_norm_var": 6.543489583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0989,
+      "loss/crossentropy": 1.968306265771389,
+      "loss/hidden": 3.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18628506902605296,
+      "step": 27100
+    },
+    {
+      "epoch": 0.22591666666666665,
+      "grad_norm": 31.5,
+      "grad_norm_var": 5.123958333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1743,
+      "loss/crossentropy": 2.0560183130204677,
+      "loss/hidden": 3.384765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1968917019665241,
+      "step": 27110
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.781184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1428,
+      "loss/crossentropy": 1.9009317338466645,
+      "loss/hidden": 3.43359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17883394416421652,
+      "step": 27120
+    },
+    {
+      "epoch": 0.22608333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 28191.665559895835,
+      "learning_rate": 0.0001,
+      "loss": 7.2408,
+      "loss/crossentropy": 1.8899662017822265,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18106959164142608,
+      "step": 27130
+    },
+    {
+      "epoch": 0.22616666666666665,
+      "grad_norm": 32.0,
+      "grad_norm_var": 28224.924739583334,
+      "learning_rate": 0.0001,
+      "loss": 7.1578,
+      "loss/crossentropy": 1.984883976727724,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809134477749467,
+      "step": 27140
+    },
+    {
+      "epoch": 0.22625,
+      "grad_norm": 29.375,
+      "grad_norm_var": 12.028580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.953794313967228,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18341592662036418,
+      "step": 27150
+    },
+    {
+      "epoch": 0.22633333333333333,
+      "grad_norm": 38.25,
+      "grad_norm_var": 12.775455729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1109,
+      "loss/crossentropy": 1.884887158870697,
+      "loss/hidden": 3.320703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1784294692799449,
+      "step": 27160
+    },
+    {
+      "epoch": 0.22641666666666665,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.453059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1741,
+      "loss/crossentropy": 1.9588986977934837,
+      "loss/hidden": 3.396875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20042267832905053,
+      "step": 27170
+    },
+    {
+      "epoch": 0.2265,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.4650390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1549,
+      "loss/crossentropy": 1.9995316296815873,
+      "loss/hidden": 3.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19012631997466087,
+      "step": 27180
+    },
+    {
+      "epoch": 0.22658333333333333,
+      "grad_norm": 31.75,
+      "grad_norm_var": 151.12337239583334,
+      "learning_rate": 0.0001,
+      "loss": 7.2233,
+      "loss/crossentropy": 1.9893013656139373,
+      "loss/hidden": 3.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17861738428473473,
+      "step": 27190
+    },
+    {
+      "epoch": 0.22666666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.8421223958333335,
+      "learning_rate": 0.0001,
+      "loss": 7.2131,
+      "loss/crossentropy": 1.995251429080963,
+      "loss/hidden": 3.316796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19560580812394618,
+      "step": 27200
+    },
+    {
+      "epoch": 0.22675,
+      "grad_norm": 29.75,
+      "grad_norm_var": 9.225518128792339e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2451,
+      "loss/crossentropy": 1.8945991858839988,
+      "loss/hidden": 3.50546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18896378241479397,
+      "step": 27210
+    },
+    {
+      "epoch": 0.22683333333333333,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.225518129032463e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1456,
+      "loss/crossentropy": 2.036046567559242,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1910065958276391,
+      "step": 27220
+    },
+    {
+      "epoch": 0.22691666666666666,
+      "grad_norm": 29.125,
+      "grad_norm_var": 3.4067057291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1233,
+      "loss/crossentropy": 1.9148334339261055,
+      "loss/hidden": 3.375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18873931430280208,
+      "step": 27230
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 42.75,
+      "grad_norm_var": 11.346875,
+      "learning_rate": 0.0001,
+      "loss": 7.1504,
+      "loss/crossentropy": 1.9534554749727249,
+      "loss/hidden": 3.438671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18276325464248658,
+      "step": 27240
+    },
+    {
+      "epoch": 0.22708333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 11.662239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1178,
+      "loss/crossentropy": 1.908184403181076,
+      "loss/hidden": 3.3984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18567621633410453,
+      "step": 27250
+    },
+    {
+      "epoch": 0.22716666666666666,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.144791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0414,
+      "loss/crossentropy": 2.005711516737938,
+      "loss/hidden": 3.259765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1818504286929965,
+      "step": 27260
+    },
+    {
+      "epoch": 0.22725,
+      "grad_norm": 33.25,
+      "grad_norm_var": 13.0103515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1907,
+      "loss/crossentropy": 1.8307042494416237,
+      "loss/hidden": 3.30859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18112731482833624,
+      "step": 27270
+    },
+    {
+      "epoch": 0.22733333333333333,
+      "grad_norm": 28.125,
+      "grad_norm_var": 15.812239583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1973,
+      "loss/crossentropy": 1.9920083984732628,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19723976906388999,
+      "step": 27280
+    },
+    {
+      "epoch": 0.22741666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.96875,
+      "learning_rate": 0.0001,
+      "loss": 7.1379,
+      "loss/crossentropy": 1.8356394201517106,
+      "loss/hidden": 3.344140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18771058432757853,
+      "step": 27290
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 30.375,
+      "grad_norm_var": 6.151811410057933e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2954,
+      "loss/crossentropy": 1.980231226980686,
+      "loss/hidden": 3.548046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19072088338434695,
+      "step": 27300
+    },
+    {
+      "epoch": 0.22758333333333333,
+      "grad_norm": 31.375,
+      "grad_norm_var": 3.0603515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0422,
+      "loss/crossentropy": 1.8550451293587684,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16044338047504425,
+      "step": 27310
+    },
+    {
+      "epoch": 0.22766666666666666,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.02890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0719,
+      "loss/crossentropy": 1.9593237459659576,
+      "loss/hidden": 3.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1956369702704251,
+      "step": 27320
+    },
+    {
+      "epoch": 0.22775,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.7666015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0704,
+      "loss/crossentropy": 1.9946182563900947,
+      "loss/hidden": 3.306640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17183393556624652,
+      "step": 27330
+    },
+    {
+      "epoch": 0.22783333333333333,
+      "grad_norm": 30.875,
+      "grad_norm_var": 2.033072916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1034,
+      "loss/crossentropy": 2.0074402779340743,
+      "loss/hidden": 3.294921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18853956907987596,
+      "step": 27340
+    },
+    {
+      "epoch": 0.22791666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.5205981727940936e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2378,
+      "loss/crossentropy": 1.8830150440335274,
+      "loss/hidden": 3.3734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1879985213279724,
+      "step": 27350
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.0322265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0323,
+      "loss/crossentropy": 1.9409915179014205,
+      "loss/hidden": 3.315234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.168220454454422,
+      "step": 27360
+    },
+    {
+      "epoch": 0.22808333333333333,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.8333333333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.101,
+      "loss/crossentropy": 1.8928554818034171,
+      "loss/hidden": 3.36484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17416304480284453,
+      "step": 27370
+    },
+    {
+      "epoch": 0.22816666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.7749348958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1218,
+      "loss/crossentropy": 1.9035050943493843,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19629047475755215,
+      "step": 27380
+    },
+    {
+      "epoch": 0.22825,
+      "grad_norm": 34.25,
+      "grad_norm_var": 3.4254557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0669,
+      "loss/crossentropy": 1.906415231525898,
+      "loss/hidden": 3.286328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1748045612126589,
+      "step": 27390
+    },
+    {
+      "epoch": 0.22833333333333333,
+      "grad_norm": 29.75,
+      "grad_norm_var": 3.5113932291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.199,
+      "loss/crossentropy": 2.006585682928562,
+      "loss/hidden": 3.372265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18857967015355825,
+      "step": 27400
+    },
+    {
+      "epoch": 0.22841666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 3.068489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.131,
+      "loss/crossentropy": 1.9577410407364368,
+      "loss/hidden": 3.273046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17287150779739022,
+      "step": 27410
+    },
+    {
+      "epoch": 0.2285,
+      "grad_norm": 32.5,
+      "grad_norm_var": 3.9926432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1261,
+      "loss/crossentropy": 1.85484227091074,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17866760212928057,
+      "step": 27420
+    },
+    {
+      "epoch": 0.22858333333333333,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.524739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1966,
+      "loss/crossentropy": 1.8921881765127182,
+      "loss/hidden": 3.385546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1881699152290821,
+      "step": 27430
+    },
+    {
+      "epoch": 0.22866666666666666,
+      "grad_norm": 28.875,
+      "grad_norm_var": 1.1177083333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0225,
+      "loss/crossentropy": 1.9804581493139266,
+      "loss/hidden": 3.32265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.189727352373302,
+      "step": 27440
+    },
+    {
+      "epoch": 0.22875,
+      "grad_norm": 30.875,
+      "grad_norm_var": 5.664518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.122,
+      "loss/crossentropy": 2.057726627588272,
+      "loss/hidden": 3.367578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17681959159672261,
+      "step": 27450
+    },
+    {
+      "epoch": 0.22883333333333333,
+      "grad_norm": 37.5,
+      "grad_norm_var": 7.671875,
+      "learning_rate": 0.0001,
+      "loss": 7.2149,
+      "loss/crossentropy": 1.9036873772740364,
+      "loss/hidden": 3.371484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18807742707431316,
+      "step": 27460
+    },
+    {
+      "epoch": 0.22891666666666666,
+      "grad_norm": 30.375,
+      "grad_norm_var": 8.478059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2383,
+      "loss/crossentropy": 1.9230199366807939,
+      "loss/hidden": 3.453515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19383216928690672,
+      "step": 27470
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.071809895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1673,
+      "loss/crossentropy": 1.8595254063606261,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17235099440440535,
+      "step": 27480
+    },
+    {
+      "epoch": 0.22908333333333333,
+      "grad_norm": 28.25,
+      "grad_norm_var": 9.06507739659257e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1966,
+      "loss/crossentropy": 1.96918486058712,
+      "loss/hidden": 3.441015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19443108905106782,
+      "step": 27490
+    },
+    {
+      "epoch": 0.22916666666666666,
+      "grad_norm": 30.5,
+      "grad_norm_var": 4.160872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1124,
+      "loss/crossentropy": 1.9865930266678333,
+      "loss/hidden": 3.378515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18117357343435286,
+      "step": 27500
+    },
+    {
+      "epoch": 0.22925,
+      "grad_norm": 35.0,
+      "grad_norm_var": 333.27916666666664,
+      "learning_rate": 0.0001,
+      "loss": 7.1528,
+      "loss/crossentropy": 1.9410887464880944,
+      "loss/hidden": 3.376171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18673182874917985,
+      "step": 27510
+    },
+    {
+      "epoch": 0.22933333333333333,
+      "grad_norm": 33.75,
+      "grad_norm_var": 5.040625,
+      "learning_rate": 0.0001,
+      "loss": 7.1241,
+      "loss/crossentropy": 1.881496834754944,
+      "loss/hidden": 3.19375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16761683002114297,
+      "step": 27520
+    },
+    {
+      "epoch": 0.22941666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 4.74375,
+      "learning_rate": 0.0001,
+      "loss": 7.2614,
+      "loss/crossentropy": 1.888000276684761,
+      "loss/hidden": 3.471484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19276460837572812,
+      "step": 27530
+    },
+    {
+      "epoch": 0.2295,
+      "grad_norm": 33.5,
+      "grad_norm_var": 1.1260582170580225e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2518,
+      "loss/crossentropy": 1.8867227509617805,
+      "loss/hidden": 3.41328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2147750338539481,
+      "step": 27540
+    },
+    {
+      "epoch": 0.22958333333333333,
+      "grad_norm": 30.75,
+      "grad_norm_var": 22.076497395833332,
+      "learning_rate": 0.0001,
+      "loss": 7.1127,
+      "loss/crossentropy": 2.0613905027508737,
+      "loss/hidden": 3.303125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18643336817622186,
+      "step": 27550
+    },
+    {
+      "epoch": 0.22966666666666666,
+      "grad_norm": 30.125,
+      "grad_norm_var": 20.876497395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1738,
+      "loss/crossentropy": 1.8951681800186635,
+      "loss/hidden": 3.34765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18999544940888882,
+      "step": 27560
+    },
+    {
+      "epoch": 0.22975,
+      "grad_norm": 29.0,
+      "grad_norm_var": 20.633333333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2042,
+      "loss/crossentropy": 1.9467280194163323,
+      "loss/hidden": 3.2921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18563881926238537,
+      "step": 27570
+    },
+    {
+      "epoch": 0.22983333333333333,
+      "grad_norm": 33.25,
+      "grad_norm_var": 5.339322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0993,
+      "loss/crossentropy": 1.963101963698864,
+      "loss/hidden": 3.296484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17428113296627998,
+      "step": 27580
+    },
+    {
+      "epoch": 0.22991666666666666,
+      "grad_norm": 30.0,
+      "grad_norm_var": 3.5634765625,
+      "learning_rate": 0.0001,
+      "loss": 7.2365,
+      "loss/crossentropy": 1.890703658759594,
+      "loss/hidden": 3.409375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20103781912475824,
+      "step": 27590
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 33.25,
+      "grad_norm_var": 1.7957682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.17,
+      "loss/crossentropy": 1.995637485384941,
+      "loss/hidden": 3.28828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1792814839631319,
+      "step": 27600
+    },
+    {
+      "epoch": 0.23008333333333333,
+      "grad_norm": 31.125,
+      "grad_norm_var": 1.5676432291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1334,
+      "loss/crossentropy": 1.9646591037511825,
+      "loss/hidden": 3.33984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19813134893774986,
+      "step": 27610
+    },
+    {
+      "epoch": 0.23016666666666666,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.481184895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2763,
+      "loss/crossentropy": 1.8000656798481942,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18027634639292955,
+      "step": 27620
+    },
+    {
+      "epoch": 0.23025,
+      "grad_norm": 29.375,
+      "grad_norm_var": 2.378125,
+      "learning_rate": 0.0001,
+      "loss": 7.1013,
+      "loss/crossentropy": 1.9079209670424462,
+      "loss/hidden": 3.339453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18559508733451366,
+      "step": 27630
+    },
+    {
+      "epoch": 0.23033333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 5.742643229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.2083,
+      "loss/crossentropy": 2.0068946465849877,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19677400682121515,
+      "step": 27640
+    },
+    {
+      "epoch": 0.23041666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 232.22024739583333,
+      "learning_rate": 0.0001,
+      "loss": 7.2233,
+      "loss/crossentropy": 1.8868644028902053,
+      "loss/hidden": 3.233203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1909199321642518,
+      "step": 27650
+    },
+    {
+      "epoch": 0.2305,
+      "grad_norm": 36.5,
+      "grad_norm_var": 11.338997395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1986,
+      "loss/crossentropy": 1.9457387745380401,
+      "loss/hidden": 3.4015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2033731983974576,
+      "step": 27660
+    },
+    {
+      "epoch": 0.23058333333333333,
+      "grad_norm": 32.5,
+      "grad_norm_var": 4.171809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1852,
+      "loss/crossentropy": 1.9001018285751343,
+      "loss/hidden": 3.394921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19174335710704327,
+      "step": 27670
+    },
+    {
+      "epoch": 0.23066666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.451041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0582,
+      "loss/crossentropy": 1.916756871342659,
+      "loss/hidden": 3.270703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17665382772684096,
+      "step": 27680
+    },
+    {
+      "epoch": 0.23075,
+      "grad_norm": 32.0,
+      "grad_norm_var": 5.047916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0935,
+      "loss/crossentropy": 1.8924582168459891,
+      "loss/hidden": 3.2984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17363310251384972,
+      "step": 27690
+    },
+    {
+      "epoch": 0.23083333333333333,
+      "grad_norm": 29.5,
+      "grad_norm_var": 4.7244140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2235,
+      "loss/crossentropy": 2.0114047855138777,
+      "loss/hidden": 3.427734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2055151153355837,
+      "step": 27700
+    },
+    {
+      "epoch": 0.23091666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 9.371875,
+      "learning_rate": 0.0001,
+      "loss": 7.2316,
+      "loss/crossentropy": 2.0463429734110834,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18475891146808862,
+      "step": 27710
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 30.125,
+      "grad_norm_var": 4.508072916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.039,
+      "loss/crossentropy": 2.0385479748249056,
+      "loss/hidden": 3.358984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17953770719468592,
+      "step": 27720
+    },
+    {
+      "epoch": 0.23108333333333334,
+      "grad_norm": 39.25,
+      "grad_norm_var": 4.906284653475966e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.145,
+      "loss/crossentropy": 1.9630020201206206,
+      "loss/hidden": 3.45234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18972373753786087,
+      "step": 27730
+    },
+    {
+      "epoch": 0.23116666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.906284651695659e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0511,
+      "loss/crossentropy": 2.0196684926748274,
+      "loss/hidden": 3.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1826843611896038,
+      "step": 27740
+    },
+    {
+      "epoch": 0.23125,
+      "grad_norm": 30.625,
+      "grad_norm_var": 3.0122395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.037,
+      "loss/crossentropy": 1.9496258698403834,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19348469637334348,
+      "step": 27750
+    },
+    {
+      "epoch": 0.23133333333333334,
+      "grad_norm": 33.5,
+      "grad_norm_var": 4.828125,
+      "learning_rate": 0.0001,
+      "loss": 7.0436,
+      "loss/crossentropy": 1.959471721947193,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18149461448192597,
+      "step": 27760
+    },
+    {
+      "epoch": 0.23141666666666666,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.2541015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0977,
+      "loss/crossentropy": 1.9670984774827958,
+      "loss/hidden": 3.194140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1750069571658969,
+      "step": 27770
+    },
+    {
+      "epoch": 0.2315,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.656184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0955,
+      "loss/crossentropy": 1.9592193782329559,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18199043311178684,
+      "step": 27780
+    },
+    {
+      "epoch": 0.23158333333333334,
+      "grad_norm": 34.75,
+      "grad_norm_var": 7.335872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1428,
+      "loss/crossentropy": 1.88242307305336,
+      "loss/hidden": 3.370703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1717664568684995,
+      "step": 27790
+    },
+    {
+      "epoch": 0.23166666666666666,
+      "grad_norm": 34.25,
+      "grad_norm_var": 5.851041666666666,
+      "learning_rate": 0.0001,
+      "loss": 6.9942,
+      "loss/crossentropy": 1.899032047390938,
+      "loss/hidden": 3.29453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1815300617367029,
+      "step": 27800
+    },
+    {
+      "epoch": 0.23175,
+      "grad_norm": 28.75,
+      "grad_norm_var": 4.037434895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1342,
+      "loss/crossentropy": 1.8215820118784904,
+      "loss/hidden": 3.36171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17661795169115066,
+      "step": 27810
+    },
+    {
+      "epoch": 0.23183333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.096809895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0974,
+      "loss/crossentropy": 1.958702638745308,
+      "loss/hidden": 3.262109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17975962162017822,
+      "step": 27820
+    },
+    {
+      "epoch": 0.23191666666666666,
+      "grad_norm": 28.625,
+      "grad_norm_var": 47.3166015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0944,
+      "loss/crossentropy": 1.8525143072009087,
+      "loss/hidden": 3.4625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19761906899511814,
+      "step": 27830
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 28.875,
+      "grad_norm_var": 46.10625,
+      "learning_rate": 0.0001,
+      "loss": 7.1034,
+      "loss/crossentropy": 1.9569281414151192,
+      "loss/hidden": 3.292578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780390538275242,
+      "step": 27840
+    },
+    {
+      "epoch": 0.23208333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.7330729166666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1192,
+      "loss/crossentropy": 1.9151904314756394,
+      "loss/hidden": 3.337109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17840450406074523,
+      "step": 27850
+    },
+    {
+      "epoch": 0.23216666666666666,
+      "grad_norm": 31.25,
+      "grad_norm_var": 1.9309895833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1703,
+      "loss/crossentropy": 1.9035947397351265,
+      "loss/hidden": 3.315625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18102405425161122,
+      "step": 27860
+    },
+    {
+      "epoch": 0.23225,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.456184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1439,
+      "loss/crossentropy": 1.9543967366218566,
+      "loss/hidden": 3.33671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1841458534821868,
+      "step": 27870
+    },
+    {
+      "epoch": 0.23233333333333334,
+      "grad_norm": 31.75,
+      "grad_norm_var": 4.641080729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0693,
+      "loss/crossentropy": 1.7356991663575172,
+      "loss/hidden": 3.325,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16167342327535153,
+      "step": 27880
+    },
+    {
+      "epoch": 0.23241666666666666,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.326041666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1102,
+      "loss/crossentropy": 1.980505609512329,
+      "loss/hidden": 3.3484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18476668316870928,
+      "step": 27890
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 32.75,
+      "grad_norm_var": 1.1710062543119583e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2854,
+      "loss/crossentropy": 1.9565278634428978,
+      "loss/hidden": 3.410546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18145386315882206,
+      "step": 27900
+    },
+    {
+      "epoch": 0.23258333333333334,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.1710062547042304e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.065,
+      "loss/crossentropy": 1.9382911235094071,
+      "loss/hidden": 3.255078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16825210005044938,
+      "step": 27910
+    },
+    {
+      "epoch": 0.23266666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.21640625,
+      "learning_rate": 0.0001,
+      "loss": 7.2105,
+      "loss/crossentropy": 2.0439094960689546,
+      "loss/hidden": 3.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1978301014751196,
+      "step": 27920
+    },
+    {
+      "epoch": 0.23275,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.1020833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.209,
+      "loss/crossentropy": 1.9764700159430504,
+      "loss/hidden": 3.29765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.181573954410851,
+      "step": 27930
+    },
+    {
+      "epoch": 0.23283333333333334,
+      "grad_norm": 50.0,
+      "grad_norm_var": 23.423958333333335,
+      "learning_rate": 0.0001,
+      "loss": 7.1603,
+      "loss/crossentropy": 1.8444764196872712,
+      "loss/hidden": 3.321875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17761189825832843,
+      "step": 27940
+    },
+    {
+      "epoch": 0.23291666666666666,
+      "grad_norm": 29.0,
+      "grad_norm_var": 25.778125,
+      "learning_rate": 0.0001,
+      "loss": 7.1092,
+      "loss/crossentropy": 1.9597385331988335,
+      "loss/hidden": 3.43515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18163458220660686,
+      "step": 27950
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 29.75,
+      "grad_norm_var": 5.237955729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0083,
+      "loss/crossentropy": 1.9658189684152603,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18846516273915767,
+      "step": 27960
+    },
+    {
+      "epoch": 0.23308333333333334,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.7957682291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.2289,
+      "loss/crossentropy": 2.031617395579815,
+      "loss/hidden": 3.469921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19132843110710382,
+      "step": 27970
+    },
+    {
+      "epoch": 0.23316666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 22.587239583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.2326,
+      "loss/crossentropy": 1.8775870144367217,
+      "loss/hidden": 3.335546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17848015744239093,
+      "step": 27980
+    },
+    {
+      "epoch": 0.23325,
+      "grad_norm": 30.75,
+      "grad_norm_var": 24.90390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0803,
+      "loss/crossentropy": 1.7753734052181245,
+      "loss/hidden": 3.396484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1669750839471817,
+      "step": 27990
+    },
+    {
+      "epoch": 0.23333333333333334,
+      "grad_norm": 35.25,
+      "grad_norm_var": 5.8509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0889,
+      "loss/crossentropy": 1.8468929290771485,
+      "loss/hidden": 3.44296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18284159786999227,
+      "step": 28000
+    },
+    {
+      "epoch": 0.23341666666666666,
+      "grad_norm": 30.625,
+      "grad_norm_var": 8.056517377549577e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0605,
+      "loss/crossentropy": 1.967296339571476,
+      "loss/hidden": 3.38359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1843493614345789,
+      "step": 28010
+    },
+    {
+      "epoch": 0.2335,
+      "grad_norm": 28.25,
+      "grad_norm_var": 8.056517378783752e+17,
+      "learning_rate": 0.0001,
+      "loss": 6.9998,
+      "loss/crossentropy": 1.9485969483852386,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17147983238101006,
+      "step": 28020
+    },
+    {
+      "epoch": 0.23358333333333334,
+      "grad_norm": 30.875,
+      "grad_norm_var": 3.870833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1837,
+      "loss/crossentropy": 1.8758643388748169,
+      "loss/hidden": 3.2671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17025366369634867,
+      "step": 28030
+    },
+    {
+      "epoch": 0.23366666666666666,
+      "grad_norm": 29.375,
+      "grad_norm_var": 88.38723958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1548,
+      "loss/crossentropy": 2.022619427740574,
+      "loss/hidden": 3.40859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18955955728888513,
+      "step": 28040
+    },
+    {
+      "epoch": 0.23375,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.428125,
+      "learning_rate": 0.0001,
+      "loss": 7.1513,
+      "loss/crossentropy": 2.019335350394249,
+      "loss/hidden": 3.257421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17942013666033746,
+      "step": 28050
+    },
+    {
+      "epoch": 0.23383333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 0.71640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1725,
+      "loss/crossentropy": 2.056171551346779,
+      "loss/hidden": 3.2640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17913210503757,
+      "step": 28060
+    },
+    {
+      "epoch": 0.23391666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.2738932291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0943,
+      "loss/crossentropy": 1.977993954718113,
+      "loss/hidden": 3.27578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17863202467560768,
+      "step": 28070
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.4229166666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0954,
+      "loss/crossentropy": 1.901370932161808,
+      "loss/hidden": 3.496875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19455235209316016,
+      "step": 28080
+    },
+    {
+      "epoch": 0.23408333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 5.102018229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0566,
+      "loss/crossentropy": 2.03947092294693,
+      "loss/hidden": 3.27421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17749351877719163,
+      "step": 28090
+    },
+    {
+      "epoch": 0.23416666666666666,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.195833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.091,
+      "loss/crossentropy": 1.934491528570652,
+      "loss/hidden": 3.244921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.164643563516438,
+      "step": 28100
+    },
+    {
+      "epoch": 0.23425,
+      "grad_norm": 29.625,
+      "grad_norm_var": 2.474739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1275,
+      "loss/crossentropy": 1.9412788435816766,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18773472849279643,
+      "step": 28110
+    },
+    {
+      "epoch": 0.23433333333333334,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.3764973958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0455,
+      "loss/crossentropy": 2.063568714261055,
+      "loss/hidden": 3.362109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17677181120961905,
+      "step": 28120
+    },
+    {
+      "epoch": 0.23441666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 4.020768229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1121,
+      "loss/crossentropy": 1.8816739857196807,
+      "loss/hidden": 3.57109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21015746258199214,
+      "step": 28130
+    },
+    {
+      "epoch": 0.2345,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.949739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1553,
+      "loss/crossentropy": 1.7247467294335366,
+      "loss/hidden": 3.416796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19285904094576836,
+      "step": 28140
+    },
+    {
+      "epoch": 0.23458333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 182.9759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1795,
+      "loss/crossentropy": 1.88283511698246,
+      "loss/hidden": 3.357421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17643056586384773,
+      "step": 28150
+    },
+    {
+      "epoch": 0.23466666666666666,
+      "grad_norm": 29.625,
+      "grad_norm_var": 183.18098958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1364,
+      "loss/crossentropy": 1.8417801693081857,
+      "loss/hidden": 3.485546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1907374018803239,
+      "step": 28160
+    },
+    {
+      "epoch": 0.23475,
+      "grad_norm": 30.125,
+      "grad_norm_var": 1.5759765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0267,
+      "loss/crossentropy": 1.8955570936203003,
+      "loss/hidden": 3.35390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1749934259802103,
+      "step": 28170
+    },
+    {
+      "epoch": 0.23483333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.8427083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1461,
+      "loss/crossentropy": 1.7635036900639534,
+      "loss/hidden": 3.269140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16406409572809935,
+      "step": 28180
+    },
+    {
+      "epoch": 0.23491666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 2.311393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1245,
+      "loss/crossentropy": 1.8766773998737336,
+      "loss/hidden": 3.394140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17512669004499912,
+      "step": 28190
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.2122395833333335,
+      "learning_rate": 0.0001,
+      "loss": 7.0458,
+      "loss/crossentropy": 1.9205218426883222,
+      "loss/hidden": 3.28671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17881693318486214,
+      "step": 28200
+    },
+    {
+      "epoch": 0.23508333333333334,
+      "grad_norm": 30.125,
+      "grad_norm_var": 2.2827473958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0986,
+      "loss/crossentropy": 1.9086599498987198,
+      "loss/hidden": 3.32421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19183070510625838,
+      "step": 28210
+    },
+    {
+      "epoch": 0.23516666666666666,
+      "grad_norm": 30.75,
+      "grad_norm_var": 3.692643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2031,
+      "loss/crossentropy": 1.9150561094284058,
+      "loss/hidden": 3.42734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20571896992623806,
+      "step": 28220
+    },
+    {
+      "epoch": 0.23525,
+      "grad_norm": 31.875,
+      "grad_norm_var": 7.201822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1181,
+      "loss/crossentropy": 1.9541972309350968,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1935982121154666,
+      "step": 28230
+    },
+    {
+      "epoch": 0.23533333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 0.7872395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0893,
+      "loss/crossentropy": 1.985153666138649,
+      "loss/hidden": 3.39921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19399791844189168,
+      "step": 28240
+    },
+    {
+      "epoch": 0.23541666666666666,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1.34140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2882,
+      "loss/crossentropy": 1.940797159075737,
+      "loss/hidden": 3.487890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19267026241868734,
+      "step": 28250
+    },
+    {
+      "epoch": 0.2355,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.2059895833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0739,
+      "loss/crossentropy": 2.008839640021324,
+      "loss/hidden": 3.33046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17725307308137417,
+      "step": 28260
+    },
+    {
+      "epoch": 0.23558333333333334,
+      "grad_norm": 31.125,
+      "grad_norm_var": 2.1020833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1245,
+      "loss/crossentropy": 2.046760416030884,
+      "loss/hidden": 3.316015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17304104641079904,
+      "step": 28270
+    },
+    {
+      "epoch": 0.23566666666666666,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.994205729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1558,
+      "loss/crossentropy": 1.917816373705864,
+      "loss/hidden": 3.3453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1739194754511118,
+      "step": 28280
+    },
+    {
+      "epoch": 0.23575,
+      "grad_norm": 38.5,
+      "grad_norm_var": 320.7619140625,
+      "learning_rate": 0.0001,
+      "loss": 7.2803,
+      "loss/crossentropy": 1.977133461833,
+      "loss/hidden": 3.287890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17967441882938145,
+      "step": 28290
+    },
+    {
+      "epoch": 0.23583333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 313.0768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.106,
+      "loss/crossentropy": 1.9159955084323883,
+      "loss/hidden": 3.282421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17695919778198005,
+      "step": 28300
+    },
+    {
+      "epoch": 0.23591666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.1754557291666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0956,
+      "loss/crossentropy": 1.9123063787817955,
+      "loss/hidden": 3.318359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1840075297281146,
+      "step": 28310
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.381705729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1154,
+      "loss/crossentropy": 1.918743271380663,
+      "loss/hidden": 3.321484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809801382943988,
+      "step": 28320
+    },
+    {
+      "epoch": 0.23608333333333334,
+      "grad_norm": 30.5,
+      "grad_norm_var": 2.87265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1334,
+      "loss/crossentropy": 1.8941898062825202,
+      "loss/hidden": 3.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17906710784882307,
+      "step": 28330
+    },
+    {
+      "epoch": 0.23616666666666666,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.0072916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0389,
+      "loss/crossentropy": 1.8745123133063317,
+      "loss/hidden": 3.454296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19220603834837674,
+      "step": 28340
+    },
+    {
+      "epoch": 0.23625,
+      "grad_norm": 32.25,
+      "grad_norm_var": 17.1150390625,
+      "learning_rate": 0.0001,
+      "loss": 7.1386,
+      "loss/crossentropy": 1.969007197022438,
+      "loss/hidden": 3.29609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17624503392726182,
+      "step": 28350
+    },
+    {
+      "epoch": 0.23633333333333334,
+      "grad_norm": 29.375,
+      "grad_norm_var": 16.59140625,
+      "learning_rate": 0.0001,
+      "loss": 7.1462,
+      "loss/crossentropy": 1.983209139108658,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1789218747988343,
+      "step": 28360
+    },
+    {
+      "epoch": 0.23641666666666666,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.481184895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1307,
+      "loss/crossentropy": 2.0268557518720627,
+      "loss/hidden": 3.37734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19520417749881744,
+      "step": 28370
+    },
+    {
+      "epoch": 0.2365,
+      "grad_norm": 35.0,
+      "grad_norm_var": 4.139583333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0894,
+      "loss/crossentropy": 2.1167716801166536,
+      "loss/hidden": 3.41953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1965515911579132,
+      "step": 28380
+    },
+    {
+      "epoch": 0.23658333333333334,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.699868155204076e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2317,
+      "loss/crossentropy": 1.9840284824371337,
+      "loss/hidden": 3.387890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1855791164562106,
+      "step": 28390
+    },
+    {
+      "epoch": 0.23666666666666666,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.4659920198839043e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1876,
+      "loss/crossentropy": 1.9338719502091408,
+      "loss/hidden": 3.37578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1777909006923437,
+      "step": 28400
+    },
+    {
+      "epoch": 0.23675,
+      "grad_norm": 40.5,
+      "grad_norm_var": 1.2334315326382715e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2184,
+      "loss/crossentropy": 1.9514154016971588,
+      "loss/hidden": 3.304296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17409740798175335,
+      "step": 28410
+    },
+    {
+      "epoch": 0.23683333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 13.929166666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1608,
+      "loss/crossentropy": 1.8895763352513313,
+      "loss/hidden": 3.442578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19579068031162022,
+      "step": 28420
+    },
+    {
+      "epoch": 0.23691666666666666,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.1260416666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1872,
+      "loss/crossentropy": 2.056773643195629,
+      "loss/hidden": 3.391015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1792487060651183,
+      "step": 28430
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 29.125,
+      "grad_norm_var": 1.4072916666666666,
+      "learning_rate": 0.0001,
+      "loss": 6.9928,
+      "loss/crossentropy": 1.9713964492082596,
+      "loss/hidden": 3.25859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19105317927896975,
+      "step": 28440
+    },
+    {
+      "epoch": 0.23708333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 1.5509765625,
+      "learning_rate": 0.0001,
+      "loss": 7.0923,
+      "loss/crossentropy": 1.9143947891891002,
+      "loss/hidden": 3.2640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18168039117008447,
+      "step": 28450
+    },
+    {
+      "epoch": 0.23716666666666666,
+      "grad_norm": 32.0,
+      "grad_norm_var": 25.730989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1441,
+      "loss/crossentropy": 2.0951196521520616,
+      "loss/hidden": 3.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16775632444769145,
+      "step": 28460
+    },
+    {
+      "epoch": 0.23725,
+      "grad_norm": 35.0,
+      "grad_norm_var": 7.832217003657878e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0638,
+      "loss/crossentropy": 1.8629944987595082,
+      "loss/hidden": 3.40234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1974106090143323,
+      "step": 28470
+    },
+    {
+      "epoch": 0.23733333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 7.832217004690376e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.109,
+      "loss/crossentropy": 1.97913136780262,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18055109083652496,
+      "step": 28480
+    },
+    {
+      "epoch": 0.23741666666666666,
+      "grad_norm": 41.25,
+      "grad_norm_var": 13.47265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1125,
+      "loss/crossentropy": 2.1005553156137466,
+      "loss/hidden": 3.255859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1802096463739872,
+      "step": 28490
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 30.625,
+      "grad_norm_var": 15.075455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1136,
+      "loss/crossentropy": 1.9337881088256836,
+      "loss/hidden": 3.3125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18060240298509597,
+      "step": 28500
+    },
+    {
+      "epoch": 0.23758333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 9.41640625,
+      "learning_rate": 0.0001,
+      "loss": 6.9693,
+      "loss/crossentropy": 1.8861358433961868,
+      "loss/hidden": 3.2765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17229306865483524,
+      "step": 28510
+    },
+    {
+      "epoch": 0.23766666666666666,
+      "grad_norm": 27.25,
+      "grad_norm_var": 9.176041666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1162,
+      "loss/crossentropy": 2.039627140760422,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18696105442941188,
+      "step": 28520
+    },
+    {
+      "epoch": 0.23775,
+      "grad_norm": 33.25,
+      "grad_norm_var": 46.24557291666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0511,
+      "loss/crossentropy": 1.9077040404081345,
+      "loss/hidden": 3.380078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.182385216280818,
+      "step": 28530
+    },
+    {
+      "epoch": 0.23783333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 50.32890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0775,
+      "loss/crossentropy": 2.0057973437011243,
+      "loss/hidden": 3.36953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16893456382676958,
+      "step": 28540
+    },
+    {
+      "epoch": 0.23791666666666667,
+      "grad_norm": 29.625,
+      "grad_norm_var": 39.055989583333336,
+      "learning_rate": 0.0001,
+      "loss": 7.1102,
+      "loss/crossentropy": 1.9796912029385567,
+      "loss/hidden": 3.401171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19601555690169334,
+      "step": 28550
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 34.75,
+      "grad_norm_var": 42.90833333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2015,
+      "loss/crossentropy": 2.041183885931969,
+      "loss/hidden": 3.275,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1937493771314621,
+      "step": 28560
+    },
+    {
+      "epoch": 0.23808333333333334,
+      "grad_norm": 31.25,
+      "grad_norm_var": 3.40625,
+      "learning_rate": 0.0001,
+      "loss": 7.1823,
+      "loss/crossentropy": 1.8784384042024613,
+      "loss/hidden": 3.2359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15812398698180913,
+      "step": 28570
+    },
+    {
+      "epoch": 0.23816666666666667,
+      "grad_norm": 33.5,
+      "grad_norm_var": 2.968489583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.2025,
+      "loss/crossentropy": 1.954780527949333,
+      "loss/hidden": 3.405859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19987259786576034,
+      "step": 28580
+    },
+    {
+      "epoch": 0.23825,
+      "grad_norm": 28.25,
+      "grad_norm_var": 3.6014973958333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2089,
+      "loss/crossentropy": 1.962501013278961,
+      "loss/hidden": 3.373046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19429302960634232,
+      "step": 28590
+    },
+    {
+      "epoch": 0.23833333333333334,
+      "grad_norm": 33.25,
+      "grad_norm_var": 2.802018229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.138,
+      "loss/crossentropy": 2.0587940007448196,
+      "loss/hidden": 3.248046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18460742980241776,
+      "step": 28600
+    },
+    {
+      "epoch": 0.23841666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.596875,
+      "learning_rate": 0.0001,
+      "loss": 7.2439,
+      "loss/crossentropy": 1.9444520816206932,
+      "loss/hidden": 3.415625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18984251283109188,
+      "step": 28610
+    },
+    {
+      "epoch": 0.2385,
+      "grad_norm": 32.0,
+      "grad_norm_var": 49.556705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1837,
+      "loss/crossentropy": 1.8991711422801019,
+      "loss/hidden": 3.405859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19750791862607003,
+      "step": 28620
+    },
+    {
+      "epoch": 0.23858333333333334,
+      "grad_norm": 31.375,
+      "grad_norm_var": 1120.10625,
+      "learning_rate": 0.0001,
+      "loss": 7.259,
+      "loss/crossentropy": 1.96170809045434,
+      "loss/hidden": 3.352734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18910102304071189,
+      "step": 28630
+    },
+    {
+      "epoch": 0.23866666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 6.827351302743972e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2781,
+      "loss/crossentropy": 1.9471130222082138,
+      "loss/hidden": 3.358203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18848676588386298,
+      "step": 28640
+    },
+    {
+      "epoch": 0.23875,
+      "grad_norm": 35.25,
+      "grad_norm_var": 3.49140625,
+      "learning_rate": 0.0001,
+      "loss": 7.0337,
+      "loss/crossentropy": 2.007591152191162,
+      "loss/hidden": 3.383203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18906123116612433,
+      "step": 28650
+    },
+    {
+      "epoch": 0.23883333333333334,
+      "grad_norm": 31.875,
+      "grad_norm_var": 4.255208333333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1309,
+      "loss/crossentropy": 1.928679521381855,
+      "loss/hidden": 3.24765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16594836842268706,
+      "step": 28660
+    },
+    {
+      "epoch": 0.23891666666666667,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.755143229166667,
+      "learning_rate": 0.0001,
+      "loss": 6.9913,
+      "loss/crossentropy": 1.8956782035529613,
+      "loss/hidden": 3.3265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18167539816349745,
+      "step": 28670
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.6247395833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0914,
+      "loss/crossentropy": 2.0241030544042586,
+      "loss/hidden": 3.30546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18768082708120346,
+      "step": 28680
+    },
+    {
+      "epoch": 0.23908333333333334,
+      "grad_norm": 31.0,
+      "grad_norm_var": 2.903580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2082,
+      "loss/crossentropy": 1.9375281184911728,
+      "loss/hidden": 3.400390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19355784971266984,
+      "step": 28690
+    },
+    {
+      "epoch": 0.23916666666666667,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.209830729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.2215,
+      "loss/crossentropy": 1.8107972949743272,
+      "loss/hidden": 3.331640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1827746696770191,
+      "step": 28700
+    },
+    {
+      "epoch": 0.23925,
+      "grad_norm": 3355443200.0,
+      "grad_norm_var": 7.036874287812882e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.116,
+      "loss/crossentropy": 1.9909470707178116,
+      "loss/hidden": 3.262890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1865258850157261,
+      "step": 28710
+    },
+    {
+      "epoch": 0.23933333333333334,
+      "grad_norm": 28.375,
+      "grad_norm_var": 7.036874284772012e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.0786,
+      "loss/crossentropy": 1.9539726614952087,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18131066150963307,
+      "step": 28720
+    },
+    {
+      "epoch": 0.23941666666666667,
+      "grad_norm": 32.25,
+      "grad_norm_var": 41.5197265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1467,
+      "loss/crossentropy": 2.0862886130809786,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1997101480141282,
+      "step": 28730
+    },
+    {
+      "epoch": 0.2395,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.034309895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2236,
+      "loss/crossentropy": 1.928671945631504,
+      "loss/hidden": 3.347265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18612642604857682,
+      "step": 28740
+    },
+    {
+      "epoch": 0.23958333333333334,
+      "grad_norm": 29.125,
+      "grad_norm_var": 8.39765625,
+      "learning_rate": 0.0001,
+      "loss": 7.132,
+      "loss/crossentropy": 1.9359063521027564,
+      "loss/hidden": 3.333984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17668631281703712,
+      "step": 28750
+    },
+    {
+      "epoch": 0.23966666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.660872395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.2023,
+      "loss/crossentropy": 2.113462457060814,
+      "loss/hidden": 3.28984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1961621779948473,
+      "step": 28760
+    },
+    {
+      "epoch": 0.23975,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.120833333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1154,
+      "loss/crossentropy": 1.9160461634397508,
+      "loss/hidden": 3.27421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18458777330815793,
+      "step": 28770
+    },
+    {
+      "epoch": 0.23983333333333334,
+      "grad_norm": 30.0,
+      "grad_norm_var": 5.55390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0269,
+      "loss/crossentropy": 1.9210067354142666,
+      "loss/hidden": 3.268359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17120173536241054,
+      "step": 28780
+    },
+    {
+      "epoch": 0.23991666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.415625,
+      "learning_rate": 0.0001,
+      "loss": 7.0378,
+      "loss/crossentropy": 1.9652040712535381,
+      "loss/hidden": 3.28125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17696336582303046,
+      "step": 28790
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 30.75,
+      "grad_norm_var": 64.45826822916666,
+      "learning_rate": 0.0001,
+      "loss": 7.1611,
+      "loss/crossentropy": 1.9047557294368744,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1698973761871457,
+      "step": 28800
+    },
+    {
+      "epoch": 0.24008333333333334,
+      "grad_norm": 29.625,
+      "grad_norm_var": 3.9875,
+      "learning_rate": 0.0001,
+      "loss": 7.1515,
+      "loss/crossentropy": 1.857559971511364,
+      "loss/hidden": 3.4484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18930272348225116,
+      "step": 28810
+    },
+    {
+      "epoch": 0.24016666666666667,
+      "grad_norm": 39.0,
+      "grad_norm_var": 8.199739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1916,
+      "loss/crossentropy": 2.057189577817917,
+      "loss/hidden": 3.334765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18551570512354373,
+      "step": 28820
+    },
+    {
+      "epoch": 0.24025,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.76015625,
+      "learning_rate": 0.0001,
+      "loss": 7.0911,
+      "loss/crossentropy": 1.942581658065319,
+      "loss/hidden": 3.308984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18429636009968817,
+      "step": 28830
+    },
+    {
+      "epoch": 0.24033333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 3.661393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1687,
+      "loss/crossentropy": 2.0083205685019494,
+      "loss/hidden": 3.3921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20171780474483966,
+      "step": 28840
+    },
+    {
+      "epoch": 0.24041666666666667,
+      "grad_norm": 28.5,
+      "grad_norm_var": 7.353580729166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0565,
+      "loss/crossentropy": 1.903568771481514,
+      "loss/hidden": 3.454296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16533924285322427,
+      "step": 28850
+    },
+    {
+      "epoch": 0.2405,
+      "grad_norm": 29.25,
+      "grad_norm_var": 22.46640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1802,
+      "loss/crossentropy": 1.945820477604866,
+      "loss/hidden": 3.412109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19535621600225567,
+      "step": 28860
+    },
+    {
+      "epoch": 0.24058333333333334,
+      "grad_norm": 36.5,
+      "grad_norm_var": 1.0388009774292666e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.199,
+      "loss/crossentropy": 1.8470682509243488,
+      "loss/hidden": 3.356640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17288553221151232,
+      "step": 28870
+    },
+    {
+      "epoch": 0.24066666666666667,
+      "grad_norm": 39.5,
+      "grad_norm_var": 1.038800976817737e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1072,
+      "loss/crossentropy": 1.880460461974144,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1683451456949115,
+      "step": 28880
+    },
+    {
+      "epoch": 0.24075,
+      "grad_norm": 31.375,
+      "grad_norm_var": 10.059309895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.0757,
+      "loss/crossentropy": 1.929717454314232,
+      "loss/hidden": 3.322265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18224339447915555,
+      "step": 28890
+    },
+    {
+      "epoch": 0.24083333333333334,
+      "grad_norm": 28.875,
+      "grad_norm_var": 9.107291666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0714,
+      "loss/crossentropy": 1.8345098130404949,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18774579502642155,
+      "step": 28900
+    },
+    {
+      "epoch": 0.24091666666666667,
+      "grad_norm": 30.5,
+      "grad_norm_var": 31.654166666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0389,
+      "loss/crossentropy": 2.0085376158356665,
+      "loss/hidden": 3.32578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18396059479564428,
+      "step": 28910
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 30.875,
+      "grad_norm_var": 24.931705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0286,
+      "loss/crossentropy": 1.9365892857313156,
+      "loss/hidden": 3.374609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17444046605378388,
+      "step": 28920
+    },
+    {
+      "epoch": 0.24108333333333334,
+      "grad_norm": 29.25,
+      "grad_norm_var": 191.18645833333332,
+      "learning_rate": 0.0001,
+      "loss": 7.1216,
+      "loss/crossentropy": 1.8819866538047791,
+      "loss/hidden": 3.226171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16597810126841067,
+      "step": 28930
+    },
+    {
+      "epoch": 0.24116666666666667,
+      "grad_norm": 35.0,
+      "grad_norm_var": 14.012955729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0173,
+      "loss/crossentropy": 2.009412518143654,
+      "loss/hidden": 3.220703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17949128895998,
+      "step": 28940
+    },
+    {
+      "epoch": 0.24125,
+      "grad_norm": 30.625,
+      "grad_norm_var": 9.7181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.008,
+      "loss/crossentropy": 1.960886961966753,
+      "loss/hidden": 3.32265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17062551975250245,
+      "step": 28950
+    },
+    {
+      "epoch": 0.24133333333333334,
+      "grad_norm": 43.25,
+      "grad_norm_var": 30.167708333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.2025,
+      "loss/crossentropy": 1.9925682991743088,
+      "loss/hidden": 3.30078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17259026784449816,
+      "step": 28960
+    },
+    {
+      "epoch": 0.24141666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 28.588997395833335,
+      "learning_rate": 0.0001,
+      "loss": 7.0311,
+      "loss/crossentropy": 1.98997260928154,
+      "loss/hidden": 3.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19723801352083684,
+      "step": 28970
+    },
+    {
+      "epoch": 0.2415,
+      "grad_norm": 34.5,
+      "grad_norm_var": 20.501822916666665,
+      "learning_rate": 0.0001,
+      "loss": 7.1469,
+      "loss/crossentropy": 1.9680213287472725,
+      "loss/hidden": 3.231640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17318310346454382,
+      "step": 28980
+    },
+    {
+      "epoch": 0.24158333333333334,
+      "grad_norm": 32.25,
+      "grad_norm_var": 6.855989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0672,
+      "loss/crossentropy": 1.8203165039420128,
+      "loss/hidden": 3.4234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18160245288163424,
+      "step": 28990
+    },
+    {
+      "epoch": 0.24166666666666667,
+      "grad_norm": 34.0,
+      "grad_norm_var": 14.589518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.3081,
+      "loss/crossentropy": 2.091747897863388,
+      "loss/hidden": 3.46796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2050326919183135,
+      "step": 29000
+    },
+    {
+      "epoch": 0.24175,
+      "grad_norm": 33.5,
+      "grad_norm_var": 15.649739583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1634,
+      "loss/crossentropy": 1.864132596552372,
+      "loss/hidden": 3.41484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19338729809969663,
+      "step": 29010
+    },
+    {
+      "epoch": 0.24183333333333334,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.575455729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0064,
+      "loss/crossentropy": 2.032469463348389,
+      "loss/hidden": 3.41171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1867183681577444,
+      "step": 29020
+    },
+    {
+      "epoch": 0.24191666666666667,
+      "grad_norm": 35.75,
+      "grad_norm_var": 6.649739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0996,
+      "loss/crossentropy": 1.8639530837535858,
+      "loss/hidden": 3.406640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18421055134385825,
+      "step": 29030
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 33.0,
+      "grad_norm_var": 4.260872395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.1809,
+      "loss/crossentropy": 2.039676921069622,
+      "loss/hidden": 3.284375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19269126188009977,
+      "step": 29040
+    },
+    {
+      "epoch": 0.24208333333333334,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.4400390625,
+      "learning_rate": 0.0001,
+      "loss": 7.0644,
+      "loss/crossentropy": 1.9928770169615746,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19527556244283914,
+      "step": 29050
+    },
+    {
+      "epoch": 0.24216666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 4.115625,
+      "learning_rate": 0.0001,
+      "loss": 7.0506,
+      "loss/crossentropy": 2.0005327865481375,
+      "loss/hidden": 3.237890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17999807205051183,
+      "step": 29060
+    },
+    {
+      "epoch": 0.24225,
+      "grad_norm": 30.75,
+      "grad_norm_var": 5.044205729166666,
+      "learning_rate": 0.0001,
+      "loss": 6.9399,
+      "loss/crossentropy": 2.010702919960022,
+      "loss/hidden": 3.217578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1866823337972164,
+      "step": 29070
+    },
+    {
+      "epoch": 0.24233333333333335,
+      "grad_norm": 33.75,
+      "grad_norm_var": 24.819791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1194,
+      "loss/crossentropy": 2.033090981841087,
+      "loss/hidden": 3.3109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18511803727596998,
+      "step": 29080
+    },
+    {
+      "epoch": 0.24241666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 22.289322916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.1307,
+      "loss/crossentropy": 2.0014316767454146,
+      "loss/hidden": 3.409765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2168583981692791,
+      "step": 29090
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 29.625,
+      "grad_norm_var": 8.610416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0001,
+      "loss/crossentropy": 1.693302933871746,
+      "loss/hidden": 3.30546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16856823153793812,
+      "step": 29100
+    },
+    {
+      "epoch": 0.24258333333333335,
+      "grad_norm": 30.0,
+      "grad_norm_var": 7.094205729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1148,
+      "loss/crossentropy": 2.0196719616651535,
+      "loss/hidden": 3.2203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16759639885276556,
+      "step": 29110
+    },
+    {
+      "epoch": 0.24266666666666667,
+      "grad_norm": 34.5,
+      "grad_norm_var": 4.561393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.1775,
+      "loss/crossentropy": 2.0070349991321565,
+      "loss/hidden": 3.44375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19561398085206746,
+      "step": 29120
+    },
+    {
+      "epoch": 0.24275,
+      "grad_norm": 31.125,
+      "grad_norm_var": 4.941666666666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0416,
+      "loss/crossentropy": 1.753032784163952,
+      "loss/hidden": 3.3671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17061343006789684,
+      "step": 29130
+    },
+    {
+      "epoch": 0.24283333333333335,
+      "grad_norm": 27.75,
+      "grad_norm_var": 48.21432291666667,
+      "learning_rate": 0.0001,
+      "loss": 6.9941,
+      "loss/crossentropy": 1.846510899811983,
+      "loss/hidden": 3.259765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17116302344948053,
+      "step": 29140
+    },
+    {
+      "epoch": 0.24291666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 28.587239583333332,
+      "learning_rate": 0.0001,
+      "loss": 7.0891,
+      "loss/crossentropy": 1.9794099092483521,
+      "loss/hidden": 3.279296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16955801863223313,
+      "step": 29150
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 32.75,
+      "grad_norm_var": 9.025,
+      "learning_rate": 0.0001,
+      "loss": 6.9463,
+      "loss/crossentropy": 1.8831994347274303,
+      "loss/hidden": 3.196875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17682631043717265,
+      "step": 29160
+    },
+    {
+      "epoch": 0.24308333333333335,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.23125,
+      "learning_rate": 0.0001,
+      "loss": 7.1007,
+      "loss/crossentropy": 1.9835622057318687,
+      "loss/hidden": 3.34296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16830469761043787,
+      "step": 29170
+    },
+    {
+      "epoch": 0.24316666666666667,
+      "grad_norm": 43.0,
+      "grad_norm_var": 16.401822916666667,
+      "learning_rate": 0.0001,
+      "loss": 6.999,
+      "loss/crossentropy": 1.8381020322442054,
+      "loss/hidden": 3.3359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17938976171426474,
+      "step": 29180
+    },
+    {
+      "epoch": 0.24325,
+      "grad_norm": 36.5,
+      "grad_norm_var": 15.02890625,
+      "learning_rate": 0.0001,
+      "loss": 7.0233,
+      "loss/crossentropy": 2.0556397050619126,
+      "loss/hidden": 3.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1735588414594531,
+      "step": 29190
+    },
+    {
+      "epoch": 0.24333333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 12.167643229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0437,
+      "loss/crossentropy": 1.9663966223597527,
+      "loss/hidden": 3.26328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1763702655211091,
+      "step": 29200
+    },
+    {
+      "epoch": 0.24341666666666667,
+      "grad_norm": 36.5,
+      "grad_norm_var": 31.372916666666665,
+      "learning_rate": 0.0001,
+      "loss": 7.0654,
+      "loss/crossentropy": 2.007933700084686,
+      "loss/hidden": 3.329296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17167334407567977,
+      "step": 29210
+    },
+    {
+      "epoch": 0.2435,
+      "grad_norm": 31.75,
+      "grad_norm_var": 74.62083333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1378,
+      "loss/crossentropy": 1.906649786233902,
+      "loss/hidden": 3.315234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18802914079278707,
+      "step": 29220
+    },
+    {
+      "epoch": 0.24358333333333335,
+      "grad_norm": 31.625,
+      "grad_norm_var": 47.697916666666664,
+      "learning_rate": 0.0001,
+      "loss": 7.1162,
+      "loss/crossentropy": 1.944581964612007,
+      "loss/hidden": 3.344921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17684601247310638,
+      "step": 29230
+    },
+    {
+      "epoch": 0.24366666666666667,
+      "grad_norm": 32.75,
+      "grad_norm_var": 58.14270833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0986,
+      "loss/crossentropy": 1.9346801176667214,
+      "loss/hidden": 3.31328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1813599731773138,
+      "step": 29240
+    },
+    {
+      "epoch": 0.24375,
+      "grad_norm": 34.5,
+      "grad_norm_var": 29.025,
+      "learning_rate": 0.0001,
+      "loss": 6.9542,
+      "loss/crossentropy": 1.9739003598690033,
+      "loss/hidden": 3.348046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18150815181434155,
+      "step": 29250
+    },
+    {
+      "epoch": 0.24383333333333335,
+      "grad_norm": 34.0,
+      "grad_norm_var": 25.415625,
+      "learning_rate": 0.0001,
+      "loss": 7.15,
+      "loss/crossentropy": 1.9636050060391426,
+      "loss/hidden": 3.340234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1886537389829755,
+      "step": 29260
+    },
+    {
+      "epoch": 0.24391666666666667,
+      "grad_norm": 34.75,
+      "grad_norm_var": 27.34765625,
+      "learning_rate": 0.0001,
+      "loss": 7.044,
+      "loss/crossentropy": 2.0582940846681597,
+      "loss/hidden": 3.233984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17936681099236013,
+      "step": 29270
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 38.5,
+      "grad_norm_var": 25.623958333333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0325,
+      "loss/crossentropy": 1.9108435377478599,
+      "loss/hidden": 3.35234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2016294825822115,
+      "step": 29280
+    },
+    {
+      "epoch": 0.24408333333333335,
+      "grad_norm": 35.25,
+      "grad_norm_var": 16.103059895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.027,
+      "loss/crossentropy": 1.9197970561683178,
+      "loss/hidden": 3.25390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.168515251763165,
+      "step": 29290
+    },
+    {
+      "epoch": 0.24416666666666667,
+      "grad_norm": 34.25,
+      "grad_norm_var": 14.414518229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0807,
+      "loss/crossentropy": 2.1058178797364233,
+      "loss/hidden": 3.3765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18827453423291446,
+      "step": 29300
+    },
+    {
+      "epoch": 0.24425,
+      "grad_norm": 29.375,
+      "grad_norm_var": 12.861393229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0362,
+      "loss/crossentropy": 1.7994571581482888,
+      "loss/hidden": 3.26015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16001259442418814,
+      "step": 29310
+    },
+    {
+      "epoch": 0.24433333333333335,
+      "grad_norm": 31.25,
+      "grad_norm_var": 7.549739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1525,
+      "loss/crossentropy": 2.01288483440876,
+      "loss/hidden": 3.362890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19834488816559315,
+      "step": 29320
+    },
+    {
+      "epoch": 0.24441666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 10.5212890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1113,
+      "loss/crossentropy": 1.8625710979104042,
+      "loss/hidden": 3.180078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.15823894497007132,
+      "step": 29330
+    },
+    {
+      "epoch": 0.2445,
+      "grad_norm": 29.625,
+      "grad_norm_var": 6.350779036021556e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1831,
+      "loss/crossentropy": 1.8274046443402767,
+      "loss/hidden": 3.2453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17070938218384982,
+      "step": 29340
+    },
+    {
+      "epoch": 0.24458333333333335,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.420247395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.0417,
+      "loss/crossentropy": 1.928498236835003,
+      "loss/hidden": 3.316015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18636565264314414,
+      "step": 29350
+    },
+    {
+      "epoch": 0.24466666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 4.759375,
+      "learning_rate": 0.0001,
+      "loss": 7.0604,
+      "loss/crossentropy": 1.8408309891819954,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1754040101543069,
+      "step": 29360
+    },
+    {
+      "epoch": 0.24475,
+      "grad_norm": 28.625,
+      "grad_norm_var": 8.255989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0692,
+      "loss/crossentropy": 2.0853366553783417,
+      "loss/hidden": 3.29296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19377084486186505,
+      "step": 29370
+    },
+    {
+      "epoch": 0.24483333333333332,
+      "grad_norm": 27.625,
+      "grad_norm_var": 16.3978515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1115,
+      "loss/crossentropy": 1.8351604774594308,
+      "loss/hidden": 3.2703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16586365969851613,
+      "step": 29380
+    },
+    {
+      "epoch": 0.24491666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 28.538541666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1788,
+      "loss/crossentropy": 1.9523347169160843,
+      "loss/hidden": 3.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17606011740863323,
+      "step": 29390
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 30.875,
+      "grad_norm_var": 6.417805252538053e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2163,
+      "loss/crossentropy": 1.7989266946911813,
+      "loss/hidden": 3.6203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19965290129184723,
+      "step": 29400
+    },
+    {
+      "epoch": 0.24508333333333332,
+      "grad_norm": 28.625,
+      "grad_norm_var": 70.6931640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1635,
+      "loss/crossentropy": 1.885206399857998,
+      "loss/hidden": 3.4265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17618857119232417,
+      "step": 29410
+    },
+    {
+      "epoch": 0.24516666666666667,
+      "grad_norm": 31.0,
+      "grad_norm_var": 170.95618489583333,
+      "learning_rate": 0.0001,
+      "loss": 7.1485,
+      "loss/crossentropy": 2.013509748876095,
+      "loss/hidden": 3.355859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17810031324625014,
+      "step": 29420
+    },
+    {
+      "epoch": 0.24525,
+      "grad_norm": 28.875,
+      "grad_norm_var": 172.98014322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.1189,
+      "loss/crossentropy": 2.0351277559995653,
+      "loss/hidden": 3.307421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17403772547841073,
+      "step": 29430
+    },
+    {
+      "epoch": 0.24533333333333332,
+      "grad_norm": 29.25,
+      "grad_norm_var": 3.312239583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0794,
+      "loss/crossentropy": 1.9156136021018029,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18918067030608654,
+      "step": 29440
+    },
+    {
+      "epoch": 0.24541666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.8947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1379,
+      "loss/crossentropy": 1.8753335192799567,
+      "loss/hidden": 3.46328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1829879242926836,
+      "step": 29450
+    },
+    {
+      "epoch": 0.2455,
+      "grad_norm": 31.125,
+      "grad_norm_var": 39.36145833333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1471,
+      "loss/crossentropy": 1.9919239401817321,
+      "loss/hidden": 3.317578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1804093485698104,
+      "step": 29460
+    },
+    {
+      "epoch": 0.24558333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 10.5212890625,
+      "learning_rate": 0.0001,
+      "loss": 7.2239,
+      "loss/crossentropy": 2.0031524688005446,
+      "loss/hidden": 3.36796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1932737486436963,
+      "step": 29470
+    },
+    {
+      "epoch": 0.24566666666666667,
+      "grad_norm": 41.0,
+      "grad_norm_var": 106.97389322916666,
+      "learning_rate": 0.0001,
+      "loss": 7.2368,
+      "loss/crossentropy": 1.9022671937942506,
+      "loss/hidden": 3.426953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18752275481820108,
+      "step": 29480
+    },
+    {
+      "epoch": 0.24575,
+      "grad_norm": 30.375,
+      "grad_norm_var": 106.02545572916667,
+      "learning_rate": 0.0001,
+      "loss": 7.1407,
+      "loss/crossentropy": 1.9851775020360947,
+      "loss/hidden": 3.343359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17676679585129024,
+      "step": 29490
+    },
+    {
+      "epoch": 0.24583333333333332,
+      "grad_norm": 35.75,
+      "grad_norm_var": 3.6113932291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.213,
+      "loss/crossentropy": 2.048147828876972,
+      "loss/hidden": 3.328515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18454352170228958,
+      "step": 29500
+    },
+    {
+      "epoch": 0.24591666666666667,
+      "grad_norm": 32.0,
+      "grad_norm_var": 17.034375,
+      "learning_rate": 0.0001,
+      "loss": 7.098,
+      "loss/crossentropy": 1.8598377980291843,
+      "loss/hidden": 3.341015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1807760160882026,
+      "step": 29510
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 38.75,
+      "grad_norm_var": 4.3362978295718445e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.2694,
+      "loss/crossentropy": 2.0400223821401595,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19468207955360411,
+      "step": 29520
+    },
+    {
+      "epoch": 0.24608333333333332,
+      "grad_norm": 32.25,
+      "grad_norm_var": 1.0307613420640623e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1781,
+      "loss/crossentropy": 2.050127571821213,
+      "loss/hidden": 3.23984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1719770599156618,
+      "step": 29530
+    },
+    {
+      "epoch": 0.24616666666666667,
+      "grad_norm": 31.875,
+      "grad_norm_var": 6.689428611981641e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1596,
+      "loss/crossentropy": 2.0742219775915145,
+      "loss/hidden": 3.219140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17910778634250163,
+      "step": 29540
+    },
+    {
+      "epoch": 0.24625,
+      "grad_norm": 32.25,
+      "grad_norm_var": 4.1462890625,
+      "learning_rate": 0.0001,
+      "loss": 7.1458,
+      "loss/crossentropy": 2.051993590593338,
+      "loss/hidden": 3.272265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17897749189287424,
+      "step": 29550
+    },
+    {
+      "epoch": 0.24633333333333332,
+      "grad_norm": 30.625,
+      "grad_norm_var": 1.8025390625,
+      "learning_rate": 0.0001,
+      "loss": 6.9734,
+      "loss/crossentropy": 1.8892281383275986,
+      "loss/hidden": 3.212890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1630697438493371,
+      "step": 29560
+    },
+    {
+      "epoch": 0.24641666666666667,
+      "grad_norm": 29.5,
+      "grad_norm_var": 6.632291666666666,
+      "learning_rate": 0.0001,
+      "loss": 6.9846,
+      "loss/crossentropy": 1.903426329791546,
+      "loss/hidden": 3.277734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17775410879403353,
+      "step": 29570
+    },
+    {
+      "epoch": 0.2465,
+      "grad_norm": 29.0,
+      "grad_norm_var": 5.68515625,
+      "learning_rate": 0.0001,
+      "loss": 7.0957,
+      "loss/crossentropy": 1.898318810760975,
+      "loss/hidden": 3.333203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18351435456424953,
+      "step": 29580
+    },
+    {
+      "epoch": 0.24658333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.5372395833333334,
+      "learning_rate": 0.0001,
+      "loss": 7.1606,
+      "loss/crossentropy": 1.9313055925071239,
+      "loss/hidden": 3.3609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18387855887413024,
+      "step": 29590
+    },
+    {
+      "epoch": 0.24666666666666667,
+      "grad_norm": 32.5,
+      "grad_norm_var": 2.479622395833333,
+      "learning_rate": 0.0001,
+      "loss": 7.3307,
+      "loss/crossentropy": 1.982286486029625,
+      "loss/hidden": 3.418359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19546118825674058,
+      "step": 29600
+    },
+    {
+      "epoch": 0.24675,
+      "grad_norm": 30.0,
+      "grad_norm_var": 2.3728515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1375,
+      "loss/crossentropy": 1.8560178518295287,
+      "loss/hidden": 3.355078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17613652739673852,
+      "step": 29610
+    },
+    {
+      "epoch": 0.24683333333333332,
+      "grad_norm": 33.5,
+      "grad_norm_var": 56.978580729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1199,
+      "loss/crossentropy": 2.0633393064141274,
+      "loss/hidden": 3.26796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19099162686616183,
+      "step": 29620
+    },
+    {
+      "epoch": 0.24691666666666667,
+      "grad_norm": 30.875,
+      "grad_norm_var": 1.7610877448142195e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.2221,
+      "loss/crossentropy": 2.004875534772873,
+      "loss/hidden": 3.540234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21292517967522145,
+      "step": 29630
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 2902458368.0,
+      "grad_norm_var": 1.2668578594134776e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.251,
+      "loss/crossentropy": 1.9059226460754872,
+      "loss/hidden": 3.267578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.16682118121534587,
+      "step": 29640
+    },
+    {
+      "epoch": 0.24708333333333332,
+      "grad_norm": 29.5,
+      "grad_norm_var": 1.2668578590795062e+18,
+      "learning_rate": 0.0001,
+      "loss": 7.1071,
+      "loss/crossentropy": 1.8222531959414483,
+      "loss/hidden": 3.366015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1752259062603116,
+      "step": 29650
+    },
+    {
+      "epoch": 0.24716666666666667,
+      "grad_norm": 45.75,
+      "grad_norm_var": 3.853392321682867e+17,
+      "learning_rate": 0.0001,
+      "loss": 7.1803,
+      "loss/crossentropy": 1.8945453360676765,
+      "loss/hidden": 3.414453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19549553301185368,
+      "step": 29660
+    },
+    {
+      "epoch": 0.24725,
+      "grad_norm": 32.75,
+      "grad_norm_var": 28.358072916666668,
+      "learning_rate": 0.0001,
+      "loss": 7.2325,
+      "loss/crossentropy": 2.0085205674171447,
+      "loss/hidden": 3.380859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19520842619240283,
+      "step": 29670
+    },
+    {
+      "epoch": 0.24733333333333332,
+      "grad_norm": 29.375,
+      "grad_norm_var": 4.576822916666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0009,
+      "loss/crossentropy": 2.0199078917503357,
+      "loss/hidden": 3.239453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17313358131796122,
+      "step": 29680
+    },
+    {
+      "epoch": 0.24741666666666667,
+      "grad_norm": 41.75,
+      "grad_norm_var": 11.720768229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.284,
+      "loss/crossentropy": 2.153177574276924,
+      "loss/hidden": 3.244140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18877132665365934,
+      "step": 29690
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 32.0,
+      "grad_norm_var": 11.106705729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0622,
+      "loss/crossentropy": 1.9068312883377074,
+      "loss/hidden": 3.287109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19678921941667796,
+      "step": 29700
+    },
+    {
+      "epoch": 0.24758333333333332,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.591080729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.0804,
+      "loss/crossentropy": 1.8381352588534354,
+      "loss/hidden": 3.368359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18390652108937502,
+      "step": 29710
+    },
+    {
+      "epoch": 0.24766666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.1181640625,
+      "learning_rate": 0.0001,
+      "loss": 7.165,
+      "loss/crossentropy": 2.007746438682079,
+      "loss/hidden": 3.341796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1871322887018323,
+      "step": 29720
+    },
+    {
+      "epoch": 0.24775,
+      "grad_norm": 30.0,
+      "grad_norm_var": 1.08515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1062,
+      "loss/crossentropy": 2.0214229375123978,
+      "loss/hidden": 3.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1926682161167264,
+      "step": 29730
+    },
+    {
+      "epoch": 0.24783333333333332,
+      "grad_norm": 32.0,
+      "grad_norm_var": 1.8080729166666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1341,
+      "loss/crossentropy": 1.9575099855661393,
+      "loss/hidden": 3.334765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17581414431333542,
+      "step": 29740
+    },
+    {
+      "epoch": 0.24791666666666667,
+      "grad_norm": 30.625,
+      "grad_norm_var": 6.298372395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1319,
+      "loss/crossentropy": 2.023171266913414,
+      "loss/hidden": 3.415625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1809300735592842,
+      "step": 29750
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 31.625,
+      "grad_norm_var": 3.280143229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.0423,
+      "loss/crossentropy": 1.9499368622899056,
+      "loss/hidden": 3.265234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17184760514646769,
+      "step": 29760
+    },
+    {
+      "epoch": 0.24808333333333332,
+      "grad_norm": 30.25,
+      "grad_norm_var": 3.497916666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1069,
+      "loss/crossentropy": 2.012082076072693,
+      "loss/hidden": 3.39296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17977374251931905,
+      "step": 29770
+    },
+    {
+      "epoch": 0.24816666666666667,
+      "grad_norm": 29.875,
+      "grad_norm_var": 1.0843098958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.1189,
+      "loss/crossentropy": 1.8774370312690736,
+      "loss/hidden": 3.413671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19751372523605823,
+      "step": 29780
+    },
+    {
+      "epoch": 0.24825,
+      "grad_norm": 34.5,
+      "grad_norm_var": 2.91875,
+      "learning_rate": 0.0001,
+      "loss": 7.0911,
+      "loss/crossentropy": 1.9262201353907584,
+      "loss/hidden": 3.25078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.17088953591883183,
+      "step": 29790
+    },
+    {
+      "epoch": 0.24833333333333332,
+      "grad_norm": 29.875,
+      "grad_norm_var": 5.454622395833334,
+      "learning_rate": 0.0001,
+      "loss": 7.1529,
+      "loss/crossentropy": 1.830883041024208,
+      "loss/hidden": 3.510546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2032022202387452,
+      "step": 29800
+    },
+    {
+      "epoch": 0.24841666666666667,
+      "grad_norm": 29.0,
+      "grad_norm_var": 41.073893229166664,
+      "learning_rate": 0.0001,
+      "loss": 7.1897,
+      "loss/crossentropy": 1.971740435063839,
+      "loss/hidden": 3.425,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20232045296579598,
+      "step": 29810
+    },
+    {
+      "epoch": 0.2485,
+      "grad_norm": 34.25,
+      "grad_norm_var": 40.583072916666666,
+      "learning_rate": 0.0001,
+      "loss": 7.227,
+      "loss/crossentropy": 2.053927531838417,
+      "loss/hidden": 3.3,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18464548140764236,
+      "step": 29820
+    },
+    {
+      "epoch": 0.24858333333333332,
+      "grad_norm": 31.875,
+      "grad_norm_var": 1.86875,
+      "learning_rate": 0.0001,
+      "loss": 7.1605,
+      "loss/crossentropy": 1.997328954935074,
+      "loss/hidden": 3.292578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18435040749609471,
+      "step": 29830
+    },
+    {
+      "epoch": 0.24866666666666667,
+      "grad_norm": 29.25,
+      "grad_norm_var": 1.9442057291666666,
+      "learning_rate": 0.0001,
+      "loss": 7.0646,
+      "loss/crossentropy": 1.814618805795908,
+      "loss/hidden": 3.428515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18235307652503252,
+      "step": 29840
+    },
+    {
+      "epoch": 0.24875,
+      "grad_norm": 31.875,
+      "grad_norm_var": 3.674934895833333,
+      "learning_rate": 0.0001,
+      "loss": 7.2329,
+      "loss/crossentropy": 2.074957937002182,
+      "loss/hidden": 3.2703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18078752495348455,
+      "step": 29850
+    },
+    {
+      "epoch": 0.24883333333333332,
+      "grad_norm": 29.5,
+      "grad_norm_var": 2.0822265625,
+      "learning_rate": 0.0001,
+      "loss": 7.0902,
+      "loss/crossentropy": 2.029583752155304,
+      "loss/hidden": 3.379296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19445701725780964,
+      "step": 29860
+    },
+    {
+      "epoch": 0.24891666666666667,
+      "grad_norm": 31.625,
+      "grad_norm_var": 2.4947265625,
+      "learning_rate": 0.0001,
+      "loss": 7.1805,
+      "loss/crossentropy": 1.9934955328702926,
+      "loss/hidden": 3.273828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1727988539263606,
+      "step": 29870
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 32.0,
+      "grad_norm_var": 6.305989583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.0855,
+      "loss/crossentropy": 1.8810793429613113,
+      "loss/hidden": 3.28671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1780039880424738,
+      "step": 29880
+    },
+    {
+      "epoch": 0.24908333333333332,
+      "grad_norm": 29.75,
+      "grad_norm_var": 7.33515625,
+      "learning_rate": 0.0001,
+      "loss": 7.1347,
+      "loss/crossentropy": 1.9258745312690735,
+      "loss/hidden": 3.53828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19972074814140797,
+      "step": 29890
+    },
+    {
+      "epoch": 0.24916666666666668,
+      "grad_norm": 29.625,
+      "grad_norm_var": 64.16712239583333,
+      "learning_rate": 0.0001,
+      "loss": 7.163,
+      "loss/crossentropy": 1.8959374621510505,
+      "loss/hidden": 3.441796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.20531977619975805,
+      "step": 29900
+    },
+    {
+      "epoch": 0.24925,
+      "grad_norm": 31.0,
+      "grad_norm_var": 81.09765625,
+      "learning_rate": 0.0001,
+      "loss": 7.1171,
+      "loss/crossentropy": 1.9798023968935012,
+      "loss/hidden": 3.26875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18218860533088446,
+      "step": 29910
+    },
+    {
+      "epoch": 0.24933333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 36.1556640625,
+      "learning_rate": 0.0001,
+      "loss": 7.1113,
+      "loss/crossentropy": 2.0541242122650147,
+      "loss/hidden": 3.250390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1778485918417573,
+      "step": 29920
+    },
+    {
+      "epoch": 0.24941666666666668,
+      "grad_norm": 31.5,
+      "grad_norm_var": 39.701822916666664,
+      "learning_rate": 0.0001,
+      "loss": 7.1558,
+      "loss/crossentropy": 2.0165991961956022,
+      "loss/hidden": 3.24765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1778322727419436,
+      "step": 29930
+    },
+    {
+      "epoch": 0.2495,
+      "grad_norm": 31.125,
+      "grad_norm_var": 30.787955729166665,
+      "learning_rate": 0.0001,
+      "loss": 7.1946,
+      "loss/crossentropy": 1.9186311453580855,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18566787131130696,
+      "step": 29940
+    },
+    {
+      "epoch": 0.24958333333333332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 24.963997395833335,
+      "learning_rate": 0.0001,
+      "loss": 7.1259,
+      "loss/crossentropy": 2.087065522372723,
+      "loss/hidden": 3.340625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1846929408609867,
+      "step": 29950
+    },
+    {
+      "epoch": 0.24966666666666668,
+      "grad_norm": 37.75,
+      "grad_norm_var": 31.292643229166668,
+      "learning_rate": 0.0001,
+      "loss": 7.1004,
+      "loss/crossentropy": 2.089809614419937,
+      "loss/hidden": 3.34609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1884284181520343,
+      "step": 29960
+    },
+    {
+      "epoch": 0.24975,
+      "grad_norm": 30.125,
+      "grad_norm_var": 36.197330729166666,
+      "learning_rate": 0.0001,
+      "loss": 7.1446,
+      "loss/crossentropy": 1.9828887164592743,
+      "loss/hidden": 3.330078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18696920815855264,
+      "step": 29970
+    },
+    {
+      "epoch": 0.24983333333333332,
+      "grad_norm": 40.5,
+      "grad_norm_var": 25.710416666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.0843,
+      "loss/crossentropy": 1.8713553190231322,
+      "loss/hidden": 3.3703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.18477697428315878,
+      "step": 29980
+    },
+    {
+      "epoch": 0.24991666666666668,
+      "grad_norm": 33.5,
+      "grad_norm_var": 10.069791666666667,
+      "learning_rate": 0.0001,
+      "loss": 7.1491,
+      "loss/crossentropy": 1.9470527619123459,
+      "loss/hidden": 3.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.19067223686724902,
+      "step": 29990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 30.5,
+      "grad_norm_var": 9.930989583333334,
+      "learning_rate": 0.0001,
+      "loss": 7.0683,
+      "loss/crossentropy": 2.0086402118206026,
+      "loss/hidden": 3.334375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.1874341730028391,
+      "step": 30000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 120000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 10000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.57253009602642e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}