diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7233 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6,
+  "eval_steps": 2000,
+  "global_step": 6000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.001,
+      "grad_norm": 4736.0,
+      "learning_rate": 1.9e-05,
+      "loss": 132.1055,
+      "loss/crossentropy": 12.246079635620116,
+      "loss/hidden": 18.7125,
+      "loss/jsd": 0.0,
+      "loss/logits": 10.372939014434815,
+      "step": 10
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 330.0,
+      "grad_norm_var": 91640269.18333334,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 95.9731,
+      "loss/crossentropy": 8.862393474578857,
+      "loss/hidden": 18.675,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.677179157733917,
+      "step": 20
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 394.0,
+      "grad_norm_var": 237715.45,
+      "learning_rate": 3.7e-05,
+      "loss": 86.3778,
+      "loss/crossentropy": 8.083840227127075,
+      "loss/hidden": 18.259375,
+      "loss/jsd": 0.0,
+      "loss/logits": 6.130921971797943,
+      "step": 30
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 924.0,
+      "grad_norm_var": 2.6757682503402172e+16,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 82.5914,
+      "loss/crossentropy": 7.802511918544769,
+      "loss/hidden": 17.440625,
+      "loss/jsd": 0.0,
+      "loss/logits": 5.772503018379211,
+      "step": 40
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 516.0,
+      "grad_norm_var": 38597.583333333336,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 75.3397,
+      "loss/crossentropy": 7.156700026988983,
+      "loss/hidden": 17.253125,
+      "loss/jsd": 0.0,
+      "loss/logits": 5.156575608253479,
+      "step": 50
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 1232.0,
+      "grad_norm_var": 68241.45,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 61.2745,
+      "loss/crossentropy": 6.0138510942459105,
+      "loss/hidden": 15.80625,
+      "loss/jsd": 0.0,
+      "loss/logits": 3.8037488579750063,
+      "step": 60
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 376.0,
+      "grad_norm_var": 626103.4,
+      "learning_rate": 7.3e-05,
+      "loss": 41.3695,
+      "loss/crossentropy": 4.422797441482544,
+      "loss/hidden": 13.1125,
+      "loss/jsd": 0.0,
+      "loss/logits": 2.4006322652101515,
+      "step": 70
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 272.0,
+      "grad_norm_var": 674923.45,
+      "learning_rate": 8.200000000000001e-05,
+      "loss": 27.4755,
+      "loss/crossentropy": 3.3576226443052293,
+      "loss/hidden": 10.7359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.3968962401151657,
+      "step": 80
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 296.0,
+      "grad_norm_var": 15426.383333333333,
+      "learning_rate": 9.1e-05,
+      "loss": 22.6607,
+      "loss/crossentropy": 3.217679074406624,
+      "loss/hidden": 9.2140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 1.055714099109173,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 328.0,
+      "grad_norm_var": 9349.666666666666,
+      "learning_rate": 0.0001,
+      "loss": 20.3108,
+      "loss/crossentropy": 2.934060016274452,
+      "loss/hidden": 8.40703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8702833190560341,
+      "step": 100
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 194.0,
+      "grad_norm_var": 5992.866666666667,
+      "learning_rate": 0.0001,
+      "loss": 18.8852,
+      "loss/crossentropy": 2.8450062334537507,
+      "loss/hidden": 8.221875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.8380498677492142,
+      "step": 110
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 244.0,
+      "grad_norm_var": 1176.5333333333333,
+      "learning_rate": 0.0001,
+      "loss": 17.97,
+      "loss/crossentropy": 2.612249107658863,
+      "loss/hidden": 7.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.686215291172266,
+      "step": 120
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 242.0,
+      "grad_norm_var": 1168.8958333333333,
+      "learning_rate": 0.0001,
+      "loss": 17.2904,
+      "loss/crossentropy": 2.8242316216230394,
+      "loss/hidden": 7.7390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.7805894792079926,
+      "step": 130
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 179.0,
+      "grad_norm_var": 1465.1333333333334,
+      "learning_rate": 0.0001,
+      "loss": 16.5581,
+      "loss/crossentropy": 2.737143725156784,
+      "loss/hidden": 7.3421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6888546235859394,
+      "step": 140
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 175.0,
+      "grad_norm_var": 1119.8625,
+      "learning_rate": 0.0001,
+      "loss": 16.0501,
+      "loss/crossentropy": 2.7599751561880113,
+      "loss/hidden": 7.05703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.6640767879784107,
+      "step": 150
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 186.0,
+      "grad_norm_var": 1044.5166666666667,
+      "learning_rate": 0.0001,
+      "loss": 15.4631,
+      "loss/crossentropy": 2.6100075274705885,
+      "loss/hidden": 6.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5824844464659691,
+      "step": 160
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 179.0,
+      "grad_norm_var": 1082.8,
+      "learning_rate": 0.0001,
+      "loss": 15.2201,
+      "loss/crossentropy": 2.4276285111904143,
+      "loss/hidden": 6.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5915141828358174,
+      "step": 170
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 153.0,
+      "grad_norm_var": 622.6625,
+      "learning_rate": 0.0001,
+      "loss": 14.9606,
+      "loss/crossentropy": 2.630460512638092,
+      "loss/hidden": 6.52578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5396774187684059,
+      "step": 180
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 176.0,
+      "grad_norm_var": 1093.2,
+      "learning_rate": 0.0001,
+      "loss": 14.6255,
+      "loss/crossentropy": 2.3158223152160646,
+      "loss/hidden": 6.50390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4905257746577263,
+      "step": 190
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 112.0,
+      "grad_norm_var": 695.7291666666666,
+      "learning_rate": 0.0001,
+      "loss": 14.3647,
+      "loss/crossentropy": 2.586851382255554,
+      "loss/hidden": 6.42265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5586091712117195,
+      "step": 200
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 118.5,
+      "grad_norm_var": 574.3072916666666,
+      "learning_rate": 0.0001,
+      "loss": 14.0867,
+      "loss/crossentropy": 2.5010055124759676,
+      "loss/hidden": 6.34453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4965482771396637,
+      "step": 210
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 88.5,
+      "grad_norm_var": 662.65,
+      "learning_rate": 0.0001,
+      "loss": 13.6551,
+      "loss/crossentropy": 2.573444625735283,
+      "loss/hidden": 6.33125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5534068010747433,
+      "step": 220
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 118.0,
+      "grad_norm_var": 412.1958333333333,
+      "learning_rate": 0.0001,
+      "loss": 13.4715,
+      "loss/crossentropy": 2.4142292886972427,
+      "loss/hidden": 5.96640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44360905699431896,
+      "step": 230
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 134.0,
+      "grad_norm_var": 242.9,
+      "learning_rate": 0.0001,
+      "loss": 13.3289,
+      "loss/crossentropy": 2.4670142769813537,
+      "loss/hidden": 5.98671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.47392544001340864,
+      "step": 240
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 137.0,
+      "grad_norm_var": 158.4625,
+      "learning_rate": 0.0001,
+      "loss": 13.0031,
+      "loss/crossentropy": 2.416000656783581,
+      "loss/hidden": 5.7859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44607544504106045,
+      "step": 250
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 109.0,
+      "grad_norm_var": 279.990625,
+      "learning_rate": 0.0001,
+      "loss": 13.0076,
+      "loss/crossentropy": 2.370332670211792,
+      "loss/hidden": 5.9984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.5006627842783928,
+      "step": 260
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 129.0,
+      "grad_norm_var": 427.37395833333335,
+      "learning_rate": 0.0001,
+      "loss": 12.8809,
+      "loss/crossentropy": 2.281908763945103,
+      "loss/hidden": 5.98671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.45061586182564495,
+      "step": 270
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 98.0,
+      "grad_norm_var": 278.1489583333333,
+      "learning_rate": 0.0001,
+      "loss": 12.8942,
+      "loss/crossentropy": 2.3922384053468706,
+      "loss/hidden": 5.6984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44376694336533545,
+      "step": 280
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 99.5,
+      "grad_norm_var": 303.55,
+      "learning_rate": 0.0001,
+      "loss": 12.7122,
+      "loss/crossentropy": 2.730095013976097,
+      "loss/hidden": 5.49140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4411045670509338,
+      "step": 290
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 112.5,
+      "grad_norm_var": 359.56666666666666,
+      "learning_rate": 0.0001,
+      "loss": 12.5618,
+      "loss/crossentropy": 2.3741705983877184,
+      "loss/hidden": 5.43203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40091707594692705,
+      "step": 300
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 84.5,
+      "grad_norm_var": 245.25729166666667,
+      "learning_rate": 0.0001,
+      "loss": 12.2525,
+      "loss/crossentropy": 2.2781229317188263,
+      "loss/hidden": 5.53515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4274128321558237,
+      "step": 310
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 108.5,
+      "grad_norm_var": 140.59583333333333,
+      "learning_rate": 0.0001,
+      "loss": 12.2935,
+      "loss/crossentropy": 2.5757294684648513,
+      "loss/hidden": 5.4609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42916890494525434,
+      "step": 320
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 108.0,
+      "grad_norm_var": 70.89895833333334,
+      "learning_rate": 0.0001,
+      "loss": 12.1545,
+      "loss/crossentropy": 2.527638339996338,
+      "loss/hidden": 5.378125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4032053742557764,
+      "step": 330
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 210.0,
+      "grad_norm_var": 1272.465625,
+      "learning_rate": 0.0001,
+      "loss": 12.2482,
+      "loss/crossentropy": 2.5401821002364158,
+      "loss/hidden": 5.390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4444709587842226,
+      "step": 340
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 79.5,
+      "grad_norm_var": 1376.5958333333333,
+      "learning_rate": 0.0001,
+      "loss": 12.08,
+      "loss/crossentropy": 2.514840933680534,
+      "loss/hidden": 5.2640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4077944982796907,
+      "step": 350
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 87.0,
+      "grad_norm_var": 418.83229166666666,
+      "learning_rate": 0.0001,
+      "loss": 12.0245,
+      "loss/crossentropy": 2.420889538526535,
+      "loss/hidden": 5.34921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.44222328886389733,
+      "step": 360
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 76.5,
+      "grad_norm_var": 138.5625,
+      "learning_rate": 0.0001,
+      "loss": 11.7097,
+      "loss/crossentropy": 2.2826619133353234,
+      "loss/hidden": 5.3296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3849468305706978,
+      "step": 370
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 96.5,
+      "grad_norm_var": 184.93229166666666,
+      "learning_rate": 0.0001,
+      "loss": 11.465,
+      "loss/crossentropy": 2.4052042722702027,
+      "loss/hidden": 5.16796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40173302926123144,
+      "step": 380
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 125.5,
+      "grad_norm_var": 183.09583333333333,
+      "learning_rate": 0.0001,
+      "loss": 11.6273,
+      "loss/crossentropy": 2.540145033597946,
+      "loss/hidden": 5.215625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.41224894523620603,
+      "step": 390
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 83.5,
+      "grad_norm_var": 258.315625,
+      "learning_rate": 0.0001,
+      "loss": 11.397,
+      "loss/crossentropy": 2.207468980550766,
+      "loss/hidden": 5.09296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3590874429792166,
+      "step": 400
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 94.5,
+      "grad_norm_var": 184.5625,
+      "learning_rate": 0.0001,
+      "loss": 11.443,
+      "loss/crossentropy": 2.4378984421491623,
+      "loss/hidden": 5.21171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.40493359677493573,
+      "step": 410
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 106.5,
+      "grad_norm_var": 125.590625,
+      "learning_rate": 0.0001,
+      "loss": 11.5678,
+      "loss/crossentropy": 2.518555220961571,
+      "loss/hidden": 5.07265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.4297170080244541,
+      "step": 420
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 87.5,
+      "grad_norm_var": 115.765625,
+      "learning_rate": 0.0001,
+      "loss": 11.3132,
+      "loss/crossentropy": 2.490597203373909,
+      "loss/hidden": 5.11171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.403754598274827,
+      "step": 430
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 92.5,
+      "grad_norm_var": 156.35729166666667,
+      "learning_rate": 0.0001,
+      "loss": 11.1476,
+      "loss/crossentropy": 2.037529316544533,
+      "loss/hidden": 5.07421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35246654506772757,
+      "step": 440
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 80.5,
+      "grad_norm_var": 210.66666666666666,
+      "learning_rate": 0.0001,
+      "loss": 11.3038,
+      "loss/crossentropy": 2.3201738983392715,
+      "loss/hidden": 5.0828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.38196625709533694,
+      "step": 450
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 107.5,
+      "grad_norm_var": 284.1666666666667,
+      "learning_rate": 0.0001,
+      "loss": 11.3625,
+      "loss/crossentropy": 2.4791718110442162,
+      "loss/hidden": 4.95546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36495909169316293,
+      "step": 460
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 91.5,
+      "grad_norm_var": 247.39895833333333,
+      "learning_rate": 0.0001,
+      "loss": 11.0542,
+      "loss/crossentropy": 2.3155667960643767,
+      "loss/hidden": 4.93828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.362844867631793,
+      "step": 470
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 95.0,
+      "grad_norm_var": 194.79895833333333,
+      "learning_rate": 0.0001,
+      "loss": 11.2413,
+      "loss/crossentropy": 2.496318203210831,
+      "loss/hidden": 4.840625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3887303464114666,
+      "step": 480
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 74.5,
+      "grad_norm_var": 243.840625,
+      "learning_rate": 0.0001,
+      "loss": 10.9416,
+      "loss/crossentropy": 2.385223904252052,
+      "loss/hidden": 4.85234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3598880790174007,
+      "step": 490
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 79.0,
+      "grad_norm_var": 105.990625,
+      "learning_rate": 0.0001,
+      "loss": 10.9114,
+      "loss/crossentropy": 2.2462552055716514,
+      "loss/hidden": 4.80859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3265662036836147,
+      "step": 500
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 96.5,
+      "grad_norm_var": 138.43229166666666,
+      "learning_rate": 0.0001,
+      "loss": 10.8821,
+      "loss/crossentropy": 2.297148121893406,
+      "loss/hidden": 4.8609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3467547960579395,
+      "step": 510
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 97.5,
+      "grad_norm_var": 129.365625,
+      "learning_rate": 0.0001,
+      "loss": 10.9299,
+      "loss/crossentropy": 2.4197026968002318,
+      "loss/hidden": 4.7921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3632193084806204,
+      "step": 520
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 81.5,
+      "grad_norm_var": 99.47395833333333,
+      "learning_rate": 0.0001,
+      "loss": 10.787,
+      "loss/crossentropy": 2.36982424557209,
+      "loss/hidden": 4.825,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3405680742114782,
+      "step": 530
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 85.5,
+      "grad_norm_var": 48.340625,
+      "learning_rate": 0.0001,
+      "loss": 10.8675,
+      "loss/crossentropy": 2.4611779801547526,
+      "loss/hidden": 4.8625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36872007288038733,
+      "step": 540
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 93.5,
+      "grad_norm_var": 84.24895833333333,
+      "learning_rate": 0.0001,
+      "loss": 10.64,
+      "loss/crossentropy": 2.1758567959070207,
+      "loss/hidden": 4.7484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3336840860545635,
+      "step": 550
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 114.0,
+      "grad_norm_var": 129.53098958333334,
+      "learning_rate": 0.0001,
+      "loss": 10.5615,
+      "loss/crossentropy": 2.3970536097884176,
+      "loss/hidden": 4.7625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34276723079383375,
+      "step": 560
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 80.0,
+      "grad_norm_var": 579.57890625,
+      "learning_rate": 0.0001,
+      "loss": 10.8999,
+      "loss/crossentropy": 2.4695185527205465,
+      "loss/hidden": 4.9453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.42829814068973066,
+      "step": 570
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 85.0,
+      "grad_norm_var": 596.9572916666667,
+      "learning_rate": 0.0001,
+      "loss": 10.8802,
+      "loss/crossentropy": 2.3520184576511385,
+      "loss/hidden": 4.790625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3662864986807108,
+      "step": 580
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 73.0,
+      "grad_norm_var": 181.69583333333333,
+      "learning_rate": 0.0001,
+      "loss": 10.6744,
+      "loss/crossentropy": 2.2842736929655074,
+      "loss/hidden": 4.71484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3500846643000841,
+      "step": 590
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 97.0,
+      "grad_norm_var": 160.58307291666668,
+      "learning_rate": 0.0001,
+      "loss": 10.6987,
+      "loss/crossentropy": 2.29906165599823,
+      "loss/hidden": 4.602734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.334361494705081,
+      "step": 600
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 89.0,
+      "grad_norm_var": 162.67682291666668,
+      "learning_rate": 0.0001,
+      "loss": 10.6143,
+      "loss/crossentropy": 2.3032930195331573,
+      "loss/hidden": 4.6703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3258141163736582,
+      "step": 610
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 77.5,
+      "grad_norm_var": 97.12916666666666,
+      "learning_rate": 0.0001,
+      "loss": 10.5946,
+      "loss/crossentropy": 2.452244046330452,
+      "loss/hidden": 4.7109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3432691916823387,
+      "step": 620
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 75.5,
+      "grad_norm_var": 227.69973958333333,
+      "learning_rate": 0.0001,
+      "loss": 10.6287,
+      "loss/crossentropy": 2.2894835874438284,
+      "loss/hidden": 4.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35672005768865345,
+      "step": 630
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 70.0,
+      "grad_norm_var": 541.2322916666667,
+      "learning_rate": 0.0001,
+      "loss": 10.6195,
+      "loss/crossentropy": 2.4114772886037827,
+      "loss/hidden": 4.70546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.35591375902295114,
+      "step": 640
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 77.0,
+      "grad_norm_var": 435.15390625,
+      "learning_rate": 0.0001,
+      "loss": 10.4142,
+      "loss/crossentropy": 2.332440134882927,
+      "loss/hidden": 4.634375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.339809150993824,
+      "step": 650
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 71.5,
+      "grad_norm_var": 118.03307291666667,
+      "learning_rate": 0.0001,
+      "loss": 10.4602,
+      "loss/crossentropy": 2.154422373324633,
+      "loss/hidden": 4.54140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3334257358685136,
+      "step": 660
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 73.5,
+      "grad_norm_var": 144.94166666666666,
+      "learning_rate": 0.0001,
+      "loss": 10.5185,
+      "loss/crossentropy": 2.3223402693867685,
+      "loss/hidden": 4.795703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.37188967503607273,
+      "step": 670
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 61.5,
+      "grad_norm_var": 169.65598958333334,
+      "learning_rate": 0.0001,
+      "loss": 10.5323,
+      "loss/crossentropy": 2.332353001832962,
+      "loss/hidden": 4.50625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31948004066944125,
+      "step": 680
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 74.0,
+      "grad_norm_var": 155.94140625,
+      "learning_rate": 0.0001,
+      "loss": 10.4359,
+      "loss/crossentropy": 2.4077556908130644,
+      "loss/hidden": 4.623828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.339173823595047,
+      "step": 690
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 82.5,
+      "grad_norm_var": 125.55416666666666,
+      "learning_rate": 0.0001,
+      "loss": 10.4493,
+      "loss/crossentropy": 2.292634981870651,
+      "loss/hidden": 4.571875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3477486100047827,
+      "step": 700
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 88.0,
+      "grad_norm_var": 155.84166666666667,
+      "learning_rate": 0.0001,
+      "loss": 10.2041,
+      "loss/crossentropy": 2.4034020826220512,
+      "loss/hidden": 4.53046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3406600248068571,
+      "step": 710
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 124.0,
+      "grad_norm_var": 230.83307291666668,
+      "learning_rate": 0.0001,
+      "loss": 10.3489,
+      "loss/crossentropy": 2.333241228759289,
+      "loss/hidden": 4.6015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3285223826766014,
+      "step": 720
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 71.0,
+      "grad_norm_var": 278.95390625,
+      "learning_rate": 0.0001,
+      "loss": 10.1548,
+      "loss/crossentropy": 2.4066421508789064,
+      "loss/hidden": 4.682421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.338771004602313,
+      "step": 730
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 84.5,
+      "grad_norm_var": 166.85729166666667,
+      "learning_rate": 0.0001,
+      "loss": 10.2647,
+      "loss/crossentropy": 2.2724754482507707,
+      "loss/hidden": 4.567578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3267147310078144,
+      "step": 740
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 67.5,
+      "grad_norm_var": 343.5247395833333,
+      "learning_rate": 0.0001,
+      "loss": 10.2815,
+      "loss/crossentropy": 2.3046080738306047,
+      "loss/hidden": 4.473828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33236319161951544,
+      "step": 750
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 68.5,
+      "grad_norm_var": 306.540625,
+      "learning_rate": 0.0001,
+      "loss": 10.2479,
+      "loss/crossentropy": 2.2831736013293265,
+      "loss/hidden": 4.62734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3329113606363535,
+      "step": 760
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 88.5,
+      "grad_norm_var": 111.57473958333334,
+      "learning_rate": 0.0001,
+      "loss": 10.2161,
+      "loss/crossentropy": 2.3853780582547186,
+      "loss/hidden": 4.541015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31959532871842383,
+      "step": 770
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 80.5,
+      "grad_norm_var": 110.65729166666667,
+      "learning_rate": 0.0001,
+      "loss": 10.2076,
+      "loss/crossentropy": 2.3982744574546815,
+      "loss/hidden": 4.55859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3542841043323278,
+      "step": 780
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 66.0,
+      "grad_norm_var": 275.6322916666667,
+      "learning_rate": 0.0001,
+      "loss": 10.1697,
+      "loss/crossentropy": 2.4292824655771255,
+      "loss/hidden": 4.632421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.36711033545434474,
+      "step": 790
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 57.25,
+      "grad_norm_var": 290.9291666666667,
+      "learning_rate": 0.0001,
+      "loss": 10.2176,
+      "loss/crossentropy": 2.380542576313019,
+      "loss/hidden": 4.509375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3368827097117901,
+      "step": 800
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 60.75,
+      "grad_norm_var": 52.67916666666667,
+      "learning_rate": 0.0001,
+      "loss": 10.2311,
+      "loss/crossentropy": 2.4212940514087675,
+      "loss/hidden": 4.55546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.355662290379405,
+      "step": 810
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 60.75,
+      "grad_norm_var": 65.81666666666666,
+      "learning_rate": 0.0001,
+      "loss": 10.1866,
+      "loss/crossentropy": 2.4809795886278154,
+      "loss/hidden": 4.49140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3553234666585922,
+      "step": 820
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 56.5,
+      "grad_norm_var": 98.2875,
+      "learning_rate": 0.0001,
+      "loss": 9.9805,
+      "loss/crossentropy": 2.306653854250908,
+      "loss/hidden": 4.40078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3146494958549738,
+      "step": 830
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 68.0,
+      "grad_norm_var": 38.51015625,
+      "learning_rate": 0.0001,
+      "loss": 10.1087,
+      "loss/crossentropy": 2.250006601214409,
+      "loss/hidden": 4.422265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30200174674391744,
+      "step": 840
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 70.5,
+      "grad_norm_var": 43.483072916666664,
+      "learning_rate": 0.0001,
+      "loss": 10.0526,
+      "loss/crossentropy": 2.211633677780628,
+      "loss/hidden": 4.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31178686060011385,
+      "step": 850
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 61.0,
+      "grad_norm_var": 41.545572916666664,
+      "learning_rate": 0.0001,
+      "loss": 10.1915,
+      "loss/crossentropy": 2.5281356513500213,
+      "loss/hidden": 4.389453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.34625968635082244,
+      "step": 860
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 72.5,
+      "grad_norm_var": 54.475,
+      "learning_rate": 0.0001,
+      "loss": 10.0007,
+      "loss/crossentropy": 2.4020907685160635,
+      "loss/hidden": 4.326171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32252500094473363,
+      "step": 870
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 142.0,
+      "grad_norm_var": 499.1375,
+      "learning_rate": 0.0001,
+      "loss": 9.99,
+      "loss/crossentropy": 2.384984764456749,
+      "loss/hidden": 4.38515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3189360786229372,
+      "step": 880
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 57.75,
+      "grad_norm_var": 527.23515625,
+      "learning_rate": 0.0001,
+      "loss": 9.9879,
+      "loss/crossentropy": 2.3401281625032424,
+      "loss/hidden": 4.46328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3382201848551631,
+      "step": 890
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 71.5,
+      "grad_norm_var": 95.97265625,
+      "learning_rate": 0.0001,
+      "loss": 9.9352,
+      "loss/crossentropy": 2.3969784706830977,
+      "loss/hidden": 4.384375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.336395762488246,
+      "step": 900
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 79.0,
+      "grad_norm_var": 144.96666666666667,
+      "learning_rate": 0.0001,
+      "loss": 10.149,
+      "loss/crossentropy": 2.4599110893905163,
+      "loss/hidden": 4.30703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3240171395242214,
+      "step": 910
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 65.5,
+      "grad_norm_var": 119.2375,
+      "learning_rate": 0.0001,
+      "loss": 9.9634,
+      "loss/crossentropy": 2.4210876494646074,
+      "loss/hidden": 4.30390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32166178375482557,
+      "step": 920
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 63.0,
+      "grad_norm_var": 41.47083333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.744,
+      "loss/crossentropy": 2.2256636448204516,
+      "loss/hidden": 4.284765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29795306362211704,
+      "step": 930
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 53.5,
+      "grad_norm_var": 192.55807291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.8636,
+      "loss/crossentropy": 2.297808923572302,
+      "loss/hidden": 4.31640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30742434673011304,
+      "step": 940
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 61.0,
+      "grad_norm_var": 81.95729166666666,
+      "learning_rate": 0.0001,
+      "loss": 9.798,
+      "loss/crossentropy": 2.3219059616327287,
+      "loss/hidden": 4.211328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30037002861499784,
+      "step": 950
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 56.75,
+      "grad_norm_var": 61.55807291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.7449,
+      "loss/crossentropy": 2.3104363679885864,
+      "loss/hidden": 4.388671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.327311984449625,
+      "step": 960
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 60.0,
+      "grad_norm_var": 56.18932291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.9668,
+      "loss/crossentropy": 2.308886554837227,
+      "loss/hidden": 4.407421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3183224782347679,
+      "step": 970
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 66.5,
+      "grad_norm_var": 42.05416666666667,
+      "learning_rate": 0.0001,
+      "loss": 9.7807,
+      "loss/crossentropy": 2.3363482102751734,
+      "loss/hidden": 4.2921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3384779039770365,
+      "step": 980
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 57.25,
+      "grad_norm_var": 56.891666666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.7501,
+      "loss/crossentropy": 2.1767295479774473,
+      "loss/hidden": 4.466015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31410733237862587,
+      "step": 990
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 50.25,
+      "grad_norm_var": 75.85598958333334,
+      "learning_rate": 0.0001,
+      "loss": 9.9273,
+      "loss/crossentropy": 2.505411845445633,
+      "loss/hidden": 4.36015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33212706074118614,
+      "step": 1000
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 77.5,
+      "grad_norm_var": 196.21848958333334,
+      "learning_rate": 0.0001,
+      "loss": 9.9237,
+      "loss/crossentropy": 2.3281257838010787,
+      "loss/hidden": 4.35546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32293859515339135,
+      "step": 1010
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 63.25,
+      "grad_norm_var": 167.42395833333333,
+      "learning_rate": 0.0001,
+      "loss": 9.7592,
+      "loss/crossentropy": 2.3165650010108947,
+      "loss/hidden": 4.32890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31759811006486416,
+      "step": 1020
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 60.0,
+      "grad_norm_var": 153.80833333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.7366,
+      "loss/crossentropy": 2.3203016728162766,
+      "loss/hidden": 4.28515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31944827549159527,
+      "step": 1030
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 66.5,
+      "grad_norm_var": 3319.3958333333335,
+      "learning_rate": 0.0001,
+      "loss": 10.0035,
+      "loss/crossentropy": 2.4188640087842943,
+      "loss/hidden": 4.38828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3581279247999191,
+      "step": 1040
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 60.25,
+      "grad_norm_var": 3338.31640625,
+      "learning_rate": 0.0001,
+      "loss": 9.6837,
+      "loss/crossentropy": 2.2860016629099844,
+      "loss/hidden": 4.325390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.318701284006238,
+      "step": 1050
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 74.0,
+      "grad_norm_var": 112.4,
+      "learning_rate": 0.0001,
+      "loss": 9.517,
+      "loss/crossentropy": 2.4143033266067504,
+      "loss/hidden": 4.319140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3072842717170715,
+      "step": 1060
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 71.5,
+      "grad_norm_var": 68.60598958333334,
+      "learning_rate": 0.0001,
+      "loss": 9.8549,
+      "loss/crossentropy": 2.351083371043205,
+      "loss/hidden": 4.398046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.33429058492183683,
+      "step": 1070
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 53.25,
+      "grad_norm_var": 43.83229166666667,
+      "learning_rate": 0.0001,
+      "loss": 9.7738,
+      "loss/crossentropy": 2.4011227190494537,
+      "loss/hidden": 4.29453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3128178097307682,
+      "step": 1080
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 72.0,
+      "grad_norm_var": 34.82890625,
+      "learning_rate": 0.0001,
+      "loss": 9.7432,
+      "loss/crossentropy": 2.310031126439571,
+      "loss/hidden": 4.38984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3273486144840717,
+      "step": 1090
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 66.5,
+      "grad_norm_var": 111.64895833333334,
+      "learning_rate": 0.0001,
+      "loss": 9.6743,
+      "loss/crossentropy": 2.3055127263069153,
+      "loss/hidden": 4.21796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32233874313533306,
+      "step": 1100
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 51.5,
+      "grad_norm_var": 46.70729166666667,
+      "learning_rate": 0.0001,
+      "loss": 9.8026,
+      "loss/crossentropy": 2.314373381435871,
+      "loss/hidden": 4.256640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3083756107836962,
+      "step": 1110
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 57.75,
+      "grad_norm_var": 7292.4375,
+      "learning_rate": 0.0001,
+      "loss": 9.7291,
+      "loss/crossentropy": 2.5138203650712967,
+      "loss/hidden": 4.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30809955932199956,
+      "step": 1120
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 56.5,
+      "grad_norm_var": 29.190625,
+      "learning_rate": 0.0001,
+      "loss": 9.6823,
+      "loss/crossentropy": 2.2719234466552733,
+      "loss/hidden": 4.294140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3143883816897869,
+      "step": 1130
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 60.5,
+      "grad_norm_var": 45.925,
+      "learning_rate": 0.0001,
+      "loss": 9.7564,
+      "loss/crossentropy": 2.4254489660263063,
+      "loss/hidden": 4.261328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3154076419770718,
+      "step": 1140
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 56.0,
+      "grad_norm_var": 71.74583333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.7001,
+      "loss/crossentropy": 2.28252642005682,
+      "loss/hidden": 4.323046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3203336976468563,
+      "step": 1150
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 67.0,
+      "grad_norm_var": 46.040625,
+      "learning_rate": 0.0001,
+      "loss": 9.7436,
+      "loss/crossentropy": 2.391976150870323,
+      "loss/hidden": 4.225390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31455044373869895,
+      "step": 1160
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 46.0,
+      "grad_norm_var": 47.06640625,
+      "learning_rate": 0.0001,
+      "loss": 9.5622,
+      "loss/crossentropy": 2.3361207604408265,
+      "loss/hidden": 4.19296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30060703232884406,
+      "step": 1170
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 56.25,
+      "grad_norm_var": 49.264322916666664,
+      "learning_rate": 0.0001,
+      "loss": 9.6834,
+      "loss/crossentropy": 2.297483670711517,
+      "loss/hidden": 4.2890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2907493541017175,
+      "step": 1180
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 52.5,
+      "grad_norm_var": 12.27890625,
+      "learning_rate": 0.0001,
+      "loss": 9.6207,
+      "loss/crossentropy": 2.2364058643579483,
+      "loss/hidden": 4.277734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3097097765654325,
+      "step": 1190
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 68.5,
+      "grad_norm_var": 35.055989583333336,
+      "learning_rate": 0.0001,
+      "loss": 9.6018,
+      "loss/crossentropy": 2.2412969201803206,
+      "loss/hidden": 4.287109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31851550191640854,
+      "step": 1200
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 60.5,
+      "grad_norm_var": 25.774739583333332,
+      "learning_rate": 0.0001,
+      "loss": 9.6979,
+      "loss/crossentropy": 2.3062032952904703,
+      "loss/hidden": 4.258984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3124631106853485,
+      "step": 1210
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 59.25,
+      "grad_norm_var": 20.026822916666667,
+      "learning_rate": 0.0001,
+      "loss": 9.7129,
+      "loss/crossentropy": 2.4036868065595627,
+      "loss/hidden": 4.20234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31101155243813994,
+      "step": 1220
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 53.25,
+      "grad_norm_var": 75.30833333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.7047,
+      "loss/crossentropy": 2.3730016142129897,
+      "loss/hidden": 4.193359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3105484452098608,
+      "step": 1230
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 62.25,
+      "grad_norm_var": 33.27682291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.6313,
+      "loss/crossentropy": 2.2872567594051363,
+      "loss/hidden": 4.319140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3244694545865059,
+      "step": 1240
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 61.25,
+      "grad_norm_var": 25.673958333333335,
+      "learning_rate": 0.0001,
+      "loss": 9.6217,
+      "loss/crossentropy": 2.3013710603117943,
+      "loss/hidden": 4.29140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.32178852558135984,
+      "step": 1250
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 49.5,
+      "grad_norm_var": 46.97265625,
+      "learning_rate": 0.0001,
+      "loss": 9.6151,
+      "loss/crossentropy": 2.2743802405893803,
+      "loss/hidden": 4.19921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3104738780297339,
+      "step": 1260
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 52.25,
+      "grad_norm_var": 318.94557291666666,
+      "learning_rate": 0.0001,
+      "loss": 9.635,
+      "loss/crossentropy": 2.2751111879944803,
+      "loss/hidden": 4.17578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2947248375043273,
+      "step": 1270
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 59.5,
+      "grad_norm_var": 201.26848958333332,
+      "learning_rate": 0.0001,
+      "loss": 9.605,
+      "loss/crossentropy": 2.3590754181146623,
+      "loss/hidden": 4.116015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29773430675268175,
+      "step": 1280
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 50.0,
+      "grad_norm_var": 25.795833333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.4314,
+      "loss/crossentropy": 2.165515697002411,
+      "loss/hidden": 4.148046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2729496695101261,
+      "step": 1290
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 51.5,
+      "grad_norm_var": 65.69557291666666,
+      "learning_rate": 0.0001,
+      "loss": 9.4579,
+      "loss/crossentropy": 2.425456903874874,
+      "loss/hidden": 4.1140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3002984166145325,
+      "step": 1300
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 55.75,
+      "grad_norm_var": 74.63515625,
+      "learning_rate": 0.0001,
+      "loss": 9.562,
+      "loss/crossentropy": 2.3212677478790282,
+      "loss/hidden": 4.209765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28645528480410576,
+      "step": 1310
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 44.75,
+      "grad_norm_var": 39.139322916666664,
+      "learning_rate": 0.0001,
+      "loss": 9.305,
+      "loss/crossentropy": 2.2911602184176445,
+      "loss/hidden": 4.133984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28404638059437276,
+      "step": 1320
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 52.25,
+      "grad_norm_var": 76.19583333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.3122,
+      "loss/crossentropy": 2.3109163105487824,
+      "loss/hidden": 4.137109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2864396806806326,
+      "step": 1330
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 47.0,
+      "grad_norm_var": 41.66015625,
+      "learning_rate": 0.0001,
+      "loss": 9.4629,
+      "loss/crossentropy": 2.353537403047085,
+      "loss/hidden": 4.08203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2971150416880846,
+      "step": 1340
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 46.25,
+      "grad_norm_var": 45.31848958333333,
+      "learning_rate": 0.0001,
+      "loss": 9.365,
+      "loss/crossentropy": 2.3774181246757506,
+      "loss/hidden": 4.09296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2799839396029711,
+      "step": 1350
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 51.0,
+      "grad_norm_var": 17.93515625,
+      "learning_rate": 0.0001,
+      "loss": 9.3498,
+      "loss/crossentropy": 2.246833881735802,
+      "loss/hidden": 4.18828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2903384942561388,
+      "step": 1360
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 51.25,
+      "grad_norm_var": 12.420833333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.4976,
+      "loss/crossentropy": 2.453240838646889,
+      "loss/hidden": 4.173828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3144164770841599,
+      "step": 1370
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 70.0,
+      "grad_norm_var": 2011.0322916666667,
+      "learning_rate": 0.0001,
+      "loss": 9.5884,
+      "loss/crossentropy": 2.174116183817387,
+      "loss/hidden": 4.24921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2923248626291752,
+      "step": 1380
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 53.75,
+      "grad_norm_var": 1988.8833333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.5249,
+      "loss/crossentropy": 2.3638354018330574,
+      "loss/hidden": 4.184765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3066251628100872,
+      "step": 1390
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 55.5,
+      "grad_norm_var": 22.779166666666665,
+      "learning_rate": 0.0001,
+      "loss": 9.3528,
+      "loss/crossentropy": 2.4166768550872804,
+      "loss/hidden": 4.123828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29636494982987643,
+      "step": 1400
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 60.5,
+      "grad_norm_var": 66.59348958333334,
+      "learning_rate": 0.0001,
+      "loss": 9.5339,
+      "loss/crossentropy": 2.3475931867957116,
+      "loss/hidden": 4.1953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30608872696757317,
+      "step": 1410
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 51.25,
+      "grad_norm_var": 62.49140625,
+      "learning_rate": 0.0001,
+      "loss": 9.3342,
+      "loss/crossentropy": 2.1785849004983904,
+      "loss/hidden": 4.161328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27641028352081776,
+      "step": 1420
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 54.25,
+      "grad_norm_var": 30.154166666666665,
+      "learning_rate": 0.0001,
+      "loss": 9.3898,
+      "loss/crossentropy": 2.3990818440914152,
+      "loss/hidden": 4.18359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2944341886788607,
+      "step": 1430
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 51.75,
+      "grad_norm_var": 38.93932291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.4628,
+      "loss/crossentropy": 2.4946817860007284,
+      "loss/hidden": 4.198828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31867978498339655,
+      "step": 1440
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 53.75,
+      "grad_norm_var": 33.9,
+      "learning_rate": 0.0001,
+      "loss": 9.3416,
+      "loss/crossentropy": 2.2067521095275877,
+      "loss/hidden": 4.235546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2976540043950081,
+      "step": 1450
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 60.75,
+      "grad_norm_var": 142.08229166666666,
+      "learning_rate": 0.0001,
+      "loss": 9.4716,
+      "loss/crossentropy": 2.4361192852258684,
+      "loss/hidden": 4.1140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2877715673297644,
+      "step": 1460
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 58.75,
+      "grad_norm_var": 44.35,
+      "learning_rate": 0.0001,
+      "loss": 9.4006,
+      "loss/crossentropy": 2.239429622516036,
+      "loss/hidden": 4.026171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27844256814569235,
+      "step": 1470
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 45.0,
+      "grad_norm_var": 33.95390625,
+      "learning_rate": 0.0001,
+      "loss": 9.3993,
+      "loss/crossentropy": 2.0759536787867545,
+      "loss/hidden": 4.068359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2688772227615118,
+      "step": 1480
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 51.75,
+      "grad_norm_var": 25.795833333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.3786,
+      "loss/crossentropy": 2.286362998187542,
+      "loss/hidden": 4.15078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2942257083952427,
+      "step": 1490
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 46.75,
+      "grad_norm_var": 20.520833333333332,
+      "learning_rate": 0.0001,
+      "loss": 9.2903,
+      "loss/crossentropy": 2.312733788788319,
+      "loss/hidden": 3.971484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2691910218447447,
+      "step": 1500
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 50.25,
+      "grad_norm_var": 28.290625,
+      "learning_rate": 0.0001,
+      "loss": 9.3076,
+      "loss/crossentropy": 2.2467628076672552,
+      "loss/hidden": 4.105078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2887777745723724,
+      "step": 1510
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 63.5,
+      "grad_norm_var": 33.73098958333333,
+      "learning_rate": 0.0001,
+      "loss": 9.4203,
+      "loss/crossentropy": 2.372379180788994,
+      "loss/hidden": 4.07578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3087839350104332,
+      "step": 1520
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 45.5,
+      "grad_norm_var": 40.108333333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.3215,
+      "loss/crossentropy": 2.3452367037534714,
+      "loss/hidden": 4.210546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3159611392766237,
+      "step": 1530
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 58.25,
+      "grad_norm_var": 27.539322916666666,
+      "learning_rate": 0.0001,
+      "loss": 9.3755,
+      "loss/crossentropy": 2.3029753446578978,
+      "loss/hidden": 3.999609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2623455457389355,
+      "step": 1540
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 51.75,
+      "grad_norm_var": 26.9,
+      "learning_rate": 0.0001,
+      "loss": 9.3578,
+      "loss/crossentropy": 2.3988554388284684,
+      "loss/hidden": 4.08828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2846154376864433,
+      "step": 1550
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 91.0,
+      "grad_norm_var": 1307.8372395833333,
+      "learning_rate": 0.0001,
+      "loss": 9.432,
+      "loss/crossentropy": 2.343544365465641,
+      "loss/hidden": 4.021875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2907770898193121,
+      "step": 1560
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 52.0,
+      "grad_norm_var": 170.62890625,
+      "learning_rate": 0.0001,
+      "loss": 9.3432,
+      "loss/crossentropy": 2.173108433187008,
+      "loss/hidden": 4.10859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28518917988985776,
+      "step": 1570
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 42.0,
+      "grad_norm_var": 47.56015625,
+      "learning_rate": 0.0001,
+      "loss": 9.367,
+      "loss/crossentropy": 2.2230691239237785,
+      "loss/hidden": 4.23515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30039387457072736,
+      "step": 1580
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 72.0,
+      "grad_norm_var": 1.226104970407838e+18,
+      "learning_rate": 0.0001,
+      "loss": 9.3564,
+      "loss/crossentropy": 2.263391149044037,
+      "loss/hidden": 4.10625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2923804897814989,
+      "step": 1590
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 52.5,
+      "grad_norm_var": 1.2261049681378806e+18,
+      "learning_rate": 0.0001,
+      "loss": 9.4959,
+      "loss/crossentropy": 2.113241518288851,
+      "loss/hidden": 4.087109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2759646028280258,
+      "step": 1600
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 66.0,
+      "grad_norm_var": 734.1489583333333,
+      "learning_rate": 0.0001,
+      "loss": 9.4743,
+      "loss/crossentropy": 2.3895165085792542,
+      "loss/hidden": 4.059765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2987998936325312,
+      "step": 1610
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 44.75,
+      "grad_norm_var": 50.00182291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.1919,
+      "loss/crossentropy": 2.251766300201416,
+      "loss/hidden": 4.04375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2781111396849155,
+      "step": 1620
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 52.75,
+      "grad_norm_var": 437.49583333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.4572,
+      "loss/crossentropy": 2.382322034239769,
+      "loss/hidden": 4.07734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31318275928497313,
+      "step": 1630
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 61.0,
+      "grad_norm_var": 40.301822916666666,
+      "learning_rate": 0.0001,
+      "loss": 9.2668,
+      "loss/crossentropy": 2.1683703124523164,
+      "loss/hidden": 4.07578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.283413190767169,
+      "step": 1640
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 42.75,
+      "grad_norm_var": 57.307291666666664,
+      "learning_rate": 0.0001,
+      "loss": 9.339,
+      "loss/crossentropy": 2.3430400043725967,
+      "loss/hidden": 4.036328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.287694800645113,
+      "step": 1650
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 46.75,
+      "grad_norm_var": 65.52395833333334,
+      "learning_rate": 0.0001,
+      "loss": 9.3768,
+      "loss/crossentropy": 2.2867416352033616,
+      "loss/hidden": 4.017578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29683431759476664,
+      "step": 1660
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 52.5,
+      "grad_norm_var": 61.18932291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.2451,
+      "loss/crossentropy": 2.3707614041864873,
+      "loss/hidden": 4.06015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29184688804671166,
+      "step": 1670
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 51.75,
+      "grad_norm_var": 20.895572916666666,
+      "learning_rate": 0.0001,
+      "loss": 9.3601,
+      "loss/crossentropy": 2.3268392831087112,
+      "loss/hidden": 4.12734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29570323824882505,
+      "step": 1680
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 44.0,
+      "grad_norm_var": 10.290625,
+      "learning_rate": 0.0001,
+      "loss": 9.4214,
+      "loss/crossentropy": 2.324131193757057,
+      "loss/hidden": 4.1984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3133995305746794,
+      "step": 1690
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 58.25,
+      "grad_norm_var": 19.124739583333334,
+      "learning_rate": 0.0001,
+      "loss": 9.2465,
+      "loss/crossentropy": 2.35849623978138,
+      "loss/hidden": 4.00703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2762619823217392,
+      "step": 1700
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 45.75,
+      "grad_norm_var": 53.89895833333333,
+      "learning_rate": 0.0001,
+      "loss": 9.1951,
+      "loss/crossentropy": 2.3914038598537446,
+      "loss/hidden": 3.9984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2871177852153778,
+      "step": 1710
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 43.25,
+      "grad_norm_var": 16.479166666666668,
+      "learning_rate": 0.0001,
+      "loss": 9.1669,
+      "loss/crossentropy": 2.152750685811043,
+      "loss/hidden": 4.100390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28708020225167274,
+      "step": 1720
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 49.25,
+      "grad_norm_var": 13.45390625,
+      "learning_rate": 0.0001,
+      "loss": 9.1015,
+      "loss/crossentropy": 2.2946193665266037,
+      "loss/hidden": 4.085546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3062314610928297,
+      "step": 1730
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 46.5,
+      "grad_norm_var": 22.473958333333332,
+      "learning_rate": 0.0001,
+      "loss": 9.1287,
+      "loss/crossentropy": 2.1538643553853034,
+      "loss/hidden": 3.9421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2666194221004844,
+      "step": 1740
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 47.0,
+      "grad_norm_var": 32.62057291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.411,
+      "loss/crossentropy": 2.387891933321953,
+      "loss/hidden": 4.14921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29542505368590355,
+      "step": 1750
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 45.25,
+      "grad_norm_var": 26.92265625,
+      "learning_rate": 0.0001,
+      "loss": 9.2833,
+      "loss/crossentropy": 2.3024097591638566,
+      "loss/hidden": 4.03984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29066667445003985,
+      "step": 1760
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 53.5,
+      "grad_norm_var": 17.832291666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.2665,
+      "loss/crossentropy": 2.4454205125570296,
+      "loss/hidden": 3.955859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2910691563040018,
+      "step": 1770
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 42.25,
+      "grad_norm_var": 29.865625,
+      "learning_rate": 0.0001,
+      "loss": 9.1701,
+      "loss/crossentropy": 2.2966391056776048,
+      "loss/hidden": 4.027734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2789210833609104,
+      "step": 1780
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 48.25,
+      "grad_norm_var": 17.548958333333335,
+      "learning_rate": 0.0001,
+      "loss": 9.1992,
+      "loss/crossentropy": 2.395502945780754,
+      "loss/hidden": 3.934765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2776679117232561,
+      "step": 1790
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 40.75,
+      "grad_norm_var": 13.282291666666667,
+      "learning_rate": 0.0001,
+      "loss": 9.1046,
+      "loss/crossentropy": 2.22285817861557,
+      "loss/hidden": 3.9046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26667180880904195,
+      "step": 1800
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 36.25,
+      "grad_norm_var": 34.90807291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.3204,
+      "loss/crossentropy": 2.3842350512743,
+      "loss/hidden": 4.009375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.300260554254055,
+      "step": 1810
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 46.75,
+      "grad_norm_var": 27.77890625,
+      "learning_rate": 0.0001,
+      "loss": 9.0943,
+      "loss/crossentropy": 2.274762773513794,
+      "loss/hidden": 4.01328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28360783979296683,
+      "step": 1820
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 55.5,
+      "grad_norm_var": 27.298958333333335,
+      "learning_rate": 0.0001,
+      "loss": 9.1699,
+      "loss/crossentropy": 2.1643219627439976,
+      "loss/hidden": 3.9921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.267458438500762,
+      "step": 1830
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 49.75,
+      "grad_norm_var": 43.94583333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.3022,
+      "loss/crossentropy": 2.464679929614067,
+      "loss/hidden": 3.9546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29758369028568266,
+      "step": 1840
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 51.0,
+      "grad_norm_var": 37.90807291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.1863,
+      "loss/crossentropy": 2.3199010998010636,
+      "loss/hidden": 3.99453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27702242247760295,
+      "step": 1850
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 46.5,
+      "grad_norm_var": 40.920833333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.2872,
+      "loss/crossentropy": 2.4041683062911035,
+      "loss/hidden": 4.09140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.30005627647042277,
+      "step": 1860
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 39.75,
+      "grad_norm_var": 40.723958333333336,
+      "learning_rate": 0.0001,
+      "loss": 9.1081,
+      "loss/crossentropy": 2.273802790045738,
+      "loss/hidden": 4.175,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3045934235677123,
+      "step": 1870
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 43.25,
+      "grad_norm_var": 33.35729166666667,
+      "learning_rate": 0.0001,
+      "loss": 9.106,
+      "loss/crossentropy": 2.3607766672968866,
+      "loss/hidden": 3.9765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28193066976964476,
+      "step": 1880
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 48.25,
+      "grad_norm_var": 14.915625,
+      "learning_rate": 0.0001,
+      "loss": 9.1437,
+      "loss/crossentropy": 2.2798361241817475,
+      "loss/hidden": 3.9890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2721746701747179,
+      "step": 1890
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 44.0,
+      "grad_norm_var": 21.223958333333332,
+      "learning_rate": 0.0001,
+      "loss": 9.0972,
+      "loss/crossentropy": 2.21695294380188,
+      "loss/hidden": 4.0015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2832322970032692,
+      "step": 1900
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 39.5,
+      "grad_norm_var": 27.808333333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.1587,
+      "loss/crossentropy": 2.1728454776108266,
+      "loss/hidden": 3.98828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27170457877218723,
+      "step": 1910
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 41.75,
+      "grad_norm_var": 13.315625,
+      "learning_rate": 0.0001,
+      "loss": 9.1326,
+      "loss/crossentropy": 2.154237084835768,
+      "loss/hidden": 4.063671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27950075305998323,
+      "step": 1920
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 43.0,
+      "grad_norm_var": 25.240625,
+      "learning_rate": 0.0001,
+      "loss": 9.1013,
+      "loss/crossentropy": 2.2507698431611063,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2808088269084692,
+      "step": 1930
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 49.25,
+      "grad_norm_var": 22.832291666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.2429,
+      "loss/crossentropy": 2.288056728243828,
+      "loss/hidden": 4.1546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.31668607220053674,
+      "step": 1940
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 48.5,
+      "grad_norm_var": 58.09557291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.1742,
+      "loss/crossentropy": 2.2107961744070055,
+      "loss/hidden": 4.05078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2858551822602749,
+      "step": 1950
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 39.25,
+      "grad_norm_var": 49.50390625,
+      "learning_rate": 0.0001,
+      "loss": 9.1293,
+      "loss/crossentropy": 2.224529256671667,
+      "loss/hidden": 3.9703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27973891496658326,
+      "step": 1960
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 39.25,
+      "grad_norm_var": 13.890625,
+      "learning_rate": 0.0001,
+      "loss": 9.0689,
+      "loss/crossentropy": 2.363737019896507,
+      "loss/hidden": 4.027734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2919711694121361,
+      "step": 1970
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 55.75,
+      "grad_norm_var": 26.655989583333334,
+      "learning_rate": 0.0001,
+      "loss": 9.2228,
+      "loss/crossentropy": 2.3380469545722007,
+      "loss/hidden": 4.0171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28581551983952524,
+      "step": 1980
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 45.0,
+      "grad_norm_var": 27.357291666666665,
+      "learning_rate": 0.0001,
+      "loss": 9.173,
+      "loss/crossentropy": 2.43135461807251,
+      "loss/hidden": 3.970703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28222124874591825,
+      "step": 1990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 44.5,
+      "grad_norm_var": 16.04140625,
+      "learning_rate": 0.0001,
+      "loss": 9.1554,
+      "loss/crossentropy": 2.4415812104940415,
+      "loss/hidden": 4.064453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2984179027378559,
+      "step": 2000
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 51.75,
+      "grad_norm_var": 18.032291666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.204,
+      "loss/crossentropy": 2.571503698825836,
+      "loss/hidden": 4.020703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3065837759524584,
+      "step": 2010
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 42.25,
+      "grad_norm_var": 16.573958333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.1575,
+      "loss/crossentropy": 2.2947281152009964,
+      "loss/hidden": 3.949609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28560531958937646,
+      "step": 2020
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 43.0,
+      "grad_norm_var": 10.068489583333333,
+      "learning_rate": 0.0001,
+      "loss": 9.1347,
+      "loss/crossentropy": 2.432750529050827,
+      "loss/hidden": 3.9953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28870879150927065,
+      "step": 2030
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 45.0,
+      "grad_norm_var": 14.683333333333334,
+      "learning_rate": 0.0001,
+      "loss": 9.0302,
+      "loss/crossentropy": 2.291880601644516,
+      "loss/hidden": 3.852734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27138952538371086,
+      "step": 2040
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 54.5,
+      "grad_norm_var": 24.0625,
+      "learning_rate": 0.0001,
+      "loss": 9.1623,
+      "loss/crossentropy": 2.1033009082078933,
+      "loss/hidden": 4.074609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29145455472171305,
+      "step": 2050
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 42.25,
+      "grad_norm_var": 29.780989583333334,
+      "learning_rate": 0.0001,
+      "loss": 9.1409,
+      "loss/crossentropy": 2.3487906470894813,
+      "loss/hidden": 3.9140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2789519714191556,
+      "step": 2060
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 42.0,
+      "grad_norm_var": 15.395572916666667,
+      "learning_rate": 0.0001,
+      "loss": 9.0027,
+      "loss/crossentropy": 2.210711918771267,
+      "loss/hidden": 3.975,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2679262701421976,
+      "step": 2070
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 43.5,
+      "grad_norm_var": 12.7125,
+      "learning_rate": 0.0001,
+      "loss": 9.0306,
+      "loss/crossentropy": 2.1945893600583077,
+      "loss/hidden": 3.948046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28115708455443383,
+      "step": 2080
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 47.0,
+      "grad_norm_var": 28.683072916666667,
+      "learning_rate": 0.0001,
+      "loss": 9.1344,
+      "loss/crossentropy": 2.237662248313427,
+      "loss/hidden": 4.035546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2781697390601039,
+      "step": 2090
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 42.0,
+      "grad_norm_var": 15.390625,
+      "learning_rate": 0.0001,
+      "loss": 9.0004,
+      "loss/crossentropy": 2.269840542972088,
+      "loss/hidden": 3.9765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29009242728352547,
+      "step": 2100
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 39.75,
+      "grad_norm_var": 8.10390625,
+      "learning_rate": 0.0001,
+      "loss": 9.0135,
+      "loss/crossentropy": 2.3315619856119154,
+      "loss/hidden": 4.028515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27684418186545373,
+      "step": 2110
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 45.75,
+      "grad_norm_var": 115.615625,
+      "learning_rate": 0.0001,
+      "loss": 9.2092,
+      "loss/crossentropy": 2.4081921339035035,
+      "loss/hidden": 4.01171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29490497298538687,
+      "step": 2120
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 43.0,
+      "grad_norm_var": 95.76432291666667,
+      "learning_rate": 0.0001,
+      "loss": 9.1058,
+      "loss/crossentropy": 2.393023744225502,
+      "loss/hidden": 3.94921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2818208742886782,
+      "step": 2130
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 39.5,
+      "grad_norm_var": 13.873958333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.0997,
+      "loss/crossentropy": 2.1606352396309374,
+      "loss/hidden": 3.843359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25883881878107784,
+      "step": 2140
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 44.0,
+      "grad_norm_var": 22.845572916666665,
+      "learning_rate": 0.0001,
+      "loss": 9.1464,
+      "loss/crossentropy": 2.179043120145798,
+      "loss/hidden": 4.056640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2761132620275021,
+      "step": 2150
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 46.5,
+      "grad_norm_var": 28.957291666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.1057,
+      "loss/crossentropy": 2.301611530780792,
+      "loss/hidden": 3.959375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29443784058094025,
+      "step": 2160
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 46.25,
+      "grad_norm_var": 36.708333333333336,
+      "learning_rate": 0.0001,
+      "loss": 9.0952,
+      "loss/crossentropy": 2.3569841012358665,
+      "loss/hidden": 3.994921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28191804718226193,
+      "step": 2170
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 81.5,
+      "grad_norm_var": 117.24895833333333,
+      "learning_rate": 0.0001,
+      "loss": 9.0595,
+      "loss/crossentropy": 2.4225870154798033,
+      "loss/hidden": 3.953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2870227605104446,
+      "step": 2180
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 36.75,
+      "grad_norm_var": 173.97265625,
+      "learning_rate": 0.0001,
+      "loss": 9.1612,
+      "loss/crossentropy": 2.34947164952755,
+      "loss/hidden": 4.0546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2929512483999133,
+      "step": 2190
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 41.75,
+      "grad_norm_var": 31.648958333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.101,
+      "loss/crossentropy": 2.218617644906044,
+      "loss/hidden": 4.029296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2752823047339916,
+      "step": 2200
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 47.0,
+      "grad_norm_var": 10.50390625,
+      "learning_rate": 0.0001,
+      "loss": 9.0496,
+      "loss/crossentropy": 2.4772594451904295,
+      "loss/hidden": 3.96953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28055914528667925,
+      "step": 2210
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 49.5,
+      "grad_norm_var": 442.3,
+      "learning_rate": 0.0001,
+      "loss": 8.9647,
+      "loss/crossentropy": 2.44835202395916,
+      "loss/hidden": 3.905859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27223448157310487,
+      "step": 2220
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 35.25,
+      "grad_norm_var": 577.0458333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.1287,
+      "loss/crossentropy": 2.2784701570868493,
+      "loss/hidden": 3.98671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28111674822866917,
+      "step": 2230
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 43.75,
+      "grad_norm_var": 212.74348958333334,
+      "learning_rate": 0.0001,
+      "loss": 9.1339,
+      "loss/crossentropy": 2.2536803498864173,
+      "loss/hidden": 3.934375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27366876490414144,
+      "step": 2240
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 42.5,
+      "grad_norm_var": 30.143489583333334,
+      "learning_rate": 0.0001,
+      "loss": 9.0536,
+      "loss/crossentropy": 2.300386372208595,
+      "loss/hidden": 3.937890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2872367199510336,
+      "step": 2250
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 40.75,
+      "grad_norm_var": 19.65390625,
+      "learning_rate": 0.0001,
+      "loss": 9.0113,
+      "loss/crossentropy": 2.318666061758995,
+      "loss/hidden": 3.8796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.259370943903923,
+      "step": 2260
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 43.75,
+      "grad_norm_var": 12.46015625,
+      "learning_rate": 0.0001,
+      "loss": 8.9343,
+      "loss/crossentropy": 2.169833867251873,
+      "loss/hidden": 3.8984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26568643413484094,
+      "step": 2270
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 38.75,
+      "grad_norm_var": 11.832291666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.0042,
+      "loss/crossentropy": 2.1532950207591055,
+      "loss/hidden": 3.987890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2669289981946349,
+      "step": 2280
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 46.5,
+      "grad_norm_var": 13.432291666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.1282,
+      "loss/crossentropy": 2.0724870592355726,
+      "loss/hidden": 4.09765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27939337231218814,
+      "step": 2290
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 35.75,
+      "grad_norm_var": 223.3875,
+      "learning_rate": 0.0001,
+      "loss": 9.1656,
+      "loss/crossentropy": 2.4021882474422456,
+      "loss/hidden": 3.941796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28748833425343034,
+      "step": 2300
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 40.5,
+      "grad_norm_var": 264.82395833333334,
+      "learning_rate": 0.0001,
+      "loss": 8.9421,
+      "loss/crossentropy": 2.201041653752327,
+      "loss/hidden": 4.012109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29431225806474687,
+      "step": 2310
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 37.25,
+      "grad_norm_var": 14.939322916666667,
+      "learning_rate": 0.0001,
+      "loss": 9.0761,
+      "loss/crossentropy": 2.156717260926962,
+      "loss/hidden": 4.16484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27824588380753995,
+      "step": 2320
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 44.0,
+      "grad_norm_var": 66.07890625,
+      "learning_rate": 0.0001,
+      "loss": 9.1021,
+      "loss/crossentropy": 2.3962995454669,
+      "loss/hidden": 3.849609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27611064203083513,
+      "step": 2330
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 37.5,
+      "grad_norm_var": 2.5671221192004644e+18,
+      "learning_rate": 0.0001,
+      "loss": 9.1022,
+      "loss/crossentropy": 2.302082321047783,
+      "loss/hidden": 3.980859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2941384054720402,
+      "step": 2340
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 42.75,
+      "grad_norm_var": 49.62473958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.7886,
+      "loss/crossentropy": 2.239869697391987,
+      "loss/hidden": 3.977734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25684802830219267,
+      "step": 2350
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 39.5,
+      "grad_norm_var": 10.540625,
+      "learning_rate": 0.0001,
+      "loss": 8.8748,
+      "loss/crossentropy": 2.2847611531615257,
+      "loss/hidden": 4.076171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2846489936113358,
+      "step": 2360
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 44.25,
+      "grad_norm_var": 33.15416666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.8811,
+      "loss/crossentropy": 2.209371344745159,
+      "loss/hidden": 3.907421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2737090703099966,
+      "step": 2370
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 42.0,
+      "grad_norm_var": 58.614322916666666,
+      "learning_rate": 0.0001,
+      "loss": 9.0391,
+      "loss/crossentropy": 2.381209687888622,
+      "loss/hidden": 3.9296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2849856551736593,
+      "step": 2380
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 39.75,
+      "grad_norm_var": 17.541666666666668,
+      "learning_rate": 0.0001,
+      "loss": 9.1216,
+      "loss/crossentropy": 2.2550544410943987,
+      "loss/hidden": 4.000390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2924022350460291,
+      "step": 2390
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 46.25,
+      "grad_norm_var": 18.491666666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.9833,
+      "loss/crossentropy": 2.2908297032117844,
+      "loss/hidden": 3.901953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26052255779504774,
+      "step": 2400
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 38.25,
+      "grad_norm_var": 23.640625,
+      "learning_rate": 0.0001,
+      "loss": 9.1123,
+      "loss/crossentropy": 2.3713207334280013,
+      "loss/hidden": 3.988671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2721697688102722,
+      "step": 2410
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 42.5,
+      "grad_norm_var": 69.77473958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.9108,
+      "loss/crossentropy": 2.0408870808780195,
+      "loss/hidden": 3.873046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24748602956533433,
+      "step": 2420
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 37.25,
+      "grad_norm_var": 72.11015625,
+      "learning_rate": 0.0001,
+      "loss": 9.0111,
+      "loss/crossentropy": 2.3027436569333077,
+      "loss/hidden": 3.908984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27137077748775484,
+      "step": 2430
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 54.75,
+      "grad_norm_var": 55.057291666666664,
+      "learning_rate": 0.0001,
+      "loss": 9.0763,
+      "loss/crossentropy": 2.2573105663061144,
+      "loss/hidden": 4.1703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3013453852385283,
+      "step": 2440
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 39.25,
+      "grad_norm_var": 50.00416666666667,
+      "learning_rate": 0.0001,
+      "loss": 9.0658,
+      "loss/crossentropy": 2.4534697026014327,
+      "loss/hidden": 4.172265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2898527968674898,
+      "step": 2450
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 40.75,
+      "grad_norm_var": 18.71640625,
+      "learning_rate": 0.0001,
+      "loss": 8.9953,
+      "loss/crossentropy": 2.357613870501518,
+      "loss/hidden": 3.9125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27832051999866964,
+      "step": 2460
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 34.0,
+      "grad_norm_var": 14.832291666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.9576,
+      "loss/crossentropy": 2.295011055469513,
+      "loss/hidden": 4.029296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29190085306763647,
+      "step": 2470
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 50.75,
+      "grad_norm_var": 24.075,
+      "learning_rate": 0.0001,
+      "loss": 8.7866,
+      "loss/crossentropy": 2.2598410531878472,
+      "loss/hidden": 3.84921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2570509884506464,
+      "step": 2480
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 38.0,
+      "grad_norm_var": 97.60416666666667,
+      "learning_rate": 0.0001,
+      "loss": 9.0289,
+      "loss/crossentropy": 2.365708181262016,
+      "loss/hidden": 3.835546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2712419513612986,
+      "step": 2490
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 44.5,
+      "grad_norm_var": 104.10729166666667,
+      "learning_rate": 0.0001,
+      "loss": 8.8744,
+      "loss/crossentropy": 2.063434064388275,
+      "loss/hidden": 3.873046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2460779383778572,
+      "step": 2500
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 41.75,
+      "grad_norm_var": 28.974739583333335,
+      "learning_rate": 0.0001,
+      "loss": 8.9272,
+      "loss/crossentropy": 2.2377428650856017,
+      "loss/hidden": 3.968359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26241020299494267,
+      "step": 2510
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 43.5,
+      "grad_norm_var": 20.84140625,
+      "learning_rate": 0.0001,
+      "loss": 8.9773,
+      "loss/crossentropy": 2.2245729833841326,
+      "loss/hidden": 3.854296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2788569286465645,
+      "step": 2520
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 37.25,
+      "grad_norm_var": 10.36640625,
+      "learning_rate": 0.0001,
+      "loss": 8.8657,
+      "loss/crossentropy": 2.19287933409214,
+      "loss/hidden": 3.873046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2571034274995327,
+      "step": 2530
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 42.75,
+      "grad_norm_var": 25.641666666666666,
+      "learning_rate": 0.0001,
+      "loss": 9.0621,
+      "loss/crossentropy": 2.344786374270916,
+      "loss/hidden": 4.005078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29999860040843485,
+      "step": 2540
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 45.25,
+      "grad_norm_var": 28.983333333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.8825,
+      "loss/crossentropy": 2.165843137353659,
+      "loss/hidden": 3.794921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24595264531672,
+      "step": 2550
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 44.0,
+      "grad_norm_var": 13.407291666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.9041,
+      "loss/crossentropy": 2.3431010633707046,
+      "loss/hidden": 3.880859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27991249822080133,
+      "step": 2560
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 49.25,
+      "grad_norm_var": 19.137239583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.7585,
+      "loss/crossentropy": 2.2741693764925004,
+      "loss/hidden": 3.848046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2573831077665091,
+      "step": 2570
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 40.75,
+      "grad_norm_var": 23.523958333333333,
+      "learning_rate": 0.0001,
+      "loss": 9.0385,
+      "loss/crossentropy": 2.3741003662347793,
+      "loss/hidden": 3.9703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3021115079522133,
+      "step": 2580
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 38.5,
+      "grad_norm_var": 20.873958333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.8554,
+      "loss/crossentropy": 2.259954023361206,
+      "loss/hidden": 3.92734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2688195243477821,
+      "step": 2590
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 33.25,
+      "grad_norm_var": 16.76015625,
+      "learning_rate": 0.0001,
+      "loss": 8.8802,
+      "loss/crossentropy": 2.155090569704771,
+      "loss/hidden": 3.91796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.256628708448261,
+      "step": 2600
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 41.5,
+      "grad_norm_var": 15.032291666666667,
+      "learning_rate": 0.0001,
+      "loss": 9.0016,
+      "loss/crossentropy": 2.365834577381611,
+      "loss/hidden": 3.85859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27286841757595537,
+      "step": 2610
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 41.75,
+      "grad_norm_var": 9.50390625,
+      "learning_rate": 0.0001,
+      "loss": 8.8993,
+      "loss/crossentropy": 2.137889374792576,
+      "loss/hidden": 3.96171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27018810212612154,
+      "step": 2620
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 44.75,
+      "grad_norm_var": 12.58515625,
+      "learning_rate": 0.0001,
+      "loss": 8.9333,
+      "loss/crossentropy": 2.279120808839798,
+      "loss/hidden": 3.8859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26760734505951406,
+      "step": 2630
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 33.75,
+      "grad_norm_var": 20.205989583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.998,
+      "loss/crossentropy": 2.133891487121582,
+      "loss/hidden": 3.978125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2548325901851058,
+      "step": 2640
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 38.25,
+      "grad_norm_var": 24.26640625,
+      "learning_rate": 0.0001,
+      "loss": 8.8773,
+      "loss/crossentropy": 2.0680222399532795,
+      "loss/hidden": 3.983984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2616095909848809,
+      "step": 2650
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 37.75,
+      "grad_norm_var": 24.65390625,
+      "learning_rate": 0.0001,
+      "loss": 8.8787,
+      "loss/crossentropy": 2.2427713751792906,
+      "loss/hidden": 3.976171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2840047996491194,
+      "step": 2660
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 50.0,
+      "grad_norm_var": 22.190625,
+      "learning_rate": 0.0001,
+      "loss": 8.878,
+      "loss/crossentropy": 2.3332558259367944,
+      "loss/hidden": 3.774609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527316328138113,
+      "step": 2670
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 58.5,
+      "grad_norm_var": 53.95,
+      "learning_rate": 0.0001,
+      "loss": 8.8498,
+      "loss/crossentropy": 2.116853891313076,
+      "loss/hidden": 3.810546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24547674022614957,
+      "step": 2680
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 37.5,
+      "grad_norm_var": 48.47057291666667,
+      "learning_rate": 0.0001,
+      "loss": 8.9221,
+      "loss/crossentropy": 2.468301197886467,
+      "loss/hidden": 3.809765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26750445999205114,
+      "step": 2690
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 46.5,
+      "grad_norm_var": 16.812239583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.7264,
+      "loss/crossentropy": 2.1068901009857655,
+      "loss/hidden": 3.970703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2827789710834622,
+      "step": 2700
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 41.25,
+      "grad_norm_var": 17.765625,
+      "learning_rate": 0.0001,
+      "loss": 8.766,
+      "loss/crossentropy": 2.4922314494848252,
+      "loss/hidden": 3.943359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28248917534947393,
+      "step": 2710
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 37.25,
+      "grad_norm_var": 4.524739583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.8324,
+      "loss/crossentropy": 2.268562327325344,
+      "loss/hidden": 3.8296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.250153512135148,
+      "step": 2720
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 38.5,
+      "grad_norm_var": 11.832291666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.8538,
+      "loss/crossentropy": 2.171993290632963,
+      "loss/hidden": 3.860546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2605165271088481,
+      "step": 2730
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 37.25,
+      "grad_norm_var": 17.5625,
+      "learning_rate": 0.0001,
+      "loss": 8.8625,
+      "loss/crossentropy": 2.3496526792645454,
+      "loss/hidden": 3.78828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532901844009757,
+      "step": 2740
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 38.5,
+      "grad_norm_var": 18.69140625,
+      "learning_rate": 0.0001,
+      "loss": 8.8941,
+      "loss/crossentropy": 2.377558296918869,
+      "loss/hidden": 3.844921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2773334577679634,
+      "step": 2750
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 46.75,
+      "grad_norm_var": 22.22890625,
+      "learning_rate": 0.0001,
+      "loss": 8.7839,
+      "loss/crossentropy": 2.248410400748253,
+      "loss/hidden": 3.980078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27028046883642676,
+      "step": 2760
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 38.75,
+      "grad_norm_var": 23.164322916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.7791,
+      "loss/crossentropy": 2.4241216853260994,
+      "loss/hidden": 3.8984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2659046190790832,
+      "step": 2770
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 53.25,
+      "grad_norm_var": 59.78098958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.8019,
+      "loss/crossentropy": 2.1859550148248674,
+      "loss/hidden": 3.828515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25679499059915545,
+      "step": 2780
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 41.5,
+      "grad_norm_var": 62.925455729166664,
+      "learning_rate": 0.0001,
+      "loss": 8.8086,
+      "loss/crossentropy": 2.312130589783192,
+      "loss/hidden": 3.91171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2820352425798774,
+      "step": 2790
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 46.75,
+      "grad_norm_var": 18.951822916666668,
+      "learning_rate": 0.0001,
+      "loss": 8.7941,
+      "loss/crossentropy": 2.258910335600376,
+      "loss/hidden": 3.902734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2627917256206274,
+      "step": 2800
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 40.25,
+      "grad_norm_var": 17.639322916666668,
+      "learning_rate": 0.0001,
+      "loss": 8.7417,
+      "loss/crossentropy": 2.2589244581758976,
+      "loss/hidden": 3.753515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2513848140835762,
+      "step": 2810
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 40.5,
+      "grad_norm_var": 12.032291666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.8627,
+      "loss/crossentropy": 2.17684805393219,
+      "loss/hidden": 3.853125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.255695578455925,
+      "step": 2820
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 40.75,
+      "grad_norm_var": 114.61432291666667,
+      "learning_rate": 0.0001,
+      "loss": 8.8412,
+      "loss/crossentropy": 2.2954238772392275,
+      "loss/hidden": 3.918359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2858715243637562,
+      "step": 2830
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 40.5,
+      "grad_norm_var": 5.641666666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.9391,
+      "loss/crossentropy": 2.329847712814808,
+      "loss/hidden": 3.7625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25534543097019197,
+      "step": 2840
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 38.5,
+      "grad_norm_var": 40.84557291666667,
+      "learning_rate": 0.0001,
+      "loss": 8.7803,
+      "loss/crossentropy": 2.2853938594460486,
+      "loss/hidden": 3.866796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24832999743521214,
+      "step": 2850
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 36.75,
+      "grad_norm_var": 39.958072916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.6796,
+      "loss/crossentropy": 2.234159553050995,
+      "loss/hidden": 3.847265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26201403168961407,
+      "step": 2860
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 39.0,
+      "grad_norm_var": 9.457291666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.7109,
+      "loss/crossentropy": 2.2819659531116487,
+      "loss/hidden": 3.894140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26489345021545885,
+      "step": 2870
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 37.25,
+      "grad_norm_var": 9.268489583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.828,
+      "loss/crossentropy": 2.4358034074306487,
+      "loss/hidden": 3.81328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27083273865282537,
+      "step": 2880
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 42.0,
+      "grad_norm_var": 12.805989583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.8211,
+      "loss/crossentropy": 2.2965500839054585,
+      "loss/hidden": 3.869921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26719480073079466,
+      "step": 2890
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 37.25,
+      "grad_norm_var": 9.020572916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.7927,
+      "loss/crossentropy": 2.3898714184761047,
+      "loss/hidden": 3.930078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2959909211844206,
+      "step": 2900
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 44.25,
+      "grad_norm_var": 10.283333333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.7893,
+      "loss/crossentropy": 2.349339473247528,
+      "loss/hidden": 3.7953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25724136754870414,
+      "step": 2910
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 35.5,
+      "grad_norm_var": 17.418489583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.7582,
+      "loss/crossentropy": 2.1329104267060757,
+      "loss/hidden": 3.83046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2572308249771595,
+      "step": 2920
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 42.0,
+      "grad_norm_var": 14.87265625,
+      "learning_rate": 0.0001,
+      "loss": 8.6388,
+      "loss/crossentropy": 2.23364320397377,
+      "loss/hidden": 3.82421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2617119399830699,
+      "step": 2930
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 42.75,
+      "grad_norm_var": 8.204166666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.6531,
+      "loss/crossentropy": 2.1770762100815775,
+      "loss/hidden": 3.833984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2418960839509964,
+      "step": 2940
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 45.5,
+      "grad_norm_var": 16.265625,
+      "learning_rate": 0.0001,
+      "loss": 8.8053,
+      "loss/crossentropy": 2.18697277456522,
+      "loss/hidden": 3.78125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26141371857374907,
+      "step": 2950
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 32.25,
+      "grad_norm_var": 19.145833333333332,
+      "learning_rate": 0.0001,
+      "loss": 8.7178,
+      "loss/crossentropy": 2.303604170680046,
+      "loss/hidden": 3.738671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2531938493251801,
+      "step": 2960
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 33.5,
+      "grad_norm_var": 10.88515625,
+      "learning_rate": 0.0001,
+      "loss": 8.5781,
+      "loss/crossentropy": 2.112999178469181,
+      "loss/hidden": 3.853515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2678428884595633,
+      "step": 2970
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 44.0,
+      "grad_norm_var": 32.84973958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.8415,
+      "loss/crossentropy": 2.407758575677872,
+      "loss/hidden": 3.925,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3035837195813656,
+      "step": 2980
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 36.5,
+      "grad_norm_var": 21.0125,
+      "learning_rate": 0.0001,
+      "loss": 8.6496,
+      "loss/crossentropy": 2.1467753663659095,
+      "loss/hidden": 3.802734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25200750436633823,
+      "step": 2990
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 37.25,
+      "grad_norm_var": 102.01848958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.5981,
+      "loss/crossentropy": 2.22710300385952,
+      "loss/hidden": 3.741015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24842255041003228,
+      "step": 3000
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 37.5,
+      "grad_norm_var": 20.633072916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.6443,
+      "loss/crossentropy": 2.3478307321667673,
+      "loss/hidden": 3.805859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25510224178433416,
+      "step": 3010
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 41.5,
+      "grad_norm_var": 8.864322916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.7802,
+      "loss/crossentropy": 2.3206353336572647,
+      "loss/hidden": 3.855078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2652175173163414,
+      "step": 3020
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 46.25,
+      "grad_norm_var": 1.6257994395224812e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.8456,
+      "loss/crossentropy": 2.3741259276866913,
+      "loss/hidden": 3.973046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2950541414320469,
+      "step": 3030
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 38.25,
+      "grad_norm_var": 1.6257994392887188e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.7725,
+      "loss/crossentropy": 2.413549691438675,
+      "loss/hidden": 3.87734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26055113933980467,
+      "step": 3040
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 46.0,
+      "grad_norm_var": 297.95807291666665,
+      "learning_rate": 0.0001,
+      "loss": 8.66,
+      "loss/crossentropy": 2.3086455732584,
+      "loss/hidden": 3.956640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2837849177420139,
+      "step": 3050
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 38.75,
+      "grad_norm_var": 326.315625,
+      "learning_rate": 0.0001,
+      "loss": 8.7568,
+      "loss/crossentropy": 2.3664418935775755,
+      "loss/hidden": 3.912890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2698039198294282,
+      "step": 3060
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 36.5,
+      "grad_norm_var": 21.515625,
+      "learning_rate": 0.0001,
+      "loss": 8.7201,
+      "loss/crossentropy": 2.1954890489578247,
+      "loss/hidden": 3.8921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2684766609221697,
+      "step": 3070
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 41.0,
+      "grad_norm_var": 7.140625,
+      "learning_rate": 0.0001,
+      "loss": 8.8511,
+      "loss/crossentropy": 2.310048124939203,
+      "loss/hidden": 3.84609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2778544146567583,
+      "step": 3080
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 35.25,
+      "grad_norm_var": 10.1875,
+      "learning_rate": 0.0001,
+      "loss": 8.7144,
+      "loss/crossentropy": 2.3563184320926664,
+      "loss/hidden": 3.8625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2816514492034912,
+      "step": 3090
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 36.75,
+      "grad_norm_var": 2.279947281849385e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.7129,
+      "loss/crossentropy": 2.439156624674797,
+      "loss/hidden": 3.85078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28767146319150927,
+      "step": 3100
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 39.5,
+      "grad_norm_var": 2.2799472807295058e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.6894,
+      "loss/crossentropy": 2.176630274951458,
+      "loss/hidden": 3.76484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25743562281131743,
+      "step": 3110
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 36.0,
+      "grad_norm_var": 42.69348958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.6887,
+      "loss/crossentropy": 2.2695231288671494,
+      "loss/hidden": 3.796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2609828669577837,
+      "step": 3120
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 38.75,
+      "grad_norm_var": 12.6875,
+      "learning_rate": 0.0001,
+      "loss": 8.7316,
+      "loss/crossentropy": 2.2914074435830116,
+      "loss/hidden": 3.970703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25941712930798533,
+      "step": 3130
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 35.5,
+      "grad_norm_var": 9.191666666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.675,
+      "loss/crossentropy": 2.2625655576586725,
+      "loss/hidden": 3.950390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26395085640251637,
+      "step": 3140
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 48.25,
+      "grad_norm_var": 64.8125,
+      "learning_rate": 0.0001,
+      "loss": 8.8068,
+      "loss/crossentropy": 2.157360579818487,
+      "loss/hidden": 3.814453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2572598461061716,
+      "step": 3150
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 36.0,
+      "grad_norm_var": 69.0212890625,
+      "learning_rate": 0.0001,
+      "loss": 8.656,
+      "loss/crossentropy": 2.24196752011776,
+      "loss/hidden": 3.821875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27792793661355975,
+      "step": 3160
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 35.0,
+      "grad_norm_var": 8.108333333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.662,
+      "loss/crossentropy": 2.2520006895065308,
+      "loss/hidden": 3.708203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2597096076235175,
+      "step": 3170
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 34.0,
+      "grad_norm_var": 12.848958333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.5281,
+      "loss/crossentropy": 2.2527818381786346,
+      "loss/hidden": 3.772265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2495524413883686,
+      "step": 3180
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 41.0,
+      "grad_norm_var": 51.365625,
+      "learning_rate": 0.0001,
+      "loss": 8.7132,
+      "loss/crossentropy": 2.2904104314744473,
+      "loss/hidden": 3.944140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2817814026027918,
+      "step": 3190
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 36.25,
+      "grad_norm_var": 22.22265625,
+      "learning_rate": 0.0001,
+      "loss": 8.4574,
+      "loss/crossentropy": 2.090634661912918,
+      "loss/hidden": 3.79921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24919861294329165,
+      "step": 3200
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 38.5,
+      "grad_norm_var": 10.030989583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.5723,
+      "loss/crossentropy": 2.117210125923157,
+      "loss/hidden": 3.672265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2293582683429122,
+      "step": 3210
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 36.25,
+      "grad_norm_var": 17.933072916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.5557,
+      "loss/crossentropy": 2.252604177594185,
+      "loss/hidden": 3.71171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2455908928066492,
+      "step": 3220
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 34.0,
+      "grad_norm_var": 15.673958333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.6303,
+      "loss/crossentropy": 2.2780065298080445,
+      "loss/hidden": 3.701953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25410398468375206,
+      "step": 3230
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 37.75,
+      "grad_norm_var": 31.186393229166665,
+      "learning_rate": 0.0001,
+      "loss": 8.6026,
+      "loss/crossentropy": 2.3745017647743225,
+      "loss/hidden": 3.819140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2740194508805871,
+      "step": 3240
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 44.75,
+      "grad_norm_var": 16.865625,
+      "learning_rate": 0.0001,
+      "loss": 8.7984,
+      "loss/crossentropy": 2.3756151482462884,
+      "loss/hidden": 3.894921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2671732004731894,
+      "step": 3250
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 44.0,
+      "grad_norm_var": 18.073893229166668,
+      "learning_rate": 0.0001,
+      "loss": 8.6327,
+      "loss/crossentropy": 2.3531317353248595,
+      "loss/hidden": 3.717578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25254391208291055,
+      "step": 3260
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 38.5,
+      "grad_norm_var": 10.248893229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.6167,
+      "loss/crossentropy": 2.283932936191559,
+      "loss/hidden": 3.96171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.28510861806571486,
+      "step": 3270
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 32.25,
+      "grad_norm_var": 7.364322916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.6054,
+      "loss/crossentropy": 2.1746396124362946,
+      "loss/hidden": 3.709375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24525153413414955,
+      "step": 3280
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 41.25,
+      "grad_norm_var": 18.978059895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.8527,
+      "loss/crossentropy": 2.270473413169384,
+      "loss/hidden": 3.8609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.274440161883831,
+      "step": 3290
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 48.5,
+      "grad_norm_var": 18.778580729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.6327,
+      "loss/crossentropy": 2.3394594848155976,
+      "loss/hidden": 3.77109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25994330234825613,
+      "step": 3300
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 34.75,
+      "grad_norm_var": 17.190625,
+      "learning_rate": 0.0001,
+      "loss": 8.7313,
+      "loss/crossentropy": 2.347683647274971,
+      "loss/hidden": 3.894921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27402915358543395,
+      "step": 3310
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 31.0,
+      "grad_norm_var": 27.712239583333332,
+      "learning_rate": 0.0001,
+      "loss": 8.7637,
+      "loss/crossentropy": 2.257617971301079,
+      "loss/hidden": 3.88046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2689027152955532,
+      "step": 3320
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 40.0,
+      "grad_norm_var": 27.319205729166665,
+      "learning_rate": 0.0001,
+      "loss": 8.829,
+      "loss/crossentropy": 2.2091607600450516,
+      "loss/hidden": 3.83515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2670388799160719,
+      "step": 3330
+    },
+    {
+      "epoch": 0.334,
+      "grad_norm": 37.5,
+      "grad_norm_var": 38.63098958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.759,
+      "loss/crossentropy": 2.410393309593201,
+      "loss/hidden": 3.73359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2507057674229145,
+      "step": 3340
+    },
+    {
+      "epoch": 0.335,
+      "grad_norm": 35.25,
+      "grad_norm_var": 9.292122395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.6462,
+      "loss/crossentropy": 2.299106788635254,
+      "loss/hidden": 3.76328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2520503532141447,
+      "step": 3350
+    },
+    {
+      "epoch": 0.336,
+      "grad_norm": 44.25,
+      "grad_norm_var": 13.2869140625,
+      "learning_rate": 0.0001,
+      "loss": 8.6432,
+      "loss/crossentropy": 2.2736548662185667,
+      "loss/hidden": 3.81796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2679216586053371,
+      "step": 3360
+    },
+    {
+      "epoch": 0.337,
+      "grad_norm": 40.25,
+      "grad_norm_var": 16.704622395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.6969,
+      "loss/crossentropy": 2.300890862941742,
+      "loss/hidden": 3.85703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27277763597667215,
+      "step": 3370
+    },
+    {
+      "epoch": 0.338,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.362239583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.6139,
+      "loss/crossentropy": 2.2713751554489137,
+      "loss/hidden": 3.739453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2698351971805096,
+      "step": 3380
+    },
+    {
+      "epoch": 0.339,
+      "grad_norm": 35.5,
+      "grad_norm_var": 7.383333333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.6652,
+      "loss/crossentropy": 2.154055279493332,
+      "loss/hidden": 3.784375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2516574438661337,
+      "step": 3390
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 46.5,
+      "grad_norm_var": 231.42083333333332,
+      "learning_rate": 0.0001,
+      "loss": 8.7521,
+      "loss/crossentropy": 2.1560363829135896,
+      "loss/hidden": 3.78828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2552214227616787,
+      "step": 3400
+    },
+    {
+      "epoch": 0.341,
+      "grad_norm": 34.0,
+      "grad_norm_var": 16.01015625,
+      "learning_rate": 0.0001,
+      "loss": 8.6719,
+      "loss/crossentropy": 2.210598033666611,
+      "loss/hidden": 3.7078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24775836057960987,
+      "step": 3410
+    },
+    {
+      "epoch": 0.342,
+      "grad_norm": 46.75,
+      "grad_norm_var": 23.555989583333332,
+      "learning_rate": 0.0001,
+      "loss": 8.732,
+      "loss/crossentropy": 2.0089349642395975,
+      "loss/hidden": 3.95625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24481147788465024,
+      "step": 3420
+    },
+    {
+      "epoch": 0.343,
+      "grad_norm": 43.5,
+      "grad_norm_var": 24.71015625,
+      "learning_rate": 0.0001,
+      "loss": 8.5835,
+      "loss/crossentropy": 2.167103961110115,
+      "loss/hidden": 3.73046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23311931267380714,
+      "step": 3430
+    },
+    {
+      "epoch": 0.344,
+      "grad_norm": 38.75,
+      "grad_norm_var": 16.239322916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.7327,
+      "loss/crossentropy": 2.223423732817173,
+      "loss/hidden": 3.916796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27787868976593016,
+      "step": 3440
+    },
+    {
+      "epoch": 0.345,
+      "grad_norm": 32.5,
+      "grad_norm_var": 12.4166015625,
+      "learning_rate": 0.0001,
+      "loss": 8.6674,
+      "loss/crossentropy": 2.190432313084602,
+      "loss/hidden": 3.933203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2527661222964525,
+      "step": 3450
+    },
+    {
+      "epoch": 0.346,
+      "grad_norm": 38.5,
+      "grad_norm_var": 239.47337239583334,
+      "learning_rate": 0.0001,
+      "loss": 8.678,
+      "loss/crossentropy": 2.121590741723776,
+      "loss/hidden": 3.746875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22831694399937988,
+      "step": 3460
+    },
+    {
+      "epoch": 0.347,
+      "grad_norm": 32.5,
+      "grad_norm_var": 24.29765625,
+      "learning_rate": 0.0001,
+      "loss": 8.5703,
+      "loss/crossentropy": 2.238359749317169,
+      "loss/hidden": 3.776171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2545921359211206,
+      "step": 3470
+    },
+    {
+      "epoch": 0.348,
+      "grad_norm": 31.375,
+      "grad_norm_var": 57.77291666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.667,
+      "loss/crossentropy": 2.183069321513176,
+      "loss/hidden": 3.723828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26235801838338374,
+      "step": 3480
+    },
+    {
+      "epoch": 0.349,
+      "grad_norm": 37.25,
+      "grad_norm_var": 47.95104166666667,
+      "learning_rate": 0.0001,
+      "loss": 8.6657,
+      "loss/crossentropy": 2.294294211268425,
+      "loss/hidden": 3.7875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2434792961925268,
+      "step": 3490
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 55.75,
+      "grad_norm_var": 32.8931640625,
+      "learning_rate": 0.0001,
+      "loss": 8.6293,
+      "loss/crossentropy": 2.2934862852096556,
+      "loss/hidden": 3.7671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2526032764464617,
+      "step": 3500
+    },
+    {
+      "epoch": 0.351,
+      "grad_norm": 41.75,
+      "grad_norm_var": 37.820247395833334,
+      "learning_rate": 0.0001,
+      "loss": 8.6558,
+      "loss/crossentropy": 2.240943320095539,
+      "loss/hidden": 3.798828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2533996310085058,
+      "step": 3510
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 34.0,
+      "grad_norm_var": 19.158333333333335,
+      "learning_rate": 0.0001,
+      "loss": 8.7265,
+      "loss/crossentropy": 2.2778089314699175,
+      "loss/hidden": 3.74765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24110115878283978,
+      "step": 3520
+    },
+    {
+      "epoch": 0.353,
+      "grad_norm": 33.75,
+      "grad_norm_var": 14.61640625,
+      "learning_rate": 0.0001,
+      "loss": 8.6184,
+      "loss/crossentropy": 2.2676281452178957,
+      "loss/hidden": 3.703515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25084604155272244,
+      "step": 3530
+    },
+    {
+      "epoch": 0.354,
+      "grad_norm": 41.25,
+      "grad_norm_var": 11.601822916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.6994,
+      "loss/crossentropy": 2.291515235602856,
+      "loss/hidden": 3.919140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.29151172675192355,
+      "step": 3540
+    },
+    {
+      "epoch": 0.355,
+      "grad_norm": 35.75,
+      "grad_norm_var": 7.620833333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.606,
+      "loss/crossentropy": 2.2359238654375075,
+      "loss/hidden": 3.81484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25299829691648484,
+      "step": 3550
+    },
+    {
+      "epoch": 0.356,
+      "grad_norm": 35.25,
+      "grad_norm_var": 7.1,
+      "learning_rate": 0.0001,
+      "loss": 8.5051,
+      "loss/crossentropy": 2.2826671570539476,
+      "loss/hidden": 3.843359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26029736921191216,
+      "step": 3560
+    },
+    {
+      "epoch": 0.357,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.422330729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.5601,
+      "loss/crossentropy": 2.29564026594162,
+      "loss/hidden": 3.790234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24953206069767475,
+      "step": 3570
+    },
+    {
+      "epoch": 0.358,
+      "grad_norm": 34.25,
+      "grad_norm_var": 9.414518229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.677,
+      "loss/crossentropy": 2.360885411500931,
+      "loss/hidden": 3.812890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2608415879309177,
+      "step": 3580
+    },
+    {
+      "epoch": 0.359,
+      "grad_norm": 41.75,
+      "grad_norm_var": 7.558333333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.591,
+      "loss/crossentropy": 2.0790357582271097,
+      "loss/hidden": 3.759375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23188311588019134,
+      "step": 3590
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 38.0,
+      "grad_norm_var": 10.890625,
+      "learning_rate": 0.0001,
+      "loss": 8.6616,
+      "loss/crossentropy": 2.289612150192261,
+      "loss/hidden": 3.878125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2747206833213568,
+      "step": 3600
+    },
+    {
+      "epoch": 0.361,
+      "grad_norm": 42.0,
+      "grad_norm_var": 14.145572916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.589,
+      "loss/crossentropy": 2.132586442679167,
+      "loss/hidden": 3.811328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25861772149801254,
+      "step": 3610
+    },
+    {
+      "epoch": 0.362,
+      "grad_norm": 35.0,
+      "grad_norm_var": 11.537239583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.6171,
+      "loss/crossentropy": 2.3269239187240602,
+      "loss/hidden": 3.8265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2712526451796293,
+      "step": 3620
+    },
+    {
+      "epoch": 0.363,
+      "grad_norm": 33.75,
+      "grad_norm_var": 20.012239583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.5134,
+      "loss/crossentropy": 2.326130175590515,
+      "loss/hidden": 3.791796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24642233476042746,
+      "step": 3630
+    },
+    {
+      "epoch": 0.364,
+      "grad_norm": 31.875,
+      "grad_norm_var": 30.459830729166665,
+      "learning_rate": 0.0001,
+      "loss": 8.5808,
+      "loss/crossentropy": 2.30035699903965,
+      "loss/hidden": 3.733984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26470062173902986,
+      "step": 3640
+    },
+    {
+      "epoch": 0.365,
+      "grad_norm": 37.25,
+      "grad_norm_var": 8.626041666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.5614,
+      "loss/crossentropy": 2.3303778156638146,
+      "loss/hidden": 3.871875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27759894989430905,
+      "step": 3650
+    },
+    {
+      "epoch": 0.366,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.374739583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.5104,
+      "loss/crossentropy": 2.3886318862438203,
+      "loss/hidden": 3.801953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26098744831979276,
+      "step": 3660
+    },
+    {
+      "epoch": 0.367,
+      "grad_norm": 28.875,
+      "grad_norm_var": 18.685416666666665,
+      "learning_rate": 0.0001,
+      "loss": 8.5148,
+      "loss/crossentropy": 2.3471581265330315,
+      "loss/hidden": 3.760546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.258730498701334,
+      "step": 3670
+    },
+    {
+      "epoch": 0.368,
+      "grad_norm": 48.5,
+      "grad_norm_var": 36.4744140625,
+      "learning_rate": 0.0001,
+      "loss": 8.5484,
+      "loss/crossentropy": 2.1698090970516204,
+      "loss/hidden": 3.70078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24077225737273694,
+      "step": 3680
+    },
+    {
+      "epoch": 0.369,
+      "grad_norm": 30.125,
+      "grad_norm_var": 25.111393229166666,
+      "learning_rate": 0.0001,
+      "loss": 8.5671,
+      "loss/crossentropy": 2.1866169169545175,
+      "loss/hidden": 3.8203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24500060379505156,
+      "step": 3690
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 30.375,
+      "grad_norm_var": 2.130484072425508e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.6635,
+      "loss/crossentropy": 2.453831446170807,
+      "loss/hidden": 4.06171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2722825076431036,
+      "step": 3700
+    },
+    {
+      "epoch": 0.371,
+      "grad_norm": 32.5,
+      "grad_norm_var": 32.78795572916667,
+      "learning_rate": 0.0001,
+      "loss": 8.5788,
+      "loss/crossentropy": 2.212860561162233,
+      "loss/hidden": 3.629296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21873269081115723,
+      "step": 3710
+    },
+    {
+      "epoch": 0.372,
+      "grad_norm": 29.625,
+      "grad_norm_var": 31.475455729166665,
+      "learning_rate": 0.0001,
+      "loss": 8.5924,
+      "loss/crossentropy": 2.3226836264133452,
+      "loss/hidden": 3.711328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2526125326752663,
+      "step": 3720
+    },
+    {
+      "epoch": 0.373,
+      "grad_norm": 40.25,
+      "grad_norm_var": 74.74680989583334,
+      "learning_rate": 0.0001,
+      "loss": 8.7309,
+      "loss/crossentropy": 2.211090712249279,
+      "loss/hidden": 3.85078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2619786085560918,
+      "step": 3730
+    },
+    {
+      "epoch": 0.374,
+      "grad_norm": 30.5,
+      "grad_norm_var": 49.24108072916667,
+      "learning_rate": 0.0001,
+      "loss": 8.4683,
+      "loss/crossentropy": 2.117314028739929,
+      "loss/hidden": 3.787890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24047958478331566,
+      "step": 3740
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 43.0,
+      "grad_norm_var": 13.481184895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.5891,
+      "loss/crossentropy": 2.3331554174423217,
+      "loss/hidden": 3.737109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.265634342469275,
+      "step": 3750
+    },
+    {
+      "epoch": 0.376,
+      "grad_norm": 70.0,
+      "grad_norm_var": 86.09837239583334,
+      "learning_rate": 0.0001,
+      "loss": 8.4671,
+      "loss/crossentropy": 2.279883709549904,
+      "loss/hidden": 3.764453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2557247843593359,
+      "step": 3760
+    },
+    {
+      "epoch": 0.377,
+      "grad_norm": 37.5,
+      "grad_norm_var": 101.2072265625,
+      "learning_rate": 0.0001,
+      "loss": 8.6263,
+      "loss/crossentropy": 2.119837614893913,
+      "loss/hidden": 3.7328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24443967882543802,
+      "step": 3770
+    },
+    {
+      "epoch": 0.378,
+      "grad_norm": 36.0,
+      "grad_norm_var": 11.41015625,
+      "learning_rate": 0.0001,
+      "loss": 8.573,
+      "loss/crossentropy": 2.2743908286094667,
+      "loss/hidden": 3.7046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24655351527035235,
+      "step": 3780
+    },
+    {
+      "epoch": 0.379,
+      "grad_norm": 30.5,
+      "grad_norm_var": 22.8337890625,
+      "learning_rate": 0.0001,
+      "loss": 8.5496,
+      "loss/crossentropy": 2.298393335938454,
+      "loss/hidden": 3.661328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2468328095972538,
+      "step": 3790
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 36.75,
+      "grad_norm_var": 21.886393229166668,
+      "learning_rate": 0.0001,
+      "loss": 8.4728,
+      "loss/crossentropy": 2.285892593860626,
+      "loss/hidden": 3.696484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22759304326027632,
+      "step": 3800
+    },
+    {
+      "epoch": 0.381,
+      "grad_norm": 37.0,
+      "grad_norm_var": 25.983268229166665,
+      "learning_rate": 0.0001,
+      "loss": 8.4654,
+      "loss/crossentropy": 2.200984264165163,
+      "loss/hidden": 3.760546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2451606505550444,
+      "step": 3810
+    },
+    {
+      "epoch": 0.382,
+      "grad_norm": 37.5,
+      "grad_norm_var": 13.956705729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.5469,
+      "loss/crossentropy": 2.2647089801728724,
+      "loss/hidden": 3.803125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2533489029854536,
+      "step": 3820
+    },
+    {
+      "epoch": 0.383,
+      "grad_norm": 34.75,
+      "grad_norm_var": 7.221809895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.579,
+      "loss/crossentropy": 2.386853316426277,
+      "loss/hidden": 3.795703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2518702711910009,
+      "step": 3830
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 42.0,
+      "grad_norm_var": 15.8650390625,
+      "learning_rate": 0.0001,
+      "loss": 8.4816,
+      "loss/crossentropy": 2.3519598811864855,
+      "loss/hidden": 3.75625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.253318839520216,
+      "step": 3840
+    },
+    {
+      "epoch": 0.385,
+      "grad_norm": 38.25,
+      "grad_norm_var": 14.728580729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.7292,
+      "loss/crossentropy": 2.164312995970249,
+      "loss/hidden": 3.8234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24283661209046842,
+      "step": 3850
+    },
+    {
+      "epoch": 0.386,
+      "grad_norm": 37.75,
+      "grad_norm_var": 14.876822916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.6236,
+      "loss/crossentropy": 2.1916888520121574,
+      "loss/hidden": 3.69140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23480207994580268,
+      "step": 3860
+    },
+    {
+      "epoch": 0.387,
+      "grad_norm": 33.25,
+      "grad_norm_var": 17.855208333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.4472,
+      "loss/crossentropy": 2.2794968456029894,
+      "loss/hidden": 3.759375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2547815594822168,
+      "step": 3870
+    },
+    {
+      "epoch": 0.388,
+      "grad_norm": 39.75,
+      "grad_norm_var": 10.063541666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.5547,
+      "loss/crossentropy": 2.2261194586753845,
+      "loss/hidden": 3.73203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24962050542235376,
+      "step": 3880
+    },
+    {
+      "epoch": 0.389,
+      "grad_norm": 33.75,
+      "grad_norm_var": 25.229622395833335,
+      "learning_rate": 0.0001,
+      "loss": 8.428,
+      "loss/crossentropy": 2.2669249922037125,
+      "loss/hidden": 3.589453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22769966274499892,
+      "step": 3890
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 34.75,
+      "grad_norm_var": 22.89765625,
+      "learning_rate": 0.0001,
+      "loss": 8.4932,
+      "loss/crossentropy": 2.2599128648638724,
+      "loss/hidden": 3.744921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24268860407173634,
+      "step": 3900
+    },
+    {
+      "epoch": 0.391,
+      "grad_norm": 37.0,
+      "grad_norm_var": 44.83483072916667,
+      "learning_rate": 0.0001,
+      "loss": 8.5202,
+      "loss/crossentropy": 2.1972982972860335,
+      "loss/hidden": 3.812109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25300098545849325,
+      "step": 3910
+    },
+    {
+      "epoch": 0.392,
+      "grad_norm": 40.25,
+      "grad_norm_var": 11.49140625,
+      "learning_rate": 0.0001,
+      "loss": 8.5002,
+      "loss/crossentropy": 2.347226142883301,
+      "loss/hidden": 3.658203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24455956518650054,
+      "step": 3920
+    },
+    {
+      "epoch": 0.393,
+      "grad_norm": 31.125,
+      "grad_norm_var": 17.219791666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.4054,
+      "loss/crossentropy": 2.2351839393377304,
+      "loss/hidden": 3.59140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22245508581399917,
+      "step": 3930
+    },
+    {
+      "epoch": 0.394,
+      "grad_norm": 37.0,
+      "grad_norm_var": 11.382747395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.4771,
+      "loss/crossentropy": 2.4160192787647246,
+      "loss/hidden": 3.700390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2585303969681263,
+      "step": 3940
+    },
+    {
+      "epoch": 0.395,
+      "grad_norm": 30.875,
+      "grad_norm_var": 10.731705729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.4064,
+      "loss/crossentropy": 2.329276342689991,
+      "loss/hidden": 3.641796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23809341453015803,
+      "step": 3950
+    },
+    {
+      "epoch": 0.396,
+      "grad_norm": 31.875,
+      "grad_norm_var": 18.456184895833335,
+      "learning_rate": 0.0001,
+      "loss": 8.4948,
+      "loss/crossentropy": 2.397703355550766,
+      "loss/hidden": 3.8703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2577156092971563,
+      "step": 3960
+    },
+    {
+      "epoch": 0.397,
+      "grad_norm": 35.25,
+      "grad_norm_var": 16.1291015625,
+      "learning_rate": 0.0001,
+      "loss": 8.5021,
+      "loss/crossentropy": 2.3261855766177177,
+      "loss/hidden": 3.694921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523797513917089,
+      "step": 3970
+    },
+    {
+      "epoch": 0.398,
+      "grad_norm": 33.0,
+      "grad_norm_var": 8.205208333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.4531,
+      "loss/crossentropy": 2.235724928975105,
+      "loss/hidden": 3.7296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2461556438356638,
+      "step": 3980
+    },
+    {
+      "epoch": 0.399,
+      "grad_norm": 35.5,
+      "grad_norm_var": 7.25390625,
+      "learning_rate": 0.0001,
+      "loss": 8.3779,
+      "loss/crossentropy": 2.258603112399578,
+      "loss/hidden": 3.731640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2523756165057421,
+      "step": 3990
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 32.5,
+      "grad_norm_var": 11.34765625,
+      "learning_rate": 0.0001,
+      "loss": 8.542,
+      "loss/crossentropy": 2.2897588342428206,
+      "loss/hidden": 3.6453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2516363400965929,
+      "step": 4000
+    },
+    {
+      "epoch": 0.401,
+      "grad_norm": 32.0,
+      "grad_norm_var": 9.889518229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.4234,
+      "loss/crossentropy": 2.1721622362732886,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25033344645053146,
+      "step": 4010
+    },
+    {
+      "epoch": 0.402,
+      "grad_norm": 35.75,
+      "grad_norm_var": 15.199934895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.4018,
+      "loss/crossentropy": 2.1795839801430703,
+      "loss/hidden": 3.78515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24097833968698978,
+      "step": 4020
+    },
+    {
+      "epoch": 0.403,
+      "grad_norm": 37.75,
+      "grad_norm_var": 162.76451822916667,
+      "learning_rate": 0.0001,
+      "loss": 8.4738,
+      "loss/crossentropy": 2.4760424941778183,
+      "loss/hidden": 3.86484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.3052004296332598,
+      "step": 4030
+    },
+    {
+      "epoch": 0.404,
+      "grad_norm": 41.5,
+      "grad_norm_var": 2.408370239548424e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.4302,
+      "loss/crossentropy": 2.1855442106723784,
+      "loss/hidden": 3.77578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24037305619567634,
+      "step": 4040
+    },
+    {
+      "epoch": 0.405,
+      "grad_norm": 32.0,
+      "grad_norm_var": 2.4083702412167086e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.3213,
+      "loss/crossentropy": 2.22409378439188,
+      "loss/hidden": 3.653515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23280739206820728,
+      "step": 4050
+    },
+    {
+      "epoch": 0.406,
+      "grad_norm": 39.5,
+      "grad_norm_var": 9.825,
+      "learning_rate": 0.0001,
+      "loss": 8.4657,
+      "loss/crossentropy": 2.28346493691206,
+      "loss/hidden": 3.773828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542352583259344,
+      "step": 4060
+    },
+    {
+      "epoch": 0.407,
+      "grad_norm": 39.75,
+      "grad_norm_var": 7.640625,
+      "learning_rate": 0.0001,
+      "loss": 8.3937,
+      "loss/crossentropy": 2.2639666229486464,
+      "loss/hidden": 3.851953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27007580138742926,
+      "step": 4070
+    },
+    {
+      "epoch": 0.408,
+      "grad_norm": 37.25,
+      "grad_norm_var": 10.493684895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.468,
+      "loss/crossentropy": 2.356726923584938,
+      "loss/hidden": 3.716796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.242316972091794,
+      "step": 4080
+    },
+    {
+      "epoch": 0.409,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.8462890625,
+      "learning_rate": 0.0001,
+      "loss": 8.3684,
+      "loss/crossentropy": 2.151221239566803,
+      "loss/hidden": 3.692578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2382324907928705,
+      "step": 4090
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 37.0,
+      "grad_norm_var": 13.3775390625,
+      "learning_rate": 0.0001,
+      "loss": 8.296,
+      "loss/crossentropy": 2.1645509719848635,
+      "loss/hidden": 3.65703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2371783286333084,
+      "step": 4100
+    },
+    {
+      "epoch": 0.411,
+      "grad_norm": 32.5,
+      "grad_norm_var": 25.270247395833334,
+      "learning_rate": 0.0001,
+      "loss": 8.3435,
+      "loss/crossentropy": 2.1995768398046494,
+      "loss/hidden": 3.6640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23499403558671475,
+      "step": 4110
+    },
+    {
+      "epoch": 0.412,
+      "grad_norm": 38.0,
+      "grad_norm_var": 13.875,
+      "learning_rate": 0.0001,
+      "loss": 8.3746,
+      "loss/crossentropy": 2.135378623008728,
+      "loss/hidden": 3.612109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24154506418854,
+      "step": 4120
+    },
+    {
+      "epoch": 0.413,
+      "grad_norm": 40.25,
+      "grad_norm_var": 12.233072916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.3505,
+      "loss/crossentropy": 2.356252074241638,
+      "loss/hidden": 3.734765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2688302733004093,
+      "step": 4130
+    },
+    {
+      "epoch": 0.414,
+      "grad_norm": 39.25,
+      "grad_norm_var": 14.333072916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.4228,
+      "loss/crossentropy": 2.334370291233063,
+      "loss/hidden": 3.648046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23012813031673432,
+      "step": 4140
+    },
+    {
+      "epoch": 0.415,
+      "grad_norm": 35.5,
+      "grad_norm_var": 8.170833333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.3324,
+      "loss/crossentropy": 2.214457754790783,
+      "loss/hidden": 3.651171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24744220934808253,
+      "step": 4150
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 37.0,
+      "grad_norm_var": 7.603580729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.3237,
+      "loss/crossentropy": 2.3053094416856768,
+      "loss/hidden": 3.62578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2290981512516737,
+      "step": 4160
+    },
+    {
+      "epoch": 0.417,
+      "grad_norm": 50.75,
+      "grad_norm_var": 23.6931640625,
+      "learning_rate": 0.0001,
+      "loss": 8.2895,
+      "loss/crossentropy": 2.100640784204006,
+      "loss/hidden": 3.72578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24771791882812977,
+      "step": 4170
+    },
+    {
+      "epoch": 0.418,
+      "grad_norm": 42.25,
+      "grad_norm_var": 25.6509765625,
+      "learning_rate": 0.0001,
+      "loss": 8.4997,
+      "loss/crossentropy": 2.4176384449005126,
+      "loss/hidden": 3.734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2665034931153059,
+      "step": 4180
+    },
+    {
+      "epoch": 0.419,
+      "grad_norm": 46.75,
+      "grad_norm_var": 16.7041015625,
+      "learning_rate": 0.0001,
+      "loss": 8.241,
+      "loss/crossentropy": 2.0916571110486983,
+      "loss/hidden": 3.640234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23109357040375472,
+      "step": 4190
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 34.0,
+      "grad_norm_var": 10.265625,
+      "learning_rate": 0.0001,
+      "loss": 8.3903,
+      "loss/crossentropy": 2.4284214213490487,
+      "loss/hidden": 3.676171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24520040042698382,
+      "step": 4200
+    },
+    {
+      "epoch": 0.421,
+      "grad_norm": 48.25,
+      "grad_norm_var": 17.694205729166665,
+      "learning_rate": 0.0001,
+      "loss": 8.4379,
+      "loss/crossentropy": 2.342070159316063,
+      "loss/hidden": 3.750390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2608378600329161,
+      "step": 4210
+    },
+    {
+      "epoch": 0.422,
+      "grad_norm": 31.875,
+      "grad_norm_var": 21.2806640625,
+      "learning_rate": 0.0001,
+      "loss": 8.2667,
+      "loss/crossentropy": 2.3339773267507553,
+      "loss/hidden": 3.68828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24873733669519424,
+      "step": 4220
+    },
+    {
+      "epoch": 0.423,
+      "grad_norm": 34.0,
+      "grad_norm_var": 10.834375,
+      "learning_rate": 0.0001,
+      "loss": 8.313,
+      "loss/crossentropy": 2.2396500378847124,
+      "loss/hidden": 3.647265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23453602455556394,
+      "step": 4230
+    },
+    {
+      "epoch": 0.424,
+      "grad_norm": 31.875,
+      "grad_norm_var": 20.712239583333332,
+      "learning_rate": 0.0001,
+      "loss": 8.3885,
+      "loss/crossentropy": 2.1176706120371818,
+      "loss/hidden": 3.818359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24183569326996804,
+      "step": 4240
+    },
+    {
+      "epoch": 0.425,
+      "grad_norm": 46.0,
+      "grad_norm_var": 18.212434895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.4311,
+      "loss/crossentropy": 2.161303213238716,
+      "loss/hidden": 3.71015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23596356846392155,
+      "step": 4250
+    },
+    {
+      "epoch": 0.426,
+      "grad_norm": 31.75,
+      "grad_norm_var": 18.470833333333335,
+      "learning_rate": 0.0001,
+      "loss": 8.389,
+      "loss/crossentropy": 2.1959901452064514,
+      "loss/hidden": 3.683203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23274125456809996,
+      "step": 4260
+    },
+    {
+      "epoch": 0.427,
+      "grad_norm": 40.0,
+      "grad_norm_var": 15.642643229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.3995,
+      "loss/crossentropy": 2.3957006752491,
+      "loss/hidden": 3.6546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23920847922563554,
+      "step": 4270
+    },
+    {
+      "epoch": 0.428,
+      "grad_norm": 32.25,
+      "grad_norm_var": 12.506705729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.3247,
+      "loss/crossentropy": 2.140459132194519,
+      "loss/hidden": 3.776953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24535099379718303,
+      "step": 4280
+    },
+    {
+      "epoch": 0.429,
+      "grad_norm": 31.125,
+      "grad_norm_var": 14.913997395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.2296,
+      "loss/crossentropy": 2.330910986661911,
+      "loss/hidden": 3.65546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25125612393021585,
+      "step": 4290
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 31.0,
+      "grad_norm_var": 9.062955729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.2789,
+      "loss/crossentropy": 2.21281051337719,
+      "loss/hidden": 3.717578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24373065643012523,
+      "step": 4300
+    },
+    {
+      "epoch": 0.431,
+      "grad_norm": 46.25,
+      "grad_norm_var": 18.794205729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.283,
+      "loss/crossentropy": 2.182158187031746,
+      "loss/hidden": 3.6703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24544784277677537,
+      "step": 4310
+    },
+    {
+      "epoch": 0.432,
+      "grad_norm": 45.75,
+      "grad_norm_var": 38.16458333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.2852,
+      "loss/crossentropy": 2.365989252924919,
+      "loss/hidden": 3.706640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24662891514599322,
+      "step": 4320
+    },
+    {
+      "epoch": 0.433,
+      "grad_norm": 35.25,
+      "grad_norm_var": 16.307291666666668,
+      "learning_rate": 0.0001,
+      "loss": 8.2794,
+      "loss/crossentropy": 2.1546150177717207,
+      "loss/hidden": 3.620703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2297368910163641,
+      "step": 4330
+    },
+    {
+      "epoch": 0.434,
+      "grad_norm": 31.625,
+      "grad_norm_var": 19.6400390625,
+      "learning_rate": 0.0001,
+      "loss": 8.3956,
+      "loss/crossentropy": 2.385764144361019,
+      "loss/hidden": 3.7015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24324760176241397,
+      "step": 4340
+    },
+    {
+      "epoch": 0.435,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.987239583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.2695,
+      "loss/crossentropy": 2.112358179688454,
+      "loss/hidden": 3.788671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24638627246022224,
+      "step": 4350
+    },
+    {
+      "epoch": 0.436,
+      "grad_norm": 34.5,
+      "grad_norm_var": 8.0875,
+      "learning_rate": 0.0001,
+      "loss": 8.448,
+      "loss/crossentropy": 2.182591002434492,
+      "loss/hidden": 3.746875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24270438468083738,
+      "step": 4360
+    },
+    {
+      "epoch": 0.437,
+      "grad_norm": 38.25,
+      "grad_norm_var": 5.695833333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.3975,
+      "loss/crossentropy": 2.401635229587555,
+      "loss/hidden": 3.65859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24400906264781952,
+      "step": 4370
+    },
+    {
+      "epoch": 0.438,
+      "grad_norm": 34.75,
+      "grad_norm_var": 8.372330729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.3822,
+      "loss/crossentropy": 2.3247624695301057,
+      "loss/hidden": 3.591015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24138722717761993,
+      "step": 4380
+    },
+    {
+      "epoch": 0.439,
+      "grad_norm": 62.75,
+      "grad_norm_var": 70.6103515625,
+      "learning_rate": 0.0001,
+      "loss": 8.3893,
+      "loss/crossentropy": 2.1605212301015855,
+      "loss/hidden": 3.75,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.270247707888484,
+      "step": 4390
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 34.5,
+      "grad_norm_var": 80.6375,
+      "learning_rate": 0.0001,
+      "loss": 8.3494,
+      "loss/crossentropy": 2.201382315158844,
+      "loss/hidden": 3.55390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22101359032094478,
+      "step": 4400
+    },
+    {
+      "epoch": 0.441,
+      "grad_norm": 37.75,
+      "grad_norm_var": 12.06875,
+      "learning_rate": 0.0001,
+      "loss": 8.3221,
+      "loss/crossentropy": 2.1608468025922773,
+      "loss/hidden": 3.570703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22636283356696368,
+      "step": 4410
+    },
+    {
+      "epoch": 0.442,
+      "grad_norm": 32.25,
+      "grad_norm_var": 16.13515625,
+      "learning_rate": 0.0001,
+      "loss": 8.2013,
+      "loss/crossentropy": 2.2836680516600607,
+      "loss/hidden": 3.6765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24737481120973825,
+      "step": 4420
+    },
+    {
+      "epoch": 0.443,
+      "grad_norm": 34.5,
+      "grad_norm_var": 14.570768229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.3241,
+      "loss/crossentropy": 2.1778072111308573,
+      "loss/hidden": 3.685546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23579915445297955,
+      "step": 4430
+    },
+    {
+      "epoch": 0.444,
+      "grad_norm": 33.5,
+      "grad_norm_var": 14.844205729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.2468,
+      "loss/crossentropy": 2.115637184679508,
+      "loss/hidden": 3.54921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22633790075778962,
+      "step": 4440
+    },
+    {
+      "epoch": 0.445,
+      "grad_norm": 40.0,
+      "grad_norm_var": 8.757291666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.3095,
+      "loss/crossentropy": 2.2457625687122347,
+      "loss/hidden": 3.678125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24750035293400288,
+      "step": 4450
+    },
+    {
+      "epoch": 0.446,
+      "grad_norm": 30.75,
+      "grad_norm_var": 9.359309895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.3604,
+      "loss/crossentropy": 2.2244989693164827,
+      "loss/hidden": 3.742578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.27126055024564266,
+      "step": 4460
+    },
+    {
+      "epoch": 0.447,
+      "grad_norm": 32.75,
+      "grad_norm_var": 8.16015625,
+      "learning_rate": 0.0001,
+      "loss": 8.293,
+      "loss/crossentropy": 2.208540087938309,
+      "loss/hidden": 3.591796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23848242741078138,
+      "step": 4470
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 31.5,
+      "grad_norm_var": 27.215625,
+      "learning_rate": 0.0001,
+      "loss": 8.4332,
+      "loss/crossentropy": 2.182788160443306,
+      "loss/hidden": 3.775,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26532087065279486,
+      "step": 4480
+    },
+    {
+      "epoch": 0.449,
+      "grad_norm": 39.5,
+      "grad_norm_var": 29.916666666666668,
+      "learning_rate": 0.0001,
+      "loss": 8.2552,
+      "loss/crossentropy": 2.172296644747257,
+      "loss/hidden": 3.66875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23516011722385882,
+      "step": 4490
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 33.75,
+      "grad_norm_var": 21.775455729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.3872,
+      "loss/crossentropy": 2.304448103904724,
+      "loss/hidden": 3.56015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23418739810585976,
+      "step": 4500
+    },
+    {
+      "epoch": 0.451,
+      "grad_norm": 31.75,
+      "grad_norm_var": 23.40390625,
+      "learning_rate": 0.0001,
+      "loss": 8.4055,
+      "loss/crossentropy": 2.2493974685668947,
+      "loss/hidden": 3.652734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2410556711256504,
+      "step": 4510
+    },
+    {
+      "epoch": 0.452,
+      "grad_norm": 32.25,
+      "grad_norm_var": 19.04375,
+      "learning_rate": 0.0001,
+      "loss": 8.2847,
+      "loss/crossentropy": 2.0756215125322344,
+      "loss/hidden": 3.658984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23735107891261578,
+      "step": 4520
+    },
+    {
+      "epoch": 0.453,
+      "grad_norm": 38.5,
+      "grad_norm_var": 11.048958333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.4058,
+      "loss/crossentropy": 2.2705332577228545,
+      "loss/hidden": 3.75703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2552911601960659,
+      "step": 4530
+    },
+    {
+      "epoch": 0.454,
+      "grad_norm": 32.25,
+      "grad_norm_var": 15.964583333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.3352,
+      "loss/crossentropy": 2.2062640622258187,
+      "loss/hidden": 3.663671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23420217223465442,
+      "step": 4540
+    },
+    {
+      "epoch": 0.455,
+      "grad_norm": 35.5,
+      "grad_norm_var": 18.8041015625,
+      "learning_rate": 0.0001,
+      "loss": 8.1398,
+      "loss/crossentropy": 2.2153613708913324,
+      "loss/hidden": 3.688671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23869724282994867,
+      "step": 4550
+    },
+    {
+      "epoch": 0.456,
+      "grad_norm": 32.25,
+      "grad_norm_var": 15.843489583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.3482,
+      "loss/crossentropy": 2.2614838272333144,
+      "loss/hidden": 3.719921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2509554075077176,
+      "step": 4560
+    },
+    {
+      "epoch": 0.457,
+      "grad_norm": 37.0,
+      "grad_norm_var": 17.045572916666668,
+      "learning_rate": 0.0001,
+      "loss": 8.3724,
+      "loss/crossentropy": 2.1315455704927446,
+      "loss/hidden": 3.720703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24938025698065758,
+      "step": 4570
+    },
+    {
+      "epoch": 0.458,
+      "grad_norm": 31.75,
+      "grad_norm_var": 18.753580729166668,
+      "learning_rate": 0.0001,
+      "loss": 8.1789,
+      "loss/crossentropy": 2.250936383008957,
+      "loss/hidden": 3.601953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23103775745257735,
+      "step": 4580
+    },
+    {
+      "epoch": 0.459,
+      "grad_norm": 31.75,
+      "grad_norm_var": 9.3087890625,
+      "learning_rate": 0.0001,
+      "loss": 8.2367,
+      "loss/crossentropy": 2.0491979137063026,
+      "loss/hidden": 3.767578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24958254247903824,
+      "step": 4590
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.034309895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.2813,
+      "loss/crossentropy": 2.1869849786162376,
+      "loss/hidden": 3.678125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2445020995102823,
+      "step": 4600
+    },
+    {
+      "epoch": 0.461,
+      "grad_norm": 33.25,
+      "grad_norm_var": 16.960416666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.2774,
+      "loss/crossentropy": 2.3401444420218467,
+      "loss/hidden": 3.644140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2346777945756912,
+      "step": 4610
+    },
+    {
+      "epoch": 0.462,
+      "grad_norm": 31.875,
+      "grad_norm_var": 20.257747395833334,
+      "learning_rate": 0.0001,
+      "loss": 8.261,
+      "loss/crossentropy": 2.2197701543569566,
+      "loss/hidden": 3.671484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24417277611792088,
+      "step": 4620
+    },
+    {
+      "epoch": 0.463,
+      "grad_norm": 39.75,
+      "grad_norm_var": 21.111458333333335,
+      "learning_rate": 0.0001,
+      "loss": 8.1879,
+      "loss/crossentropy": 2.2979017451405523,
+      "loss/hidden": 3.584765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24459463655948638,
+      "step": 4630
+    },
+    {
+      "epoch": 0.464,
+      "grad_norm": 37.75,
+      "grad_norm_var": 18.424739583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.2222,
+      "loss/crossentropy": 2.1899428203701974,
+      "loss/hidden": 3.63828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23625375218689443,
+      "step": 4640
+    },
+    {
+      "epoch": 0.465,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.676822916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.2134,
+      "loss/crossentropy": 2.173795387148857,
+      "loss/hidden": 3.653515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21926050689071416,
+      "step": 4650
+    },
+    {
+      "epoch": 0.466,
+      "grad_norm": 32.75,
+      "grad_norm_var": 5.024739583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.2134,
+      "loss/crossentropy": 2.2807129830121995,
+      "loss/hidden": 3.5890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22838456649333239,
+      "step": 4660
+    },
+    {
+      "epoch": 0.467,
+      "grad_norm": 32.75,
+      "grad_norm_var": 7.588541666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.2508,
+      "loss/crossentropy": 2.19968124628067,
+      "loss/hidden": 3.590625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23003219701349736,
+      "step": 4670
+    },
+    {
+      "epoch": 0.468,
+      "grad_norm": 32.25,
+      "grad_norm_var": 8.268489583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.2723,
+      "loss/crossentropy": 2.3073908984661102,
+      "loss/hidden": 3.621484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23195548579096795,
+      "step": 4680
+    },
+    {
+      "epoch": 0.469,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.398372395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.3115,
+      "loss/crossentropy": 2.251572087407112,
+      "loss/hidden": 3.615234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24007561076432465,
+      "step": 4690
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 37.25,
+      "grad_norm_var": 12.067708333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.215,
+      "loss/crossentropy": 2.179315264523029,
+      "loss/hidden": 3.614453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23048642594367266,
+      "step": 4700
+    },
+    {
+      "epoch": 0.471,
+      "grad_norm": 33.5,
+      "grad_norm_var": 9.384830729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.4657,
+      "loss/crossentropy": 2.304041627049446,
+      "loss/hidden": 3.705859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2532807156443596,
+      "step": 4710
+    },
+    {
+      "epoch": 0.472,
+      "grad_norm": 34.75,
+      "grad_norm_var": 2.5403116373864177e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.3845,
+      "loss/crossentropy": 2.28207755535841,
+      "loss/hidden": 3.6125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24173217974603176,
+      "step": 4720
+    },
+    {
+      "epoch": 0.473,
+      "grad_norm": 31.375,
+      "grad_norm_var": 2.8580729166666665,
+      "learning_rate": 0.0001,
+      "loss": 8.1924,
+      "loss/crossentropy": 2.317093315720558,
+      "loss/hidden": 3.50859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22850329093635083,
+      "step": 4730
+    },
+    {
+      "epoch": 0.474,
+      "grad_norm": 38.5,
+      "grad_norm_var": 55.86354166666667,
+      "learning_rate": 0.0001,
+      "loss": 8.205,
+      "loss/crossentropy": 2.2346624046564103,
+      "loss/hidden": 3.76015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2535475058481097,
+      "step": 4740
+    },
+    {
+      "epoch": 0.475,
+      "grad_norm": 32.75,
+      "grad_norm_var": 6.105989583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.2004,
+      "loss/crossentropy": 2.205992843210697,
+      "loss/hidden": 3.5,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2157002430409193,
+      "step": 4750
+    },
+    {
+      "epoch": 0.476,
+      "grad_norm": 42.5,
+      "grad_norm_var": 7.551041666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.2493,
+      "loss/crossentropy": 2.2944780766963957,
+      "loss/hidden": 3.553515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2302501540631056,
+      "step": 4760
+    },
+    {
+      "epoch": 0.477,
+      "grad_norm": 34.5,
+      "grad_norm_var": 54.297916666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.2422,
+      "loss/crossentropy": 2.349091801047325,
+      "loss/hidden": 3.726953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25370817482471464,
+      "step": 4770
+    },
+    {
+      "epoch": 0.478,
+      "grad_norm": 35.25,
+      "grad_norm_var": 48.0119140625,
+      "learning_rate": 0.0001,
+      "loss": 8.1697,
+      "loss/crossentropy": 2.419427090883255,
+      "loss/hidden": 3.5984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23952311277389526,
+      "step": 4780
+    },
+    {
+      "epoch": 0.479,
+      "grad_norm": 44.0,
+      "grad_norm_var": 34.6509765625,
+      "learning_rate": 0.0001,
+      "loss": 8.2147,
+      "loss/crossentropy": 2.2663041442632674,
+      "loss/hidden": 3.512890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22062067724764348,
+      "step": 4790
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 30.375,
+      "grad_norm_var": 38.708333333333336,
+      "learning_rate": 0.0001,
+      "loss": 8.1478,
+      "loss/crossentropy": 2.3492169111967085,
+      "loss/hidden": 3.63359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23971957936882973,
+      "step": 4800
+    },
+    {
+      "epoch": 0.481,
+      "grad_norm": 36.75,
+      "grad_norm_var": 9.3056640625,
+      "learning_rate": 0.0001,
+      "loss": 8.165,
+      "loss/crossentropy": 2.232720893621445,
+      "loss/hidden": 3.580078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22206582501530647,
+      "step": 4810
+    },
+    {
+      "epoch": 0.482,
+      "grad_norm": 34.25,
+      "grad_norm_var": 6.8197265625,
+      "learning_rate": 0.0001,
+      "loss": 8.1216,
+      "loss/crossentropy": 2.1783568069338797,
+      "loss/hidden": 3.4515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21557580903172494,
+      "step": 4820
+    },
+    {
+      "epoch": 0.483,
+      "grad_norm": 31.5,
+      "grad_norm_var": 6.039322916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.2857,
+      "loss/crossentropy": 2.3961785644292832,
+      "loss/hidden": 3.640234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.257023797929287,
+      "step": 4830
+    },
+    {
+      "epoch": 0.484,
+      "grad_norm": 33.0,
+      "grad_norm_var": 12.29375,
+      "learning_rate": 0.0001,
+      "loss": 8.1802,
+      "loss/crossentropy": 2.310921123623848,
+      "loss/hidden": 3.526953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23252013735473157,
+      "step": 4840
+    },
+    {
+      "epoch": 0.485,
+      "grad_norm": 33.25,
+      "grad_norm_var": 23.69140625,
+      "learning_rate": 0.0001,
+      "loss": 8.2647,
+      "loss/crossentropy": 2.152234472334385,
+      "loss/hidden": 3.6578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2241989640519023,
+      "step": 4850
+    },
+    {
+      "epoch": 0.486,
+      "grad_norm": 32.75,
+      "grad_norm_var": 26.65390625,
+      "learning_rate": 0.0001,
+      "loss": 8.2819,
+      "loss/crossentropy": 2.1197937928140163,
+      "loss/hidden": 3.702734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24686675220727922,
+      "step": 4860
+    },
+    {
+      "epoch": 0.487,
+      "grad_norm": 37.25,
+      "grad_norm_var": 13.0228515625,
+      "learning_rate": 0.0001,
+      "loss": 8.3156,
+      "loss/crossentropy": 2.345510223507881,
+      "loss/hidden": 3.571484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22903156131505967,
+      "step": 4870
+    },
+    {
+      "epoch": 0.488,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.5337890625,
+      "learning_rate": 0.0001,
+      "loss": 8.1946,
+      "loss/crossentropy": 2.1908657550811768,
+      "loss/hidden": 3.682421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23578502163290976,
+      "step": 4880
+    },
+    {
+      "epoch": 0.489,
+      "grad_norm": 33.75,
+      "grad_norm_var": 24.545572916666668,
+      "learning_rate": 0.0001,
+      "loss": 8.1165,
+      "loss/crossentropy": 2.180183355510235,
+      "loss/hidden": 3.661328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23667961843311786,
+      "step": 4890
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 32.75,
+      "grad_norm_var": 58.805989583333336,
+      "learning_rate": 0.0001,
+      "loss": 8.127,
+      "loss/crossentropy": 2.206071509420872,
+      "loss/hidden": 3.51796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21433540284633637,
+      "step": 4900
+    },
+    {
+      "epoch": 0.491,
+      "grad_norm": 38.0,
+      "grad_norm_var": 50.01223958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1544,
+      "loss/crossentropy": 2.1096328511834144,
+      "loss/hidden": 3.61484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24193457532674073,
+      "step": 4910
+    },
+    {
+      "epoch": 0.492,
+      "grad_norm": 39.25,
+      "grad_norm_var": 17.08125,
+      "learning_rate": 0.0001,
+      "loss": 8.202,
+      "loss/crossentropy": 2.2372330710291863,
+      "loss/hidden": 3.63203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23599924352020024,
+      "step": 4920
+    },
+    {
+      "epoch": 0.493,
+      "grad_norm": 33.25,
+      "grad_norm_var": 12.742643229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.1759,
+      "loss/crossentropy": 2.196950948238373,
+      "loss/hidden": 3.638671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24246960394084455,
+      "step": 4930
+    },
+    {
+      "epoch": 0.494,
+      "grad_norm": 40.5,
+      "grad_norm_var": 6.883072916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.1278,
+      "loss/crossentropy": 2.3560511782765388,
+      "loss/hidden": 3.633984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2382162045687437,
+      "step": 4940
+    },
+    {
+      "epoch": 0.495,
+      "grad_norm": 35.0,
+      "grad_norm_var": 11.6603515625,
+      "learning_rate": 0.0001,
+      "loss": 8.2982,
+      "loss/crossentropy": 2.3035822331905367,
+      "loss/hidden": 3.64140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2485156562179327,
+      "step": 4950
+    },
+    {
+      "epoch": 0.496,
+      "grad_norm": 34.25,
+      "grad_norm_var": 14.192708333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.1579,
+      "loss/crossentropy": 2.184766189754009,
+      "loss/hidden": 3.637109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23721186630427837,
+      "step": 4960
+    },
+    {
+      "epoch": 0.497,
+      "grad_norm": 33.5,
+      "grad_norm_var": 13.871809895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.1975,
+      "loss/crossentropy": 2.1983452700078487,
+      "loss/hidden": 3.697265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24081590361893176,
+      "step": 4970
+    },
+    {
+      "epoch": 0.498,
+      "grad_norm": 43.25,
+      "grad_norm_var": 13.948372395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.2262,
+      "loss/crossentropy": 2.3249034196138383,
+      "loss/hidden": 3.56875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.234759721159935,
+      "step": 4980
+    },
+    {
+      "epoch": 0.499,
+      "grad_norm": 31.5,
+      "grad_norm_var": 11.167708333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.1488,
+      "loss/crossentropy": 2.2298896074295045,
+      "loss/hidden": 3.6546875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2505023546516895,
+      "step": 4990
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 38.75,
+      "grad_norm_var": 10.43125,
+      "learning_rate": 0.0001,
+      "loss": 8.1769,
+      "loss/crossentropy": 2.201507803052664,
+      "loss/hidden": 3.520703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21926793903112413,
+      "step": 5000
+    },
+    {
+      "epoch": 0.501,
+      "grad_norm": 44.75,
+      "grad_norm_var": 16.2853515625,
+      "learning_rate": 0.0001,
+      "loss": 8.039,
+      "loss/crossentropy": 2.270238833874464,
+      "loss/hidden": 3.613671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23682384472340345,
+      "step": 5010
+    },
+    {
+      "epoch": 0.502,
+      "grad_norm": 31.75,
+      "grad_norm_var": 14.38515625,
+      "learning_rate": 0.0001,
+      "loss": 8.1599,
+      "loss/crossentropy": 2.108400362730026,
+      "loss/hidden": 3.718359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2417622933164239,
+      "step": 5020
+    },
+    {
+      "epoch": 0.503,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.8989583333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1698,
+      "loss/crossentropy": 2.3586475804448126,
+      "loss/hidden": 3.571875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23084877729415892,
+      "step": 5030
+    },
+    {
+      "epoch": 0.504,
+      "grad_norm": 32.5,
+      "grad_norm_var": 1.962513882217063e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.293,
+      "loss/crossentropy": 2.308009374141693,
+      "loss/hidden": 3.631640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25030199717730284,
+      "step": 5040
+    },
+    {
+      "epoch": 0.505,
+      "grad_norm": 34.25,
+      "grad_norm_var": 1.962513880139065e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.2269,
+      "loss/crossentropy": 2.2477807879447935,
+      "loss/hidden": 3.63046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22722288742661476,
+      "step": 5050
+    },
+    {
+      "epoch": 0.506,
+      "grad_norm": 31.0,
+      "grad_norm_var": 230.7962890625,
+      "learning_rate": 0.0001,
+      "loss": 8.0995,
+      "loss/crossentropy": 2.113031893968582,
+      "loss/hidden": 3.625390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2260434988886118,
+      "step": 5060
+    },
+    {
+      "epoch": 0.507,
+      "grad_norm": 29.5,
+      "grad_norm_var": 226.95462239583333,
+      "learning_rate": 0.0001,
+      "loss": 8.1254,
+      "loss/crossentropy": 2.1494341671466826,
+      "loss/hidden": 3.628125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2316014662384987,
+      "step": 5070
+    },
+    {
+      "epoch": 0.508,
+      "grad_norm": 42.25,
+      "grad_norm_var": 17.206184895833335,
+      "learning_rate": 0.0001,
+      "loss": 8.3222,
+      "loss/crossentropy": 2.409487584233284,
+      "loss/hidden": 3.51953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24138148501515388,
+      "step": 5080
+    },
+    {
+      "epoch": 0.509,
+      "grad_norm": 33.0,
+      "grad_norm_var": 15.231184895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.0576,
+      "loss/crossentropy": 2.285913223773241,
+      "loss/hidden": 3.528515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2232502717524767,
+      "step": 5090
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 33.5,
+      "grad_norm_var": 15.065559895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.1747,
+      "loss/crossentropy": 2.180917738378048,
+      "loss/hidden": 3.56328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23142165634781123,
+      "step": 5100
+    },
+    {
+      "epoch": 0.511,
+      "grad_norm": 32.25,
+      "grad_norm_var": 13.1166015625,
+      "learning_rate": 0.0001,
+      "loss": 8.1417,
+      "loss/crossentropy": 2.226011593639851,
+      "loss/hidden": 3.506640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22545368764549495,
+      "step": 5110
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 33.25,
+      "grad_norm_var": 4.7572265625,
+      "learning_rate": 0.0001,
+      "loss": 8.1568,
+      "loss/crossentropy": 2.2642487674951552,
+      "loss/hidden": 3.626953125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24627051521092652,
+      "step": 5120
+    },
+    {
+      "epoch": 0.513,
+      "grad_norm": 29.375,
+      "grad_norm_var": 8.576822916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.1412,
+      "loss/crossentropy": 2.1210575878620146,
+      "loss/hidden": 3.603125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22232303582131863,
+      "step": 5130
+    },
+    {
+      "epoch": 0.514,
+      "grad_norm": 36.25,
+      "grad_norm_var": 7.51015625,
+      "learning_rate": 0.0001,
+      "loss": 8.1105,
+      "loss/crossentropy": 2.338705539703369,
+      "loss/hidden": 3.61796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2322886861860752,
+      "step": 5140
+    },
+    {
+      "epoch": 0.515,
+      "grad_norm": 32.25,
+      "grad_norm_var": 9.846809895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.1908,
+      "loss/crossentropy": 2.2111464768648146,
+      "loss/hidden": 3.72109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.26425624899566175,
+      "step": 5150
+    },
+    {
+      "epoch": 0.516,
+      "grad_norm": 28.875,
+      "grad_norm_var": 9.946809895833333,
+      "learning_rate": 0.0001,
+      "loss": 8.0375,
+      "loss/crossentropy": 2.176833947002888,
+      "loss/hidden": 3.54765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22021548971533775,
+      "step": 5160
+    },
+    {
+      "epoch": 0.517,
+      "grad_norm": 34.0,
+      "grad_norm_var": 4.706705729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.1521,
+      "loss/crossentropy": 2.1654283188283445,
+      "loss/hidden": 3.48203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2115080550312996,
+      "step": 5170
+    },
+    {
+      "epoch": 0.518,
+      "grad_norm": 30.25,
+      "grad_norm_var": 5.541666666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.13,
+      "loss/crossentropy": 2.2458599150180816,
+      "loss/hidden": 3.596484375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22666897978633643,
+      "step": 5180
+    },
+    {
+      "epoch": 0.519,
+      "grad_norm": 44.75,
+      "grad_norm_var": 14.142122395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.145,
+      "loss/crossentropy": 2.317767137289047,
+      "loss/hidden": 3.5671875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23820882234722376,
+      "step": 5190
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 32.5,
+      "grad_norm_var": 23.218684895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.1323,
+      "loss/crossentropy": 2.279913380742073,
+      "loss/hidden": 3.558984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24836393278092145,
+      "step": 5200
+    },
+    {
+      "epoch": 0.521,
+      "grad_norm": 31.875,
+      "grad_norm_var": 16.846875,
+      "learning_rate": 0.0001,
+      "loss": 8.2041,
+      "loss/crossentropy": 2.201883518695831,
+      "loss/hidden": 3.57890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22954922150820495,
+      "step": 5210
+    },
+    {
+      "epoch": 0.522,
+      "grad_norm": 34.25,
+      "grad_norm_var": 221.16223958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.273,
+      "loss/crossentropy": 2.2926857471466064,
+      "loss/hidden": 3.541796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22256891019642353,
+      "step": 5220
+    },
+    {
+      "epoch": 0.523,
+      "grad_norm": 75.0,
+      "grad_norm_var": 233.03098958333334,
+      "learning_rate": 0.0001,
+      "loss": 8.1559,
+      "loss/crossentropy": 2.1607275292277337,
+      "loss/hidden": 3.65859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24259125851094723,
+      "step": 5230
+    },
+    {
+      "epoch": 0.524,
+      "grad_norm": 33.5,
+      "grad_norm_var": 170.5353515625,
+      "learning_rate": 0.0001,
+      "loss": 8.1314,
+      "loss/crossentropy": 2.2264937654137613,
+      "loss/hidden": 3.653125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23244266752153636,
+      "step": 5240
+    },
+    {
+      "epoch": 0.525,
+      "grad_norm": 34.0,
+      "grad_norm_var": 69.98515625,
+      "learning_rate": 0.0001,
+      "loss": 8.0757,
+      "loss/crossentropy": 2.263314816355705,
+      "loss/hidden": 3.6921875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2497670866549015,
+      "step": 5250
+    },
+    {
+      "epoch": 0.526,
+      "grad_norm": 32.75,
+      "grad_norm_var": 15.91640625,
+      "learning_rate": 0.0001,
+      "loss": 8.1746,
+      "loss/crossentropy": 2.2392116367816923,
+      "loss/hidden": 3.659765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24461503997445105,
+      "step": 5260
+    },
+    {
+      "epoch": 0.527,
+      "grad_norm": 30.875,
+      "grad_norm_var": 118.5931640625,
+      "learning_rate": 0.0001,
+      "loss": 8.145,
+      "loss/crossentropy": 2.184428018331528,
+      "loss/hidden": 3.6265625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24789600986987353,
+      "step": 5270
+    },
+    {
+      "epoch": 0.528,
+      "grad_norm": 40.5,
+      "grad_norm_var": 127.34791666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.0442,
+      "loss/crossentropy": 2.1109130561351774,
+      "loss/hidden": 3.528125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21775138471275568,
+      "step": 5280
+    },
+    {
+      "epoch": 0.529,
+      "grad_norm": 34.5,
+      "grad_norm_var": 6.623958333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1308,
+      "loss/crossentropy": 2.386467677354813,
+      "loss/hidden": 3.607421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24021831918507813,
+      "step": 5290
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 34.25,
+      "grad_norm_var": 7.5666015625,
+      "learning_rate": 0.0001,
+      "loss": 8.1256,
+      "loss/crossentropy": 2.3109169751405716,
+      "loss/hidden": 3.630859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2501339312642813,
+      "step": 5300
+    },
+    {
+      "epoch": 0.531,
+      "grad_norm": 31.875,
+      "grad_norm_var": 12.195572916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.1229,
+      "loss/crossentropy": 2.346800622344017,
+      "loss/hidden": 3.608203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2536924373358488,
+      "step": 5310
+    },
+    {
+      "epoch": 0.532,
+      "grad_norm": 31.875,
+      "grad_norm_var": 9.997330729166666,
+      "learning_rate": 0.0001,
+      "loss": 8.0104,
+      "loss/crossentropy": 2.1395985931158066,
+      "loss/hidden": 3.471875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21648342311382293,
+      "step": 5320
+    },
+    {
+      "epoch": 0.533,
+      "grad_norm": 32.25,
+      "grad_norm_var": 54.80358072916667,
+      "learning_rate": 0.0001,
+      "loss": 8.2808,
+      "loss/crossentropy": 2.072698312997818,
+      "loss/hidden": 3.70625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21971321273595096,
+      "step": 5330
+    },
+    {
+      "epoch": 0.534,
+      "grad_norm": 37.75,
+      "grad_norm_var": 582.04140625,
+      "learning_rate": 0.0001,
+      "loss": 8.4396,
+      "loss/crossentropy": 2.168602865189314,
+      "loss/hidden": 3.68828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22807842567563058,
+      "step": 5340
+    },
+    {
+      "epoch": 0.535,
+      "grad_norm": 36.0,
+      "grad_norm_var": 17.0619140625,
+      "learning_rate": 0.0001,
+      "loss": 8.2354,
+      "loss/crossentropy": 2.212322035431862,
+      "loss/hidden": 3.579296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21836955063045024,
+      "step": 5350
+    },
+    {
+      "epoch": 0.536,
+      "grad_norm": 37.75,
+      "grad_norm_var": 18.780989583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.2718,
+      "loss/crossentropy": 2.1710034780204297,
+      "loss/hidden": 3.656640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2387496206909418,
+      "step": 5360
+    },
+    {
+      "epoch": 0.537,
+      "grad_norm": 29.125,
+      "grad_norm_var": 27.276041666666668,
+      "learning_rate": 0.0001,
+      "loss": 8.2691,
+      "loss/crossentropy": 2.097558119148016,
+      "loss/hidden": 3.680078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2329912935383618,
+      "step": 5370
+    },
+    {
+      "epoch": 0.538,
+      "grad_norm": 37.5,
+      "grad_norm_var": 18.0181640625,
+      "learning_rate": 0.0001,
+      "loss": 8.057,
+      "loss/crossentropy": 2.275981144607067,
+      "loss/hidden": 3.55703125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22314830794930457,
+      "step": 5380
+    },
+    {
+      "epoch": 0.539,
+      "grad_norm": 34.25,
+      "grad_norm_var": 19.930208333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.182,
+      "loss/crossentropy": 2.184271165728569,
+      "loss/hidden": 3.6609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23054859917610884,
+      "step": 5390
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 34.0,
+      "grad_norm_var": 17.333072916666666,
+      "learning_rate": 0.0001,
+      "loss": 8.3439,
+      "loss/crossentropy": 2.480144701898098,
+      "loss/hidden": 3.701171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2607876468449831,
+      "step": 5400
+    },
+    {
+      "epoch": 0.541,
+      "grad_norm": 33.75,
+      "grad_norm_var": 14.249739583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.2028,
+      "loss/crossentropy": 2.2421235501766206,
+      "loss/hidden": 3.637890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23026540800929068,
+      "step": 5410
+    },
+    {
+      "epoch": 0.542,
+      "grad_norm": 33.0,
+      "grad_norm_var": 14.408072916666667,
+      "learning_rate": 0.0001,
+      "loss": 8.122,
+      "loss/crossentropy": 2.173307144641876,
+      "loss/hidden": 3.64453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22846251353621483,
+      "step": 5420
+    },
+    {
+      "epoch": 0.543,
+      "grad_norm": 33.0,
+      "grad_norm_var": 5.3,
+      "learning_rate": 0.0001,
+      "loss": 8.2194,
+      "loss/crossentropy": 2.3099235713481905,
+      "loss/hidden": 3.629296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23289041519165038,
+      "step": 5430
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 36.0,
+      "grad_norm_var": 28.040625,
+      "learning_rate": 0.0001,
+      "loss": 8.1837,
+      "loss/crossentropy": 2.2322505958378316,
+      "loss/hidden": 3.643359375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2281514243222773,
+      "step": 5440
+    },
+    {
+      "epoch": 0.545,
+      "grad_norm": 30.0,
+      "grad_norm_var": 69.69166666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.1516,
+      "loss/crossentropy": 2.199453258514404,
+      "loss/hidden": 3.5890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22161313518881798,
+      "step": 5450
+    },
+    {
+      "epoch": 0.546,
+      "grad_norm": 36.25,
+      "grad_norm_var": 88.72890625,
+      "learning_rate": 0.0001,
+      "loss": 8.305,
+      "loss/crossentropy": 2.3247879207134248,
+      "loss/hidden": 3.74609375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2331052988767624,
+      "step": 5460
+    },
+    {
+      "epoch": 0.547,
+      "grad_norm": 50.25,
+      "grad_norm_var": 49.67265625,
+      "learning_rate": 0.0001,
+      "loss": 8.1121,
+      "loss/crossentropy": 2.2907343961298468,
+      "loss/hidden": 3.598828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22796698454767467,
+      "step": 5470
+    },
+    {
+      "epoch": 0.548,
+      "grad_norm": 34.5,
+      "grad_norm_var": 27.73125,
+      "learning_rate": 0.0001,
+      "loss": 8.1348,
+      "loss/crossentropy": 2.4852760285139084,
+      "loss/hidden": 3.5296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23372339643537998,
+      "step": 5480
+    },
+    {
+      "epoch": 0.549,
+      "grad_norm": 33.0,
+      "grad_norm_var": 2.703285648974309e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.0635,
+      "loss/crossentropy": 2.3434904247522352,
+      "loss/hidden": 3.725390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21989205628633499,
+      "step": 5490
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 30.5,
+      "grad_norm_var": 6.481184895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.1213,
+      "loss/crossentropy": 2.048020973801613,
+      "loss/hidden": 3.53046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21088404105976225,
+      "step": 5500
+    },
+    {
+      "epoch": 0.551,
+      "grad_norm": 41.0,
+      "grad_norm_var": 16.025,
+      "learning_rate": 0.0001,
+      "loss": 8.143,
+      "loss/crossentropy": 2.246925861388445,
+      "loss/hidden": 3.616796875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2380803508684039,
+      "step": 5510
+    },
+    {
+      "epoch": 0.552,
+      "grad_norm": 34.0,
+      "grad_norm_var": 19.940625,
+      "learning_rate": 0.0001,
+      "loss": 8.1266,
+      "loss/crossentropy": 2.306339371204376,
+      "loss/hidden": 3.707421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2542409796267748,
+      "step": 5520
+    },
+    {
+      "epoch": 0.553,
+      "grad_norm": 30.25,
+      "grad_norm_var": 25.34375,
+      "learning_rate": 0.0001,
+      "loss": 8.1711,
+      "loss/crossentropy": 2.172018714249134,
+      "loss/hidden": 3.602734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24370079562067987,
+      "step": 5530
+    },
+    {
+      "epoch": 0.554,
+      "grad_norm": 42.25,
+      "grad_norm_var": 30.013541666666665,
+      "learning_rate": 0.0001,
+      "loss": 8.1492,
+      "loss/crossentropy": 2.22998360991478,
+      "loss/hidden": 3.625390625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23992773257195948,
+      "step": 5540
+    },
+    {
+      "epoch": 0.555,
+      "grad_norm": 32.25,
+      "grad_norm_var": 28.613541666666666,
+      "learning_rate": 0.0001,
+      "loss": 8.0695,
+      "loss/crossentropy": 2.2180883288383484,
+      "loss/hidden": 3.598828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22950777132064104,
+      "step": 5550
+    },
+    {
+      "epoch": 0.556,
+      "grad_norm": 33.0,
+      "grad_norm_var": 34.692708333333336,
+      "learning_rate": 0.0001,
+      "loss": 8.1162,
+      "loss/crossentropy": 2.174109402298927,
+      "loss/hidden": 3.702734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24184305276721715,
+      "step": 5560
+    },
+    {
+      "epoch": 0.557,
+      "grad_norm": 34.0,
+      "grad_norm_var": 7.314583333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.0114,
+      "loss/crossentropy": 2.1175956279039383,
+      "loss/hidden": 3.53828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21476623937487602,
+      "step": 5570
+    },
+    {
+      "epoch": 0.558,
+      "grad_norm": 35.25,
+      "grad_norm_var": 19.8119140625,
+      "learning_rate": 0.0001,
+      "loss": 8.1025,
+      "loss/crossentropy": 2.2917901635169984,
+      "loss/hidden": 3.558984375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22450571469962596,
+      "step": 5580
+    },
+    {
+      "epoch": 0.559,
+      "grad_norm": 38.5,
+      "grad_norm_var": 21.97890625,
+      "learning_rate": 0.0001,
+      "loss": 8.2385,
+      "loss/crossentropy": 2.3675080120563505,
+      "loss/hidden": 3.63828125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23681335002183915,
+      "step": 5590
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 33.0,
+      "grad_norm_var": 9.764583333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1467,
+      "loss/crossentropy": 2.3383423417806624,
+      "loss/hidden": 3.623046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23495447412133216,
+      "step": 5600
+    },
+    {
+      "epoch": 0.561,
+      "grad_norm": 34.25,
+      "grad_norm_var": 11.692643229166666,
+      "learning_rate": 0.0001,
+      "loss": 8.1075,
+      "loss/crossentropy": 2.2219532161951063,
+      "loss/hidden": 3.581640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21968780737370253,
+      "step": 5610
+    },
+    {
+      "epoch": 0.562,
+      "grad_norm": 29.875,
+      "grad_norm_var": 19.882747395833334,
+      "learning_rate": 0.0001,
+      "loss": 8.0262,
+      "loss/crossentropy": 2.121725457906723,
+      "loss/hidden": 3.564453125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22616409026086332,
+      "step": 5620
+    },
+    {
+      "epoch": 0.563,
+      "grad_norm": 31.625,
+      "grad_norm_var": 1.869627140463989e+18,
+      "learning_rate": 0.0001,
+      "loss": 8.2844,
+      "loss/crossentropy": 2.2004577577114106,
+      "loss/hidden": 3.651171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22547926437109708,
+      "step": 5630
+    },
+    {
+      "epoch": 0.564,
+      "grad_norm": 32.0,
+      "grad_norm_var": 4.611458333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1244,
+      "loss/crossentropy": 2.1516154944896697,
+      "loss/hidden": 3.55078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2172885647043586,
+      "step": 5640
+    },
+    {
+      "epoch": 0.565,
+      "grad_norm": 30.5,
+      "grad_norm_var": 7.317708333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.0514,
+      "loss/crossentropy": 2.2103342309594156,
+      "loss/hidden": 3.6140625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22790345773100854,
+      "step": 5650
+    },
+    {
+      "epoch": 0.566,
+      "grad_norm": 34.25,
+      "grad_norm_var": 13.6587890625,
+      "learning_rate": 0.0001,
+      "loss": 8.0864,
+      "loss/crossentropy": 2.2035325288772585,
+      "loss/hidden": 3.57734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23142583221197127,
+      "step": 5660
+    },
+    {
+      "epoch": 0.567,
+      "grad_norm": 31.625,
+      "grad_norm_var": 9.45625,
+      "learning_rate": 0.0001,
+      "loss": 8.0716,
+      "loss/crossentropy": 2.1144707940518854,
+      "loss/hidden": 3.555078125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21162977050989867,
+      "step": 5670
+    },
+    {
+      "epoch": 0.568,
+      "grad_norm": 32.5,
+      "grad_norm_var": 31.39140625,
+      "learning_rate": 0.0001,
+      "loss": 8.096,
+      "loss/crossentropy": 2.1229008197784425,
+      "loss/hidden": 3.554296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2217079123482108,
+      "step": 5680
+    },
+    {
+      "epoch": 0.569,
+      "grad_norm": 35.5,
+      "grad_norm_var": 30.71875,
+      "learning_rate": 0.0001,
+      "loss": 8.0824,
+      "loss/crossentropy": 2.1256051540374754,
+      "loss/hidden": 3.53046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22356193587183953,
+      "step": 5690
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 30.125,
+      "grad_norm_var": 6.362239583333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1334,
+      "loss/crossentropy": 2.1438381403684614,
+      "loss/hidden": 3.51875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2077854923903942,
+      "step": 5700
+    },
+    {
+      "epoch": 0.571,
+      "grad_norm": 32.5,
+      "grad_norm_var": 14.85390625,
+      "learning_rate": 0.0001,
+      "loss": 8.0718,
+      "loss/crossentropy": 2.2198398813605307,
+      "loss/hidden": 3.580859375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22509159836918116,
+      "step": 5710
+    },
+    {
+      "epoch": 0.572,
+      "grad_norm": 34.75,
+      "grad_norm_var": 13.4759765625,
+      "learning_rate": 0.0001,
+      "loss": 8.0583,
+      "loss/crossentropy": 2.4060455739498137,
+      "loss/hidden": 3.708203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.25263102930039166,
+      "step": 5720
+    },
+    {
+      "epoch": 0.573,
+      "grad_norm": 28.625,
+      "grad_norm_var": 6.946809895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.0726,
+      "loss/crossentropy": 2.1238791063427924,
+      "loss/hidden": 3.578125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21989983841776847,
+      "step": 5730
+    },
+    {
+      "epoch": 0.574,
+      "grad_norm": 33.75,
+      "grad_norm_var": 9.4884765625,
+      "learning_rate": 0.0001,
+      "loss": 8.1374,
+      "loss/crossentropy": 2.1622555539011956,
+      "loss/hidden": 3.583203125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21868109367787839,
+      "step": 5740
+    },
+    {
+      "epoch": 0.575,
+      "grad_norm": 33.0,
+      "grad_norm_var": 16.939518229166666,
+      "learning_rate": 0.0001,
+      "loss": 7.9816,
+      "loss/crossentropy": 2.203578273952007,
+      "loss/hidden": 3.57109375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22708230018615722,
+      "step": 5750
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 33.5,
+      "grad_norm_var": 7.49765625,
+      "learning_rate": 0.0001,
+      "loss": 8.0534,
+      "loss/crossentropy": 2.0832700729370117,
+      "loss/hidden": 3.5890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2166902620345354,
+      "step": 5760
+    },
+    {
+      "epoch": 0.577,
+      "grad_norm": 38.5,
+      "grad_norm_var": 7.069205729166667,
+      "learning_rate": 0.0001,
+      "loss": 8.1567,
+      "loss/crossentropy": 2.1000006228685377,
+      "loss/hidden": 3.6328125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2189541209489107,
+      "step": 5770
+    },
+    {
+      "epoch": 0.578,
+      "grad_norm": 34.0,
+      "grad_norm_var": 10.34140625,
+      "learning_rate": 0.0001,
+      "loss": 7.9902,
+      "loss/crossentropy": 2.2755674168467523,
+      "loss/hidden": 3.5046875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2158666817471385,
+      "step": 5780
+    },
+    {
+      "epoch": 0.579,
+      "grad_norm": 31.0,
+      "grad_norm_var": 7.744791666666667,
+      "learning_rate": 0.0001,
+      "loss": 8.049,
+      "loss/crossentropy": 2.3444782361388206,
+      "loss/hidden": 3.565625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22829483300447465,
+      "step": 5790
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 31.75,
+      "grad_norm_var": 6.076497395833333,
+      "learning_rate": 0.0001,
+      "loss": 8.0271,
+      "loss/crossentropy": 2.202429711073637,
+      "loss/hidden": 3.4640625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21492673214524985,
+      "step": 5800
+    },
+    {
+      "epoch": 0.581,
+      "grad_norm": 31.0,
+      "grad_norm_var": 3.6811848958333333,
+      "learning_rate": 0.0001,
+      "loss": 7.8948,
+      "loss/crossentropy": 2.3126792818307877,
+      "loss/hidden": 3.46015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22085475884377956,
+      "step": 5810
+    },
+    {
+      "epoch": 0.582,
+      "grad_norm": 33.0,
+      "grad_norm_var": 3.542708333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1146,
+      "loss/crossentropy": 2.264694780111313,
+      "loss/hidden": 3.47421875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21074176728725433,
+      "step": 5820
+    },
+    {
+      "epoch": 0.583,
+      "grad_norm": 31.5,
+      "grad_norm_var": 2.3452473958333333,
+      "learning_rate": 0.0001,
+      "loss": 8.079,
+      "loss/crossentropy": 2.2641124561429025,
+      "loss/hidden": 3.5125,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22780102472752334,
+      "step": 5830
+    },
+    {
+      "epoch": 0.584,
+      "grad_norm": 32.5,
+      "grad_norm_var": 14.5947265625,
+      "learning_rate": 0.0001,
+      "loss": 8.0337,
+      "loss/crossentropy": 2.218563383817673,
+      "loss/hidden": 3.62734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24179403018206358,
+      "step": 5840
+    },
+    {
+      "epoch": 0.585,
+      "grad_norm": 32.75,
+      "grad_norm_var": 3.1705729166666665,
+      "learning_rate": 0.0001,
+      "loss": 8.0352,
+      "loss/crossentropy": 2.2127060025930403,
+      "loss/hidden": 3.51015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22846792116761208,
+      "step": 5850
+    },
+    {
+      "epoch": 0.586,
+      "grad_norm": 31.0,
+      "grad_norm_var": 5.418489583333334,
+      "learning_rate": 0.0001,
+      "loss": 8.0163,
+      "loss/crossentropy": 2.2808433353900908,
+      "loss/hidden": 3.579296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23100998885929586,
+      "step": 5860
+    },
+    {
+      "epoch": 0.587,
+      "grad_norm": 33.75,
+      "grad_norm_var": 7.083333333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.0979,
+      "loss/crossentropy": 2.335478585958481,
+      "loss/hidden": 3.541015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2336500260978937,
+      "step": 5870
+    },
+    {
+      "epoch": 0.588,
+      "grad_norm": 31.5,
+      "grad_norm_var": 46.71243489583333,
+      "learning_rate": 0.0001,
+      "loss": 8.065,
+      "loss/crossentropy": 2.1413629055023193,
+      "loss/hidden": 3.46171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21667953655123712,
+      "step": 5880
+    },
+    {
+      "epoch": 0.589,
+      "grad_norm": 33.25,
+      "grad_norm_var": 90.58515625,
+      "learning_rate": 0.0001,
+      "loss": 8.1265,
+      "loss/crossentropy": 2.3575997933745385,
+      "loss/hidden": 3.5171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23772844970226287,
+      "step": 5890
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 42.25,
+      "grad_norm_var": 108.78020833333333,
+      "learning_rate": 0.0001,
+      "loss": 8.1898,
+      "loss/crossentropy": 2.31534286737442,
+      "loss/hidden": 3.5765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23516609705984592,
+      "step": 5900
+    },
+    {
+      "epoch": 0.591,
+      "grad_norm": 30.125,
+      "grad_norm_var": 96.6228515625,
+      "learning_rate": 0.0001,
+      "loss": 8.2098,
+      "loss/crossentropy": 2.3408817887306212,
+      "loss/hidden": 3.58515625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.234616519510746,
+      "step": 5910
+    },
+    {
+      "epoch": 0.592,
+      "grad_norm": 34.5,
+      "grad_norm_var": 5.761458333333334,
+      "learning_rate": 0.0001,
+      "loss": 8.0412,
+      "loss/crossentropy": 2.3171244740486143,
+      "loss/hidden": 3.690234375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.24309027940034866,
+      "step": 5920
+    },
+    {
+      "epoch": 0.593,
+      "grad_norm": 31.875,
+      "grad_norm_var": 18.249934895833334,
+      "learning_rate": 0.0001,
+      "loss": 8.2448,
+      "loss/crossentropy": 2.3113586097955703,
+      "loss/hidden": 3.537890625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2302385514602065,
+      "step": 5930
+    },
+    {
+      "epoch": 0.594,
+      "grad_norm": 29.625,
+      "grad_norm_var": 16.43125,
+      "learning_rate": 0.0001,
+      "loss": 7.9336,
+      "loss/crossentropy": 2.325835222005844,
+      "loss/hidden": 3.434765625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.21180371306836604,
+      "step": 5940
+    },
+    {
+      "epoch": 0.595,
+      "grad_norm": 30.5,
+      "grad_norm_var": 10.299739583333333,
+      "learning_rate": 0.0001,
+      "loss": 7.9796,
+      "loss/crossentropy": 2.0509180039167405,
+      "loss/hidden": 3.604296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2253203097730875,
+      "step": 5950
+    },
+    {
+      "epoch": 0.596,
+      "grad_norm": 35.5,
+      "grad_norm_var": 2.948958333333333,
+      "learning_rate": 0.0001,
+      "loss": 8.0105,
+      "loss/crossentropy": 2.320768731832504,
+      "loss/hidden": 3.509375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.2229237537831068,
+      "step": 5960
+    },
+    {
+      "epoch": 0.597,
+      "grad_norm": 28.375,
+      "grad_norm_var": 5.462434895833334,
+      "learning_rate": 0.0001,
+      "loss": 7.9407,
+      "loss/crossentropy": 2.3377193987369536,
+      "loss/hidden": 3.551171875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22489294074475766,
+      "step": 5970
+    },
+    {
+      "epoch": 0.598,
+      "grad_norm": 37.25,
+      "grad_norm_var": 10.767643229166667,
+      "learning_rate": 0.0001,
+      "loss": 7.9551,
+      "loss/crossentropy": 2.26409173309803,
+      "loss/hidden": 3.5734375,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.23901313543319702,
+      "step": 5980
+    },
+    {
+      "epoch": 0.599,
+      "grad_norm": 35.0,
+      "grad_norm_var": 14.5375,
+      "learning_rate": 0.0001,
+      "loss": 7.8592,
+      "loss/crossentropy": 2.2190980166196823,
+      "loss/hidden": 3.554296875,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22315293960273266,
+      "step": 5990
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 31.0,
+      "grad_norm_var": 11.592643229166667,
+      "learning_rate": 0.0001,
+      "loss": 8.0295,
+      "loss/crossentropy": 2.3040059447288512,
+      "loss/hidden": 3.491015625,
+      "loss/jsd": 0.0,
+      "loss/logits": 0.22013801857829093,
+      "step": 6000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 2000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7145060192052838e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}